Authors: Yiheng Tao, Yihe Zhang, Matthew T. Dearing, Xin Wang, Yuping Fan, Zhiling Lan
Abstract: Efficient scheduling of LLM inference tasks is essential for achieving low latency and high throughput, particularly with the growing use of reasoning-capable LLMs. Traditional strategies like First-Come-First-Serve (FCFS) often suffer from Head-of-Line (HOL) blocking, where long-running tasks delay shorter ones queued behind them. In this paper, we introduce PARS, a prompt-aware LLM task scheduler that improves serving efficiency by approximating shortest-job-first (SJF) scheduling through pairwise ranking with margin ranking loss. PARS focuses on impactful scheduling decisions and is seamlessly integrated into the state-of-the-art LLM serving system vLLM. It effectively predicts response-length-based task ordering, reducing latency with minimal overhead. Extensive experiments across multiple LLMs and real-world inference datasets show that PARS significantly improves performance, including for reasoning workloads. Furthermore, our cross-model evaluations demonstrate that the design generalizes well, enabling effective scheduling even when predictors are trained on different LLMs.
Authors: Yanlong Wang, Hang Yu, Jian Xu, Fei Ma, Hongkang Zhang, Tongtong Feng, Zijian Zhang, Shao-Lun Huang, Danny Dongning Sun, Xiao-Ping Zhang
Abstract: Large time series foundation models often adopt channel-independent architectures to handle varying data dimensions, but this design ignores crucial cross-channel dependencies. Concurrently, existing multimodal approaches have not fully exploited the power of large vision models (LVMs) to interpret spatiotemporal data. Additionally, there remains significant unexplored potential in leveraging the advantages of information extraction from different modalities to enhance time series forecasting performance. To address these gaps, we propose the VIFO, a cross-modal forecasting model. VIFO uniquely renders multivariate time series into image, enabling pre-trained LVM to extract complex cross-channel patterns that are invisible to channel-independent models. These visual features are then aligned and fused with representations from the time series modality. By freezing the LVM and training only 7.45% of its parameters, VIFO achieves competitive performance on multiple benchmarks, offering an efficient and effective solution for capturing cross-variable relationships in
Authors: Ali Yavari, Alireza Mohamadi, Elham Beydaghi, Rainer A. Leitgeb
Abstract: Ensuring the reliability of deep neural networks (DNNs) in the presence of real world noise and intentional perturbations remains a significant challenge. To address this, attribution methods have been proposed, though their efficacy remains suboptimal and necessitates further refinement. In this paper, we propose a novel category of transferable adversarial attacks, called transferable frequency-aware attacks, enabling frequency-aware exploration via both high-and low-frequency components. Based on this type of attacks, we also propose a novel attribution method, named Frequency-Aware Model Parameter Explorer (FAMPE), which improves the explainability for DNNs. Relative to the current state-of-the-art method AttEXplore, our FAMPE attains an average gain of 13.02% in Insertion Score, thereby outperforming existing approaches. Through detailed ablation studies, we also investigate the role of both high- and low-frequency components in explainability.
Authors: Xinyuan Song, Guangji Bai, Liang Zhao
Abstract: Pruning is critical for scaling large language models (LLMs). Global pruning achieves strong performance but requires $\mathcal{O}(N)$ memory, which is infeasible for billion-parameter models. Local pruning reduces GPU memory usage to that of a single layer by pruning layers independently, but it neglects inter-layer dependencies and often leads to suboptimal performance in high-sparsity regimes. Unlike unstructured pruning, structured pruning produces regular sparsity patterns that align well with GPU kernels and library optimizations, making it more hardware-efficient. However, structured pruning typically relies on global pruning, since structured patterns are more prone to severe performance degradation under local optimization. To jointly achieve structured pruning and the memory efficiency of local pruning, we propose a divide-and-conquer strategy that decomposes the global pruning problem into coordinated subproblems across different modules, each of which fits within limited GPU memory. Building on this idea, we design \textbf{STRUPRUNE}, an ADMM-based framework that integrates structured sparsity into the pruning process, combining the memory efficiency of local pruning with the hardware compatibility of structured methods. We derive a closed-form analytical solution for structured pruning masks that provides an explicit rule for layer-wise sparsity allocation, and further develop an energy-based asymptotic framework yielding a softmax-form allocation scheme that simplifies optimization while adapting to heterogeneous layer importance. Experiments demonstrate that STRUPRUNE matches the perplexity of global structured pruning while reducing memory cost from $\mathcal{O}(N)$ to $\mathcal{O}(\sqrt{N})$, enabling practical deployment at the billion-parameter scale.
Authors: Jiancheng Zhang, Yinglun Zhu
Abstract: Active learning (AL) is a principled strategy to reduce annotation cost in data-hungry deep learning. However, existing AL algorithms focus almost exclusively on unimodal data, overlooking the substantial annotation burden in multimodal learning. We introduce the first framework for multimodal active learning with unaligned data, where the learner must actively acquire cross-modal alignments rather than labels on pre-aligned pairs. This setting captures the practical bottleneck in modern multimodal pipelines such as CLIP and SigLIP, where unimodal features are easy to obtain but high-quality alignment is costly. We develop a new algorithm that combines uncertainty and diversity principles in a modality-aware design, achieves linear-time acquisition, and applies seamlessly to both pool-based and streaming-based settings. Extensive experiments on benchmark datasets demonstrate that our approach consistently reduces multimodal annotation cost while preserving performance; for instance, on the ColorSwap dataset it cuts annotation requirements by up to $40\%$ without loss in accuracy.
Authors: Anusha Agarwal, Dibakar Roy Sarkar, Somdatta Goswami
Abstract: Traumatic brain injury (TBI) remains a major public health concern, with over 69 million cases annually worldwide. Finite element (FE) models offer high-fidelity predictions of brain deformation but are computationally expensive, requiring hours per simulation and limiting their clinical utility for rapid decision-making. This study benchmarks state-of-the-art neural operator (NO) architectures for rapid, patient-specific prediction of brain displacement fields, aiming to enable real-time TBI modeling in clinical and translational settings. We formulated TBI modeling as an operator learning problem, mapping subject-specific anatomical MRI, magnetic resonance elastography (MRE) stiffness maps, and demographic features to full-field 3D brain displacement predictions. Four architectures - Fourier Neural Operator (FNO), Factorized FNO (F-FNO), Multi-Grid FNO (MG-FNO), and Deep Operator Network (DeepONet) were trained and evaluated on 249 MRE datasets across physiologically relevant frequencies (20 - 90 Hz). MG-FNO achieved the highest accuracy (MSE = 0.0023, 94.3\% spatial fidelity) and preserved fine-scale features, while F-FNO converged 2$\times$ faster than standard FNO. DeepONet offered the fastest inference (14.5 iterations/s) with a 7$\times$ computational speed-up over MG-FNO, suggesting utility for embedded or edge computing applications. All NOs reduced computation time from hours to milliseconds without sacrificing anatomical realism. NOs provide an efficient, resolution-invariant approach for predicting brain deformation, opening the door to real-time, patient-specific TBI risk assessment, clinical triage support, and optimization of protective equipment. These results highlight the potential for NO-based digital twins of the human brain, enabling scalable, on-demand biomechanical modeling in both clinical and population health contexts.
Authors: Lukas R\"uttgers, Till Aczel, Andreas Plesner, Roger Wattenhofer
Abstract: Differentiable logic gate networks (DLGNs) exhibit extraordinary efficiency at inference while sustaining competitive accuracy. But vanishing gradients, discretization errors, and high training cost impede scaling these networks. Even with dedicated parameter initialization schemes from subsequent works, increasing depth still harms accuracy. We show that the root cause of these issues lies in the underlying parametrization of logic gate neurons themselves. To overcome this issue, we propose a reparametrization that also shrinks the parameter size logarithmically in the number of inputs per gate. For binary inputs, this already reduces the model size by 4x, speeds up the backward pass by up to 1.86x, and converges in 8.5x fewer training steps. On top of that, we show that the accuracy on CIFAR-100 remains stable and sometimes superior to the original parametrization.
Authors: Hanzhong Cao, Wenbo Yan, Ying Tan
Abstract: Many methods aim to enhance time series forecasting by decomposing the series through intricate model structures and prior knowledge, yet they are inevitably limited by computational complexity and the robustness of the assumptions. Our research uncovers that in the complex domain and higher-order hypercomplex spaces, the characteristic frequencies of time series naturally decrease. Leveraging this insight, we propose Numerion, a time series forecasting model based on multiple hypercomplex spaces. Specifically, grounded in theoretical support, we generalize linear layers and activation functions to hypercomplex spaces of arbitrary power-of-two dimensions and introduce a novel Real-Hypercomplex-Real Domain Multi-Layer Perceptron (RHR-MLP) architecture. Numerion utilizes multiple RHR-MLPs to map time series into hypercomplex spaces of varying dimensions, naturally decomposing and independently modeling the series, and adaptively fuses the latent patterns exhibited in different spaces through a dynamic fusion mechanism. Experiments validate the model`s performance, achieving state-of-the-art results on multiple public datasets. Visualizations and quantitative analyses comprehensively demonstrate the ability of multi-dimensional RHR-MLPs to naturally decompose time series and reveal the tendency of higher dimensional hypercomplex spaces to capture lower frequency features.
Authors: Duc Kieu, Kien Do, Tuan Hoang, Thao Minh Le, Tung Kieu, Dang Nguyen, Thin Nguyen
Abstract: Multi-domain translation (MDT) aims to learn translations between multiple domains, yet existing approaches either require fully aligned tuples or can only handle domain pairs seen in training, limiting their practicality and excluding many cross-domain mappings. We introduce universal MDT (UMDT), a generalization of MDT that seeks to translate between any pair of $K$ domains using only $K-1$ paired datasets with a central domain. To tackle this problem, we propose Diffusion Router (DR), a unified diffusion-based framework that models all central$\leftrightarrow$non-central translations with a single noise predictor conditioned on the source and target domain labels. DR enables indirect non-central translations by routing through the central domain. We further introduce a novel scalable learning strategy with a variational-bound objective and an efficient Tweedie refinement procedure to support direct non-central mappings. Through evaluation on three large-scale UMDT benchmarks, DR achieves state-of-the-art results for both indirect and direct translations, while lowering sampling cost and unlocking novel tasks such as sketch$\leftrightarrow$segmentation. These results establish DR as a scalable and versatile framework for universal translation across multiple domains.
Authors: Heyang Gao, Zexu Sun, Erxue Min, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, Xu Chen
Abstract: Large Language Models (LLMs) as autonomous agents are increasingly tasked with solving complex, long-horizon problems. Aligning these agents via preference-based offline methods like Direct Preference Optimization (DPO) is a promising direction, yet it faces a critical granularity mismatch. Trajectory-level DPO provides a signal that is too coarse for precise credit assignment, while step-level DPO is often too myopic to capture the value of multi-step behaviors. To resolve this challenge, we introduce Hierarchical Preference Learning (HPL), a hierarchical framework that optimizes LLM agents by leveraging preference signals at multiple, synergistic granularities. While HPL incorporates trajectory- and step-level DPO for global and local policy stability, its core innovation lies in group-level preference optimization guided by a dual-layer curriculum. Our approach first decomposes expert trajectories into semantically coherent action groups and then generates contrasting suboptimal groups to enable preference learning at a fine-grained, sub-task level. Then, instead of treating all preference pairs equally, HPL introduces a curriculum scheduler that organizes the learning process from simple to complex. This curriculum is structured along two axes: the group length, representing sub-task complexity, and the sample difficulty, defined by the reward gap between preferred and dispreferred action groups. Experiments on three challenging agent benchmarks show that HPL outperforms existing state-of-the-art methods. Our analyses demonstrate that the hierarchical DPO loss effectively integrates preference signals across multiple granularities, while the dual-layer curriculum is crucial for enabling the agent to solve a wide range of tasks, from simple behaviors to complex multi-step sequences.
Authors: David Benfield, Stefano Coniglio, Phan Tu Vuong, Alain Zemkoho
Abstract: Adversarial machine learning concerns situations in which learners face attacks from active adversaries. Such scenarios arise in applications such as spam email filtering, malware detection and fake image generation, where security methods must be actively updated to keep up with the everimproving generation of malicious data. Pessimistic Bilevel optimisation has been shown to be an effective method of training resilient classifiers against such adversaries. By modelling these scenarios as a game between the learner and the adversary, we anticipate how the adversary will modify their data and then train a resilient classifier accordingly. However, since existing pessimistic bilevel approaches feature an unrestricted adversary, the model is vulnerable to becoming overly pessimistic and unrealistic. When finding the optimal solution that defeats the classifier, it is possible that the adversary's data becomes nonsensical and loses its intended nature. Such an adversary will not properly reflect reality, and consequently, will lead to poor classifier performance when implemented on real-world data. By constructing a constrained pessimistic bilevel optimisation model, we restrict the adversary's movements and identify a solution that better reflects reality. We demonstrate through experiments that this model performs, on average, better than the existing approach.
Authors: Wen Wu, Ziyang Zhang, Liwei Liu, Xuenan Xu, Junlin Liu, Ke Fan, Qitan Lv, Jimin Zhuang, Chen Zhang, Zheqi Yuan, Siyuan Hou, Tianyi Lin, Kai Chen, Bowen Zhou, Chao Zhang
Abstract: The scientific reasoning ability of large language models (LLMs) has recently attracted significant attention. Time series, as a fundamental modality in scientific data, presents unique challenges that are often overlooked in current multimodal LLMs, which either encode numerical sequences as text or convert them into images. Such approaches may be insufficient for comprehensive scientific time series understanding and generation. Existing unified time series models typically specialise in either forecasting or analysis, and their effectiveness on non-periodic, heterogeneous scientific signals remains unclear. To address these gaps, we introduce SciTS, a benchmark spanning 12 scientific domains and 43 tasks, with over 50k+ instances, both univariate and multivariate signals ranging from $10^0$ to $10^7$ in length and up to 10~MHz in frequency. We benchmark 17 models, including text-only LLMs, multimodal LLMs, and unified time series models, and find that general-purpose LLMs exhibit stronger generalisability than specialised time series models, while representing time series as text or images limits their performance due to excessively long sequences and loss of numerical precision, respectively. We then introduce TimeOmni, a framework that equips LLMs with the ability to understand and generate time series while remaining compatible with general-purpose LLM training. This work fills a gap in both dedicated benchmarks and modelling frameworks for scientific time series, paving the way for LLMs to understand and generate complex temporal scientific data.
Authors: Zijian Zhao, Sen Li
Abstract: On-demand ride-sharing platforms, such as Uber and Lyft, face the intricate real-time challenge of bundling and matching passengers-each with distinct origins and destinations-to available vehicles, all while navigating significant system uncertainties. Due to the extensive observation space arising from the large number of drivers and orders, order dispatching, though fundamentally a centralized task, is often addressed using Multi-Agent Reinforcement Learning (MARL). However, independent MARL methods fail to capture global information and exhibit poor cooperation among workers, while Centralized Training Decentralized Execution (CTDE) MARL methods suffer from the curse of dimensionality. To overcome these challenges, we propose Triple-BERT, a centralized Single Agent Reinforcement Learning (MARL) method designed specifically for large-scale order dispatching on ride-sharing platforms. Built on a variant TD3, our approach addresses the vast action space through an action decomposition strategy that breaks down the joint action probability into individual driver action probabilities. To handle the extensive observation space, we introduce a novel BERT-based network, where parameter reuse mitigates parameter growth as the number of drivers and orders increases, and the attention mechanism effectively captures the complex relationships among the large pool of driver and orders. We validate our method using a real-world ride-hailing dataset from Manhattan. Triple-BERT achieves approximately an 11.95% improvement over current state-of-the-art methods, with a 4.26% increase in served orders and a 22.25% reduction in pickup times. Our code, trained model parameters, and processed data are publicly available at the repository https://github.com/RS2002/Triple-BERT .
Authors: Chang'an Yi, Xiaohui Deng, Shuaicheng Niu, Yan Zhou
Abstract: Test-time adaptation (TTA) aims to transfer knowledge from a source model to unknown test data with potential distribution shifts in an online manner. Many existing TTA methods rely on entropy as a confidence metric to optimize the model. However, these approaches are sensitive to the predefined entropy threshold, influencing which samples are chosen for model adaptation. Consequently, potentially reliable target samples are often overlooked and underutilized. For instance, a sample's entropy might slightly exceed the threshold initially, but fall below it after the model is updated. Such samples can provide stable supervised information and offer a normal range of gradients to guide model adaptation. In this paper, we propose a general approach, \underline{POEM}, to promote TTA via ex\underline{\textbf{p}}loring the previously unexpl\underline{\textbf{o}}red reliabl\underline{\textbf{e}} sa\underline{\textbf{m}}ples. Additionally, we introduce an extra Adapt Branch network to strike a balance between extracting domain-agnostic representations and achieving high performance on target data. Comprehensive experiments across multiple architectures demonstrate that POEM consistently outperforms existing TTA methods in both challenging scenarios and real-world domain shifts, while remaining computationally efficient. The effectiveness of POEM is evaluated through extensive analyses and thorough ablation studies. Moreover, the core idea behind POEM can be employed as an augmentation strategy to boost the performance of existing TTA approaches. The source code is publicly available at \emph{https://github.com/ycarobot/POEM}
Authors: Yoonjeon Kim, Doohyuk Jang, Eunho Yang
Abstract: Recent studies on reasoning models explore the meta-awareness of language models, the ability to know how to think by itself. We argue that large reasoning models lack this meta-awareness property by proving severe misalignment between true rollouts and predicted meta information. We posit that aligning meta-prediction with true rollouts will lead to significant performance gains. To verify this hypothesis, we design a training pipeline that boosts Meta-Awareness via Self-Alignment (MASA), and prove that enhanced meta-awareness directly translates to improved accuracy. Unlike existing meta-cognitive reasoning models, our method does not require external training sources but leverages self-generated signals to train meta-awareness. Moreover, our method enables efficient training by i) filtering out zero-variance prompts that are either trivial or unsolvable and ii) cutting off lengthy rollouts when they are unlikely to lead to correct answers. The results are inspiring: our strategy yields significant improvements in both accuracy and training efficiency on in-domain tasks and shows strong generalization to out-of-domain benchmarks. More specifically, our method can speed up GRPO training by over 1.28x to reach the same performance, and achieve a 19.3% gain in accuracy on AIME25, and a 6.2 % average gain over six mathematics benchmarks. Training with meta-cognitive guidance enhances out-of-domain generalization, giving a 3.87 % boost on GPQA-Diamond and a 2.08 % overall accuracy gain across 13 benchmarks spanning logical, scientific, and coding domains.
Authors: Juan Jose Herrera-Aranda, Guillermo Gomez-Trenado, Francisco Herrera, Isaac Triguero
Abstract: Zero-Shot Learning is an important paradigm within General-Purpose Artificial Intelligence Systems, particularly in those that operate in open-world scenarios where systems must adapt to new tasks dynamically. Semantic spaces play a pivotal role as they bridge seen and unseen classes, but whether human-annotated or generated by a machine learning model, they often contain noisy, redundant, or irrelevant attributes that hinder performance. To address this, we introduce a partitioning scheme that simulates unseen conditions in an inductive setting (which is the most challenging), allowing attribute relevance to be assessed without access to semantic information from unseen classes. Within this framework, we study two complementary feature-selection strategies and assess their generalisation. The first adapts embedded feature selection to the particular demands of ZSL, turning model-driven rankings into meaningful semantic pruning; the second leverages evolutionary computation to directly explore the space of attribute subsets more broadly. Experiments on five benchmark datasets (AWA2, CUB, SUN, aPY, FLO) show that both methods consistently improve accuracy on unseen classes by reducing redundancy, but in complementary ways: RFS is efficient and competitive though dependent on critical hyperparameters, whereas GA is more costly yet explores the search space more broadly and avoids such dependence. These results confirm that semantic spaces are inherently redundant and highlight the proposed partitioning scheme as an effective tool to refine them under inductive conditions.
Authors: C. Coelho, M. Hohmann, D. Fern\'andez, L. Penter, S. Ihlenfeldt, O. Niggemann
Abstract: Thermal errors in machine tools significantly impact machining precision and productivity. Traditional thermal error correction/compensation methods rely on measured temperature-deformation fields or on transfer functions. Most existing data-driven compensation strategies employ neural networks (NNs) to directly predict thermal errors or specific compensation values. While effective, these approaches are tightly bound to particular error types, spatial locations, or machine configurations, limiting their generality and adaptability. In this work, we introduce a novel paradigm in which NNs are trained to predict high-fidelity temperature and heat flux fields within the machine tool. The proposed framework enables subsequent computation and correction of a wide range of error types using modular, swappable downstream components. The NN is trained using data obtained with the finite element method under varying initial conditions and incorporates a correlation-based selection strategy that identifies the most informative measurement points, minimising hardware requirements during inference. We further benchmark state-of-the-art time-series NN architectures, namely Recurrent NN, Gated Recurrent Unit, Long-Short Term Memory (LSTM), Bidirectional LSTM, Transformer, and Temporal Convolutional Network, by training both specialised models, tailored for specific initial conditions, and general models, capable of extrapolating to unseen scenarios. The results show accurate and low-cost prediction of temperature and heat flux fields, laying the basis for enabling flexible and generalisable thermal error correction in machine tool environments.
Authors: Andi Zhang, Xuan Ding, Haofan Wang, Steven McDonagh, Samuel Kaski
Abstract: We propose Orthogonal Monte Carlo Dropout, a mechanism that enforces strict orthogonality when combining sparse semantic vectors without extra time complexity. LoRA, a popular fine-tuning method for large models, typically trains a module to represent a specific concept such as an object or a style. When multiple LoRAs are merged, for example to generate an object in a particular style, their semantic vectors may interfere with each other. Our method guarantees, at the theoretical and runtime levels, that merged LoRAs remain orthogonal and thus free from direct interference. However, empirical analysis reveals that such orthogonality does not lead to the semantic disentanglement or compositionality highlighted in prior work on compositional adaptation. This finding suggests that inter-LoRA orthogonality alone may be insufficient for achieving true semantic compositionality, prompting a re-examination of its role in adapter merging.
Authors: Agnieszka Polowczyk, Alicja Polowczyk, Joanna Waczy\'nska, Piotr Borycki, Przemys{\l}aw Spurek
Abstract: The impressive capability of modern text-to-image models to generate realistic visuals has come with a serious drawback: they can be misused to create harmful, deceptive or unlawful content. This has accelerated the push for machine unlearning. This new field seeks to selectively remove specific knowledge from a model's training data without causing a drop in its overall performance. However, it turns out that actually forgetting a given concept is an extremely difficult task. Models exposed to attacks using adversarial prompts show the ability to generate so-called unlearned concepts, which can be not only harmful but also illegal. In this paper, we present considerations regarding the ability of models to forget and recall knowledge, introducing the Memory Self-Regeneration task. Furthermore, we present MemoRa strategy, which we consider to be a regenerative approach supporting the effective recovery of previously lost knowledge. Moreover, we propose that robustness in knowledge retrieval is a crucial yet underexplored evaluation measure for developing more robust and effective unlearning techniques. Finally, we demonstrate that forgetting occurs in two distinct ways: short-term, where concepts can be quickly recalled, and long-term, where recovery is more challenging.
Authors: Syeda Nahida Akter, Shrimai Prabhumoye, Eric Nyberg, Mostofa Patwary, Mohammad Shoeybi, Yejin Choi, Bryan Catanzaro
Abstract: The prevailing paradigm for enhancing the reasoning abilities of LLMs revolves around post-training on high-quality, reasoning-intensive data. While emerging literature suggests that reasoning data is increasingly incorporated also during the mid-training stage-a practice that is relatively more proprietary and less openly characterized-the role of such data in pretraining remains unclear. In particular, due to the opaqueness of pretraining corpora in most frontier models, the effect of reasoning data introduced at different phases of pre- and/or post-training is relatively less reported in the scientific literature. This raises several important questions: Is adding reasoning data earlier during pretraining any better than introducing it during post-training? Could earlier inclusion risk overfitting and harm generalization, or instead establish durable foundations that later fine-tuning cannot recover? We conduct the first systematic study of how reasoning data-varying in scale, diversity, and quality-affects LLM performance when introduced at different stages of training. We find that front-loading reasoning data into pretraining is critical (19% avg gain), establishing foundational capabilities that cannot be fully replicated by later-stage SFT, even with more data. We uncover an asymmetric principle for optimal data allocation: pretraining benefits most from broad diversity in reasoning patterns (11% avg gain), while SFT is more sensitive to data quality (15% avg gain). We show that high-quality pretraining data has latent effects, activated only after SFT, and that naively scaling SFT data can be detrimental, washing away the benefits of early reasoning injection. Our results challenge the conventional separation of language modeling and reasoning, providing a principled guide for strategically allocating data across the entire training pipeline to build more capable models.
Authors: Bowei Tian, Yexiao He, Wanghao Ye, Ziyao Wang, Meng Liu, Ang Li
Abstract: Large-scale foundation models demonstrate strong performance across language, vision, and reasoning tasks. However, how they internally structure and stabilize concepts remains elusive. Inspired by causal inference, we introduce the MindCraft framework built upon Concept Trees. By applying spectral decomposition at each layer and linking principal directions into branching Concept Paths, Concept Trees reconstruct the hierarchical emergence of concepts, revealing exactly when they diverge from shared representations into linearly separable subspaces. Empirical evaluations across diverse scenarios across disciplines, including medical diagnosis, physics reasoning, and political decision-making, show that Concept Trees recover semantic hierarchies, disentangle latent concepts, and can be widely applied across multiple domains. The Concept Tree establishes a widely applicable and powerful framework that enables in-depth analysis of conceptual representations in deep models, marking a significant step forward in the foundation of interpretable AI.
Authors: Bharat Sharma, Jitendra Kumar
Abstract: Climate anomalies significantly impact terrestrial carbon cycle dynamics, necessitating robust methods for detecting and analyzing anomalous behavior in plant productivity. This study presents a novel application of variational autoencoders (VAE) for identifying extreme events in gross primary productivity (GPP) from Community Earth System Model version 2 simulations across four AR6 regions in the Continental United States. We compare VAE-based anomaly detection with traditional singular spectral analysis (SSA) methods across three time periods: 1850-80, 1950-80, and 2050-80 under the SSP585 scenario. The VAE architecture employs three dense layers and a latent space with an input sequence length of 12 months, trained on a normalized GPP time series to reconstruct the GPP and identifying anomalies based on reconstruction errors. Extreme events are defined using 5th percentile thresholds applied to both VAE and SSA anomalies. Results demonstrate strong regional agreement between VAE and SSA methods in spatial patterns of extreme event frequencies, despite VAE producing higher threshold values (179-756 GgC for VAE vs. 100-784 GgC for SSA across regions and periods). Both methods reveal increasing magnitudes and frequencies of negative carbon cycle extremes toward 2050-80, particularly in Western and Central North America. The VAE approach shows comparable performance to established SSA techniques, while offering computational advantages and enhanced capability for capturing non-linear temporal dependencies in carbon cycle variability. Unlike SSA, the VAE method does not require one to define the periodicity of the signals in the data; it discovers them from the data.
Authors: Xianglong Yan, Chengzhu Bao, Zhiteng Li, Tianao Zhang, Kaicheng Yang, Haotong Qin, Ruobing Xie, Xingwu Sun, Yulun Zhang
Abstract: Large Language Models (LLMs) have shown impressive capabilities across diverse tasks, but their large memory and compute demands hinder deployment. Ternarization has gained attention as a promising compression technique, delivering substantial size reduction and high computational efficiency. However, its potential in the post-training quantization (PTQ) setting remains underexplored, due to the challenge of training-free parameter optimization and the quantization difficulty posed by outliers and dispersed weights. To address these issues, we propose PT$^2$-LLM, a post-training ternarization framework tailored for LLMs. At its core is an Asymmetric Ternary Quantizer equipped with a two-stage refinement pipeline: (1) Iterative Ternary Fitting (ITF), which alternates between optimal ternary grid construction and flexible rounding to minimize quantization error, and (2) Activation-aware Grid Alignment (AGA), which further refines the ternary grid to better match full-precision outputs. In addition, we propose a plug-and-play Structural Similarity-based Reordering (SSR) strategy that leverages inter-column structural similarity to ease quantization and mitigate outlier effects, further enhancing overall performance. Extensive experiments demonstrate that PT$^2$-LLM delivers competitive performance against state-of-the-art (SOTA) 2-bit PTQ methods with lower memory cost, while also accelerating both prefill and decoding to achieve end-to-end speedup. The code and models will be available at https://github.com/XIANGLONGYAN/PT2-LLM.
Authors: Lingjie Yi, Raphael Douady, Chao Chen
Abstract: Multimodal contrastive learning (MCL) aims to embed data from different modalities in a shared embedding space. However, empirical evidence shows that representations from different modalities occupy completely separate regions of embedding space, a phenomenon referred to as the modality gap. Moreover, experimental findings on how the size of the modality gap influences downstream performance are inconsistent. These observations raise two key questions: (1) What causes the modality gap? (2) How does it affect downstream tasks? To address these questions, this paper introduces the first theoretical framework for analyzing the convergent optimal representations of MCL and the modality alignment when training is optimized. Specifically, we prove that without any constraint or under the cone constraint, the modality gap converges to zero. Under the subspace constraint (i.e., representations of two modalities fall into two distinct hyperplanes due to dimension collapse), the modality gap converges to the smallest angle between the two hyperplanes. This result identifies \emph{dimension collapse} as the fundamental origin of the modality gap. Furthermore, our theorems demonstrate that paired samples cannot be perfectly aligned under the subspace constraint. The modality gap influences downstream performance by affecting the alignment between sample pairs. We prove that, in this case, perfect alignment between two modalities can still be achieved via two ways: hyperplane rotation and shared space projection.
Authors: Wendi Li, Changdae Oh, Yixuan Li
Abstract: Optimistic exploration is central to improving sample efficiency in reinforcement learning with human feedback, yet existing exploratory bonus methods to incentivize exploration often fail to realize optimism. We provide a theoretical analysis showing that current formulations, under KL or $\alpha$-divergence regularization, unintentionally bias exploration toward high-probability regions of the reference model, thereby reinforcing conservative behavior instead of promoting discovery of uncertain regions. To address this pitfall, we introduce the General Exploratory Bonus (GEB), a novel theoretical framework that provably satisfies the optimism principle. GEB counteracts divergence-induced bias via reference-dependent reward regulation and unifies prior heuristic bonuses as special cases, while extending naturally across the full $\alpha$-divergence family. Empirically, GEB consistently outperforms baselines on alignment tasks across multiple divergence settings and large language model backbones. These results demonstrate that GEB offers both a principled and practical solution for optimistic exploration in RLHF.
Authors: Haolin Chen, Shiyu Wang, Can Qin, Bo Pang, Zuxin Liu, Jielin Qiu, Jianguo Zhang, Yingbo Zhou, Zeyuan Chen, Ran Xu, Shelby Heinecke, Silvio Savarese, Caiming Xiong, Huan Wang, Weiran Yao
Abstract: Diffusion language models promise bidirectional context and infilling capabilities that autoregressive coders lack, yet practical systems remain heavyweight. We introduce CoDA, a 1.7B-parameter diffusion coder trained on TPU with a fully open-source training pipeline. CoDA pairs large-scale diffusion pre-training with code-centric mid-training and instruction tuning, enabling confidence-guided sampling that keeps inference latency competitive. On Humaneval, MBPP, and EvalPlus, CoDA-1.7B-Instruct matches or surpasses diffusion models up to 7B parameters. Our release includes model checkpoints, evaluation harnesses, and TPU training pipelines to accelerate research on lightweight diffusion-based coding assistants.
Authors: Zi Liang, Zhiyao Wu, Haoyang Shang, Yulin Jin, Qingqing Ye, Huadi Zheng, Peizhao Hu, Haibo Hu
Abstract: Decision boundary, the subspace of inputs where a machine learning model assigns equal classification probabilities to two classes, is pivotal in revealing core model properties and interpreting behaviors. While analyzing the decision boundary of large language models (LLMs) has raised increasing attention recently, constructing it for mainstream LLMs remains computationally infeasible due to the enormous vocabulary-sequence sizes and the auto-regressive nature of LLMs. To address this issue, in this paper we propose Decision Potential Surface (DPS), a new notion for analyzing LLM decision boundary. DPS is defined on the confidences in distinguishing different sampling sequences for each input, which naturally captures the potential of decision boundary. We prove that the zero-height isohypse in DPS is equivalent to the decision boundary of an LLM, with enclosed regions representing decision regions. By leveraging DPS, for the first time in the literature, we propose an approximate decision boundary construction algorithm, namely $K$-DPS, which only requires K-finite times of sequence sampling to approximate an LLM's decision boundary with negligible error. We theoretically derive the upper bounds for the absolute error, expected error, and the error concentration between K-DPS and the ideal DPS, demonstrating that such errors can be trade-off with sampling times. Our results are empirically validated by extensive experiments across various LLMs and corpora.
Authors: Yukun Zhang, Xueqing Zhou
Abstract: The Transformer architecture has revolutionized artificial intelligence, yet a principled theoretical understanding of its internal mechanisms remains elusive. This paper introduces a novel analytical framework that reconceptualizes the Transformer's discrete, layered structure as a continuous spatiotemporal dynamical system governed by a master Partial Differential Equation (PDE). Within this paradigm, we map core architectural components to distinct mathematical operators: self-attention as a non-local interaction, the feed-forward network as a local reaction, and, critically, residual connections and layer normalization as indispensable stabilization mechanisms. We do not propose a new model, but rather employ the PDE system as a theoretical probe to analyze the mathematical necessity of these components. By comparing a standard Transformer with a PDE simulator that lacks explicit stabilizers, our experiments provide compelling empirical evidence for our central thesis. We demonstrate that without residual connections, the system suffers from catastrophic representational drift, while the absence of layer normalization leads to unstable, explosive training dynamics. Our findings reveal that these seemingly heuristic "tricks" are, in fact, fundamental mathematical stabilizers required to tame an otherwise powerful but inherently unstable continuous system. This work offers a first-principles explanation for the Transformer's design and establishes a new paradigm for analyzing deep neural networks through the lens of continuous dynamics.
Authors: Chenhao Ye, Ming Tang
Abstract: Backpropagation (BP), while foundational to deep learning, imposes two critical scalability bottlenecks: update locking, where network modules remain idle until the entire backward pass completes, and high memory consumption due to storing activations for gradient computation. To address these limitations, we introduce Synergistic Information Distillation (SID), a novel training framework that reframes deep learning as a cascade of local cooperative refinement problems. In SID, a deep network is structured as a pipeline of modules, each imposed with a local objective to refine a probabilistic belief about the ground-truth target. This objective balances fidelity to the target with consistency to the belief from its preceding module. By decoupling the backward dependencies between modules, SID enables parallel training and hence eliminates update locking and drastically reduces memory requirements. Meanwhile, this design preserves the standard feed-forward inference pass, making SID a versatile drop-in replacement for BP. We provide a theoretical foundation, proving that SID guarantees monotonic performance improvement with network depth. Empirically, SID consistently matches or surpasses the classification accuracy of BP, exhibiting superior scalability and pronounced robustness to label noise.Code is available at: https://github.com/ychAlbert/sid-bp
Authors: Tianao Zhang, Zhiteng Li, Xianglong Yan, Haotong Qin, Yong Guo, Yulun Zhang
Abstract: Diffusion large language models (dLLMs), which offer bidirectional context and flexible masked-denoising generation, are emerging as a compelling alternative to autoregressive (AR) LLMs. However, like AR LLMs, their model sizes continue to grow, motivating weight compression for deployment. Although post-training quantization (PTQ) is effective for AR LLMs, directly transferring it to dLLMs at 2-bit leads to unsatisfactory performance. To tackle these challenges, we propose Quant-dLLM, an ultra-low-bit PTQ framework tailored to dLLMs. Since masked-denoising activations in dLLMs differ from the fully visible signals assumed by standard PTQ methods, we introduce Masked Calibration Simulation (MCS) to align calibration with the timestep-dependent masking, which yields more reliable calibrations. Moreover, we propose a Data-aware Any-order Quantizer (DAQ) that learns ultra-low-bit weight representations via an optimization algorithm. It performs iterative approximation guided by our simulated calibration data. In addition, under a strict 2-bit budget, we introduce Adaptive Blockwise Mixed Precision (ABMP), a sensitivity-based precision allocation scheme that adaptively assigns bit width across channel groups. When restricted to 2-bit precision, Quant-dLLM consistently achieves higher accuracy than state-of-the-art (SOTA) AR-transfer PTQ methods on dLLMs. The code and models will be available at: https://github.com/ZTA2785/Quant-dLLM.
Authors: Junhao Xia, Ming Zhao, Limin Xiao, Xiujun Zhang
Abstract: Large language models (LLMs) face significant computational and memory challenges, making extremely low-bit quantization crucial for their efficient deployment. In this work, we introduce SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size, a novel framework that enables extremely low-bit quantization of LLMs while preserving their linguistic reasoning capabilities. A distinctive feature of SDQ-LLM is the continuous adjustability of the Over-Sampling Ratio (OSR), enabling dynamic adaptation to memory or VRAM constraints by selecting fractional OSR (e.g. 2.5 times) for an optimal trade-off between model size and accuracy. SDQ-LLM uses upsampling combined with Sigma-Delta Quantizer to binarize or ternarize LLMs weights, encoding high-precision parameters into 1-bit or 1.58-bit representations, replacing the multiplication operations within linear layers with addition. This approach significantly enhances inference efficiency under extremely low-bit quantization. To further reduce the loss of quantization precision, we incorporate Hadamard-based weight smoothing prior to quantization, improving the stability and robustness of the weight representations. Furthermore, to fully leverage the continuity of the OSR and reduce precision loss, recognizing the correlation between quantization sensitivity and weight variance, we propose a fine-grained, layer- and linear-wise OSR allocation strategy, MultiOSR. This strategy distributes OSR both across layers and within each layer, based on weight variance and parameter scale. Finally, extensive experiments on OPT and LLaMA model families demonstrate that SDQ-LLM achieves a more efficient and high-precision performance even under highly aggressive low-OSR settings. Our code is available at https://github.com/Dreamlittlecat/LLM-Quant-Factory.
Authors: Qian Chen, Linxin Yang, Akang Wang, Xiaodong Luo, Yin Zhang
Abstract: The combination of linear transformations and non-linear activation functions forms the foundation of most modern deep neural networks, enabling them to approximate highly complex functions. This paper explores the introduction of quadratic transformations to further increase nonlinearity in neural networks, with the aim of enhancing the performance of existing architectures. To reduce parameter complexity and computational complexity, we propose a lightweight quadratic enhancer that uses low-rankness, weight sharing, and sparsification techniques. For a fixed architecture, the proposed approach introduces quadratic interactions between features at every layer, while only adding negligible amounts of additional model parameters and forward computations. We conduct a set of proof-of-concept experiments for the proposed method across three tasks: image classification, text classification, and fine-tuning large-language models. In all tasks, the proposed approach demonstrates clear and substantial performance gains.
Authors: Filip Landgren
Abstract: Bayesian physics-informed neural networks (B-PINNs) merge data with governing equations to solve differential equations under uncertainty. However, interpreting uncertainty and overconfidence in B-PINNs requires care due to the poorly understood effects the physical constraints have on the network; overconfidence could reflect warranted precision, enforced by the constraints, rather than miscalibration. Motivated by the need to further clarify how individual physical constraints shape these networks, we introduce a scalable, matrix-free Laplace framework that decomposes the posterior Hessian into contributions from each constraint and provides metrics to quantify their relative influence on the loss landscape. Applied to the Van der Pol equation, our method tracks how constraints sculpt the network's geometry and shows, directly through the Hessian, how changing a single loss weight non-trivially redistributes curvature and effective dominance across the others.
Authors: Youjin Wang, Yangjingyi Chen, Jiahao Yan, Jiaxuan Lu, Xiao Sun
Abstract: With the explosive growth of data, long-sequence modeling has become increasingly important in tasks such as natural language processing and bioinformatics. However, existing methods face inherent trade-offs between efficiency and memory. Recurrent neural networks suffer from gradient vanishing and explosion, making them hard to scale. Transformers can model global dependencies but are constrained by quadratic complexity. Recently, selective state-space models such as Mamba have demonstrated high efficiency with O(n) time and O(1) recurrent inference, yet their long-range memory decays exponentially. In this work, we conduct mathematical derivations and information-theoretic analysis to systematically uncover the memory decay mechanism of Mamba, answering a fundamental question: what is the nature of Mamba's long-range memory and how does it retain information? To quantify key information loss, we further introduce horizontal-vertical memory fidelity metrics that capture degradation both within and across layers. Inspired by how humans distill and retain salient information when reading long documents, we propose MemMamba, a novel architectural framework that integrates state summarization mechanism together with cross-layer and cross-token attention, which alleviates long-range forgetting while preserving linear complexity. MemMamba achieves significant improvements over existing Mamba variants and Transformers on long-sequence benchmarks such as PG19 and Passkey Retrieval, while delivering a 48% speedup in inference efficiency. Both theoretical analysis and empirical results demonstrate that MemMamba achieves a breakthrough in the complexity-memory trade-off, offering a new paradigm for ultra-long sequence modeling.
Authors: Jinjie Ni, Qian Liu, Chao Du, Longxu Dou, Hang Yan, Zili Wang, Tianyu Pang, Michael Qizhe Shieh
Abstract: We introduce Quokka, the first systematic scaling law for diffusion language models (DLMs), encompassing both compute-constrained and data-constrained regimes, and studying the key modeling and optimization designs. Quokka is a good friend of Chinchilla and provides wider scopes. We hope the results would bring short-term practical guidance in DLMs training and long-term inspirations for the whole AI community.
Authors: Hao Gu, Vibhas Nair, Amrithaa Ashok Kumar, Jayvart Sharma, Ryan Lagasse
Abstract: Interpreting language models often involves circuit analysis, which aims to identify sparse subnetworks, or circuits, that accomplish specific tasks. Existing circuit discovery algorithms face a fundamental trade-off: attribution patching is fast but unfaithful to the full model, while edge pruning is faithful but computationally expensive. This research proposes a hybrid attribution and pruning (HAP) framework that uses attribution patching to identify a high-potential subgraph, then applies edge pruning to extract a faithful circuit from it. We show that HAP is 46\% faster than baseline algorithms without sacrificing circuit faithfulness. Furthermore, we present a case study on the Indirect Object Identification task, showing that our method preserves cooperative circuit components (e.g. S-inhibition heads) that attribution patching methods prune at high sparsity. Our results show that HAP could be an effective approach for improving the scalability of mechanistic interpretability research to larger models. Our code is available at https://anonymous.4open.science/r/HAP-circuit-discovery.
URLs: https://anonymous.4open.science/r/HAP-circuit-discovery.
Authors: Yufei Li, Yu Fu, Yue Dong, Cong Liu
Abstract: Large language models (LLMs) deployed on edge servers are increasingly used in latency-sensitive applications such as personalized assistants, recommendation, and content moderation. However, the non-stationary nature of user data necessitates frequent retraining, which introduces a fundamental tension between inference latency and model accuracy under constrained GPU resources. Existing retraining strategies either delay model updates, over-commit resources to retraining, or overlook iteration-level retraining granularity. In this paper, we identify that iteration-level scheduling is crucial for adapting retraining frequency to model drift without violating service-level objectives (SLOs). We propose MACE, a hybrid LLM system that colocates concurrent inference (prefill, decode) and fine-tuning, with intelligent memory management to maximize task performance while promising inference throughput. MACE leverages the insight that not all model updates equally affect output alignment and allocates GPU cycles accordingly to balance throughput, latency, and update freshness. Our trace-driven evaluation shows that MACE matches or exceeds continuous retraining while reducing inference latency by up to 63% and maintaining throughput under resource constraints. Compared to periodic retraining, MACE improves latency breakdown across prefill, decode, and finetune stages, and sustains GPU utilization above 85% in NVIDIA AGX Orin. These results demonstrate that iteration-level hybrid scheduling is a promising direction for deploying LLMs with continual learning capabilities on edge platforms.
Authors: Vinay Venkatesh, Vamsidhar R Kamanuru, Lav Kumar, Nikita Kothari
Abstract: This paper proposes Edge-FIT (Federated Instruction Tuning on the Edge), a scalable framework for Federated Instruction Tuning (FIT) of Large Language Models (LLMs). Traditional Federated Learning (TFL) methods, like FedAvg, fail when confronted with the massive parameter size of LLMs [3], [6]. Our Edge-FIT framework combines federated learning with 4-bit Quantized Low-Rank Adaptation (QLORA), mitigating the core issues of communication and computational overhead. We demonstrate this by filtering the general-purpose Databricks Dolly 15k dataset for the IoT domain. Experimental results show the Edge-FIT tuned Llama 2(7B) achieves an F1-Score of 0.89. We also demonstrate a viable trade-off using the 3.8B Phi-3-mini model, validating Edge-FIT as a scalable framework for decentralized LLM deployment on home compute gateways.
Authors: Chiming Duan, Minghua He, Pei Xiao, Tong Jia, Xin Zhang, Zhewei Zhong, Xiang Luo, Yan Niu, Lingzhe Zhang, Yifan Wu, Siyu Yu, Weijie Hong, Ying Li, Gang Huang
Abstract: Log-based anomaly detection is a essential task for ensuring the reliability and performance of software systems. However, the performance of existing anomaly detection methods heavily relies on labeling, while labeling a large volume of logs is highly challenging. To address this issue, many approaches based on transfer learning and active learning have been proposed. Nevertheless, their effectiveness is hindered by issues such as the gap between source and target system data distributions and cold-start problems. In this paper, we propose LogAction, a novel log-based anomaly detection model based on active domain adaptation. LogAction integrates transfer learning and active learning techniques. On one hand, it uses labeled data from a mature system to train a base model, mitigating the cold-start issue in active learning. On the other hand, LogAction utilize free energy-based sampling and uncertainty-based sampling to select logs located at the distribution boundaries for manual labeling, thus addresses the data distribution gap in transfer learning with minimal human labeling efforts. Experimental results on six different combinations of datasets demonstrate that LogAction achieves an average 93.01% F1 score with only 2% of manual labels, outperforming some state-of-the-art methods by 26.28%. Website: https://logaction.github.io
Authors: Haocheng Sun, Cynthia Xin Wen, Edward Hong Wang
Abstract: The main advantages of diffusion language models over autoregressive (AR) models lie in their ability to support parallel generation and bidirectional attention, enabling a more controllable generation process. In recent years, open-source mask diffusion language models have emerged, most of which are based on a variant known as absorbing diffusion. However, this paper demonstrates why mask diffusion faces inherent difficulties in achieving parallel generation and bidirectional attention. We also propose the most effective training and inference strategies for mask diffusion.
Authors: X. Angelo Huang, Ruben Ciranni, Giovanni Spadaccini, Carla J. L\'opez Zurita
Abstract: Within the growing interest in the physical sciences in developing networks with equivariance properties, Clifford neural layers shine as one approach that delivers $E(n)$ and $O(n)$ equivariances given specific group actions. In this paper, we analyze the inner structure of the computation within Clifford convolutional layers and propose and implement several optimizations to speed up the inference process while maintaining correctness. In particular, we begin by analyzing the theoretical foundations of Clifford algebras to eliminate redundant matrix allocations and computations, then systematically apply established optimization techniques to enhance performance further. We report a final average speedup of 21.35x over the baseline implementation of eleven functions and runtimes comparable to and faster than the original PyTorch implementation in six cases. In the remaining cases, we achieve performance in the same order of magnitude as the original library.
Authors: Yizhuo Ding, Wanying Qu, Jiawei Geng, Wenqi Shao, Yanwei Fu
Abstract: Large Language Models (LLMs) achieve strong performance across diverse tasks but face prohibitive computational and memory costs. Pruning offers a promising path by inducing sparsity while preserving architectural flexibility. However, existing methods struggle to balance efficiency and robustness: local metric approaches prune layer by layer but often collapse under high sparsity, whereas global feedback methods enforce consistency at the cost of expensive weight updates or restrictive semi-structured formats. We present UniPruning, a unified post-training pruning framework that combines the speed of local saliency metrics with the stability of global coordination, enabled by a mirror descent based optimization, all without updating model weights. UniPruning leverages fast layer-wise scoring and a lightweight global controller to allocate a single sparsity budget, supporting both unstructured and semi-structured N :M pruning within one framework. After a brief calibration, it can generate pruning masks for arbitrary sparsity levels in one shot, and adapts seamlessly to hardware-aware constraints. Extensive experiments on multiple pretrained LLM families and standard benchmarks show that UniPruning consistently delivers competitive or superior perplexity and zero-shot accuracy. Ablation studies further highlight the importance of mirror descent and local saliency anchoring. Overall, UniPruning provides an efficient, principled, and scalable solution for sparsifying large-scale LLMs. Our code is available at: https://github.com/RainbowQTT/UniPruning.
Authors: Rana Shahout, Colin Cai, Yilun Du, Minlan Yu, Michael Mitzenmacher
Abstract: Mixture-of-Experts (MoE) models can scale parameter capacity by routing each token to a subset of experts through a learned gate function. While conditional routing reduces training costs, it shifts the burden on inference memory: expert parameters and activations consume memory, limiting the number of experts per device. As tokens are routed, some experts become overloaded while others are underutilized. Because experts are mapped to GPUs, this imbalance translates directly into degraded system performance in terms of latency, throughput, and cost. We present LASER, a plug-and-play, inference-time routing algorithm that balances load while preserving accuracy. LASER adapts to the shape of the gate's score distribution. When scores provide a clear preference, it routes to the strongest experts; when scores are more uniform, it broadens the set of viable experts and routes to the least-loaded among them. Because LASER relies only on gate scores from a trained model, it integrates directly into existing MoE inference pipelines without retraining or finetuning. We evaluate LASER on Mixtral-8x7B and DeepSeek-MoE-16b-chat across four datasets (ARC-Easy, ARC-Challenge, MMLU, and GSM8K). LASER improves load balancing, translating into lower latency and higher throughput, while keeping the accuracy changes negligible.
Authors: Dongqi Zheng, Wenjin Fu
Abstract: We introduce Constraint-Aware Federated Learning with Lagrangian Dual Optimization (CAFL-L), a principled extension of FedAvg that explicitly incorporates device-level resource constraints including energy, communication, memory, and thermal budgets. CAFL-L employs Lagrangian dual optimization to dynamically adapt training hyperparameters -- freezing depth, local steps, batch size, and communication compression -- while preserving training stability through token-budget preservation via gradient accumulation. Experiments on a character-level language model demonstrate that CAFL-L achieves superior constraint satisfaction compared to standard FedAvg (reducing memory usage by 20% and communication by 95%) while maintaining competitive validation performance, making it practical for deployment on resource-constrained edge devices.
Authors: Arthur Sedek
Abstract: This paper introduces a novel adaptive ensemble framework that synergistically combines XGBoost and neural networks through sophisticated meta-learning. The proposed method leverages advanced uncertainty quantification techniques and feature importance integration to dynamically orchestrate model selection and combination. Experimental results demonstrate superior predictive performance and enhanced interpretability across diverse datasets, contributing to the development of more intelligent and flexible machine learning systems.
Authors: Daiheng Gao, Nanxiang Jiang, Andi Zhang, Shilin Lu, Yufei Tang, Wenbo Zhou, Weiming Zhang, Zhaoxin Fan
Abstract: Concept erasure techniques have been widely deployed in T2I diffusion models to prevent inappropriate content generation for safety and copyright considerations. However, as models evolve to next-generation architectures like Flux, established erasure methods (\textit{e.g.}, ESD, UCE, AC) exhibit degraded effectiveness, raising questions about their true mechanisms. Through systematic analysis, we reveal that concept erasure creates only an illusion of ``amnesia": rather than genuine forgetting, these methods bias sampling trajectories away from target concepts, making the erasure fundamentally reversible. This insight motivates the need to distinguish superficial safety from genuine concept removal. In this work, we propose \textbf{RevAm} (\underline{Rev}oking \underline{Am}nesia), an RL-based trajectory optimization framework that resurrects erased concepts by dynamically steering the denoising process without modifying model weights. By adapting Group Relative Policy Optimization (GRPO) to diffusion models, RevAm explores diverse recovery trajectories through trajectory-level rewards, overcoming local optima that limit existing methods. Extensive experiments demonstrate that RevAm achieves superior concept resurrection fidelity while reducing computational time by 10$\times$, exposing critical vulnerabilities in current safety mechanisms and underscoring the need for more robust erasure techniques beyond trajectory manipulation.
Authors: Tian Zheng, Subashree Venkatasubramanian, Shuolin Li, Amy Braverman, Xinyi Ke, Zhewen Hou, Peter Jin, Samarth Sanjay Agrawal
Abstract: Machine learning has been increasingly applied in climate modeling on system emulation acceleration, data-driven parameter inference, forecasting, and knowledge discovery, addressing challenges such as physical consistency, multi-scale coupling, data sparsity, robust generalization, and integration with scientific workflows. This paper analyzes a series of case studies from applied machine learning research in climate modeling, with a focus on design choices and workflow structure. Rather than reviewing technical details, we aim to synthesize workflow design patterns across diverse projects in ML-enabled climate modeling: from surrogate modeling, ML parameterization, probabilistic programming, to simulation-based inference, and physics-informed transfer learning. We unpack how these workflows are grounded in physical knowledge, informed by simulation data, and designed to integrate observations. We aim to offer a framework for ensuring rigor in scientific machine learning through more transparent model development, critical evaluation, informed adaptation, and reproducibility, and to contribute to lowering the barrier for interdisciplinary collaboration at the interface of data science and climate modeling.
Authors: Mallikarjuna Tupakula
Abstract: Multimodal foundation models hold promise for drug discovery and biomedical applications, but most existing approaches rely on heavy pretraining or large scale multimodal corpora. We investigate whether thin contrastive bridges, lightweight projection heads over frozen unimodal encoders can align chemical and textual representations without training a full multimodal model. Using paired mechanisms from ChEMBL, we align ECFP4 molecular fingerprints with biomedical sentence embeddings through dual linear projections trained with a contrastive objective. To better handle drugs sharing the same therapeutic target, we incorporate hard negative weighting and a margin loss. Evaluation under scaffold based splits, which require generalization across disjoint chemical cores, demonstrates that our approach achieves non-trivial cross modal alignment and substantially improves within target discrimination compared to frozen baselines. These results suggest that thin bridges offer a compute efficient alternative to large scale multimodal pretraining, enabling scaffold aware drug text alignment and target specific retrieval in precision medicine.
Authors: Runze Zhang, Xiaowei Zhang, Mingyang Zhao
Abstract: LLMs are emerging tools for simulating human behavior in business, economics, and social science, offering a lower-cost complement to laboratory experiments, field studies, and surveys. This paper evaluates how well LLMs replicate human behavior in operations management. Using nine published experiments in behavioral operations, we assess two criteria: replication of hypothesis-test outcomes and distributional alignment via Wasserstein distance. LLMs reproduce most hypothesis-level effects, capturing key decision biases, but their response distributions diverge from human data, including for strong commercial models. We also test two lightweight interventions -- chain-of-thought prompting and hyperparameter tuning -- which reduce misalignment and can sometimes let smaller or open-source models match or surpass larger systems.
Authors: Anirudh Subramanyam, Yuxin Chen, Robert L. Grossman
Abstract: Scaling laws for language model training traditionally characterize how performance scales with model size and dataset volume. Prior work has explored architecture variants and data treatments such as dataset filtering and noise injection in language model pretraining; however, these studies have not formalized data quality within a principled scaling law. We introduce a dimensionless data-quality parameter Q, and propose a quality-aware scaling law extending the Chinchilla framework to predict loss as a joint function of model size, data volume, and data quality. The law is motivated by an effective-sample-size and information-theoretic view of noisy or redundant corpora, and it admits two practical estimators for Q: (i) a corruption rate proxy and (ii) a deficiency measure. Through synthetic experiments in neural machine translation and autoregressive modeling -- where we systematically control data quality via multiple levels of noise injection and coverage variation -- we show that loss scales predictably with data quality and that higher-quality data can substantially reduce model size and hence compute requirements. Our results demonstrate a sublinear decay of effective data with quality and robustness to moderate data corruption; out-of-sample evaluations further validate the predictive form of the law. Unlike prior empirical analyses, our work establishes an explicit, generalizable law for data quality, offering concrete guidance for balancing data curation effort and model scale in large-scale pretraining.
Authors: Giuseppe Di Somma, Giorgio Carelli, Angela D. V. Di Virgilio, Francesco Fuso, Enrico Maccioni, Paolo Marsili
Abstract: The reconstruction of a frequency with minimal delay from a sinusoidal signal is a common task in several fields; for example Ring Laser Gyroscopes, since their output signal is a beat frequency. While conventional methods require several seconds of data, we present a neural network approach capable of reconstructing frequencies of several hundred Hertz within approximately 10 milliseconds. This enables rapid trigger generation. The method outperforms standard Fourier-based techniques, improving frequency estimation precision by a factor of 2 in the operational range of GINGERINO, our Ring Laser Gyroscope.\\ In addition to fast frequency estimation, we introduce an automated classification framework to identify physical disturbances in the signal, such as laser instabilities and seismic events, achieving accuracy rates between 99\% and 100\% on independent test datasets for the seismic class. These results mark a step forward in integrating artificial intelligence into signal analysis for geophysical applications.
Authors: Andy Wu, Chun-Cheng Lin, Yuehua Huang, Rung-Tzuo Liaw
Abstract: The training process of reinforcement learning often suffers from severe oscillations, leading to instability and degraded performance. In this paper, we propose a Constant in an Ever-Changing World (CIC) framework that enhances algorithmic stability to improve performance. CIC maintains both a representative policy and a current policy. Instead of updating the representative policy blindly, CIC selectively updates it only when the current policy demonstrates superiority. Furthermore, CIC employs an adaptive adjustment mechanism, enabling the representative and current policies to jointly facilitate critic training. We evaluate CIC on five MuJoCo environments, and the results show that CIC improves the performance of conventional algorithms without incurring additional computational cost.
Authors: Zerui Wang, Qinghao Hu, Ana Klimovic, Tianwei Zhang, Yonggang Wen, Peng Sun, Dahua Lin
Abstract: Deep learning (DL) schedulers are pivotal in optimizing resource allocation in GPU clusters, but operate with a critical limitation: they are largely blind to the semantic context of the jobs they manage. This forces them to rely on limited metadata, leading to high profiling overhead, unreliable duration estimation, inadequate failure handling, and poor observability. To this end, we propose SchedMate, a framework that bridges this semantic gap by systematically extracting deep insights from overlooked, unstructured data sources: source code, runtime logs, and historical jobs. SchedMate enhances existing schedulers non-intrusively through three LLM-based components. Our implementation integrates seamlessly with existing deep learning schedulers. Evaluations on a 128-GPU physical cluster and extensive simulations on production traces show SchedMate reduces average job completion times by up to 1.91x, substantially enhancing the scheduling performance, demonstrating the critical role of semantic-awareness in modern DL scheduling.
Authors: Ameya Daigavane, YuQing Xie, Bodhi P. Vani, Saeed Saremi, Joseph Kleinhenz, Tess Smidt
Abstract: Diffusion models are a popular class of generative models trained to reverse a noising process starting from a target data distribution. Training a diffusion model consists of learning how to denoise noisy samples at different noise levels. When training diffusion models for point clouds such as molecules and proteins, there is often no canonical orientation that can be assigned. To capture this symmetry, the true data samples are often augmented by transforming them with random rotations sampled uniformly over $SO(3)$. Then, the denoised predictions are often rotationally aligned via the Kabsch-Umeyama algorithm to the ground truth samples before computing the loss. However, the effect of this alignment step has not been well studied. Here, we show that the optimal denoiser can be expressed in terms of a matrix Fisher distribution over $SO(3)$. Alignment corresponds to sampling the mode of this distribution, and turns out to be the zeroth order approximation for small noise levels, explaining its effectiveness. We build on this perspective to derive better approximators to the optimal denoiser in the limit of small noise. Our experiments highlight that alignment is often a `good enough' approximation for the noise levels that matter most for training diffusion models.
Authors: Sofiane Ennadir, Levente Z\'olyomi, Oleg Smirnov, Tianze Wang, John Pertoft, Filip Cornell, Lele Cao
Abstract: Transformer models have become the dominant backbone for sequence modeling, leveraging self-attention to produce contextualized token representations. These are typically aggregated into fixed-size vectors via pooling operations for downstream tasks. While much of the literature has focused on attention mechanisms, the role of pooling remains underexplored despite its critical impact on model behavior. In this paper, we introduce a theoretical framework that rigorously characterizes the expressivity of Transformer-based models equipped with widely used pooling methods by deriving closed-form bounds on their representational capacity and the ability to distinguish similar inputs. Our analysis extends to different variations of attention formulations, demonstrating that these bounds hold across diverse architectural variants. We empirically evaluate pooling strategies across tasks requiring both global and local contextual understanding, spanning three major modalities: computer vision, natural language processing, and time-series analysis. Results reveal consistent trends in how pooling choices affect accuracy, sensitivity, and optimization behavior. Our findings unify theoretical and empirical perspectives, providing practical guidance for selecting or designing pooling mechanisms suited to specific tasks. This work positions pooling as a key architectural component in Transformer models and lays the foundation for more principled model design beyond attention alone.
Authors: Marian Chen, Miri Zilka
Abstract: The COVID-19 pandemic underscored a critical need for intervention strategies that balance disease containment with socioeconomic stability. We approach this challenge by designing a framework for modeling and evaluating disease-spread prevention strategies. Our framework leverages multi-objective reinforcement learning (MORL) - a formulation necessitated by competing objectives - combined with a new stochastic differential equation (SDE) pandemic simulator, calibrated and validated against global COVID-19 data. Our simulator reproduces national-scale pandemic dynamics with orders of magnitude higher fidelity than other models commonly used in reinforcement learning (RL) approaches to pandemic intervention. Training a Pareto-Conditioned Network (PCN) agent on this simulator, we illustrate the direct policy trade-offs between epidemiological control and economic stability for COVID-19. Furthermore, we demonstrate the framework's generality by extending it to pathogens with different epidemiological profiles, such as polio and influenza, and show how these profiles lead the agent to discover fundamentally different intervention policies. To ground our work in contemporary policymaking challenges, we apply the model to measles outbreaks, quantifying how a modest 5% drop in vaccination coverage necessitates significantly more stringent and costly interventions to curb disease spread. This work provides a robust and adaptable framework to support transparent, evidence-based policymaking for mitigating public health crises.
Authors: Luoma Ke, Guangpeng Zhang, Jibo He, Yajing Li, Yan Li, Xufeng Liu, Peng Fang
Abstract: With the rapid growth of the aviation industry, there is a need for a large number of flight crew. How to select the right pilots in a cost-efficient manner has become an important research question. In the current study, twenty-three pilots were recruited from China Eastern Airlines, and 23 novices were from the community of Tsinghua University. A novel approach incorporating machine learning and virtual reality technology was applied to distinguish features between these participants with different flight skills. Results indicate that SVM with the MIC feature selection method consistently achieved the highest prediction performance on all metrics with an Accuracy of 0.93, an AUC of 0.96, and an F1 of 0.93, which outperforms four other classifier algorithms and two other feature selection methods. From the perspective of feature selection methods, the MIC method can select features with a nonlinear relationship to sampling labels, instead of a simple filter-out. Our new implementation of the SVM + MIC algorithm outperforms all existing pilot selection algorithms and perhaps provides the first implementation based on eye tracking and flight dynamics data. This study's VR simulation platforms and algorithms can be used for pilot selection and training.
Authors: Xiangyu Shi, Marco Chiesa, Gerald Q. Maguire Jr., Dejan Kostic
Abstract: Large Language Models (LLMs) are increasingly deployed in multi-agent systems, where effective inter-model communication is crucial. Existing communication protocols either rely on natural language, incurring high inference costs and information loss, or on hidden states, which suffer from information concentration bias and inefficiency. To address these limitations, we propose KVComm, a novel communication framework that enables efficient communication between LLMs through selective sharing of KV pairs. KVComm leverages the rich information encoded in the KV pairs while avoiding the pitfalls of hidden states. We introduce a KV layer-wise selection strategy based on attention importance scores with a Gaussian prior to identify the most informative KV pairs for communication. Extensive experiments across diverse tasks and model pairs demonstrate that KVComm achieves comparable performance to the upper-bound method, which directly merges inputs to one model without any communication, while transmitting as few as 30\% of layers' KV pairs. Our study highlights the potential of KV pairs as an effective medium for inter-LLM communication, paving the way for scalable and efficient multi-agent systems.
Authors: Michael Chen
Abstract: There is a growing need to evaluate Large Language Models (LLMs) on complex, high-impact, real-world tasks to assess their true readiness as reasoning agents. To address this gap, we introduce AgentCaster, a contamination-free framework employing multimodal LLMs end-to-end for the challenging, long-horizon task of tornado forecasting. Within AgentCaster, models interpret heterogeneous spatiotemporal data from a high-resolution convection-allowing forecast archive. We assess model performance over a 40-day period featuring diverse historical data, spanning several major tornado outbreaks and including over 500 tornado reports. Each day, models query interactively from a pool of 3,625 forecast maps and 40,125 forecast soundings for a forecast horizon of 12-36 hours. Probabilistic tornado-risk polygon predictions are verified against ground truths derived from geometric comparisons across disjoint risk bands in projected coordinate space. To quantify accuracy, we propose domain-specific TornadoBench and TornadoHallucination metrics, with TornadoBench highly challenging for both LLMs and domain expert human forecasters. Notably, human experts significantly outperform state-of-the-art models, which demonstrate a strong tendency to hallucinate and overpredict risk intensity, struggle with precise geographic placement, and exhibit poor spatiotemporal reasoning in complex, dynamically evolving systems. AgentCaster aims to advance research on improving LLM agents for challenging reasoning tasks in critical domains.
Authors: Song Wang, Zhenyu Lei, Zhen Tan, Jundong Li, Javier Rasero, Aiying Zhang, Chirag Agarwal
Abstract: Nearly one in five adolescents currently live with a diagnosed mental or behavioral health condition, such as anxiety, depression, or conduct disorder, underscoring the urgency of developing accurate and interpretable diagnostic tools. Resting-state functional magnetic resonance imaging (rs-fMRI) provides a powerful lens into large-scale functional connectivity, where brain regions are modeled as nodes and inter-regional synchrony as edges, offering clinically relevant biomarkers for psychiatric disorders. While prior works use graph neural network (GNN) approaches for disorder prediction, they remain complex black-boxes, limiting their reliability and clinical translation. In this work, we propose CONCEPTNEURO, a concept-based diagnosis framework that leverages large language models (LLMs) and neurobiological domain knowledge to automatically generate, filter, and encode interpretable functional connectivity concepts. Each concept is represented as a structured subgraph linking specific brain regions, which are then passed through a concept classifier. Our design ensures predictions through clinically meaningful connectivity patterns, enabling both interpretability and strong predictive performance. Extensive experiments across multiple psychiatric disorder datasets demonstrate that CONCEPTNEURO-augmented GNNs consistently outperform their vanilla counterparts, improving accuracy while providing transparent, clinically aligned explanations. Furthermore, concept analyses highlight disorder-specific connectivity patterns that align with expert knowledge and suggest new hypotheses for future investigation, establishing CONCEPTNEURO as an interpretable, domain-informed framework for psychiatric disorder diagnosis.
Authors: Aryan Patel
Abstract: Aluminum is a widely used alloy, which is susceptible to fatigue failure. Characterizing fatigue performance for materials is extremely time and cost demanding, especially for high cycle data. To help mitigate this, a transfer learning based framework has been developed using Long short-term memory networks (LSTMs) in which a source LSTM model is trained based on pure axial fatigue data for Aluminum 7075-T6 alloy which is then transferred to predict high cycle torsional S-N curves. The framework was able to accurately predict Al torsional S-N curves for a much higher cycle range. It is the belief that this framework will help to drastically mitigate the cost of gathering fatigue characteristics for different materials and help prioritize tests with better cost and time constraints.
Authors: Annan Yu, Danielle C. Maddix, Boran Han, Xiyuan Zhang, Abdul Fatir Ansari, Oleksandr Shchur, Christos Faloutsos, Andrew Gordon Wilson, Michael W. Mahoney, Yuyang Wang
Abstract: Transformers are widely used across data modalities, and yet the principles distilled from text models often transfer imperfectly to models trained to other modalities. In this paper, we analyze Transformers through the lens of rank structure. Our focus is on the time series setting, where the structural properties of the data differ remarkably from those of text or vision. We show that time-series embeddings, unlike text or vision, exhibit sharply decaying singular value spectra: small patch sizes and smooth continuous mappings concentrate the data into low-rank subspaces. From this, we prove that the associated $Q/K/V$ projections admit accurate low-rank approximations, and that attention layers become compressible in proportion to the decay of the embedding spectrum. We introduce the concept of flow-of-ranks, a phenomenon by which nonlinear mixing across depth inflates the rank, explaining why early layers are most amenable to compression and why ranks grow with depth. Guided by these theoretical and empirical results, we use these insights to compress Chronos, a large time series foundation model, achieving a reduction of $65\%$ in inference time and $81\%$ in memory, without loss of accuracy. Our findings provide principled guidance for allocating width, depth, and heads in time series foundation models, and for exploiting their inherent compressibility.
Authors: Zelin Zhao, Zongyi Li, Kimia Hassibi, Kamyar Azizzadenesheli, Junchi Yan, H. Jane Bae, Di Zhou, Anima Anandkumar
Abstract: Assessing turbulence control effects for wall friction numerically is a significant challenge since it requires expensive simulations of turbulent fluid dynamics. We instead propose an efficient deep reinforcement learning (RL) framework for modeling and control of turbulent flows. It is model-based RL for predictive control (PC), where both the policy and the observer models for turbulence control are learned jointly using Physics Informed Neural Operators (PINO), which are discretization invariant and can capture fine scales in turbulent flows accurately. Our PINO-PC outperforms prior model-free reinforcement learning methods in various challenging scenarios where the flows are of high Reynolds numbers and unseen, i.e., not provided during model training. We find that PINO-PC achieves a drag reduction of 39.0\% under a bulk-velocity Reynolds number of 15,000, outperforming previous fluid control methods by more than 32\%.
Authors: Lijiao Wang, Muhammad Usama, Haris N. Koutsopoulos, Zhengbing He
Abstract: Open-source data offers a scalable and transparent foundation for estimating vehicle activity and emissions in urban regions. In this study, we propose a data-driven framework that integrates MOVES and open-source GPS trajectory data, OpenStreetMap (OSM) road networks, regional traffic datasets and satellite imagery-derived feature vectors to estimate the link level operating mode distribution and traffic emissions. A neural network model is trained to predict the distribution of MOVES-defined operating modes using only features derived from readily available data. The proposed methodology was applied using open-source data related to 45 municipalities in the Boston Metropolitan area. The "ground truth" operating mode distribution was established using OSM open-source GPS trajectories. Compared to the MOVES baseline, the proposed model reduces RMSE by over 50% for regional scale traffic emissions of key pollutants including CO, NOx, CO2, and PM2.5. This study demonstrates the feasibility of low-cost, replicable, and data-driven emissions estimation using fully open data sources.
Authors: Xiaolong Ma, Xu Dong, Ashley Tarrant, Lei Yang, Rao Kotamarthi, Jiali Wang, Feng Yan, Rajkumar Kettimuthu
Abstract: High-quality observations of hub-height winds are valuable but sparse in space and time. Simulations are widely available on regular grids but are generally biased and too coarse to inform wind-farm siting or to assess extreme-weather-related risks (e.g., gusts) at infrastructure scales. To fully utilize both data types for generating high-quality, high-resolution hub-height wind speeds (tens to ~100m above ground), this study introduces WindSR, a diffusion model with data assimilation for super-resolution downscaling of hub-height winds. WindSR integrates sparse observational data with simulation fields during downscaling using state-of-the-art diffusion models. A dynamic-radius blending method is introduced to merge observations with simulations, providing conditioning for the diffusion process. Terrain information is incorporated during both training and inference to account for its role as a key driver of winds. Evaluated against convolutional-neural-network and generative-adversarial-network baselines, WindSR outperforms them in both downscaling efficiency and accuracy. Our data assimilation reduces WindSR's model bias by approximately 20% relative to independent observations.
Authors: Harshwardhan Fartale, Ashish Kattamuri, Rahul Raja, Arpita Vats, Ishita Prasad, Akshata Kishore Moharir
Abstract: Transformer-based language models excel at both recall (retrieving memorized facts) and reasoning (performing multi-step inference), but whether these abilities rely on distinct internal mechanisms remains unclear. Distinguishing recall from reasoning is crucial for predicting model generalization, designing targeted evaluations, and building safer interventions that affect one ability without disrupting the other.We approach this question through mechanistic interpretability, using controlled datasets of synthetic linguistic puzzles to probe transformer models at the layer, head, and neuron level. Our pipeline combines activation patching and structured ablations to causally measure component contributions to each task type. Across two model families (Qwen and LLaMA), we find that interventions on distinct layers and attention heads lead to selective impairments: disabling identified "recall circuits" reduces fact-retrieval accuracy by up to 15\% while leaving reasoning intact, whereas disabling "reasoning circuits" reduces multi-step inference by a comparable margin. At the neuron level, we observe task-specific firing patterns, though these effects are less robust, consistent with neuronal polysemanticity.Our results provide the first causal evidence that recall and reasoning rely on separable but interacting circuits in transformer models. These findings advance mechanistic interpretability by linking circuit-level structure to functional specialization and demonstrate how controlled datasets and causal interventions can yield mechanistic insights into model cognition, informing safer deployment of large language models.
Authors: Sasho Nedelkoski, Alexander Acker, Odej Kao, Soeren Becker, Dominik Scheinert
Abstract: The training of large models demands substantial computational resources, typically available only in data centers with high-bandwidth interconnects. However, reducing the reliance on high-bandwidth interconnects between nodes enables the use of distributed compute resources as an alternative to centralized data center training. Building on recent advances in distributed model training, we propose an approach that further reduces communication by combining infrequent synchronizations across distributed model replicas with gradient momentum compression. In particular, we treat the optimizer momentum as a signal and decompose the Nesterov momentum into high- and low-frequency components via the discrete cosine transform (DCT). Only the high-frequency components are synchronized across model replicas every $H$ steps. Empirically, our method achieves up to a $16\times$ reduction in communication compared to the baseline DiLoCo, and it generalizes across architectures, including transformer-based language models and convolutional neural networks for images. Overall, this work advances the feasibility of training large models on distributed nodes with low-bandwidth interconnects.
Authors: Renrong Shao, Wei Zhang, Jun wang
Abstract: Data-free knowledge distillation~(DFKD) is an effective manner to solve model compression and transmission restrictions while retaining privacy protection, which has attracted extensive attention in recent years. Currently, the majority of existing methods utilize a generator to synthesize images to support the distillation. Although the current methods have achieved great success, there are still many issues to be explored. Firstly, the outstanding performance of supervised learning in deep learning drives us to explore a pseudo-supervised paradigm on DFKD. Secondly, current synthesized methods cannot distinguish the distributions of different categories of samples, thus producing ambiguous samples that may lead to an incorrect evaluation by the teacher. Besides, current methods cannot optimize the category-wise diversity samples, which will hinder the student model learning from diverse samples and further achieving better performance. In this paper, to address the above limitations, we propose a novel learning paradigm, i.e., conditional pseudo-supervised contrast for data-free knowledge distillation~(CPSC-DFKD). The primary innovations of CPSC-DFKD are: (1) introducing a conditional generative adversarial network to synthesize category-specific diverse images for pseudo-supervised learning, (2) improving the modules of the generator to distinguish the distributions of different categories, and (3) proposing pseudo-supervised contrastive learning based on teacher and student views to enhance diversity. Comprehensive experiments on three commonly-used datasets validate the performance lift of both the student and generator brought by CPSC-DFKD. The code is available at https://github.com/RoryShao/CPSC-DFKD.git
Authors: Michael Ben Ali (IRIT, IRIT-SIG, UT3), Imen Megdiche (IRIT, IRIT-SIG, INUC), Andr\'e Peninou (IRIT, IRIT-SIG, UT2J), Olivier Teste (IRIT-SIG, IRIT, UT2J, Comue de Toulouse)
Abstract: Federated Learning (FL) is a decentralized paradigm that enables a client-server architecture to collaboratively train a global Artificial Intelligence model without sharing raw data, thereby preserving privacy. A key challenge in FL is Non-IID data. Quantity Skew (QS) is a particular problem of Non-IID, where clients hold highly heterogeneous data volumes. Clustered Federated Learning (CFL) is an emergent variant of FL that presents a promising solution to Non-IID problem. It improves models' performance by grouping clients with similar data distributions into clusters. CFL methods generally fall into two operating strategies. In the first strategy, clients select the cluster that minimizes the local training loss. In the second strategy, the server groups clients based on local model similarities. However, most CFL methods lack systematic evaluation under QS but present significant challenges because of it. In this paper, we present two main contributions. The first one is an evaluation of state-of-the-art CFL algorithms under various Non-IID settings, applying multiple QS scenarios to assess their robustness. Our second contribution is a novel iterative CFL algorithm, named CORNFLQS, which proposes an optimal coordination between both operating strategies of CFL. Our approach is robust against the different variations of QS settings. We conducted intensive experiments on six image classification datasets, resulting in 270 Non-IID configurations. The results show that CORNFLQS achieves the highest average ranking in both accuracy and clustering quality, as well as strong robustness to QS perturbations. Overall, our approach outperforms actual CFL algorithms.
Authors: Yongchao Li, Jun Chen, Zhuoxuan Li, Chao Gao, Yang Li, Chu Zhang, Changyin Dong
Abstract: Interchanges are crucial nodes for vehicle transfers between highways, yet the lack of real-time ramp detectors creates blind spots in traffic prediction. To address this, we propose a Spatio-Temporal Decoupled Autoencoder (STDAE), a two-stage framework that leverages cross-modal reconstruction pretraining. In the first stage, STDAE reconstructs historical ramp flows from mainline data, forcing the model to capture intrinsic spatio-temporal relations. Its decoupled architecture with parallel spatial and temporal autoencoders efficiently extracts heterogeneous features. In the prediction stage, the learned representations are integrated with models such as GWNet to enhance accuracy. Experiments on three real-world interchange datasets show that STDAE-GWNET consistently outperforms thirteen state-of-the-art baselines and achieves performance comparable to models using historical ramp data. This demonstrates its effectiveness in overcoming detector scarcity and its plug-and-play potential for diverse forecasting pipelines.
Authors: Donghwan Rho
Abstract: Reinforcement learning with verifiable rewards (RLVR) is a promising approach for training large language models (LLMs) with stronger reasoning abilities. It has also been applied to a variety of logic puzzles. In this work, we study the Korean word-chain game using RLVR. We show that rule-derived rewards can naturally conflict, and demonstrate through experiments that a curriculum-learning scheme mitigates these conflicts. Our findings motivate further studies of puzzle tasks in diverse languages.
Authors: Ashley Lenau, Dennis Dimiduk, Stephen R. Niezgoda
Abstract: A successful deep learning network is highly dependent not only on the training dataset, but the training algorithm used to condition the network for a given task. The loss function, dataset, and tuning of hyperparameters all play an essential role in training a network, yet there is not much discussion on the reliability or reproducibility of a training algorithm. With the rise in popularity of physics-informed loss functions, this raises the question of how reliable one's loss function is in conditioning a network to enforce a particular boundary condition. Reporting the model variation is needed to assess a loss function's ability to consistently train a network to obey a given boundary condition, and provides a fairer comparison among different methods. In this work, a Pix2Pix network predicting the stress fields of high elastic contrast composites is used as a case study. Several different loss functions enforcing stress equilibrium are implemented, with each displaying different levels of variation in convergence, accuracy, and enforcing stress equilibrium across many training sessions. Suggested practices in reporting model variation are also shared.
Authors: Joseph Rawson, Domniki Ladopoulou, Petros Dellaportas
Abstract: Uncertainty-aware wind power prediction is essential for grid integration and reliable wind farm operation. We apply neural diffusion processes (NDPs)-a recent class of models that learn distributions over functions-and extend them to a multi-task NDP (MT-NDP) framework for wind power prediction. We provide the first empirical evaluation of NDPs in real supervisory control and data acquisition (SCADA) data. We introduce a task encoder within MT-NDPs to capture cross-turbine correlations and enable few-shot adaptation to unseen turbines. The proposed MT-NDP framework outperforms single-task NDPs and GPs in terms of point accuracy and calibration, particularly for wind turbines whose behaviour deviates from the fleet average. In general, NDP-based models deliver calibrated and scalable predictions suitable for operational deployment, offering sharper, yet trustworthy, predictive intervals that can support dispatch and maintenance decisions in modern wind farms.
Authors: Congzheng Song, Xinyu Tang
Abstract: Fine-tuning large language models (LLMs) with backpropagation\textemdash even for a subset of parameters such as LoRA\textemdash can be much more memory-consuming than inference and is often deemed impractical for resource-constrained mobile devices. Alternative methods, such as zeroth-order optimization (ZO), can greatly reduce the memory footprint but come at the cost of significantly slower model convergence (10$\times$ to 100$\times$ more steps than backpropagation). We propose a memory-efficient implementation of backpropagation (MeBP) on mobile devices that provides better trade-off between memory usage and compute time, while converging faster and achieving better performance than the ZO baseline. We verify the effectiveness of MeBP on an iPhone 15 Pro Max and show that various LLMs, ranging from 0.5B to 4B parameters, can be fine-tuned using less than 1GB of memory. We release an example of the MeBP implementation at https://github.com/apple/ml-mebp.
Authors: Franz A. Heinsen, Leo Kozachkov
Abstract: Many domains, from deep learning to finance, require compounding real numbers over long sequences, often leading to catastrophic numerical underflow or overflow. We introduce generalized orders of magnitude (GOOMs), a principled extension of traditional orders of magnitude that incorporates floating-point numbers as a special case, and which in practice enables stable computation over significantly larger dynamic ranges of real numbers than previously possible. We implement GOOMs, along with an efficient custom parallel prefix scan, to support native execution on parallel hardware such as GPUs. We demonstrate that our implementation of GOOMs outperforms traditional approaches with three representative experiments, all of which were previously considered impractical or impossible, and now become possible and practical: (1) compounding real matrix products far beyond standard floating-point limits; (2) estimating spectra of Lyapunov exponents in parallel, orders of magnitude faster than with previous methods, applying a novel selective-resetting method to prevent state colinearity; and (3) capturing long-range dependencies in deep recurrent neural networks with non-diagonal recurrent states, computed in parallel via a prefix scan, without requiring any form of stabilization. Our results show that our implementation of GOOMs, combined with efficient parallel scanning, offers a scalable and numerically robust alternative to conventional floating-point numbers for high-dynamic-range applications.
Authors: Jiajun Shen, Yufei Jin, Yi He, Xingquan Zhu
Abstract: Learning from large heterogeneous graphs presents significant challenges due to the scale of networks, heterogeneity in node and edge types, variations in nodal features, and complex local neighborhood structures. This paper advocates for ensemble learning as a natural solution to this problem, whereby training multiple graph learners under distinct sampling conditions, the ensemble inherently captures different aspects of graph heterogeneity. Yet, the crux lies in combining these learners to meet global optimization objective while maintaining computational efficiency on large-scale graphs. In response, we propose LHGEL, an ensemble framework that addresses these challenges through batch sampling with three key components, namely batch view aggregation, residual attention, and diversity regularization. Specifically, batch view aggregation samples subgraphs and forms multiple graph views, while residual attention adaptively weights the contributions of these views to guide node embeddings toward informative subgraphs, thereby improving the accuracy of base learners. Diversity regularization encourages representational disparity across embedding matrices derived from different views, promoting model diversity and ensemble robustness. Our theoretical study demonstrates that residual attention mitigates gradient vanishing issues commonly faced in ensemble learning. Empirical results on five real heterogeneous networks validate that our LHGEL approach consistently outperforms its state-of-the-art competitors by substantial margin. Codes and datasets are available at https://github.com/Chrisshen12/LHGEL.
Authors: Jairo Diaz-Rodriguez, Mumin Jia
Abstract: Kernel change-point detection (KCPD) has become a widely used tool for identifying structural changes in complex data. While existing theory establishes consistency under independence assumptions, real-world sequential data such as text exhibits strong dependencies. We establish new guarantees for KCPD under $m$-dependent data: specifically, we prove consistency in the number of detected change points and weak consistency in their locations under mild additional assumptions. We perform an LLM-based simulation that generates synthetic $m$-dependent text to validate the asymptotics. To complement these results, we present the first comprehensive empirical study of KCPD for text segmentation with modern embeddings. Across diverse text datasets, KCPD with text embeddings outperforms baselines in standard text segmentation metrics. We demonstrate through a case study on Taylor Swift's tweets that KCPD not only provides strong theoretical and simulated reliability but also practical effectiveness for text segmentation tasks.
Authors: Ege Cakar, Per Ola Kristensson
Abstract: Humans are black boxes -- we cannot observe their neural processes, yet society functions by evaluating verifiable arguments. AI explainability should follow this principle: stakeholders need verifiable reasoning chains, not mechanistic transparency. We propose using structured argumentation to provide a level of explanation and verification neither interpretability nor LLM-generated explanation is able to offer. Our pipeline achieves state-of-the-art 94.44 macro F1 on the AAEC published train/test split (5.7 points above prior work) and $0.81$ macro F1, $\sim$0.07 above previous published results with comparable data setups, for Argumentative MicroTexts relation classification, converting LLM text into argument graphs and enabling verification at each inferential step. We demonstrate this idea on multi-agent risk assessment using the Structured What-If Technique, where specialized agents collaborate transparently to carry out risk assessment otherwise achieved by humans alone. Using Bipolar Assumption-Based Argumentation, we capture support/attack relationships, thereby enabling automatic hallucination detection via fact nodes attacking arguments. We also provide a verification mechanism that enables iterative refinement through test-time feedback without retraining. For easy deployment, we provide a Docker container for the fine-tuned AMT model, and the rest of the code with the Bipolar ABA Python package on GitHub.
Authors: Benoit Dherin, Michael Munn
Abstract: Deep residual architectures, such as ResNet and the Transformer, have enabled models of unprecedented depth, yet a formal understanding of why depth is so effective remains an open question. A popular intuition, following Veit et al. (2016), is that these residual networks behave like ensembles of many shallower models. Our key finding is an explicit analytical formula that verifies this ensemble perspective, proving that increasing network depth is mathematically equivalent to expanding the size of this implicit ensemble. Furthermore, our expansion reveals a hierarchical ensemble structure in which the combinatorial growth of computation paths leads to an explosion in the output signal, explaining the historical necessity of normalization layers in training deep models. This insight offers a first principles explanation for the historical dependence on normalization layers and sheds new light on a family of successful normalization-free techniques like SkipInit and Fixup. However, while these previous approaches infer scaling factors through optimizer analysis or a heuristic analogy to Batch Normalization, our work offers the first explanation derived directly from the network's inherent functional structure. Specifically, our Residual Expansion Theorem reveals that scaling each residual module provides a principled solution to taming the combinatorial explosion inherent to these architectures. We further show that this scaling acts as a capacity controls that also implicitly regularizes the model's complexity.
Authors: Quan Nguyen
Abstract: While Adam is one of the most effective optimizer for training large-scale machine learning models, a theoretical understanding of how to optimally set its momentum factors, $\beta_1$ and $\beta_2$, remains largely incomplete. Prior works have shown that Adam can be seen as an instance of Follow-the-Regularized-Leader (FTRL), one of the most important class of algorithms in online learning. The prior analyses in these works required setting $\beta_1 = \sqrt{\beta_2}$, which does not cover the more practical cases with $\beta_1 \neq \sqrt{\beta_2}$. We derive novel, more general analyses that hold for both $\beta_1 \geq \sqrt{\beta_2}$ and $\beta_1 \leq \sqrt{\beta_2}$. In both cases, our results strictly generalize the existing bounds. Furthermore, we show that our bounds are tight in the worst case. We also prove that setting $\beta_1 = \sqrt{\beta_2}$ is optimal for an oblivious adversary, but sub-optimal for an non-oblivious adversary.
Authors: Anupam Panwar, Himadri Pal, Jiali Chen, Kyle Cho, Riddick Jiang, Miao Zhao, Rajiv Krishnamurthy
Abstract: Detecting anomalies in large, distributed systems presents several challenges. The first challenge arises from the sheer volume of data that needs to be processed. Flagging anomalies in a high-throughput environment calls for a careful consideration of both algorithm and system design. The second challenge comes from the heterogeneity of time-series datasets that leverage such a system in production. In practice, anomaly detection systems are rarely deployed for a single use case. Typically, there are several metrics to monitor, often across several domains (e.g. engineering, business and operations). A one-size-fits-all approach rarely works, so these systems need to be fine-tuned for every application - this is often done manually. The third challenge comes from the fact that determining the root-cause of anomalies in such settings is akin to finding a needle in a haystack. Identifying (in real time) a time-series dataset that is associated causally with the anomalous time-series data is a very difficult problem. In this paper, we describe a unified framework that addresses these challenges. Reasoning based Anomaly Detection Framework (RADF) is designed to perform real time anomaly detection on very large datasets. This framework employs a novel technique (mSelect) that automates the process of algorithm selection and hyper-parameter tuning for each use case. Finally, it incorporates a post-detection capability that allows for faster triaging and root-cause determination. Our extensive experiments demonstrate that RADF, powered by mSelect, surpasses state-of-the-art anomaly detection models in AUC performance for 5 out of 9 public benchmarking datasets. RADF achieved an AUC of over 0.85 for 7 out of 9 datasets, a distinction unmatched by any other state-of-the-art model.
Authors: Volodymyr Tkachuk, Csaba Szepesv\'ari, Xiaoqi Tan
Abstract: We study finite-horizon offline reinforcement learning (RL) with function approximation for both policy evaluation and policy optimization. Prior work established that statistically efficient learning is impossible for either of these problems when the only assumptions are that the data has good coverage (concentrability) and the state-action value function of every policy is linearly realizable ($q^\pi$-realizability) (Foster et al., 2021). Recently, Tkachuk et al. (2024) gave a statistically efficient learner for policy optimization, if in addition the data is assumed to be given as trajectories. In this work we present a statistically efficient learner for policy evaluation under the same assumptions. Further, we show that the sample complexity of the learner used by Tkachuk et al. (2024) for policy optimization can be improved by a tighter analysis.
Authors: Lunjun Zhang, Shuo Han, Hanrui Lyu, Bradly C Stadie
Abstract: We introduce D2AC, a new model-free reinforcement learning (RL) algorithm designed to train expressive diffusion policies online effectively. At its core is a policy improvement objective that avoids the high variance of typical policy gradients and the complexity of backpropagation through time. This stable learning process is critically enabled by our second contribution: a robust distributional critic, which we design through a fusion of distributional RL and clipped double Q-learning. The resulting algorithm is highly effective, achieving state-of-the-art performance on a benchmark of eighteen hard RL tasks, including Humanoid, Dog, and Shadow Hand domains, spanning both dense-reward and goal-conditioned RL scenarios. Beyond standard benchmarks, we also evaluate a biologically motivated predator-prey task to examine the behavioral robustness and generalization capacity of our approach.
Authors: Kristi Topollai, Anna Choromanska
Abstract: Few-shot classification and meta-learning methods typically struggle to generalize across diverse domains, as most approaches focus on a single dataset, failing to transfer knowledge across various seen and unseen domains. Existing solutions often suffer from low accuracy, high computational costs, and rely on restrictive assumptions. In this paper, we introduce the notion of task-level contrastiveness, a novel approach designed to address issues of existing methods. We start by introducing simple ways to define task augmentations, and thereafter define a task-level contrastive loss that encourages unsupervised clustering of task representations. Our method is lightweight and can be easily integrated within existing few-shot/meta-learning algorithms while providing significant benefits. Crucially, it leads to improved generalization and computational efficiency without requiring prior knowledge of task domains. We demonstrate the effectiveness of our approach through different experiments on the MetaDataset benchmark, where it achieves superior performance without additional complexity.
Authors: Taha M. Mahmoud, Naima Kaabouch
Abstract: The rapid growth of the Internet of Things (IoT) has expanded opportunities for innovation but also increased exposure to botnet-driven cyberattacks. Conventional detection methods often struggle with scalability, privacy, and adaptability in resource-constrained IoT environments. To address these challenges, we present a lightweight and privacy-preserving botnet detection framework based on federated learning. This approach enables distributed devices to collaboratively train models without exchanging raw data, thus maintaining user privacy while preserving detection accuracy. A communication-efficient aggregation strategy is introduced to reduce overhead, ensuring suitability for constrained IoT networks. Experiments on benchmark IoT botnet datasets demonstrate that the framework achieves high detection accuracy while substantially reducing communication costs. These findings highlight federated learning as a practical path toward scalable, secure, and privacy-aware intrusion detection for IoT ecosystems.
Authors: Lianghuan Huang, Sagnik Anupam, Insup Lee, Shuo Li, Osbert Bastani
Abstract: Reinforcement learning (RL) has emerged as a promising strategy for finetuning small language models (SLMs) to solve targeted tasks such as math and coding. However, RL algorithms tend to be resource-intensive, taking a significant amount of time to train. We propose RAPID, a novel RL algorithm that can substantially reduce the running time of RL. Our key insight is that RL tends to be costly due to the need to perform both inference and backpropagation during training. To maximize use of computational resources, our algorithm performs inference in large batches, and then performs off-policy policy gradient updates in mini-batches. For off-policy updates, we incorporate group advantage estimation into the policy gradient algorithm, and derive an importance weighted estimator to correct for the bias arising from off-policy learning. Our experiments demonstrate that our algorithm can reduce running time by 11%-34% on three benchmarks compared to state-of-the-art RL algorithms while maintaining similar or better accuracy.
Authors: Kartik Pandit, Sourav Ganguly, Arnesh Banerjee, Shaahin Angizi, Arnob Ghosh
Abstract: Ensuring safety is a foundational requirement for large language models (LLMs). Achieving an appropriate balance between enhancing the utility of model outputs and mitigating their potential for harm is a complex and persistent challenge. Contemporary approaches frequently formalize this problem within the framework of Constrained Markov Decision Processes (CMDPs) and employ established CMDP optimization techniques. However, these methods exhibit two notable limitations. First, their reliance on reward and cost functions renders performance highly sensitive to the underlying scoring mechanism, which must capture semantic meaning rather than being triggered by superficial keywords. Second, CMDP-based training entails tuning dual-variable, a process that is both computationally expensive and does not provide any provable safety guarantee for a fixed dual variable that can be exploitable through adversarial jailbreaks. To overcome these limitations, we introduce Certifiable Safe-RLHF (CS-RLHF) that introduces a cost model trained on a large-scale corpus to assign semantically grounded safety scores. In contrast to the lagrangian-based approach, CS-RLHF adopts a rectified penalty-based formulation. This design draws on the theory of exact penalty functions in constrained optimization, wherein constraint satisfaction is enforced directly through a suitably chosen penalty term. With an appropriately scaled penalty, feasibility of the safety constraints can be guaranteed at the optimizer, eliminating the need for dual-variable updates. Empirical evaluation demonstrates that CS-RLHF outperforms state-of-the-art LLM model responses rendering at-least 5 times efficient against nominal and jail-breaking prompts
Authors: William Anderson, Seung Whan Chung, Youngsoo Choi
Abstract: Accurate numerical solutions of partial differential equations are essential in many scientific fields but often require computationally expensive solvers, motivating reduced-order models (ROMs). Latent Space Dynamics Identification (LaSDI) is a data-driven ROM framework that combines autoencoders with equation discovery to learn interpretable latent dynamics. However, enforcing latent dynamics during training can compromise reconstruction accuracy of the model for simulation data. We introduce multi-stage LaSDI (mLaSDI), a framework that improves reconstruction and prediction accuracy by sequentially learning additional decoders to correct residual errors from previous stages. Applied to the 1D-1V Vlasov equation, mLaSDI consistently outperforms standard LaSDI, achieving lower prediction errors and reduced training time across a wide range of architectures.
Authors: Ashwin Prabu, Nhat Thanh Tran, Guofa Zhou, Jack Xin
Abstract: A variety of models have been developed to forecast dengue cases to date. However, it remains a challenge to predict major dengue outbreaks that need timely public warnings the most. In this paper, we introduce CrossLag, an environmentally informed attention that allows for the incorporation of lagging endogenous signals behind the significant events in the exogenous data into the architecture of the transformer at low parameter counts. Outbreaks typically lag behind major changes in climate and oceanic anomalies. We use TimeXer, a recent general-purpose transformer distinguishing exogenous-endogenous inputs, as the baseline for this study. Our proposed model outperforms TimeXer by a considerable margin in detecting and predicting major outbreaks in Singapore dengue data over a 24-week prediction window.
Authors: Fatmazohra Rezkellah, Ramzi Dakhmouche
Abstract: With the increasing adoption of Large Language Models (LLMs), more customization is needed to ensure privacy-preserving and safe generation. We address this objective from two critical aspects: unlearning of sensitive information and robustness to jail-breaking attacks. We investigate various constrained optimization formulations that address both aspects in a \emph{unified manner}, by finding the smallest possible interventions on LLM weights that either make a given vocabulary set unreachable or embed the LLM with robustness to tailored attacks by shifting part of the weights to a \emph{safer} region. Beyond unifying two key properties, this approach contrasts with previous work in that it doesn't require an oracle classifier that is typically not available or represents a computational overhead. Surprisingly, we find that the simplest point-wise constraint-based intervention we propose leads to better performance than max-min interventions, while having a lower computational cost. Comparison against state-of-the-art defense methods demonstrates superior performance of the proposed approach.
Authors: Mohammad Mohaiminul Islam, Thijs P. Kuipers, Sharvaree Vadgama, Coen de Vente, Afsana Khan, Clara I. S\'anchez, Erik J. Bekkers
Abstract: Generative models for sequential data often struggle with sparsely sampled and high-dimensional trajectories, typically reducing the learning of dynamics to pairwise transitions. We propose \textit{Interpolative Multi-Marginal Flow Matching} (IMMFM), a framework that learns continuous stochastic dynamics jointly consistent with multiple observed time points. IMMFM employs a piecewise-quadratic interpolation path as a smooth target for flow matching and jointly optimizes drift and a data-driven diffusion coefficient, supported by a theoretical condition for stable learning. This design captures intrinsic stochasticity, handles irregular sparse sampling, and yields subject-specific trajectories. Experiments on synthetic benchmarks and real-world longitudinal neuroimaging datasets show that IMMFM outperforms existing methods in both forecasting accuracy and further downstream tasks.
Authors: Burak Karabulut, Carlo Manna, Chris Develder
Abstract: Fault detection in power distribution grids is critical for ensuring system reliability and preventing costly outages. Moreover, fault detection methodologies should remain robust to evolving grid topologies caused by factors such as reconfigurations, equipment failures, and Distributed Energy Resource (DER) integration. Current data-driven state-of-the-art methods use Recurrent Neural Networks (RNNs) for temporal modeling and Graph Neural Networks (GNNs) for spatial learning, in an RNN+GNN pipeline setting (RGNN in short). Specifically, for power system fault diagnosis, Graph Convolutional Networks (GCNs) have been adopted. Yet, various more advanced GNN architectures have been proposed and adopted in domains outside of power systems. In this paper, we set out to systematically and consistently benchmark various GNN architectures in an RNN+GNN pipeline model. Specifically, to the best of our knowledge, we are the first to (i) propose to use GraphSAGE and Graph Attention (GAT, GATv2) in an RGNN for fault diagnosis, and (ii) provide a comprehensive benchmark against earlier proposed RGNN solutions (RGCN) as well as pure RNN models (especially Gated Recurrent Unit (GRU)), particularly (iii) exploring their generalization potential for deployment in different settings than those used for training them. Our experimental results on the IEEE 123-node distribution network show that RGATv2 has superior generalization capabilities, maintaining high performance with an F1-score reduction of $\sim$12% across different topology settings. In contrast, pure RNN models largely fail, experiencing an F1-score reduction of up to $\sim$60%, while other RGNN variants also exhibit significant performance degradation, i.e., up to $\sim$25% lower F1-scores.
Authors: Mehmet Onurcan Kaya, Desmond Elliott, Dim P. Papadopoulos
Abstract: Small Vision-Language Models (VLMs) provide a computationally efficient alternative to larger models, at the cost of weaker generalization abilities and downstream task performance. These shortcomings could be addressed by test-time scaling techniques, but existing methods are typically computationally demanding, contradicting the resource-efficient design goals of small models. To address these limitations, we propose two novel and efficient test-time scaling strategies that leverage the model-internal features rather than external supervision: (i) Test-Time Augmentation (TTAug), which generates multiple augmented inputs and aggregates outputs at the token level without parameter updates, and (ii) Test-Time Adaptation (TTAdapt), which adapts model parameters during inference using consensus-based pseudolabels from TTAug. Through extensive experiments across nine benchmarks, we demonstrate consistent performance improvements while maintaining computational efficiency suitable for resource-constrained environments. The generality of our approach is demonstrated both within models at different scales and across different VLMs without additional tuning.
Authors: Bongseok Kim, Jiahao Zhang, Guang Lin
Abstract: Deep learning has gained attention for solving PDEs, but the black-box nature of neural networks hinders precise enforcement of boundary conditions. To address this, we propose a boundary condition-guaranteed evolutionary Kolmogorov-Arnold Network (KAN) with radial basis functions (BEKAN). In BEKAN, we propose three distinct and combinable approaches for incorporating Dirichlet, periodic, and Neumann boundary conditions into the network. For Dirichlet problem, we use smooth and global Gaussian RBFs to construct univariate basis functions for approximating the solution and to encode boundary information at the activation level of the network. To handle periodic problems, we employ a periodic layer constructed from a set of sinusoidal functions to enforce the boundary conditions exactly. For a Neumann problem, we devise a least-squares formulation to guide the parameter evolution toward satisfying the Neumann condition. By virtue of the boundary-embedded RBFs, the periodic layer, and the evolutionary framework, we can perform accurate PDE simulations while rigorously enforcing boundary conditions. For demonstration, we conducted extensive numerical experiments on Dirichlet, Neumann, periodic, and mixed boundary value problems. The results indicate that BEKAN outperforms both multilayer perceptron (MLP) and B-splines KAN in terms of accuracy. In conclusion, the proposed approach enhances the capability of KANs in solving PDE problems while satisfying boundary conditions, thereby facilitating advancements in scientific computing and engineering applications.
Authors: Haoran Li, Chenhan Xiao, Muhao Guo, Yang Weng
Abstract: Learning dynamics is essential for model-based control and Reinforcement Learning in engineering systems, such as robotics and power systems. However, limited system measurements, such as those from low-resolution sensors, demand sample-efficient learning. Symmetry provides a powerful inductive bias by characterizing equivariant relations in system states to improve sample efficiency. While recent methods attempt to discover symmetries from data, they typically assume a single global symmetry group and treat symmetry discovery and dynamic learning as separate tasks, leading to limited expressiveness and error accumulation. In this paper, we propose the Latent Mixture of Symmetries (Latent MoS), an expressive model that captures a mixture of symmetry-governed latent factors from complex dynamical measurements. Latent MoS focuses on dynamic learning while locally and provably preserving the underlying symmetric transformations. To further capture long-term equivariance, we introduce a hierarchical architecture that stacks MoS blocks. Numerical experiments in diverse physical systems demonstrate that Latent MoS outperforms state-of-the-art baselines in interpolation and extrapolation tasks while offering interpretable latent representations suitable for future geometric and safety-critical analyses.
Authors: Ankit Bhardwaj, Ananth Balashankar, Lakshminarayanan Subramanian
Abstract: Spatio-temporal sensor data is often sparse, noisy, and irregular, and existing interpolation or learning methods struggle here because they either ignore governing PDEs or do not scale. We introduce FieldFormer, a transformer-based framework for mesh-free spatio-temporal field reconstruction that combines data-driven flexibility with physics-based structure. For each query, FieldFormer gathers a local neighborhood using a learnable velocity-scaled distance metric, enabling anisotropic adaptation to different propagation regimes. Neighborhoods are built efficiently via per-batch offset recomputation, and refined in an expectation-maximization style as the velocity scales evolve. Predictions are made by a local transformer encoder, and physics consistency is enforced through autograd-based PDE residuals and boundary-specific penalties. Across three benchmarks--a scalar anisotropic heat equation, a vector-valued shallow-water system, and a realistic advection-diffusion pollution simulation--FieldFormer consistently outperforms strong baselines by more than 40%. Our results demonstrate that FieldFormer enables accurate (RMSE$<10^{-2}$), efficient, and physically consistent field reconstruction from sparse (0.4%-2%) and noisy(10%) data.
Authors: Kehinde O. Aina, Sehoon Ha
Abstract: We address the challenge of coordinating multiple robots in narrow and confined environments, where congestion and interference often hinder collective task performance. Drawing inspiration from insect colonies, which achieve robust coordination through stigmergy -- modifying and interpreting environmental traces -- we propose a Stigmergic Multi-Agent Deep Reinforcement Learning (S-MADRL) framework that leverages virtual pheromones to model local and social interactions, enabling decentralized emergent coordination without explicit communication. To overcome the convergence and scalability limitations of existing algorithms such as MADQN, MADDPG, and MAPPO, we leverage curriculum learning, which decomposes complex tasks into progressively harder sub-problems. Simulation results show that our framework achieves the most effective coordination of up to eight agents, where robots self-organize into asymmetric workload distributions that reduce congestion and modulate group performance. This emergent behavior, analogous to strategies observed in nature, demonstrates a scalable solution for decentralized multi-agent coordination in crowded environments with communication constraints.
Authors: Wei-Lung Mao, Chun-Chi Wang, Po-Heng Chou, Kai-Chun Liu, Yu Tsao
Abstract: The rising aging population has increased the importance of fall detection (FD) systems as an assistive technology, where deep learning techniques are widely applied to enhance accuracy. FD systems typically use edge devices (EDs) worn by individuals to collect real-time data, which are transmitted to a cloud center (CC) or processed locally. However, this architecture faces challenges such as a limited ED model size and data transmission latency to the CC. Mobile edge computing (MEC), which allows computations at MEC servers deployed between EDs and CC, has been explored to address these challenges. We propose a multilayer MEC (MLMEC) framework to balance accuracy and latency. The MLMEC splits the architecture into stations, each with a neural network model. If front-end equipment cannot detect falls reliably, data are transmitted to a station with more robust back-end computing. The knowledge distillation (KD) approach was employed to improve front-end detection accuracy by allowing high-power back-end stations to provide additional learning experiences, enhancing precision while reducing latency and processing loads. Simulation results demonstrate that the KD approach improved accuracy by 11.65% on the SisFall dataset and 2.78% on the FallAllD dataset. The MLMEC with KD also reduced the data latency rate by 54.15% on the FallAllD dataset and 46.67% on the SisFall dataset compared to the MLMEC without KD. In summary, the MLMEC FD system exhibits improved accuracy and reduced latency.
Authors: Yucheng Wang, Mohamed Ragab, Yubo Hou, Zhenghua Chen, Min Wu, Xiaoli Li
Abstract: Remaining Useful Life (RUL) prediction for turbofan engines plays a vital role in predictive maintenance, ensuring operational safety and efficiency in aviation. Although data-driven approaches using machine learning and deep learning have shown potential, they face challenges such as limited data and distribution shifts caused by varying operating conditions. Domain Adaptation (DA) has emerged as a promising solution, enabling knowledge transfer from source domains with abundant data to target domains with scarce data while mitigating distributional shifts. Given the unique properties of turbofan engines, such as complex operating conditions, high-dimensional sensor data, and slower-changing signals, it is essential to conduct a focused review of DA techniques specifically tailored to turbofan engines. To address this need, this paper provides a comprehensive review of DA solutions for turbofan engine RUL prediction, analyzing key methodologies, challenges, and recent advancements. A novel taxonomy tailored to turbofan engines is introduced, organizing approaches into methodology-based (how DA is applied), alignment-based (where distributional shifts occur due to operational variations), and problem-based (why certain adaptations are needed to address specific challenges). This taxonomy offers a multidimensional view that goes beyond traditional classifications by accounting for the distinctive characteristics of turbofan engine data and the standard process of applying DA techniques to this area. Additionally, we evaluate selected DA techniques on turbofan engine datasets, providing practical insights for practitioners and identifying key challenges. Future research directions are identified to guide the development of more effective DA techniques, advancing the state of RUL prediction for turbofan engines.
Authors: Meenakshi Manikandan, Leilani Gilpin
Abstract: This paper introduces Hill-ADAM. Hill-ADAM is an optimizer with its focus towards escaping local minima in prescribed loss landscapes to find the global minimum. Hill-ADAM escapes minima by deterministically exploring the state space. This eliminates uncertainty from random gradient updates in stochastic algorithms while seldom converging at the first minimum that visits. In the paper we first derive an analytical approximation of the ADAM Optimizer step size at a particular model state. From there define the primary condition determining ADAM limitations in escaping local minima. The proposed optimizer algorithm Hill-ADAM alternates between error minimization and maximization. It maximizes to escape the local minimum and minimizes again afterward. This alternation provides an overall exploration throughout the loss space. This allows the deduction of the global minimum's state. Hill-ADAM was tested with 5 loss functions and 12 amber-saturated to cooler-shade image color correction instances.
Authors: Christopher Solinas, Radovan Haluska, David Sychrovsky, Finbarr Timbers, Nolan Bard, Michael Buro, Martin Schmid, Nathan R. Sturtevant, Michael Bowling
Abstract: We present Neural Bayesian Filtering (NBF), an algorithm for maintaining distributions over hidden states, called beliefs, in partially observable systems. NBF is trained to find a good latent representation of the beliefs induced by a task. It maps beliefs to fixed-length embedding vectors, which condition generative models for sampling. During filtering, particle-style updates compute posteriors in this embedding space using incoming observations and the environment's dynamics. NBF combines the computational efficiency of classical filters with the expressiveness of deep generative models - tracking rapidly shifting, multimodal beliefs while mitigating the risk of particle impoverishment. We validate NBF in state estimation tasks in three partially observable environments.
Authors: An Vuong, Susan Gauch
Abstract: Accurately predicting short-term stock price movement remains a challenging task due to the market's inherent volatility and sensitivity to investor sentiment. This paper discusses a deep learning framework that integrates emotion features extracted from tweet data with historical stock price information to forecast significant price changes on the following day. We utilize Meta's Llama 3.1-8B-Instruct model to preprocess tweet data, thereby enhancing the quality of emotion features derived from three emotion analysis approaches: a transformer-based DistilRoBERTa classifier from the Hugging Face library and two lexicon-based methods using National Research Council Canada (NRC) resources. These features are combined with previous-day stock price data to train a Long Short-Term Memory (LSTM) model. Experimental results on TSLA, AAPL, and AMZN stocks show that all three emotion analysis methods improve the average accuracy for predicting significant price movements, compared to the baseline model using only historical stock prices, which yields an accuracy of 13.5%. The DistilRoBERTa-based stock prediction model achieves the best performance, with accuracy rising from 23.6% to 38.5% when using LLaMA-enhanced emotion analysis. These results demonstrate that using large language models to preprocess tweet content enhances the effectiveness of emotion analysis which in turn improves the accuracy of predicting significant stock price movements.
Authors: Rabeya Amin Jhuma, Mostafa Mohaimen Akand Faisal
Abstract: This study explored how in-context learning (ICL) in large language models can be disrupted by data poisoning attacks in the setting of public health sentiment analysis. Using tweets of Human Metapneumovirus (HMPV), small adversarial perturbations such as synonym replacement, negation insertion, and randomized perturbation were introduced into the support examples. Even these minor manipulations caused major disruptions, with sentiment labels flipping in up to 67% of cases. To address this, a Spectral Signature Defense was applied, which filtered out poisoned examples while keeping the data's meaning and sentiment intact. After defense, ICL accuracy remained steady at around 46.7%, and logistic regression validation reached 100% accuracy, showing that the defense successfully preserved the dataset's integrity. Overall, the findings extend prior theoretical studies of ICL poisoning to a practical, high-stakes setting in public health discourse analysis, highlighting both the risks and potential defenses for robust LLM deployment. This study also highlights the fragility of ICL under attack and the value of spectral defenses in making AI systems more reliable for health-related social media monitoring.
Authors: Jialin Liu, Lisang Ding, Stanley Osher, Wotao Yin
Abstract: Implicit models, an emerging model class, compute outputs by iterating a single parameter block to a fixed point. This architecture realizes an infinite-depth, weight-tied network that trains with constant memory, significantly reducing memory needs for the same level of performance compared to explicit models. While it is empirically known that these compact models can often match or even exceed larger explicit networks by allocating more test-time compute, the underlying mechanism remains poorly understood. We study this gap through a nonparametric analysis of expressive power. We provide a strict mathematical characterization, showing that a simple and regular implicit operator can, through iteration, progressively express more complex mappings. We prove that for a broad class of implicit models, this process lets the model's expressive power scale with test-time compute, ultimately matching a much richer function class. The theory is validated across three domains: image reconstruction, scientific computing, and operations research, demonstrating that as test-time iterations increase, the complexity of the learned mapping rises, while the solution quality simultaneously improves and stabilizes.
Authors: Nicholas Carter, Arkaprava Gupta, Prateek Ganguli, Benedikt Dietrich, Vibhor Krishna, Samarjit Chakraborty
Abstract: Deep Brain Stimulation (DBS) is a highly effective treatment for Parkinson's Disease (PD). Recent research uses reinforcement learning (RL) for DBS, with RL agents modulating the stimulation frequency and amplitude. But, these models rely on biomarkers that are not measurable in patients and are only present in brain-on-chip (BoC) simulations. In this work, we present an RL-based DBS approach that adapts these stimulation parameters according to brain activity measurable in vivo. Using a TD3 based RL agent trained on a model of the basal ganglia region of the brain, we see a greater suppression of biomarkers correlated with PD severity compared to modern clinical DBS implementations. Our agent outperforms the standard clinical approaches in suppressing PD biomarkers while relying on information that can be measured in a real world environment, thereby opening up the possibility of training personalized RL agents specific to individual patient needs.
Authors: Huijing Zhang, Muyang Cao, Linshan Jiang, Xin Du, Di Yu, Changze Lv, Shuiguang Deng
Abstract: Continuous learning of novel classes is crucial for edge devices to preserve data privacy and maintain reliable performance in dynamic environments. However, the scenario becomes particularly challenging when data samples are insufficient, requiring on-device few-shot class-incremental learning (FSCIL) to maintain consistent model performance. Although existing work has explored parameter-efficient FSCIL frameworks based on artificial neural networks (ANNs), their deployment is still fundamentally constrained by limited device resources. Inspired by neural mechanisms, Spiking neural networks (SNNs) process spatiotemporal information efficiently, offering lower energy consumption, greater biological plausibility, and compatibility with neuromorphic hardware than ANNs. In this work, we present an SNN-based method for On-Device FSCIL, i.e., Sparsity-Aware and Fast Adaptive SNN (SAFA-SNN). We first propose sparsity-conditioned neuronal dynamics, in which most neurons remain stable while a subset stays active, thereby mitigating catastrophic forgetting. To further cope with spike non-differentiability in gradient estimation, we employ zeroth-order optimization. Moreover, during incremental learning sessions, we enhance the discriminability of new classes through subspace projection, which alleviates overfitting to novel classes. Extensive experiments conducted on two standard benchmark datasets (CIFAR100 and Mini-ImageNet) and three neuromorphic datasets (CIFAR-10-DVS, DVS128gesture, and N-Caltech101) demonstrate that SAFA-SNN outperforms baseline methods, specifically achieving at least 4.01% improvement at the last incremental session on Mini-ImageNet and 20% lower energy cost over baseline methods with practical implementation.
Authors: Amir Sadikov
Abstract: Low-discrepancy point sets and digital sequences underpin quasi-Monte Carlo (QMC) methods for high-dimensional integration. We cast two long-standing QMC design problems as program synthesis and solve them with an LLM-guided evolutionary loop that mutates and selects code under task-specific fitness: (i) constructing finite 2D/3D point sets with low star discrepancy, and (ii) choosing Sobol' direction numbers that minimize randomized QMC error on downstream integrands. Our two-phase procedure combines constructive code proposals with iterative numerical refinement. On finite sets, we rediscover known optima in small 2D cases and set new best-known 2D benchmarks for N >= 40, while matching most known 3D optima up to the proven frontier (N <= 8) and reporting improved 3D benchmarks beyond. On digital sequences, evolving Sobol' parameters yields consistent reductions in randomized quasi-Monte Carlo (rQMC) mean-squared error for several 32-dimensional option-pricing tasks relative to widely used Joe--Kuo parameters, while preserving extensibility to any sample size and compatibility with standard randomizations. Taken together, the results demonstrate that LLM-driven evolutionary program synthesis can automate the discovery of high-quality QMC constructions, recovering classical designs where they are optimal and improving them where finite-N structure matters. Data and code are available at https://github.com/hockeyguy123/openevolve-star-discrepancy.git.
URLs: https://github.com/hockeyguy123/openevolve-star-discrepancy.git.
Authors: Aymeric Fabre
Abstract: In electricity markets around the world, the ability to anticipate price movements with precision can be the difference between profit and loss, especially for fast-acting assets like battery energy storage systems (BESS). As grid volatility increases due to renewables and market decentralisation, operators and forecasters alike face growing pressure to transform prediction into strategy. Yet while forecast data is abundant, especially in advanced markets like Australia's National Electricity Market (NEM), its practical value in driving real-world BESS trading decisions remains largely unexplored. This thesis dives into that gap. This work addresses a key research question: Can the accuracy of the Australian Energy Market Operator (AEMO) energy price forecasts be systematically leveraged to develop a reliable and profitable battery energy storage system trading algorithm? Despite the availability of AEMO price forecasts, no existing framework evaluates their reliability or incorporates them into practical BESS trading strategies. By analysing patterns in forecast accuracy based on time of day, forecast horizon, and regional variations, this project creates a novel, forecast-informed BESS trading model to optimise arbitrage financial returns. The performance of this forecast-driven algorithm is benchmarked against a basic trading algorithm with no knowledge of forecast data. The study further explores the potential of machine learning techniques to predict future energy prices by enhancing AEMO forecasts to govern a more advanced trading strategy. The research outcomes will inform future improvements in energy market trading models and promote more efficient BESS integration into market operations.
Authors: Xu Wang, Yan Hu, Benyou Wang, Difan Zou
Abstract: Sparse Autoencoders (SAEs) are widely used to steer large language models (LLMs), based on the assumption that their interpretable features naturally enable effective model behavior steering. Yet, a fundamental question remains unanswered: does higher interpretability indeed imply better steering utility? To answer this question, we train 90 SAEs across three LLMs (Gemma-2-2B, Qwen-2.5-3B, Gemma-2-9B), spanning five architectures and six sparsity levels, and evaluate their interpretability and steering utility based on SAEBench (arXiv:2501.12345) and AxBench (arXiv:2502.23456) respectively, and perform a rank-agreement analysis via Kendall's rank coefficients (tau b). Our analysis reveals only a relatively weak positive association (tau b approx 0.298), indicating that interpretability is an insufficient proxy for steering performance. We conjecture the interpretability utility gap may stem from the selection of SAE features, as not all of them are equally effective for steering. To further find features that truly steer the behavior of LLMs, we propose a novel selection criterion called Delta Token Confidence, which measures how much amplifying a feature changes the next token distribution. We show that our method improves the steering performance of three LLMs by 52.52 percent compared to the current best output score based criterion (arXiv:2503.34567). Strikingly, after selecting features with high Delta Token Confidence, the correlation between interpretability and utility vanishes (tau b approx 0), and can even become negative. This further highlights the divergence between interpretability and utility for the most effective steering features.
Authors: Jijie Zhou, Niloofar Mireshghallah, Tianshi Li
Abstract: The rapid deployment of large language models (LLMs) in consumer applications has led to frequent exchanges of personal information. To obtain useful responses, users often share more than necessary, increasing privacy risks via memorization, context-based personalization, or security breaches. We present a framework to formally define and operationalize data minimization: for a given user prompt and response model, quantifying the least privacy-revealing disclosure that maintains utility, and we propose a priority-queue tree search to locate this optimal point within a privacy-ordered transformation space. We evaluated the framework on four datasets spanning open-ended conversations (ShareGPT, WildChat) and knowledge-intensive tasks with single-ground-truth answers (CaseHold, MedQA), quantifying achievable data minimization with nine LLMs as the response model. Our results demonstrate that larger frontier LLMs can tolerate stronger data minimization while maintaining task quality than smaller open-source models (85.7% redaction for GPT-5 vs. 19.3% for Qwen2.5-0.5B). By comparing with our search-derived benchmarks, we find that LLMs struggle to predict optimal data minimization directly, showing a bias toward abstraction that leads to oversharing. This suggests not just a privacy gap, but a capability gap: models may lack awareness of what information they actually need to solve a task.
Authors: Wenlong Deng, Yi Ren, Yushu Li, Boying Gong, Danica J. Sutherland, Xiaoxiao Li, Christos Thrampoulidis
Abstract: Reinforcement learning with verifiable rewards has significantly advanced the reasoning capabilities of large language models, yet how to explicitly steer training toward exploration or exploitation remains an open problem. We introduce Token Hidden Reward (THR), a token-level metric that quantifies each token's influence on the likelihood of correct responses under Group Relative Policy Optimization (GRPO). We find that training dynamics are dominated by a small subset of tokens with high absolute THR values. Most interestingly, tokens with positive THR strengthen confidence in correct outputs, thus favoring exploitation, while tokens with negative THR preserve probability mass for alternative outputs, enabling exploration. This insight suggests a natural intervention: a THR-guided reweighting algorithm that modulates GRPO's learning signals to explicitly bias training toward exploitation or exploration. We validate the efficacy of this algorithm on diverse math reasoning benchmarks. By amplifying tokens with positive THR value and weakening negative ones, our algorithm improves greedy-decoding accuracy, favoring exploitation. The reverse strategy yields consistent gains in Pass@K accuracy, favoring exploration. We further demonstrate that our algorithm integrates seamlessly with other RL objectives such as GSPO and generalizes across architectures including Llama. These findings establish THR as a principled and fine-grained mechanism for dynamically controlling exploration and exploitation in RL-tuned LLMs, providing new tools for targeted fine-tuning in reasoning-intensive applications.
Authors: Zhao Song, Shenghao Xie, Samson Zhou
Abstract: This paper studies the computational challenges of large-scale attention-based models in artificial intelligence by utilizing importance sampling methods in the streaming setting. Inspired by the classical definition of the $\ell_2$ sampler and the recent progress of the attention scheme in Large Language Models (LLMs), we propose the definition of the attention sampler. Our approach significantly reduces the computational burden of traditional attention mechanisms. We analyze the effectiveness of the attention sampler from a theoretical perspective, including space and update time. Additionally, our framework exhibits scalability and broad applicability across various model architectures and domains.
Authors: Junhua Chen, Zixi Zhang, Hantao Zhong, Rika Antonova
Abstract: We introduce Group Policy Gradient (GPG), a family of critic-free policy-gradient estimators for general MDPs. Inspired by the success of GRPO's approach in Reinforcement Learning from Human Feedback (RLHF), GPG replaces a learned value function with a group-based Monte Carlo advantage estimator, removing the memory, compute, and hyperparameter costs of training a critic while preserving PPO's clipped-objective structure. We prove the consistency of the GPG estimator, analyze the bias-variance tradeoffs, and demonstrate empirically that GPG matches or outperforms PPO on standard benchmarks. GPG makes better use of parallel simulations, which, together with its critic-free design, results in more efficient use of computational resources than PPO.
Authors: Ali Azizpour, Reza Ramezanpour, Ashutosh Sabharwal, Santiago Segarra
Abstract: Real-world graph datasets often consist of mixtures of populations, where graphs are generated from multiple distinct underlying distributions. However, modern representation learning approaches, such as graph contrastive learning (GCL) and augmentation methods like Mixup, typically overlook this mixture structure. In this work, we propose a unified framework that explicitly models data as a mixture of underlying probabilistic graph generative models represented by graphons. To characterize these graphons, we leverage graph moments (motif densities) to cluster graphs arising from the same model. This enables us to disentangle the mixture components and identify their distinct generative mechanisms. This model-aware partitioning benefits two key graph learning tasks: 1) It enables a graphon-mixture-aware mixup (GMAM), a data augmentation technique that interpolates in a semantically valid space guided by the estimated graphons, instead of assuming a single graphon per class. 2) For GCL, it enables model-adaptive and principled augmentations. Additionally, by introducing a new model-aware objective, our proposed approach (termed MGCL) improves negative sampling by restricting negatives to graphs from other models. We establish a key theoretical guarantee: a novel, tighter bound showing that graphs sampled from graphons with small cut distance will have similar motif densities with high probability. Extensive experiments on benchmark datasets demonstrate strong empirical performance. In unsupervised learning, MGCL achieves state-of-the-art results, obtaining the top average rank across eight datasets. In supervised learning, GMAM consistently outperforms existing strategies, achieving new state-of-the-art accuracy in 6 out of 7 datasets.
Authors: Zehua Liu, Han Wu, Xiaojin Fu, Shuqi Liu, Xiongwei Han, Tao Zhong, Mingxuan Yuan
Abstract: Optimizers are crucial for the efficient training of Large Language Models (LLMs). While AdamW is the de facto standard, recent structure-aware optimizers like Muon have emerged, which regularize gradient updates by operating on entire weight matrices. The Muon optimizer balances the gradient updates along all the directions. However, Muon's reliance on the matrix sign function can lead to training instability, exhibits incompatibility when fine-tuning models pre-trained with AdamW. To address these limitations, we propose \textbf{REG}, a novel optimizer that replaces Muon's aggressive matrix sign operator with the Row-and-Column-Scaling (RACS) operator. Theoretically grounded in balancing a matrix, the RACS operator regularizes the update steps in a less drastic manner, making it simpler to implement and more compatible with established training dynamics. Through extensive empirical experiments on LLM training, we demonstrate that our REG optimizer not only achieves superior performance and stability over AdamW, but also maintains consistency with the AdamW training paradigm. This consistency is particularly evident during the fine-tuning stage, where REG optimizer avoids the performance degradation observed with Muon.
Authors: Qianxin Yi, Shao-Bo Lin, Jun Fan, Yao Wang
Abstract: Reinforcement learning (RL) has been widely applied to sequential decision making, where interpretability and performance are both critical for practical adoption. Current approaches typically focus on performance and rely on post hoc explanations to account for interpretability. Different from these approaches, we focus on designing an interpretability-oriented yet performance-enhanced RL approach. Specifically, we propose a spectral based linear RL method that extends the ridge regression-based approach through a spectral filter function. The proposed method clarifies the role of regularization in controlling estimation error and further enables the design of an adaptive regularization parameter selection strategy guided by the bias-variance trade-off principle. Theoretical analysis establishes near-optimal bounds for both parameter estimation and generalization error. Extensive experiments on simulated environments and real-world datasets from Kuaishou and Taobao demonstrate that our method either outperforms or matches existing baselines in decision quality. We also conduct interpretability analyses to illustrate how the learned policies make decisions, thereby enhancing user trust. These results highlight the potential of our approach to bridge the gap between RL theory and practical decision making, providing interpretability, accuracy, and adaptability in management contexts.
Authors: Jiahao Zeng, Wolong Xing, Liangtao Shi, Xin Huang, Jialin Wang, Zhile Cao, Zhenkui Shi
Abstract: Federated learning has received significant attention for its ability to simultaneously protect customer privacy and leverage distributed data from multiple devices for model training. However, conventional approaches often focus on isolated heterogeneous scenarios, resulting in skewed feature distributions or label distributions. Meanwhile, data heterogeneity is actually a key factor in improving model performance. To address this issue, we propose a new approach called PFPL in mixed heterogeneous scenarios. The method provides richer domain knowledge and unbiased convergence targets by constructing personalized, unbiased prototypes for each client. Moreover, in the local update phase, we introduce consistent regularization to align local instances with their personalized prototypes, which significantly improves the convergence of the loss function. Experimental results on Digits and Office Caltech datasets validate the effectiveness of our approach and successfully reduce the communication cost.
Authors: Yongfu Xue
Abstract: The rapid development of parameter-efficient fine-tuning methods has noticeably improved the efficiency of adapting large language models. Among these, LoRA has gained widespread popularity due to its strong balance of effectiveness and parameter efficiency. However, LoRA relies on initializing two low-rank matrices whose product is zero, which limits its ability to effectively activate and leverage the original model weights-creating a potential bottleneck for optimal performance. To address this limitation, we propose \textbf{IniLoRA}, a novel initialization strategy that initializes the low-rank matrices to closely approximate the original model weights. Experimental results indicate that IniLoRA achieves better performance than LoRA across a range of models and tasks. Additionally, we introduce two variants, IniLoRA-$\alpha$ and IniLoRA-$\beta$, both leveraging distinct initialization methods to enhance performance further.
Authors: Nirjhar Das, Mohit Sharma, Praharsh Nanavati, Kirankumar Shiragur, Amit Deshpande
Abstract: We study the problem of auditing the fairness of a given classifier under partial feedback, where true labels are available only for positively classified individuals, (e.g., loan repayment outcomes are observed only for approved applicants). We introduce a novel cost model for acquiring additional labeled data, designed to more accurately reflect real-world costs such as credit assessment, loan processing, and potential defaults. Our goal is to find optimal fairness audit algorithms that are more cost-effective than random exploration and natural baselines. In our work, we consider two audit settings: a black-box model with no assumptions on the data distribution, and a mixture model, where features and true labels follow a mixture of exponential family distributions. In the black-box setting, we propose a near-optimal auditing algorithm under mild assumptions and show that a natural baseline can be strictly suboptimal. In the mixture model setting, we design a novel algorithm that achieves significantly lower audit cost than the black-box case. Our approach leverages prior work on learning from truncated samples and maximum-a-posteriori oracles, and extends known results on spherical Gaussian mixtures to handle exponential family mixtures, which may be of independent interest. Moreover, our algorithms apply to popular fairness metrics including demographic parity, equal opportunity, and equalized odds. Empirically, we demonstrate strong performance of our algorithms on real-world fair classification datasets like Adult Income and Law School, consistently outperforming natural baselines by around 50% in terms of audit cost.
Authors: Qianfei Fan, Jiayu Wei, Peijun Zhu, Wensheng Ye, Meie Fang
Abstract: Accurate decade-scale daily runoff forecasting in small watersheds is difficult because signals blend drifting trends, multi-scale seasonal cycles, regime shifts, and sparse extremes. Prior deep models (DLinear, TimesNet, PatchTST, TiDE, Nonstationary Transformer, LSTNet, LSTM) usually target single facets and under-utilize unlabeled spans, limiting regime adaptivity. We propose HydroFusion-LMF, a unified framework that (i) performs a learnable trend-seasonal-residual decomposition to reduce non-stationarity, (ii) routes residuals through a compact heterogeneous expert set (linear refinement, frequency kernel, patch Transformer, recurrent memory, dynamically normalized attention), (iii) fuses expert outputs via a hydrologic context-aware gate conditioned on day-of-year phase, antecedent precipitation, local variance, flood indicators, and static basin attributes, and (iv) augments supervision with a semi-supervised multi-task objective (composite MSE/MAE + extreme emphasis + NSE/KGE, masked reconstruction, multi-scale contrastive alignment, augmentation consistency, variance-filtered pseudo-labeling). Optional adapter / LoRA layers inject a frozen foundation time-series encoder efficiently. On a ~10-year daily dataset HydroFusion-LMF attains MSE 1.0128 / MAE 0.5818, improving the strongest baseline (DLinear) by 10.2% / 10.3% and the mean baseline by 24.6% / 17.1%. We observe simultaneous MSE and MAE reductions relative to baselines. The framework balances interpretability (explicit components, sparse gating) with performance, advancing label-efficient hydrologic forecasting under non-stationarity.
Authors: Michael Etienne Van Huffel, Nathan Kirk, Makram Chahine, Daniela Rus, T. Konstantin Rusch
Abstract: Low-discrepancy points are designed to efficiently fill the space in a uniform manner. This uniformity is highly advantageous in many problems in science and engineering, including in numerical integration, computer vision, machine perception, computer graphics, machine learning, and simulation. Whereas most previous low-discrepancy constructions rely on abstract algebra and number theory, Message-Passing Monte Carlo (MPMC) was recently introduced to exploit machine learning methods for generating point sets with lower discrepancy than previously possible. However, MPMC is limited to generating point sets and cannot be extended to low-discrepancy sequences (LDS), i.e., sequences of points in which every prefix has low discrepancy, a property essential for many applications. To address this limitation, we introduce Neural Low-Discrepancy Sequences ($NeuroLDS$), the first machine learning-based framework for generating LDS. Drawing inspiration from classical LDS, we train a neural network to map indices to points such that the resulting sequences exhibit minimal discrepancy across all prefixes. To this end, we deploy a two-stage learning process: supervised approximation of classical constructions followed by unsupervised fine-tuning to minimize prefix discrepancies. We demonstrate that $NeuroLDS$ outperforms all previous LDS constructions by a significant margin with respect to discrepancy measures. Moreover, we demonstrate the effectiveness of $NeuroLDS$ across diverse applications, including numerical integration, robot motion planning, and scientific machine learning. These results highlight the promise and broad significance of Neural Low-Discrepancy Sequences. Our code can be found at https://github.com/camail-official/neuro-lds.
Authors: Ping Guo, Chenyu Zhu, Siyuan Chen, Fei Liu, Xi Lin, Zhichao Lu, Qingfu Zhang
Abstract: CUDA kernel optimization has become a critical bottleneck for AI performance, as deep learning training and inference efficiency directly depends on highly optimized GPU kernels. Despite the promise of Large Language Models (LLMs) for automating kernel optimization, this field suffers from a fragmented ecosystem of isolated and incomparable approaches with unclear problem formulations. Furthermore, general-purpose LLM code evolution methods cannot meet strict correctness requirements of CUDA kernel optimization. We address these fundamental challenges by first formalizing CUDA kernel optimization as a code optimization task with a clear objective, constraints, and evaluation metrics. We then establish the first systematic LLM-based code evolution framework, EvoEngineer, that provides guidance for designing and adapting optimization strategies to achieve a balance between performance and correctness. Finally, we implement a kernel optimization system based on this framework and conduct extensive experiments on 91 real-world CUDA kernels. Our results demonstrate that EvoEngineer achieves a principled balance between performance and correctness, with the highest averaged median speedup of \textbf{2.72}$\times$ over baseline CUDA kernels and a code validity rate of \textbf{69.8}\%, outperforming existing methods on both dimensions. Our method achieves a maximum speedup of \textbf{36.75}$\times$ among all operations over PyTorch kernels and delivers the highest speedup on \textbf{28} (\textbf{56.0\%}) of 50 operations that achieve over \textbf{2$\times$} acceleration.
Authors: Guofu Xie, Chen Zhang, Xiao Zhang, Yunsheng Shi, Ting Yao, Jun Xu
Abstract: Adapting to diverse user needs at test time is a key challenge in controllable multi-objective generation. Existing methods are insufficient: merging-based approaches provide indirect, suboptimal control at the parameter level, often disregarding the impacts of multiple objectives. While decoding-based guidance is more direct, it typically requires aggregating logits from multiple expert models, incurring significant space overhead and relying heavily on individual model capacity. To address these issues, we introduce Merge-And-GuidE (MAGE), a two-stage framework that leverages model merging for guided decoding. We first identify a critical compatibility problem between the guidance and base models. In Stage 1, MAGE resolves this by dynamically constructing a more robust base model, merging a series of backbone models that account for multiple objectives. In Stage 2, we merge explicit and implicit value models into a unified guidance proxy, which then steers the decoding of the base model from Stage 1. Our analysis empirically validates Linear Mode Connectivity (LMC) in value models, explores the relationship between model merging and prediction ensembling, and demonstrates the enhanced controllability afforded by our approach. Extensive experiments show that our method outperforms existing approaches, achieving superior controllability, Pareto-optimal performance, and enhanced adaptability.
Authors: Ruoxi Yu, Haotian Jiang, Jingpu Cheng, Penghao Yu, Qianxiao Li, Zhong Li
Abstract: Transformers have achieved remarkable successes across a wide range of applications, yet the theoretical foundation of their model efficiency remains underexplored. In this work, we investigate how the model parameters -- mainly attention heads and head dimensions -- should be allocated across layers to balance expressivity and efficiency. We first provide mathematical analysis on the role of early layers in information extraction from an approximation perspective, with a theoretical characterization on the trade-off between the number of heads and head dimension under a fixed parameter budget. In addition, we uncover and prove the \emph{saturation} behavior of softmax activations: Continuously increasing head dimensions can lead to diminishing returns in learning errors, particularly for long sequences. Supported by both theory and experiments, this saturation pattern suggests that later layers can operate more efficiently with reduced parameters. Combining these insights, we propose principled strategies for allocating attention heads and dimensions across Transformers' layers, shedding light on theoretically-grounded model efficiency of Transformer-based architectures.
Authors: Yunwen Guo, Yunlun Shu, Gongyi Zhuo, Tianyu Wang
Abstract: The batched multi-armed bandit (MAB) problem, in which rewards are collected in batches, is crucial for applications such as clinical trials. Existing research predominantly assumes light-tailed reward distributions, yet many real-world scenarios, including clinical outcomes, exhibit heavy-tailed characteristics. This paper bridges this gap by proposing robust batched bandit algorithms designed for heavy-tailed rewards, within both finite-arm and Lipschitz-continuous settings. We reveal a surprising phenomenon: in the instance-independent regime, as well as in the Lipschitz setting, heavier-tailed rewards necessitate a smaller number of batches to achieve near-optimal regret. In stark contrast, for the instance-dependent setting, the required number of batches to attain near-optimal regret remains invariant with respect to tail heaviness.
Authors: Aya Laajil, Abduragim Shtanchaev, Sajan Muhammad, Eric Moulines, Salem Lahlou
Abstract: Designing mRNA sequences is a major challenge in developing next-generation therapeutics, since it involves exploring a vast space of possible nucleotide combinations while optimizing sequence properties like stability, translation efficiency, and protein expression. While Generative Flow Networks are promising for this task, their training is hindered by sparse, long-horizon rewards and multi-objective trade-offs. We propose Curriculum-Augmented GFlowNets (CAGFN), which integrate curriculum learning with multi-objective GFlowNets to generate de novo mRNA sequences. CAGFN integrates a length-based curriculum that progressively adapts the maximum sequence length guiding exploration from easier to harder subproblems. We also provide a new mRNA design environment for GFlowNets which, given a target protein sequence and a combination of biological objectives, allows for the training of models that generate plausible mRNA candidates. This provides a biologically motivated setting for applying and advancing GFlowNets in therapeutic sequence design. On different mRNA design tasks, CAGFN improves Pareto performance and biological plausibility, while maintaining diversity. Moreover, CAGFN reaches higher-quality solutions faster than a GFlowNet trained with random sequence sampling (no curriculum), and enables generalization to out-of-distribution sequences.
Authors: Lukas Eisenmann, Alena Br\"andle, Zahra Monfared, Daniel Durstewitz
Abstract: Recurrent Neural Networks (RNNs) have found widespread applications in machine learning for time series prediction and dynamical systems reconstruction, and experienced a recent renaissance with improved training algorithms and architectural designs. Understanding why and how trained RNNs produce their behavior is important for scientific and medical applications, and explainable AI more generally. An RNN's dynamical repertoire depends on the topological and geometrical properties of its state space. Stable and unstable manifolds of periodic points play a particularly important role: They dissect a dynamical system's state space into different basins of attraction, and their intersections lead to chaotic dynamics with fractal geometry. Here we introduce a novel algorithm for detecting these manifolds, with a focus on piecewise-linear RNNs (PLRNNs) employing rectified linear units (ReLUs) as their activation function. We demonstrate how the algorithm can be used to trace the boundaries between different basins of attraction, and hence to characterize multistability, a computationally important property. We further show its utility in finding so-called homoclinic points, the intersections between stable and unstable manifolds, and thus establish the existence of chaos in PLRNNs. Finally we show for an empirical example, electrophysiological recordings from a cortical neuron, how insights into the underlying dynamics could be gained through our method.
Authors: Philipp Becker, Niklas Freymuth, Serge Thilges, Fabian Otto, Gerhard Neumann
Abstract: On-policy Reinforcement Learning (RL) with PPO-like clip objectives has become the standard choice for reward-based fine-tuning of large language models (LLMs). Although recent work has explored improved estimators of advantages and normalization, the clipping mechanism itself has remained untouched. Originally introduced as a proxy for principled KL-based trust regions, clipping is a crude approximation that often causes unstable updates and suboptimal performance. We replace the clip objective with a novel discrete differentiable trust region projection, which provides principled token-level KL constraints. The projection operates on a sparse subset of the model's most important token logits to balance computational cost and projection effectiveness. Our approach, Trust Region Optimization for Large Language Models (TROLL), serves as a direct replacement for PPO-like clipping during training and does not alter the model's inference behavior. Across datasets, model families, and advantage-estimation methods, TROLL consistently outperforms PPO-like clipping in terms of training speed, stability, and final success rates.
Authors: Adam Haroon, Tristan Schuler
Abstract: High Altitude Balloons (HABs) can leverage stratospheric wind layers for limited horizontal control, enabling applications in reconnaissance, environmental monitoring, and communications networks. Existing multi-agent HAB coordination approaches use deterministic methods like Voronoi partitioning and extremum seeking control for large global constellations, which perform poorly for smaller teams and localized missions. While single-agent HAB control using reinforcement learning has been demonstrated on HABs, coordinated multi-agent reinforcement learning (MARL) has not yet been investigated. This work presents the first systematic application of multi-agent reinforcement learning (MARL) to HAB coordination for distributed area coverage. We extend our previously developed reinforcement learning simulation environment (RLHAB) to support cooperative multi-agent learning, enabling multiple agents to operate simultaneously in realistic atmospheric conditions. We adapt QMIX for HAB area coverage coordination, leveraging Centralized Training with Decentralized Execution to address atmospheric vehicle coordination challenges. Our approach employs specialized observation spaces providing individual state, environmental context, and teammate data, with hierarchical rewards prioritizing coverage while encouraging spatial distribution. We demonstrate that QMIX achieves similar performance to the theoretically optimal geometric deterministic method for distributed area coverage, validating the MARL approach and providing a foundation for more complex autonomous multi-HAB missions where deterministic methods become intractable.
Authors: Wei Guo, Jaemoo Choi, Yuchen Zhu, Molei Tao, Yongxin Chen
Abstract: The task of learning a diffusion-based neural sampler for drawing samples from an unnormalized target distribution can be viewed as a stochastic optimal control problem on path measures. However, the training of neural samplers can be challenging when the target distribution is multimodal with significant barriers separating the modes, potentially leading to mode collapse. We propose a framework named \textbf{Proximal Diffusion Neural Sampler (PDNS)} that addresses these challenges by tackling the stochastic optimal control problem via proximal point method on the space of path measures. PDNS decomposes the learning process into a series of simpler subproblems that create a path gradually approaching the desired distribution. This staged procedure traces a progressively refined path to the desired distribution and promotes thorough exploration across modes. For a practical and efficient realization, we instantiate each proximal step with a proximal weighted denoising cross-entropy (WDCE) objective. We demonstrate the effectiveness and robustness of PDNS through extensive experiments on both continuous and discrete sampling tasks, including challenging scenarios in molecular dynamics and statistical physics.
Authors: Alex Durkin, Jasper Stolte, Mehmet Mercang\"oz
Abstract: Start-ups and product grade-changes are critical steps in continuous-process plant operation, because any misstep immediately affects product quality and drives operational losses. These transitions have long relied on manual operation by a handful of expert operators, but the progressive retirement of that workforce is leaving plant owners without the tacit know-how needed to execute them consistently. In the absence of a process model, offline reinforcement learning (RL) promises to capture and even surpass human expertise by mining historical start-up and grade-change logs, yet standard offline RL struggles with distribution shift and value-overestimation whenever a learned policy ventures outside the data envelope. We introduce HOFLON (Hybrid Offline Learning + Online Optimization) to overcome those limitations. Offline, HOFLON learns (i) a latent data manifold that represents the feasible region spanned by past transitions and (ii) a long-horizon Q-critic that predicts the cumulative reward from state-action pairs. Online, it solves a one-step optimization problem that maximizes the Q-critic while penalizing deviations from the learned manifold and excessive rates of change in the manipulated variables. We test HOFLON on two industrial case studies: a polymerization reactor start-up and a paper-machine grade-change problem, and benchmark it against Implicit Q-Learning (IQL), a leading offline-RL algorithm. In both plants HOFLON not only surpasses IQL but also delivers, on average, better cumulative rewards than the best start-up or grade-change observed in the historical data, demonstrating its potential to automate transition operations beyond current expert capability.
Authors: Behraj Khan, Tahir Qasim Syed
Abstract: When training data are fragmented across batches or federated-learned across different geographic locations, trained models manifest performance degradation. That degradation partly owes to covariate shift induced by data having been fragmented across time and space and producing dissimilar empirical training distributions. Each fragment's distribution is slightly different to a hypothetical unfragmented training distribution of covariates, and to the single validation distribution. To address this problem, we propose Fisher Information for Robust fEderated validation (\textbf{FIRE}). This method accumulates fragmentation-induced covariate shift divergences from the global training distribution via an approximate Fisher information. That term, which we prove to be a more computationally-tractable estimate, is then used as a per-fragment loss penalty, enabling scalable distribution alignment. FIRE outperforms importance weighting benchmarks by $5.1\%$ at maximum and federated learning (FL) benchmarks by up to $5.3\%$ on shifted validation sets.
Authors: Behraj Khan, Tahir Qasim Syed
Abstract: We present a theoretical framework for M-FISHER, a method for sequential distribution shift detection and stable adaptation in streaming data. For detection, we construct an exponential martingale from non-conformity scores and apply Ville's inequality to obtain time-uniform guarantees on false alarm control, ensuring statistical validity at any stopping time. Under sustained shifts, we further bound the expected detection delay as $\mathcal{O}(\log(1/\delta)/\Gamma)$, where $\Gamma$ reflects the post-shift information gain, thereby linking detection efficiency to distributional divergence. For adaptation, we show that Fisher-preconditioned updates of prompt parameters implement natural gradient descent on the distributional manifold, yielding locally optimal updates that minimize KL divergence while preserving stability and parameterization invariance. Together, these results establish M-FISHER as a principled approach for robust, anytime-valid detection and geometrically stable adaptation in sequential decision-making under covariate shift.
Authors: Sarah C. Lotspeich, Abbey Collins, Brian J. Wells, Ashish K. Khanna, Joseph Rigdon, Lucy D'Agostino McGowan
Abstract: Objective: Electronic health records (EHR) data are prone to missingness and errors. Previously, we devised an "enriched" chart review protocol where a "roadmap" of auxiliary diagnoses (anchors) was used to recover missing values in EHR data (e.g., a diagnosis of impaired glycemic control might imply that a missing hemoglobin A1c value would be considered unhealthy). Still, chart reviews are expensive and time-intensive, which limits the number of patients whose data can be reviewed. Now, we investigate the accuracy and scalability of a roadmap-driven algorithm, based on ICD-10 codes (International Classification of Diseases, 10th revision), to mimic expert chart reviews and recover missing values. Materials and Methods: In addition to the clinicians' original roadmap from our previous work, we consider new versions that were iteratively refined using large language models (LLM) in conjunction with clinical expertise to expand the list of auxiliary diagnoses. Using chart reviews for 100 patients from the EHR at an extensive learning health system, we examine algorithm performance with different roadmaps. Using the larger study of $1000$ patients, we applied the final algorithm, which used a roadmap with clinician-approved additions from the LLM. Results: The algorithm recovered as much, if not more, missing data as the expert chart reviewers, depending on the roadmap. Discussion: Clinically-driven algorithms (enhanced by LLM) can recover missing EHR data with similar accuracy to chart reviews and can feasibly be applied to large samples. Extending them to monitor other dimensions of data quality (e.g., plausability) is a promising future direction.
Authors: Wenhao Deng, Long Wei, Chenglei Yu, Tailin Wu
Abstract: Reinforcement learning with verifiable rewards (RLVR) has recently enhanced the reasoning capabilities of large language models (LLMs), particularly for mathematical problem solving. However, a fundamental limitation remains: as the sampling budget increases, the advantage of RLVR-trained models over their pretrained bases often diminishes or even vanishes, revealing a strong dependence on the base model's restricted search space. We attribute this phenomenon to the widespread use of the reverse Kullback-Leibler (KL) divergence regularizer, whose mode-seeking behavior keeps the policy trapped inside the base model's support region and hampers wider exploration. To address this issue, we propose RAPO (Rewards-Aware Policy Optimization), an algorithm to promote broader yet focused exploration. Our method (i) utilizes the forward KL penalty to replace the reverse KL penalty for out-of-distribution exploration, and (ii) reweights the reference policy to facilitate adaptive in-distribution exploration. We train Qwen2.5-3B and 7B models with RAPO on the 8K SimpleRL-Zero dataset, without supervised fine-tuning, and evaluate them on AIME2024 and AIME2025. Results show that RAPO consistently improves problem-solving performance. Notably, RAPO enables models to surpass the base model's performance ceiling and solves previously intractable problems, advancing the frontier of RLVR for challenging reasoning tasks.
Authors: Xinwen Zhang, Hongchang Gao
Abstract: The recently introduced optimizer, Muon, has gained increasing attention due to its superior performance across a wide range of applications. However, its effectiveness in federated learning remains unexplored. To address this gap, this paper investigates the performance of Muon in the federated learning setting. Specifically, we propose a new algorithm, FedMuon, and establish its convergence rate for nonconvex problems. Our theoretical analysis reveals multiple favorable properties of FedMuon. In particular, due to its orthonormalized update direction, the learning rate of FedMuon is independent of problem-specific parameters, and, importantly, it can naturally accommodate heavy-tailed noise. The extensive experiments on a variety of neural network architectures validate the effectiveness of the proposed algorithm.
Authors: Oleg Filatov, Jiangtao Wang, Jan Ebert, Stefan Kesselheim
Abstract: Despite recent progress in optimal hyperparameter transfer under model and dataset scaling, no unifying explanatory principle has been established. Using the Scion optimizer, we discover that joint optimal scaling across model and dataset sizes is governed by a single invariant: the operator norm of the output layer. Across models with up to 1.3B parameters trained on up to 138B tokens, the optimal learning rate/batch size pair $(\eta^{\ast}, B^{\ast})$ consistently has the same operator norm value - a phenomenon we term norm transfer. This constant norm condition is necessary but not sufficient: while for each dataset size, multiple $(\eta, B)$ reach the optimal norm, only a unique $(\eta^{\ast}, B^{\ast})$ achieves the best loss. As a sufficient condition, we provide the first measurement of $(\eta^{\ast}, B^{\ast})$ scaling with dataset size for Scion, and find that the scaling rules are consistent with those of the Adam optimizer. Tuning per-layer-group learning rates also improves model performance, with the output layer being the most sensitive and hidden layers benefiting from lower learning rates. We provide practical insights on norm-guided optimal scaling and release our Distributed Scion (Disco) implementation with logs from over two thousand runs to support research on LLM training dynamics at scale.
Authors: Akshay Kudva, Joel A. Paulson
Abstract: Optimal design under uncertainty remains a fundamental challenge in advancing reliable, next-generation process systems. Robust optimization (RO) offers a principled approach by safeguarding against worst-case scenarios across a range of uncertain parameters. However, traditional RO methods typically require known problem structure, which limits their applicability to high-fidelity simulation environments. To overcome these limitations, recent work has explored robust Bayesian optimization (RBO) as a flexible alternative that can accommodate expensive, black-box objectives. Existing RBO methods, however, generally ignore available structural information and struggle to scale to high-dimensional settings. In this work, we introduce BONSAI (Bayesian Optimization of Network Systems under uncertAInty), a new RBO framework that leverages partial structural knowledge commonly available in simulation-based models. Instead of treating the objective as a monolithic black box, BONSAI represents it as a directed graph of interconnected white- and black-box components, allowing the algorithm to utilize intermediate information within the optimization process. We further propose a scalable Thompson sampling-based acquisition function tailored to the structured RO setting, which can be efficiently optimized using gradient-based methods. We evaluate BONSAI across a diverse set of synthetic and real-world case studies, including applications in process systems engineering. Compared to existing simulation-based RO algorithms, BONSAI consistently delivers more sample-efficient and higher-quality robust solutions, highlighting its practical advantages for uncertainty-aware design in complex engineering systems.
Authors: Hangting Ye, Jinmeng Li, He Zhao, Mingchen Zhuge, Dandan Guo, Yi Chang, Hongyuan Zha
Abstract: Existing anomaly detection (AD) methods for tabular data usually rely on some assumptions about anomaly patterns, leading to inconsistent performance in real-world scenarios. While Large Language Models (LLMs) show remarkable reasoning capabilities, their direct application to tabular AD is impeded by fundamental challenges, including difficulties in processing heterogeneous data and significant privacy risks. To address these limitations, we propose LLM-DAS, a novel framework that repositions the LLM from a ``data processor'' to an ``algorithmist''. Instead of being exposed to raw data, our framework leverages the LLM's ability to reason about algorithms. It analyzes a high-level description of a given detector to understand its intrinsic weaknesses and then generates detector-specific, data-agnostic Python code to synthesize ``hard-to-detect'' anomalies that exploit these vulnerabilities. This generated synthesis program, which is reusable across diverse datasets, is then instantiated to augment training data, systematically enhancing the detector's robustness by transforming the problem into a more discriminative two-class classification task. Extensive experiments on 36 TAD benchmarks show that LLM-DAS consistently boosts the performance of mainstream detectors. By bridging LLM reasoning with classic AD algorithms via programmatic synthesis, LLM-DAS offers a scalable, effective, and privacy-preserving approach to patching the logical blind spots of existing detectors.
Authors: Yadav Mahesh Lorik, Kaushik Sarveswaran, Nagaraj Sundaramahalingam, Aravindakumar Venugopalan
Abstract: Time series anomaly detection forms a very crucial area in several domains but poses substantial challenges. Due to time series data possessing seasonality, trends, noise, and evolving patterns (concept drift), it becomes very difficult to set a general notion of what constitutes normal behavior. Anomalies themselves could be varied, ranging from a single outlier to contextual or collective anomalies, and are normally very rare; hence, the dataset is largely imbalanced. Additional layers of complexities arise due to the problems of increased dimensionality of modern time series, real-time detection criteria, setting up appropriate detection thresholds, and arriving at results that are interpretable. To embrace these multifaceted challenges, very strong, flexible, and interpretable approaches are required. This paper presents THEMIS, a new framework for time series anomaly detection that exploits pretrained knowledge from foundation models. THEMIS extracts embeddings from the encoder of the Chronos time series foundation model and applies outlier detection techniques like Local Outlier Factor and Spectral Decomposition on the self-similarity matrix, to spot anomalies in the data. Our experiments show that this modular method achieves SOTA results on the MSL dataset and performs quite competitively on the SMAP and SWAT$^*$ datasets. Notably, THEMIS exceeds models trained specifically for anomaly detection, presenting hyperparameter robustness and interpretability by default. This paper advocates for pretrained representations from foundation models for performing efficient and adaptable anomaly detection for time series data.
Authors: Liyuan Hu, Jitao Wang, Zhenke Wu, Chengchun Shi
Abstract: This paper focuses on reinforcement learning (RL) with clustered data, which is commonly encountered in healthcare applications. We propose a generalized fitted Q-iteration (FQI) algorithm that incorporates generalized estimating equations into policy learning to handle the intra-cluster correlations. Theoretically, we demonstrate (i) the optimalities of our Q-function and policy estimators when the correlation structure is correctly specified, and (ii) their consistencies when the structure is mis-specified. Empirically, through simulations and analyses of a mobile health dataset, we find the proposed generalized FQI achieves, on average, a half reduction in regret compared to the standard FQI.
Authors: Vinod Raman, Shenghao Xie, Samson Zhou
Abstract: Motivated by the predictable nature of real-life in data streams, we study online regression when the learner has access to predictions about future examples. In the extreme case, called transductive online learning, the sequence of examples is revealed to the learner before the game begins. For this setting, we fully characterize the minimax expected regret in terms of the fat-shattering dimension, establishing a separation between transductive online regression and (adversarial) online regression. Then, we generalize this setting by allowing for noisy or \emph{imperfect} predictions about future examples. Using our results for the transductive online setting, we develop an online learner whose minimax expected regret matches the worst-case regret, improves smoothly with prediction quality, and significantly outperforms the worst-case regret when future example predictions are precise, achieving performance similar to the transductive online learner. This enables learnability for previously unlearnable classes under predictable examples, aligning with the broader learning-augmented model paradigm.
Authors: Mingsong Yan, Charles Kulick, Sui Tang
Abstract: Continuous-depth graph neural networks, also known as Graph Neural Differential Equations (GNDEs), combine the structural inductive bias of Graph Neural Networks (GNNs) with the continuous-depth architecture of Neural ODEs, offering a scalable and principled framework for modeling dynamics on graphs. In this paper, we present a rigorous convergence analysis of GNDEs with time-varying parameters in the infinite-node limit, providing theoretical insights into their size transferability. To this end, we introduce Graphon Neural Differential Equations (Graphon-NDEs) as the infinite-node limit of GNDEs and establish their well-posedness. Leveraging tools from graphon theory and dynamical systems, we prove the trajectory-wise convergence of GNDE solutions to Graphon-NDE solutions. Moreover, we derive explicit convergence rates under two deterministic graph sampling regimes: (1) weighted graphs sampled from smooth graphons, and (2) unweighted graphs sampled from $\{0,1\}$-valued (discontinuous) graphons. We further establish size transferability bounds, providing theoretical justification for the practical strategy of transferring GNDE models trained on moderate-sized graphs to larger, structurally similar graphs without retraining. Numerical experiments using synthetic and real data support our theoretical findings.
Authors: Huascar Sanchez, Briland Hitaj
Abstract: Multi-LLM collaboration promises accurate, robust, and context-aware solutions, yet existing approaches rely on implicit selection and output assessment without analyzing whether collaborating models truly complement or conflict. We introduce LLM Chemistry -- a framework that measures when LLM combinations exhibit synergistic or antagonistic behaviors that shape collective performance beyond individual capabilities. We formalize the notion of chemistry among LLMs, propose algorithms that quantify it by analyzing interaction dependencies, and recommend optimal model ensembles accordingly. Our theoretical analysis shows that chemistry among collaborating LLMs is most evident under heterogeneous model profiles, with its outcome impact shaped by task type, group size, and complexity. Evaluation on classification, summarization, and program repair tasks provides initial evidence for these task-dependent effects, thereby reinforcing our theoretical results. This establishes LLM Chemistry as both a diagnostic factor in multi-LLM systems and a foundation for ensemble recommendation.
Authors: Weiqing He, Xiang Li, Tianqi Shang, Li Shen, Weijie Su, Qi Long
Abstract: Large language models (LLMs) raise concerns about content authenticity and integrity because they can generate human-like text at scale. Text watermarks, which embed detectable statistical signals into generated text, offer a provable way to verify content origin. Many detection methods rely on pivotal statistics that are i.i.d. under human-written text, making goodness-of-fit (GoF) tests a natural tool for watermark detection. However, GoF tests remain largely underexplored in this setting. In this paper, we systematically evaluate eight GoF tests across three popular watermarking schemes, using three open-source LLMs, two datasets, various generation temperatures, and multiple post-editing methods. We find that general GoF tests can improve both the detection power and robustness of watermark detectors. Notably, we observe that text repetition, common in low-temperature settings, gives GoF tests a unique advantage not exploited by existing methods. Our results highlight that classic GoF tests are a simple yet powerful and underused tool for watermark detection in LLMs.
Authors: Shahriar Kabir Nahin, Wenxiao Xiao, Joshua Liu, Anshuman Chhabra, Hongfu Liu
Abstract: Data-centric learning seeks to improve model performance from the perspective of data quality, and has been drawing increasing attention in the machine learning community. Among its key tools, influence functions provide a powerful framework to quantify the impact of individual training samples on model predictions, enabling practitioners to identify detrimental samples and retrain models on a cleaner dataset for improved performance. However, most existing work focuses on the question: "what data benefits the learning model?" In this paper, we take a step further and investigate a more fundamental question: "what is the performance ceiling of the learning model?" Unlike prior studies that primarily measure improvement through overall accuracy, we emphasize category-wise accuracy and aim for Pareto improvements, ensuring that every class benefits, rather than allowing tradeoffs where some classes improve at the expense of others. To address this challenge, we propose category-wise influence functions and introduce an influence vector that quantifies the impact of each training sample across all categories. Leveraging these influence vectors, we develop a principled criterion to determine whether a model can still be improved, and further design a linear programming-based sample reweighting framework to achieve Pareto performance improvements. Through extensive experiments on synthetic datasets, vision, and text benchmarks, we demonstrate the effectiveness of our approach in estimating and achieving a model's performance improvement across multiple categories of interest.
Authors: Tim Bary, Tiffanie Godelaine, Axel Abels, Beno\^it Macq
Abstract: Accurate ground truth estimation in medical screening programs often relies on coalitions of experts and peer second opinions. Algorithms that efficiently aggregate noisy annotations can enhance screening workflows, particularly when data arrive continuously and expert proficiency is initially unknown. However, existing algorithms do not meet the requirements for seamless integration into screening pipelines. We therefore propose an adaptive approach for real-time annotation that (I) supports on-the-fly labeling of incoming data, (II) operates without prior knowledge of medical experts or pre-labeled data, and (III) dynamically queries additional experts based on the latent difficulty of each instance. The method incrementally gathers expert opinions until a confidence threshold is met, providing accurate labels with reduced annotation overhead. We evaluate our approach on three multi-annotator classification datasets across different modalities. Results show that our adaptive querying strategy reduces the number of expert queries by up to 50% while achieving accuracy comparable to a non-adaptive baseline. Our code is available at https://github.com/tbary/MEDICS
Authors: Iryna Stanishevska
Abstract: Thunderstorm-driven outages are difficult to predict because most storms do not cause damage, convective processes occur rapidly and chaotically, and the available public data are both noisy and incomplete. We develop a 24-48 h early-warning model for summer, thunderstorm-related outages in Michigan using only open sources (EAGLE-I for ground truth; METAR for weather). We use the publicly released EAGLE-I outage dataset (2014-2022), maintained by Oak Ridge National Laboratory for the U.S. Department of Energy. The pipeline preserves convective micro-signals from a sparse station network via parameter-specific kriging with hourly variograms and targeted overdrafting to retain extremes, and builds causal spatio-temporal features (lags/rolling statistics; k-NN/IDW spatial aggregates) capturing precursors of severe convection (moisture advection, wind shifts, and pressure drops). The two-stage model design, combining a logistic gate and an LSTM regressor, limits routine periods and reduces noise exposure. The study uses event-centric metrics (cluster-based hits/misses/false alarms) and peak-conditional MASE (cMASE) in +/-Delta-hour windows around state-level peaks (>= 50,000), with uncertainty quantified by hourly moving-block bootstrap. On the test sample, Two-Stage detects more reference peaks across all windows (e.g., at +/-48 h it records 3/4 vs. 2/4; F1 66.7% vs. 57.1%) with one extra false alarm. Near peaks, it shows modest amplitude gains (2-3% lower cMASE at +/-0-12 h; bootstrap medians +9-13% at +/-6-12 h) but small losses at +/-36-48 h (~3-4%). Overall, errors are comparable to the one-step LSTM baseline. SHAP analysis confirms moisture-advection and wind/gust precursors, underscoring the value of the feature engineering. Despite open-data noise, the feature-driven pipeline yields actionable, event-focused early warnings for thunderstorm outages.
Authors: Hanzhe Wei, Jiajun Wu, Jialin Yang, Henry Leung, Steve Drew
Abstract: Time series anomaly detection plays a crucial role in a wide range of fields, such as healthcare and internet traffic monitoring. The emergence of large language models (LLMs) offers new opportunities for detecting anomalies in the ubiquitous time series data. Traditional approaches struggle with variable-length time series sequences and context-based anomalies. We propose Soft Prompt Enhanced Anomaly Recognition (SPEAR), a novel approach to leverage LLMs for anomaly detection with soft prompts and quantization. Our methodology involves quantizing and transforming the time series data into input embeddings and combining them with learnable soft prompt embeddings. These combined embeddings are then fed into a frozen LLM. The soft prompts are updated iteratively based on a cross-entropy loss, allowing the model to adapt to time series anomaly detection. The use of soft prompts helps adapt LLMs effectively to time series tasks, while quantization ensures optimal handling of sequences, as LLMs are designed to handle discrete sequences. Our experimental results demonstrate that soft prompts effectively increase LLMs' performance in downstream tasks regarding time series anomaly detection.
Authors: Jatin Prakash, Anirudh Buvanesh
Abstract: Reinforcement learning (RL) with outcome-based rewards has proven effective for improving large language models (LLMs) on complex reasoning tasks. However, its success often depends on the base model occasionally sampling correct solutions. When no correct solutions are sampled, training encounters a zero-reward barrier where learning stalls due to zero gradients. We study this scenario through the graph search task introduced in Bachmann et al. (2024) and evaluate recent methods that incorporate desirable components such as dense rewards, diversity incentives, and improved credit assignment. Our experiments show that none of these approaches overcome the zero-reward barrier if the base model never produces a correct answer. In contrast, we find that a simple data-centric intervention of adding easier samples to the training set enables the model to eventually solve the original hard task despite starting from zero reward. Importantly, this succeeds without modifying the RL algorithm itself. Because official implementations of several baselines were unavailable, we developed our own, which allowed us to conduct a detailed analysis of their failure modes. We release these implementations to support further research at: https://github.com/rl4reasoning/rl-baselines
Authors: Emerson Melo, David M\"uller
Abstract: We establish a link between a class of discrete choice models and the theory of online learning and multi-armed bandits. Our contributions are: (i) sublinear regret bounds for a broad algorithmic family, encompassing Exp3 as a special case; (ii) a new class of adversarial bandit algorithms derived from generalized nested logit models \citep{wen:2001}; and (iii) \textcolor{black}{we introduce a novel class of generalized gradient bandit algorithms that extends beyond the widely used softmax formulation. By relaxing the restrictive independence assumptions inherent in softmax, our framework accommodates correlated learning dynamics across actions, thereby broadening the applicability of gradient bandit methods.} Overall, the proposed algorithms combine flexible model specification with computational efficiency via closed-form sampling probabilities. Numerical experiments in stochastic bandit settings demonstrate their practical effectiveness.
Authors: Michael Yang
Abstract: Hierarchical Pooling Models have demonstrated strong performance in classifying graph-structured data. While numerous innovative methods have been proposed to design cluster assignments and coarsening strategies, the relationships between clusters are often overlooked. In this paper, we introduce Inter-cluster Connectivity Enhancement Pooling (ICEPool), a novel hierarchical pooling framework designed to enhance model's understanding of inter-cluster connectivity and ability of preserving the structural integrity in the original graph. ICEPool is compatible with a wide range of pooling-based GNN models. The deployment of ICEPool as an enhancement to existing models effectively combines the strengths of the original model with ICEPool's capability to emphasize the integration of inter-cluster connectivity, resulting in a more comprehensive and robust graph-level representation. Moreover, we make theoretical analysis to ICEPool's ability of graph reconstruction to demonstrate its effectiveness in learning inter-cluster relationship that is overlooked by conventional models. Finally, the experimental results show the compatibility of ICEPool with wide varieties of models and its potential to boost the performance of existing graph neural network architectures.
Authors: Hoang Anh Just, Myeongseob Ko, Ruoxi Jia
Abstract: Distilling long reasoning traces (10K+ tokens) from stronger teacher models into smaller student LLMs via SFT has emerged as a standard paradigm. This approach is practical and efficient: it leverages the ease of generating abundant reasoning data from stronger models and provides a direct, data-driven way to teach less capable models better reasoning. While previous work has largely focused on prompt selection with responses from a single teacher, the equally important problem of choosing the best response when multiple teacher outputs are available for a single prompt remains underexplored. This challenge becomes important in a multi-teacher setting, where different students may benefit from the outputs of different teachers. This paper fills that gap with a systematic study of response selection for reasoning distillation. We first show that the current method, which picks responses the student assigns the highest global log-probability (global naturalness), fails when responses come from multiple teachers, i.e., global naturalness no longer correlates with downstream performance, especially as the reasoning traces from strong teachers become longer. To overcome this problem, we introduce Local Naturalness, which measures the student's log-probabilities over short, sequential reasoning steps conditioned only on a small local window. Local Naturalness enables two applications: 1) Teacher Selection: Aggregating local scores across prompts reliably identifies the most helpful teacher. 2) Response Selection from a Multiple Teachers: When mixing answers from many teachers, Local Naturalness boosts a 32B student's accuracy on math benchmarks by 9.4pp over global selection, also surpassing the performance achieved by training on data from the single best teacher. These results highlight the power of localized data quality evaluation and data mixing for more effective reasoning distillation.
Authors: Xue-Cheng Tai, Hao Liu, Lingfeng Li, Raymond H. Chan
Abstract: The Transformer architecture has revolutionized the field of sequence modeling and underpins the recent breakthroughs in large language models (LLMs). However, a comprehensive mathematical theory that explains its structure and operations remains elusive. In this work, we propose a novel continuous framework that rigorously interprets the Transformer as a discretization of a structured integro-differential equation. Within this formulation, the self-attention mechanism emerges naturally as a non-local integral operator, and layer normalization is characterized as a projection to a time-dependent constraint. This operator-theoretic and variational perspective offers a unified and interpretable foundation for understanding the architecture's core components, including attention, feedforward layers, and normalization. Our approach extends beyond previous theoretical analyses by embedding the entire Transformer operation in continuous domains for both token indices and feature dimensions. This leads to a principled and flexible framework that not only deepens theoretical insight but also offers new directions for architecture design, analysis, and control-based interpretations. This new interpretation provides a step toward bridging the gap between deep learning architectures and continuous mathematical modeling, and contributes a foundational perspective to the ongoing development of interpretable and theoretically grounded neural network models.
Authors: Hang Fan, Yi Xiao, Yongquan Qu, Fenghua Ling, Ben Fei, Lei Bai, Pierre Gentine
Abstract: Data-driven machine learning (ML) models have recently shown promise in surpassing traditional physics-based approaches for weather forecasting, leading to a so-called second revolution in weather forecasting. However, most ML-based forecast models treat reanalysis as the truth and are trained under variable-specific loss weighting, ignoring their physical coupling and spatial structure. Over long time horizons, the forecasts become blurry and physically unrealistic under rollout training. To address this, we reinterpret model training as a weak-constraint four-dimensional variational data assimilation (WC-4DVar) problem, treating reanalysis data as imperfect observations. This allows the loss function to incorporate reanalysis error covariance and capture multivariate dependencies. In practice, we compute the loss in a latent space learned by an autoencoder (AE), where the reanalysis error covariance becomes approximately diagonal, thus avoiding the need to explicitly model it in the high-dimensional model space. We show that rollout training with latent-space constraints improves long-term forecast skill and better preserves fine-scale structures and physical realism compared to training with model-space loss. Finally, we extend this framework to accommodate heterogeneous data sources, enabling the forecast model to be trained jointly on reanalysis and multi-source observations within a unified theoretical formulation.
Authors: Sahil Joshi, Agniva Chowdhury, Amar Kanakamedala, Ekam Singh, Evan Tu, Anshumali Shrivastava
Abstract: Softmax Attention has a quadratic time complexity, which becomes prohibitive to run at long contexts, even with highly optimized GPU kernels. For example, FlashAttention (an exact, GPU-optimized implementation of Softmax Attention) cannot complete a single forward-backward pass of a multi-head attention layer once the context exceeds ~4 million tokens on an NVIDIA GH200 (96 GB). We introduce RACE Attention, a kernel-inspired alternative to Softmax Attention that is linear in sequence length and embedding dimension. RACE Attention replaces the exponential kernel with a sharpened angular (cosine) similarity, and approximates attention outputs via randomized projections and soft Locality-Sensitive Hashing (LSH). Across language modeling, masked language modeling, and text classification, RACE Attention matches the accuracy of strong baselines while reducing runtime and memory. In a controlled scale test, it processes up to 12 million tokens during a single forward-backward pass on an NVIDIA GH200 GPU and 75 million tokens on an Intel Xeon Gold 5220R CPU, well beyond the practical limits of the current state-of-the-art attention implementations. RACE Attention thus offers a practical, theoretically grounded mechanism for outrageously long context windows on today's hardware. We hope that it gets adopted in practice.
Authors: Anthony Zhan
Abstract: Diffusion large language models (dLLMs) are a new paradigm of non-autoregressive language models that are trained to predict multiple tokens in parallel and generate text via iterative unmasking. Recent works have successfully pretrained dLLMs to parity with autoregressive LLMs at the 8B scale, but dLLMs have yet to benefit from modern post-training techniques, e.g. reinforcement learning (RL), that have proven effective for autoregressive models. Crucially, algorithms designed for traditional LLMs aren't directly compatible with diffusion frameworks due to inherent differences in modeling assumptions. Moreover, existing attempts at dLLM post-training with RL rely on heuristic-based objectives with no theoretical grounding. In this work, we present Amortized Group Relative Policy Optimization (AGRPO), a principled on-policy RL algorithm designed specifically for dLLMs. AGRPO uses Monte Carlo sampling to compute an unbiased policy gradient estimate, making it the first tractable, faithful adaptation of policy gradient methods for dLLMs. We demonstrate AGRPO's effectiveness on different math/reasoning tasks, a common setting for RL with LLMs, achieving up to +7.6% absolute gain on GSM8K and 3.8x performance on the Countdown task over the baseline LLaDA-8B-Instruct model and 1.3x performance gains over comparable RL methods such as diffu-GRPO. Furthermore, these gains persist across different numbers of sampling steps at inference time, achieving better tradeoffs between compute and performance. Our results demonstrate that online RL algorithms can be extended to diffusion LLMs in principled ways, maintaining both theoretical soundness and practical effectiveness.
Authors: Hao Wu, Yuan Gao, Xingjian Shi, Shuaipeng Li, Fan Xu, Fan Zhang, Zhihong Zhu, Weiyan Wang, Xiao Luo, Kun Wang, Xian Wu, Xiaomeng Huang
Abstract: To address the dual challenges of inherent stochasticity and non-differentiable metrics in physical spatiotemporal forecasting, we propose Spatiotemporal Forecasting as Planning (SFP), a new paradigm grounded in Model-Based Reinforcement Learning. SFP constructs a novel Generative World Model to simulate diverse, high-fidelity future states, enabling an "imagination-based" environmental simulation. Within this framework, a base forecasting model acts as an agent, guided by a beam search-based planning algorithm that leverages non-differentiable domain metrics as reward signals to explore high-return future sequences. These identified high-reward candidates then serve as pseudo-labels to continuously optimize the agent's policy through iterative self-training, significantly reducing prediction error and demonstrating exceptional performance on critical domain metrics like capturing extreme events.
Authors: Jinseong Park, Yujin Choi, Jaewook Lee
Abstract: With the increasing need to safeguard data privacy in machine learning models, differential privacy (DP) is one of the major frameworks to build privacy-preserving models. Support Vector Machines (SVMs) are widely used traditional machine learning models due to their robust margin guarantees and strong empirical performance in binary classification. However, applying DP to multi-class SVMs is inadequate, as the standard one-versus-rest (OvR) and one-versus-one (OvO) approaches repeatedly query each data sample when building multiple binary classifiers, thus consuming the privacy budget proportionally to the number of classes. To overcome this limitation, we explore all-in-one SVM approaches for DP, which access each data sample only once to construct multi-class SVM boundaries with margin maximization properties. We propose a novel differentially Private Multi-class SVM (PMSVM) with weight and gradient perturbation methods, providing rigorous sensitivity and convergence analyses to ensure DP in all-in-one SVMs. Empirical results demonstrate that our approach surpasses existing DP-SVM methods in multi-class scenarios.
Authors: Xinhao Yao, Lu Yu, Xiaolin Hu, Fengwei Teng, Qing Cui, Jun Zhou, Yong Liu
Abstract: The ongoing debate on whether reinforcement learning with verifiable rewards (RLVR) expands or shrinks the reasoning capabilities of large language models (LLMs) remains unresolved. Some studies contend that RLVR mainly improves sampling efficiency but at the expense of diversity and exploratory capacity, resulting in capability boundary shrinkage. In contrast, others demonstrate that prolonged training can lead to the emergence of novel reasoning strategies, suggesting capability boundary expansion. To reconcile these contradictory findings, we theoretically and empirically show that both perspectives are partially valid-each aligning with a separate phase in an inherent two-stage probability mass dynamic: (1) Exploitation stage: initially, the model primarily samples explored high-reward and low-reward tokens, while rarely selecting the potentially optimal token. Positive advantage estimates increase the probability of high-reward tokens and decrease those of low-reward tokens, yet the optimal token's probability remains largely unchanged during this stage. (2) Exploration stage: as training advances, the growth rate of previously acquired high-reward tokens slows as their probabilities approach saturation. When a potentially optimal token-now receiving positive advantage estimates-is occasionally sampled, its probability increases, while those of the originally high-reward tokens decrease. This dynamic suggests that over-exploitation during the exploitation stage may lead to capability boundary shrinkage, whereas prolonged training into the exploration stage can promote an expansion of the reasoning capability boundary. Building upon our insights, we revisit the potential of only using relative negative gradients for prolonging training, providing a theoretical and empirical foundation for the development of more advanced reasoning capabilities.
Authors: Kotaro J. Nishimura, Yuichi Sakumura, Kazushi Ikeda
Abstract: Class imbalance is a common challenge in real-world binary classification tasks, often leading to predictions biased toward the majority class and reduced recognition of the minority class. This issue is particularly critical in domains such as medical diagnosis and anomaly detection, where correct classification of minority classes is essential. Conventional methods often fail to deliver satisfactory performance when the imbalance ratio is extremely severe. To address this challenge, we propose a novel approach called Kernel-density-Oriented Threshold Adjustment with Regional Optimization (KOTARO), which extends the framework of kernel density estimation (KDE) by adaptively adjusting decision boundaries according to local sample density. In KOTARO, the bandwidth of Gaussian basis functions is dynamically tuned based on the estimated density around each sample, thereby enhancing the classifier's ability to capture minority regions. We validated the effectiveness of KOTARO through experiments on both synthetic and real-world imbalanced datasets. The results demonstrated that KOTARO outperformed conventional methods, particularly under conditions of severe imbalance, highlighting its potential as a promising solution for a wide range of imbalanced classification problems
Authors: Subhodip Panda, MS Varun, Shreyans Jain, Sarthak Kumar Maharana, Prathosh A. P
Abstract: For a responsible and safe deployment of diffusion models in various domains, regulating the generated outputs from these models is desirable because such models could generate undesired, violent, and obscene outputs. To tackle this problem, recent works use machine unlearning methodology to forget training data points containing these undesired features from pre-trained generative models. However, these methods proved to be ineffective in data-constrained settings where the whole training dataset is inaccessible. Thus, the principal objective of this work is to propose a machine unlearning methodology that can prevent the generation of outputs containing undesired features from a pre-trained diffusion model in such a data-constrained setting. Our proposed method, termed as Variational Diffusion Unlearning (VDU), is a computationally efficient method that only requires access to a subset of training data containing undesired features. Our approach is inspired by the variational inference framework with the objective of minimizing a loss function consisting of two terms: plasticity inducer and stability regularizer. Plasticity inducer reduces the log-likelihood of the undesired training data points, while the stability regularizer, essential for preventing loss of image generation quality, regularizes the model in parameter space. We validate the effectiveness of our method through comprehensive experiments for both class unlearning and feature unlearning. For class unlearning, we unlearn some user-identified classes from MNIST, CIFAR-10, and tinyImageNet datasets from a pre-trained unconditional denoising diffusion probabilistic model (DDPM). Similarly, for feature unlearning, we unlearn the generation of certain high-level features from a pre-trained Stable Diffusion model
Authors: Junxi Yan, Zixi Wei, Jingtao Zhan, Qingyao Ai, Yiqun Liu
Abstract: The cross-entropy scaling law has long served as a key tool for guiding the development of large language models. It shows that cross-entropy loss decreases in a predictable power-law rate as the model size increases. However, recent evidence indicates that this law breaks down at very large scales: the loss decreases more slowly than expected, which causes significant trouble for developing large language models. In this paper, we hypothesize that the root cause lies in the fact that cross-entropy itself does not truly scale; instead, only one of its hidden components does. To investigate this, we introduce a novel decomposition of cross-entropy into three parts: Error-Entropy, Self-Alignment, and Confidence. We show both theoretically and empirically that this decomposition precisely captures the training dynamics and optimization objectives. Through extensive experiments on multiple datasets and 32 models spanning five orders of magnitude in size, we find that only error-entropy follows a robust power-law scaling, while the other two terms remain largely invariant. Moreover, error-entropy constitutes the dominant share of cross-entropy in small models but diminishes in proportion as models grow larger. This explains why the cross-entropy scaling law appears accurate at small scales but fails at very large ones. Our findings establish the error-entropy scaling law as a more accurate description of model behavior. We believe it will have wide applications in the training, understanding, and future development of large language models.
Authors: Ziyan Wang, Zheng Wang, Jie Fu, Xingwei Qu, Qi Cheng, Shengpu Tang, Minjia Zhang, Xiaoming Huo
Abstract: Reinforcement learning (RL) has become central to enhancing reasoning in large language models (LLMs). Yet on-policy algorithms such as Group Relative Policy Optimization (GRPO) often suffer in early training: noisy gradients from low-quality rollouts lead to unstable updates and inefficient exploration. We introduce Slow-Fast Policy Optimization (SFPO), a simple yet efficient framework to address these limitations via decomposing each step into three stages: a short fast trajectory of inner steps on the same batch, a reposition mechanism to control off-policy drift, and a final slow correction. This reposition-before-update design preserves the objective and rollout process unchanged, making SFPO plug-compatible with existing policy-gradient pipelines. Extensive experiments demonstrate that SFPO consistently improves stability, reduces rollouts, and accelerates convergence of reasoning RL training. Specifically, it outperforms GRPO by up to 2.80 points in average on math reasoning benchmarks. It also achieves up to 4.93\texttimes{} fewer rollouts and a 4.19\texttimes{} reduction in wall-clock time to match GRPO's best accuracy.
Authors: Nan Jiang, Tengyang Xie
Abstract: This article introduces the theory of offline reinforcement learning in large state spaces, where good policies are learned from historical data without online interactions with the environment. Key concepts introduced include expressivity assumptions on function approximation (e.g., Bellman completeness vs. realizability) and data coverage (e.g., all-policy vs. single-policy coverage). A rich landscape of algorithms and results is described, depending on the assumptions one is willing to make and the sample and computational complexity guarantees one wishes to achieve. We also discuss open questions and connections to adjacent areas.
Authors: Nikita Gabdullin
Abstract: Supervised learning (SL) methods are indispensable for neural network (NN) training used to perform classification tasks. While resulting in very high accuracy, SL training often requires making NN parameter number dependent on the number of classes, limiting their applicability when the number of classes is extremely large or unknown in advance. In this paper we propose a methodology that allows one to train the same NN architecture regardless of the number of classes. This is achieved by using predefined vector systems as the target latent space configuration (LSC) during NN training. We discuss the desired properties of target configurations and choose randomly perturbed vectors of An root system for our experiments. These vectors are used to successfully train encoders and visual transformers (ViT) on Cinic-10 and ImageNet-1K in low- and high-dimensional cases by matching NN predictions with the predefined vectors. Finally, ViT is trained on a dataset with 1.28 million classes illustrating the applicability of the method to training on datasets with extremely large number of classes. In addition, potential applications of LSC in lifelong learning and NN distillation are discussed illustrating versatility of the proposed methodology.
Authors: Wei Wang, Tianhao Ma, Ming-Kun Xie, Gang Niu, Masashi Sugiyama
Abstract: Partial multi-label learning and complementary multi-label learning are two popular weakly supervised multi-label classification paradigms that aim to alleviate the high annotation costs of collecting precisely annotated multi-label data. In partial multi-label learning, each instance is annotated with a candidate label set, among which only some labels are relevant; in complementary multi-label learning, each instance is annotated with complementary labels indicating the classes to which the instance does not belong. Existing consistent approaches for the two paradigms either require accurate estimation of the generation process of candidate or complementary labels or assume a uniform distribution to eliminate the estimation problem. However, both conditions are usually difficult to satisfy in real-world scenarios. In this paper, we propose consistent approaches that do not rely on the aforementioned conditions to handle both problems in a unified way. Specifically, we propose two unbiased risk estimators based on first- and second-order strategies. Theoretically, we prove consistency w.r.t. two widely used multi-label classification evaluation metrics and derive convergence rates for the estimation errors of the proposed risk estimators. Empirically, extensive experimental results validate the effectiveness of our proposed approaches against state-of-the-art methods.
Authors: Ramzi Dakhmouche, Hossein Gorji
Abstract: Motivated by the remarkable success of Foundation Models (FMs) in language modeling, there has been growing interest in developing FMs for time series prediction, given the transformative power such models hold for science and engineering. This culminated in significant success of FMs in short-range forecasting settings. However, extrapolation or long-range forecasting remains elusive for FMs, which struggle to outperform even simple baselines. This contrasts with physical laws which have strong extrapolation properties, and raises the question of the fundamental difference between the structure of neural networks and physical laws. In this work, we identify and formalize a fundamental property characterizing the ability of statistical learning models to predict more accurately outside of their training domain, hence explaining performance deterioration for deep learning models in extrapolation settings. In addition to a theoretical analysis, we present empirical results showcasing the implications of this property on current deep learning architectures. Our results not only clarify the root causes of the extrapolation gap but also suggest directions for designing next-generation forecasting models capable of mastering extrapolation.
Authors: Ramzi Dakhmouche, Adrien Letellier, Hossein Gorji
Abstract: Effective Uncertainty Quantification (UQ) represents a key aspect for reliable deployment of Large Language Models (LLMs) in automated decision-making and beyond. Yet, for LLM generation with multiple choice structure, the state-of-the-art in UQ is still dominated by the naive baseline given by the maximum softmax score. To address this shortcoming, we demonstrate that taking a principled approach via Bayesian statistics leads to improved performance despite leveraging the simplest possible model, namely linear regression. More precisely, we propose to train multiple Bayesian linear models, each predicting the output of a layer given the output of the previous one. Based on the obtained layer-level posterior distributions, we infer the global uncertainty level of the LLM by identifying a sparse combination of distributional features, leading to an efficient UQ scheme. Numerical experiments on various LLMs show consistent improvement over state-of-the-art baselines.
Authors: Wanxin Li, Yongjin P. Park, Khanh Dao Duc
Abstract: Fairness in machine learning is a critical concern, yet most research has focused on classification tasks, leaving regression models underexplored. This paper introduces a Wasserstein projection-based framework for fairness testing in regression models, focusing on expectation-based criteria. We propose a hypothesis-testing approach and an optimal data perturbation method to improve fairness while balancing accuracy. Theoretical results include a detailed categorization of fairness criteria for regression, a dual reformulation of the Wasserstein projection test statistic, and the derivation of asymptotic bounds and limiting distributions. Experiments on synthetic and real-world datasets demonstrate that the proposed method offers higher specificity compared to permutation-based tests, and effectively detects and mitigates biases in real applications such as student performance and housing price prediction.
Authors: George Giapitzakis, Kimon Fountoulakis, Eshaan Nichani, Jason D. Lee
Abstract: Semiautomata form a rich class of sequence-processing algorithms with applications in natural language processing, robotics, computational biology, and data mining. We establish the first Statistical Query hardness result for semiautomata under the uniform distribution over input words and initial states. We show that Statistical Query hardness can be established when both the alphabet size and input length are polynomial in the number of states. Unlike the case of deterministic finite automata, where hardness typically arises through the hardness of the language they recognize (e.g., parity), our result is derived solely from the internal state-transition structure of semiautomata. Our analysis reduces the task of distinguishing the final states of two semiautomata to studying the behavior of a random walk on the group $S_{N} \times S_{N}$. By applying tools from Fourier analysis and the representation theory of the symmetric group, we obtain tight spectral gap bounds, demonstrating that after a polynomial number of steps in the number of states, distinct semiautomata become nearly uncorrelated, yielding the desired hardness result.
Authors: Ziying Zhang, Yaqing Wang, Yuxuan Sun, Min Ye, Quanming Yao
Abstract: Cold-start drug-target interaction (DTI) prediction focuses on interaction between novel drugs and proteins. Previous methods typically learn transferable interaction patterns between structures of drug and proteins to tackle it. However, insight from proteomics suggest that protein have multi-level structures and they all influence the DTI. Existing works usually represent protein with only primary structures, limiting their ability to capture interactions involving higher-level structures. Inspired by this insight, we propose ColdDTI, a framework attending on protein multi-level structure for cold-start DTI prediction. We employ hierarchical attention mechanism to mine interaction between multi-level protein structures (from primary to quaternary) and drug structures at both local and global granularities. Then, we leverage mined interactions to fuse structure representations of different levels for final prediction. Our design captures biologically transferable priors, avoiding the risk of overfitting caused by excessive reliance on representation learning. Experiments on benchmark datasets demonstrate that ColdDTI consistently outperforms previous methods in cold-start settings.
Authors: Yang Chen, Yitao Liang, Zhouchen Lin
Abstract: In Transformers, Position Embeddings (PEs) significantly influence Length Generalization (LG) performance, yet their fundamental role remains unclear. In this work, we investigate the limitations and capabilities of PEs in achieving LG. We theoretically analyze PEs in Position-Only Linear Attentions (POLAs), introducing Linear Representation Complexity (LRC) to characterize when PEs enable LG. Our analysis shows that PEs do not expand computational capabilities but structure learned computations across positions. Extending to practical Transformers, we propose Sequential Representation Complexity (SRC) and conjecture that LG is possible if and only if SRC remains invariant across scales. We support this hypothesis with empirical evidence in various reasoning tasks. To enhance LG, we introduce Scale Hint, allowing flexible instance scaling, and a Learning-Based Position Embedding framework that automatically learns positional relations. Our work provides theoretical insights and practical strategies for improving LG in Transformers.
Authors: Muhao Guo, Yang Weng
Abstract: Neural ODEs (NODEs) have emerged as powerful tools for modeling time series data, offering the flexibility to adapt to varying input scales and capture complex dynamics. However, they face significant challenges: first, their reliance on time-domain representations often limits their ability to capture long-term dependencies and periodic structures; second, the inherent mismatch between their continuous-time formulation and the discrete nature of real-world data can lead to loss of granularity and predictive accuracy. To address these limitations, we propose Fourier Ordinary Differential Equations (FODEs), an approach that embeds the dynamics in the Fourier domain. By transforming time-series data into the frequency domain using the Fast Fourier Transform (FFT), FODEs uncover global patterns and periodic behaviors that remain elusive in the time domain. Additionally, we introduce a learnable element-wise filtering mechanism that aligns continuous model outputs with discrete observations, preserving granularity and enhancing accuracy. Experiments on various time series datasets demonstrate that FODEs outperform existing methods in terms of both accuracy and efficiency. By effectively capturing both long- and short-term patterns, FODEs provide a robust framework for modeling time series dynamics.
Authors: Yiming Niu, Jinliang Deng, Yongxin Tong
Abstract: Periodicity is a fundamental characteristic of time series data and has long played a central role in forecasting. Recent deep learning methods strengthen the exploitation of periodicity by treating patches as basic tokens, thereby improving predictive effectiveness. However, their efficiency remains a bottleneck due to large parameter counts and heavy computational costs. This paper provides, for the first time, a clear explanation of why patch-level processing is inherently inefficient, supported by strong evidence from real-world data. To address these limitations, we introduce a phase perspective for modeling periodicity and present an efficient yet effective solution, PhaseFormer. PhaseFormer features phase-wise prediction through compact phase embeddings and efficient cross-phase interaction enabled by a lightweight routing mechanism. Extensive experiments demonstrate that PhaseFormer achieves state-of-the-art performance with around 1k parameters, consistently across benchmark datasets. Notably, it excels on large-scale and complex datasets, where models with comparable efficiency often struggle. This work marks a significant step toward truly efficient and effective time series forecasting. Code is available at this repository: https://github.com/neumyor/PhaseFormer_TSL
Authors: Muhao Guo, Haoran Li, Yang Weng
Abstract: Neural ordinary differential equations (NODE) have garnered significant attention for their design of continuous-depth neural networks and the ability to learn data/feature dynamics. However, for high-dimensional systems, estimating dynamics requires extensive calculations and suffers from high truncation errors for the ODE solvers. To address the issue, one intuitive approach is to consider the non-trivial topological space of the data distribution, i.e., a low-dimensional manifold. Existing methods often rely on knowledge of the manifold for projection or implicit transformation, restricting the ODE solutions on the manifold. Nevertheless, such knowledge is usually unknown in realistic scenarios. Therefore, we propose a novel approach to explore the underlying manifold to restrict the ODE process. Specifically, we employ a structure-preserved encoder to process data and find the underlying graph to approximate the manifold. Moreover, we propose novel methods to combine the NODE learning with the manifold, resulting in significant gains in computational speed and accuracy. Our experimental evaluations encompass multiple datasets, where we compare the accuracy, number of function evaluations (NFEs), and convergence speed of our model against existing baselines. Our results demonstrate superior performance, underscoring the effectiveness of our approach in addressing the challenges of high-dimensional datasets.
Authors: Minseo Kim, Coleman Hooper, Aditya Tomar, Chenfeng Xu, Mehrdad Farajtabar, Michael W. Mahoney, Kurt Keutzer, Amir Gholami
Abstract: Large Language Models (LLMs) have achieved state-of-the-art performance on a broad range of Natural Language Processing (NLP) tasks, including document processing and coding. Autoregressive Language Models (ARMs), which generate tokens sequentially conditioned on all previous tokens, have been the predominant paradigm for LLMs. However, while these networks have achieved high accuracy across a range of downstream tasks, they exhibit low arithmetic intensity due to the inherent sequential dependency with next-token prediction. Recently, Diffusion Language Models (DLMs) have emerged as a promising alternative architecture. DLMs generate output text in parallel, breaking the limitations of sequential dependency. However, the performance implications of DLMs relative to commonly deployed ARMs are not fully understood. In this work, we present a comprehensive performance study analyzing the performance characteristics of ARMs and DLMs, using both theoretical analysis and profiling data to characterize the trade-offs between these approaches. We illustrate that although DLMs exhibit higher arithmetic intensity compared to ARMs because of their capability to utilize parallelism across sequence lengths, they fail to scale effectively to longer contexts. We then explore DLMs with block-wise decoding, outlining how this approach allows for increased arithmetic intensity, while still scaling well to long contexts (similar to ARMs). We also show interesting trade-offs for batched inference, where we find that ARMs exhibit superior throughput, as they benefit more from parallelism across sequences in the batch. Finally, we highlight opportunities for accelerating DLM inference, and, in particular, highlight the importance of reducing the number of sampling steps for allowing open-source DLMs to provide improved latency relative to ARMs.
Authors: Prashansa Panda, Shalabh Bhatnagar
Abstract: Recent studies have increasingly focused on non-asymptotic convergence analyses for actor-critic (AC) algorithms. One such effort introduced a two-timescale critic-actor algorithm for the discounted cost setting using a tabular representation, where the usual roles of the actor and critic are reversed. However, only asymptotic convergence was established there. Subsequently, both asymptotic and non-asymptotic analyses of the critic-actor algorithm with linear function approximation were conducted. In our work, we introduce the first natural critic-actor algorithm with function approximation for the long-run average cost setting and under inequality constraints. We provide the non-asymptotic convergence guarantees for this algorithm. Our analysis establishes optimal learning rates and we also propose a modification to enhance sample complexity. We further show the results of experiments on three different Safety-Gym environments where our algorithm is found to be competitive in comparison with other well known algorithms.
Authors: Haiquan Qiu, You Wu, Yingjie Tan, Yaqing Wang, Quanming Yao
Abstract: Loss explosions in training deep neural networks can nullify multi-million dollar training runs. Conventional monitoring metrics like weight and gradient norms are often lagging and ambiguous predictors, as their values vary dramatically across different models and even between layers of the same model, making it difficult to establish a unified standard for detecting impending failure. We introduce Spectral Alignment (SA), a novel, theoretically-grounded metric that monitors the distributional alignment between layer inputs and the principal singular vectors of weight matrices. We show that a collapse in the sign diversity of this alignment is a powerful early predictor of representational collapse and training divergence. Empirical results on language models demonstrate that monitoring the SA distribution provides a significantly earlier and clearer warning of loss explosions than traditional scalar metrics. SA's low computational overhead makes it a practical tool for safeguarding model training.
Authors: Aayushya Agarwal, Larry Pileggi, Gauri Joshi
Abstract: Hyperparameter selection is critical for stable and efficient convergence of heterogeneous federated learning, where clients differ in computational capabilities, and data distributions are non-IID. Tuning hyperparameters is a manual and computationally expensive process as the hyperparameter space grows combinatorially with the number of clients. To address this, we introduce an end-to-end adaptive federated learning method in which both clients and central agents adaptively select their local learning rates and momentum parameters. Our approach models federated learning as a dynamical system, allowing us to draw on principles from numerical simulation and physical design. Through this perspective, selecting momentum parameters equates to critically damping the system for fast, stable convergence, while learning rates for clients and central servers are adaptively selected to satisfy accuracy properties from numerical simulation. The result is an adaptive, momentum-based federated learning algorithm in which the learning rates for clients and servers are dynamically adjusted and controlled by a single, global hyperparameter. By designing a fully integrated solution for both adaptive client updates and central agent aggregation, our method is capable of handling key challenges of heterogeneous federated learning, including objective inconsistency and client drift. Importantly, our approach achieves fast convergence while being insensitive to the choice of the global hyperparameter, making it well-suited for rapid prototyping and scalable deployment. Compared to state-of-the-art adaptive methods, our framework is shown to deliver superior convergence for heterogeneous federated learning while eliminating the need for hyperparameter tuning both client and server updates.
Authors: Di Zhang
Abstract: Kolmogorov-Arnold Networks (KANs) have emerged as a promising alternative to traditional Multi-Layer Perceptrons (MLPs), offering enhanced interpretability and a strong mathematical foundation. However, their parameter efficiency remains a significant challenge for practical deployment. This paper introduces PolyKAN, a novel theoretical framework for KAN compression that provides formal guarantees on both model size reduction and approximation error. By leveraging the inherent piecewise polynomial structure of KANs, we formulate the compression problem as one of optimal polyhedral region merging. We establish a rigorous polyhedral characterization of KANs, develop a complete theory of $\epsilon$-equivalent compression, and design an optimal dynamic programming algorithm that guarantees minimal compression under specified error bounds. Our theoretical analysis demonstrates that PolyKAN achieves provably minimal compression while maintaining strict error control, with polynomial-time complexity in all network parameters. The framework provides the first formal foundation for KAN compression with mathematical guarantees, opening new directions for efficient deployment of interpretable neural architectures.
Authors: Haiquan Qiu, Quanming Yao
Abstract: The pursuit of computational efficiency has driven the adoption of low-precision formats for training transformer models. However, this progress is often hindered by notorious training instabilities. This paper provides the first mechanistic explanation for a long-standing and unresolved failure case where training with flash attention in low-precision settings leads to catastrophic loss explosions. Our in-depth analysis reveals that the failure is not a random artifact but caused by two intertwined phenomena: the emergence of similar low-rank representations within the attention mechanism and the compounding effect of biased rounding errors inherent in low-precision arithmetic. We demonstrate how these factors create a vicious cycle of error accumulation that corrupts weight updates, ultimately derailing the training dynamics. To validate our findings, we introduce a minimal modification to the flash attention that mitigates the bias in rounding errors. This simple change stabilizes the training process, confirming our analysis and offering a practical solution to this persistent problem.
Authors: Chenlu Ding, Jiancan Wu, Leheng Sheng, Fan Zhang, Yancheng Yuan, Xiang Wang, Xiangnan He
Abstract: Multimodal large language models (MLLMs) have demonstrated remarkable capabilities across vision-language tasks, yet their large-scale deployment raises pressing concerns about memorized private data, outdated knowledge, and harmful content. Existing unlearning approaches for MLLMs typically adapt training-based strategies such as gradient ascent or preference optimization, but these methods are computationally expensive, irreversible, and often distort retained knowledge. In this work, we propose MLLMEraser, an input-aware, training-free framework for test-time unlearning. Our approach leverages activation steering to enable dynamic knowledge erasure without parameter updates. Specifically, we construct a multimodal erasure direction by contrasting adversarially perturbed, knowledge-recall image-text pairs with knowledge-erasure counterparts, capturing both textual and visual discrepancies. To prevent unnecessary interference, we further design an input-aware steering mechanism that adaptively determines when and how the erasure direction should be applied, preserving utility on retained knowledge while enforcing forgetting on designated content. Experiments on LLaVA-1.5 and Qwen-2.5-VL demonstrate that MLLMEraser consistently outperforms state-of-the-art MLLM unlearning baselines, achieving stronger forgetting performance with lower computational cost and minimal utility degradation.
Authors: Kai Yang, Yuqi Huang, Junheng Tao, Wanyu Wang, Qitian Wu
Abstract: Modeling 3D dynamics is a fundamental problem in multi-body systems across scientific and engineering domains and has important practical implications in trajectory prediction and simulation. While recent GNN-based approaches have achieved strong performance by enforcing geometric symmetries, encoding high-order features or incorporating neural-ODE mechanics, they typically depend on explicitly observed structures and inherently fail to capture the unobserved interactions that are crucial to complex physical behaviors and dynamics mechanism. In this paper, we propose PAINET, a principled SE(3)-equivariant neural architecture for learning all-pair interactions in multi-body systems. The model comprises: (1) a novel physics-inspired attention network derived from the minimization trajectory of an energy function, and (2) a parallel decoder that preserves equivariance while enabling efficient inference. Empirical results on diverse real-world benchmarks, including human motion capture, molecular dynamics, and large-scale protein simulations, show that PAINET consistently outperforms recently proposed models, yielding 4.7% to 41.5% error reductions in 3D dynamics prediction with comparable computation costs in terms of time and memory.
Authors: Jinhui Bai, Andreas Christmann, Lei Shi
Abstract: In this paper, we propose a novel kernel stochastic gradient descent (SGD) algorithm for large-scale supervised learning with general losses. Compared to traditional kernel SGD, our algorithm improves efficiency and scalability through an innovative regularization strategy. By leveraging the infinite series expansion of spherical radial basis functions, this strategy projects the stochastic gradient onto a finite-dimensional hypothesis space, which is adaptively scaled according to the bias-variance trade-off, thereby enhancing generalization performance. Based on a new estimation of the spectral structure of the kernel-induced covariance operator, we develop an analytical framework that unifies optimization and generalization analyses. We prove that both the last iterate and the suffix average converge at minimax-optimal rates, and we further establish optimal strong convergence in the reproducing kernel Hilbert space. Our framework accommodates a broad class of classical loss functions, including least-squares, Huber, and logistic losses. Moreover, the proposed algorithm significantly reduces computational complexity and achieves optimal storage complexity by incorporating coordinate-wise updates from linear SGD, thereby avoiding the costly pairwise operations typical of kernel SGD and enabling efficient processing of streaming data. Finally, extensive numerical experiments demonstrate the efficiency of our approach.
Authors: Seong Jin Ahn, Myoung-Ho Kim
Abstract: For large-scale applications, there is growing interest in replacing Graph Neural Networks (GNNs) with lightweight Multi-Layer Perceptrons (MLPs) via knowledge distillation. However, distilling GNNs for self-supervised graph representation learning into MLPs is more challenging. This is because the performance of self-supervised learning is more related to the model's inductive bias than supervised learning. This motivates us to design a new distillation method to bridge a huge capacity gap between GNNs and MLPs in self-supervised graph representation learning. In this paper, we propose \textbf{D}iffusion-\textbf{A}ssisted \textbf{D}istillation for \textbf{S}elf-supervised \textbf{G}raph representation learning with \textbf{M}LPs (DAD-SGM). The proposed method employs a denoising diffusion model as a teacher assistant to better distill the knowledge from the teacher GNN into the student MLP. This approach enhances the generalizability and robustness of MLPs in self-supervised graph representation learning. Extensive experiments demonstrate that DAD-SGM effectively distills the knowledge of self-supervised GNNs compared to state-of-the-art GNN-to-MLP distillation methods. Our implementation is available at https://github.com/SeongJinAhn/DAD-SGM.
Authors: Joseph Ramsey, Bryan Andrews
Abstract: Learning causal structure from observational data is especially challenging when latent variables or selection bias are present. The Fast Causal Inference (FCI) algorithm addresses this setting but often performs exhaustive conditional independence tests across many subsets, leading to spurious independence claims, extra or missing edges, and unreliable orientations. We present a family of score-guided mixed-strategy causal search algorithms that build on this tradition. First, we introduce BOSS-FCI and GRaSP-FCI, straightforward variants of GFCI that substitute BOSS or GRaSP for FGES, thereby retaining correctness while incurring different scalability tradeoffs. Second, we develop FCI Targeted-testing (FCIT), a novel mixed-strategy method that improves upon these variants by replacing exhaustive all-subsets testing with targeted tests guided by BOSS, yielding well-formed PAGs with higher precision and efficiency. Finally, we propose a simple heuristic, LV-Dumb (also known as BOSS-POD), which bypasses latent-variable-specific reasoning and directly returns the PAG of the BOSS DAG. Although not strictly correct in the FCI sense, it scales better and often achieves superior accuracy in practice. Simulations and real-data analyses demonstrate that BOSS-FCI and GRaSP-FCI provide sound baselines, FCIT improves both efficiency and reliability, and LV-Dumb offers a practical heuristic with strong empirical performance. Together, these method highlight the value of score-guided and targeted strategies for scalable latent-variable causal discovery.
Authors: Paul Strang, Zacharie Al\`es, C\^ome Bissuel, Olivier Juan, Safia Kedad-Sidhoum, Emmanuel Rachelson
Abstract: On the occasion of the 20th Mixed Integer Program Workshop's computational competition, this work introduces a new approach for learning to solve MIPs online. Influence branching, a new graph-oriented variable selection strategy, is applied throughout the first iterations of the branch and bound algorithm. This branching heuristic is optimized online with Thompson sampling, which ranks the best graph representations of MIP's structure according to computational speed up over SCIP. We achieve results comparable to state of the art online learning methods. Moreover, our results indicate that our method generalizes well to more general online frameworks, where variations in constraint matrix, constraint vector and objective coefficients can all occur and where more samples are available.
Authors: \'Alvaro Serra-Gomez, Daniel Jarne Ornia, Dhruva Tirumala, Thomas Moerland
Abstract: Effective exploration remains a central challenge in model-based reinforcement learning (MBRL), particularly in high-dimensional continuous control tasks where sample efficiency is crucial. A prominent line of recent work leverages learned policies as proposal distributions for Model-Predictive Path Integral (MPPI) planning. Initial approaches update the sampling policy independently of the planner distribution, typically maximizing a learned value function with deterministic policy gradient and entropy regularization. However, because the states encountered during training depend on the MPPI planner, aligning the sampling policy with the planner improves the accuracy of value estimation and long-term performance. To this end, recent methods update the sampling policy by minimizing KL divergence to the planner distribution or by introducing planner-guided regularization into the policy update. In this work, we unify these MPPI-based reinforcement learning methods under a single framework by introducing Policy Optimization-Model Predictive Control (PO-MPC), a family of KL-regularized MBRL methods that integrate the planner's action distribution as a prior in policy optimization. By aligning the learned policy with the planner's behavior, PO-MPC allows more flexibility in the policy updates to trade off Return maximization and KL divergence minimization. We clarify how prior approaches emerge as special cases of this family, and we explore previously unstudied variations. Our experiments show that these extended configurations yield significant performance improvements, advancing the state of the art in MPPI-based RL.
Authors: Nghiem T. Diep, Dung Le, Tuan Truong, Tan Dinh, Huy Nguyen, Nhat Ho
Abstract: Low-Rank Adaptation (LoRA) is a parameter-efficient fine-tuning (PEFT) technique that adapts large pre-trained models by adding low-rank matrices to their weight updates. However, in the context of fine-tuning multi-head self-attention (MHA), LoRA has been employed to adapt each attention head separately, thereby overlooking potential synergies across different heads. To mitigate this issue, we propose a novel Hyper-shared Low-Rank Adaptation (HoRA) method, which utilizes joint hypernetworks to generate low-rank matrices across attention heads. By coupling their adaptation through a shared generator, HoRA encourages cross-head information sharing, and thus directly addresses the aforementioned limitation of LoRA. By comparing LoRA and HoRA through the lens of hierarchical mixture of experts, our theoretical findings reveal that the latter achieves superior sample efficiency to the former. Furthermore, through extensive experiments across diverse language and vision benchmarks, we demonstrate that HoRA outperforms LoRA and other PEFT methods while requiring only a marginal increase in the number of trainable parameters.
Authors: Harshil Vejendla
Abstract: We introduce Wave-PDE Nets, a neural architecture whose elementary operation is a differentiable simulation of the second-order wave equation. Each layer propagates its hidden state as a continuous field through a medium with trainable spatial velocity c(x) and damping {\gamma}(x). A symplectic spectral solver based on FFTs realises this propagation in O(nlog n) time. This oscillatory, global mechanism provides a powerful alternative to attention and first-order state-space models. We prove that a single Wave-PDE layer is a universal approximator. On language and vision benchmarks, Wave-PDE Nets match or exceed Transformer performance while demonstrating superior practical efficiency, reducing wall-clock time by up to 30% and peak memory by 25%. Ablation studies confirm the critical role of symplectic integration and a spectral Laplacian for stability and performance. Visualizations of the learned physical parameters reveal that the model learns intuitive strategies for information propagation. These results position Wave-PDE Nets as a computationally efficient and robust architecture with a strong physical inductive bias.
Authors: Dung V. Nguyen, Hieu M. Vu, Nhi Y. Pham, Lei Zhang, Tan M. Nguyen
Abstract: Controlling the behaviors of large language models (LLM) is fundamental to their safety alignment and reliable deployment. However, existing steering methods are primarily driven by empirical insights and lack theoretical performance guarantees. In this work, we develop a control-theoretic foundation for activation steering by showing that popular steering methods correspond to the proportional (P) controllers, with the steering vector serving as the feedback signal. Building on this finding, we propose Proportional-Integral-Derivative (PID) Steering, a principled framework that leverages the full PID controller for activation steering in LLMs. The proportional (P) term aligns activations with target semantic directions, the integral (I) term accumulates errors to enforce persistent corrections across layers, and the derivative (D) term mitigates overshoot by counteracting rapid activation changes. This closed-loop design yields interpretable error dynamics and connects activation steering to classical stability guarantees in control theory. Moreover, PID Steering is lightweight, modular, and readily integrates with state-of-the-art steering methods. Extensive experiments across multiple LLM families and benchmarks demonstrate that PID Steering consistently outperforms existing approaches, achieving more robust and reliable behavioral control.
Authors: Sahar Koohfar
Abstract: Accurate and timely prediction of crash severity is crucial in mitigating the severe consequences of traffic accidents. Accurate and timely prediction of crash severity is crucial in mitigating the severe consequences of traffic accidents. In order to provide appropriate levels of medical assistance and transportation services, an intelligent transportation system relies on effective prediction methods. Deep learning models have gained popularity in this domain due to their capability to capture non-linear relationships among variables. In this research, we have implemented a hybrid CNN-RNN deep learning model for crash severity prediction and compared its performance against widely used statistical and machine learning models such as logistic regression, na\"ive bayes classifier, K-Nearest Neighbors (KNN), decision tree, and individual deep learning models: RNN and CNN. This study employs a methodology that considers the interconnected relationships between various features of traffic accidents. The study was conducted using a dataset of 15,870 accident records gathered over a period of seven years between 2015 and 2021 on Virginia highway I-64. The findings demonstrate that the proposed CNN-RNN hybrid model has outperformed all benchmark models in terms of predicting crash severity. This result illustrates the effectiveness of the hybrid model as it combines the advantages of both RNN and CNN models in order to achieve greater accuracy in the prediction process.
Authors: Yucong Dai, Lu Zhang, Feng Luo, Mashrur Chowdhury, Yongkai Wu
Abstract: Training fair and unbiased machine learning models is crucial for high-stakes applications, yet it presents significant challenges. Effective bias mitigation requires deep expertise in fairness definitions, metrics, data preprocessing, and machine learning techniques. In addition, the complex process of balancing model performance with fairness requirements while properly handling sensitive attributes makes fairness-aware model development inaccessible to many practitioners. To address these challenges, we introduce FairAgent, an LLM-powered automated system that significantly simplifies fairness-aware model development. FairAgent eliminates the need for deep technical expertise by automatically analyzing datasets for potential biases, handling data preprocessing and feature engineering, and implementing appropriate bias mitigation strategies based on user requirements. Our experiments demonstrate that FairAgent achieves significant performance improvements while significantly reducing development time and expertise requirements, making fairness-aware machine learning more accessible to practitioners.
Authors: Kenechukwu Ogbuagu, Sepehr Maleki, Giuseppe Bruni, Senthil Krishnababu
Abstract: The accurate prediction of flow fields around airfoils is crucial for aerodynamic design and optimisation. Computational Fluid Dynamics (CFD) models are effective but computationally expensive, thus inspiring the development of surrogate models to enable quicker predictions. These surrogate models can be based on deep learning architectures, such as Convolutional Neural Networks (CNNs), Graph Neural Networks (GNNs), and Diffusion Models (DMs). Diffusion models have shown significant promise in predicting complex flow fields. In this work, we propose FoilDiff, a diffusion-based surrogate model with a hybrid-backbone denoising network. This hybrid design combines the power of convolutional feature extraction and transformer-based global attention to generate more adaptable and accurate representations of flow structures. FoilDiff takes advantage of Denoising Diffusion Implicit Model (DDIM) sampling to optimise the efficiency of the sampling process at no additional cost to model generalisation. We used encoded representations of Reynolds number, angle of attack, and airfoil geometry to define the input space for generalisation across a wide range of aerodynamic conditions. When evaluated against state-of-the-art models, FoilDiff shows significant performance improvements, with mean prediction errors reducing by up to 85\% on the same datasets. The results have demonstrated that FoilDiff can provide both more accurate predictions and better-calibrated predictive uncertainty than existing diffusion-based models.
Authors: Haosong Zhang, Shenxi Wu, Yichi Zhang, Wei Lin
Abstract: Choosing an appropriate learning rate remains a key challenge in scaling depth of modern deep networks. The classical maximal update parameterization ($\mu$P) enforces a fixed per-layer update magnitude, which is well suited to homogeneous multilayer perceptrons (MLPs) but becomes ill-posed in heterogeneous architectures where residual accumulation and convolutions introduce imbalance across layers. We introduce Arithmetic-Mean $\mu$P (AM-$\mu$P), which constrains not each individual layer but the network-wide average one-step pre-activation second moment to a constant scale. Combined with a residual-aware He fan-in initialization - scaling residual-branch weights by the number of blocks ($\mathrm{Var}[W]=c/(K\cdot \mathrm{fan\text{-}in})$) - AM-$\mu$P yields width-robust depth laws that transfer consistently across depths. We prove that, for one- and two-dimensional convolutional networks, the maximal-update learning rate satisfies $\eta^\star(L)\propto L^{-3/2}$; with zero padding, boundary effects are constant-level as $N\gg k$. For standard residual networks with general conv+MLP blocks, we establish $\eta^\star(L)=\Theta(L^{-3/2})$, with $L$ the minimal depth. Empirical results across a range of depths confirm the $-3/2$ scaling law and enable zero-shot learning-rate transfer, providing a unified and practical LR principle for convolutional and deep residual networks without additional tuning overhead.
Authors: Nghiem T. Diep, Hien Dang, Tuan Truong, Tan Dinh, Huy Nguyen, Nhat Ho
Abstract: Parameter-efficient fine-tuning (PEFT) methods have become the standard paradigm for adapting large-scale models. Among these techniques, Weight-Decomposed Low-Rank Adaptation (DoRA) has been shown to improve both the learning capacity and training stability of the vanilla Low-Rank Adaptation (LoRA) method by explicitly decomposing pre-trained weights into magnitude and directional components. In this work, we propose DoRAN, a new variant of DoRA designed to further stabilize training and boost the sample efficiency of DoRA. Our approach includes two key stages: (i) injecting noise into the denominator of DoRA's weight decomposition, which serves as an adaptive regularizer to mitigate instabilities; and (ii) replacing static low-rank matrices with auxiliary networks that generate them dynamically, enabling parameter coupling across layers and yielding better sample efficiency in both theory and practice. Comprehensive experiments on vision and language benchmarks show that DoRAN consistently outperforms LoRA, DoRA, and other PEFT baselines. These results underscore the effectiveness of combining stabilization through noise-based regularization with network-based parameter generation, offering a promising direction for robust and efficient fine-tuning of foundation models.
Authors: G. Niklas Noren, Eva-Lisa Meldau, Johan Ellenius
Abstract: Many high-stakes AI applications target low-prevalence events, where apparent accuracy can conceal limited real-world value. Relevant AI models range from expert-defined rules and traditional machine learning to generative LLMs constrained for classification. We outline key considerations for critical appraisal of AI in rare-event recognition, including problem framing and test set design, prevalence-aware statistical evaluation, robustness assessment, and integration into human workflows. In addition, we propose an approach to structured case-level examination (SCLE), to complement statistical performance evaluation, and a comprehensive checklist to guide procurement or development of AI models for rare-event recognition. We instantiate the framework in pharmacovigilance, drawing on three studies: rule-based retrieval of pregnancy-related reports; duplicate detection combining machine learning with probabilistic record linkage; and automated redaction of person names using an LLM. We highlight pitfalls specific to the rare-event setting including optimism from unrealistic class balance and lack of difficult positive controls in test sets - and show how cost-sensitive targets align model performance with operational value. While grounded in pharmacovigilance practice, the principles generalize to domains where positives are scarce and error costs may be asymmetric.
Authors: Harshil Vejendla
Abstract: Forecasting chaotic systems is a cornerstone challenge in many scientific fields, complicated by the exponential amplification of even infinitesimal prediction errors. Modern machine learning approaches often falter due to two opposing pitfalls: over-specializing on a single, well-known chaotic system (e.g., Lorenz-63), which limits generalizability, or indiscriminately mixing vast, unrelated time-series, which prevents the model from learning the nuances of any specific dynamical regime. We propose Curriculum Chaos Forecasting (CCF), a training paradigm that bridges this gap. CCF organizes training data based on fundamental principles of dynamical systems theory, creating a curriculum that progresses from simple, periodic behaviors to highly complex, chaotic dynamics. We quantify complexity using the largest Lyapunov exponent and attractor dimension, two well-established metrics of chaos. By first training a sequence model on predictable systems and gradually introducing more chaotic trajectories, CCF enables the model to build a robust and generalizable representation of dynamical behaviors. We curate a library of over 50 synthetic ODE/PDE systems to build this curriculum. Our experiments show that pre-training with CCF significantly enhances performance on unseen, real-world benchmarks. On datasets including Sunspot numbers, electricity demand, and human ECG signals, CCF extends the valid prediction horizon by up to 40% compared to random-order training and more than doubles it compared to training on real-world data alone. We demonstrate that this benefit is consistent across various neural architectures (GRU, Transformer) and provide extensive ablations to validate the importance of the curriculum's structure.
Authors: Anoushka Harit, Zhongtian Sun, Jongmin Yu
Abstract: We propose the Causal Sphere Hypergraph Transformer (CSHT), a novel architecture for interpretable financial time-series forecasting that unifies \emph{Granger-causal hypergraph structure}, \emph{Riemannian geometry}, and \emph{causally masked Transformer attention}. CSHT models the directional influence of financial news and sentiment on asset returns by extracting multivariate Granger-causal dependencies, which are encoded as directional hyperedges on the surface of a hypersphere. Attention is constrained via angular masks that preserve both temporal directionality and geometric consistency. Evaluated on S\&P 500 data from 2018 to 2023, including the 2020 COVID-19 shock, CSHT consistently outperforms baselines across return prediction, regime classification, and top-asset ranking tasks. By enforcing predictive causal structure and embedding variables in a Riemannian manifold, CSHT delivers both \emph{robust generalisation across market regimes} and \emph{transparent attribution pathways} from macroeconomic events to stock-level responses. These results suggest that CSHT is a principled and practical solution for trustworthy financial forecasting under uncertainty.
Authors: Christopher Klugmann, Daniel Kondermann
Abstract: Human-generated categorical annotations frequently produce empirical response distributions (soft labels) that reflect ambiguity rather than simple annotator error. We introduce an ambiguity measure that maps a discrete response distribution to a scalar in the unit interval, designed to quantify aleatoric uncertainty in categorical tasks. The measure bears a close relationship to quadratic entropy (Gini-style impurity) but departs from those indices by treating an explicit "can't solve" category asymmetrically, thereby separating uncertainty arising from class-level indistinguishability from uncertainty due to explicit unresolvability. We analyze the measure's formal properties and contrast its behavior with a representative ambiguity measure from the literature. Moving beyond description, we develop statistical tools for inference: we propose frequentist point estimators for population ambiguity and derive the Bayesian posterior over ambiguity induced by Dirichlet priors on the underlying probability vector, providing a principled account of epistemic uncertainty. Numerical examples illustrate estimation, calibration, and practical use for dataset-quality assessment and downstream machine-learning workflows.
Authors: Tejal Patwardhan, Rachel Dias, Elizabeth Proehl, Grace Kim, Michele Wang, Olivia Watkins, Sim\'on Posada Fishman, Marwan Aljubeh, Phoebe Thacker, Laurance Fauconnet, Natalie S. Kim, Patrick Chao, Samuel Miserendino, Gildas Chabot, David Li, Michael Sharman, Alexandra Barr, Amelia Glaese, Jerry Tworek
Abstract: We introduce GDPval, a benchmark evaluating AI model capabilities on real-world economically valuable tasks. GDPval covers the majority of U.S. Bureau of Labor Statistics Work Activities for 44 occupations across the top 9 sectors contributing to U.S. GDP (Gross Domestic Product). Tasks are constructed from the representative work of industry professionals with an average of 14 years of experience. We find that frontier model performance on GDPval is improving roughly linearly over time, and that the current best frontier models are approaching industry experts in deliverable quality. We analyze the potential for frontier models, when paired with human oversight, to perform GDPval tasks cheaper and faster than unaided experts. We also demonstrate that increased reasoning effort, increased task context, and increased scaffolding improves model performance on GDPval. Finally, we open-source a gold subset of 220 tasks and provide a public automated grading service at evals.openai.com to facilitate future research in understanding real-world model capabilities.
Authors: Akshay Mittal, Vinay Venkatesh, Krishna Kandi, Shalini Sudarshan
Abstract: The effectiveness of single-model sequential recommendation architectures, while scalable, is often limited when catering to "power users" in sparse or niche domains. Our previous research, PinnerFormerLite, addressed this by using a fixed weighted loss to prioritize specific domains. However, this approach can be sub-optimal, as a single, uniform weight may not be sufficient for domains with very few interactions, where the training signal is easily diluted by the vast, generic dataset. This paper proposes a novel, data-driven approach: a Dynamic Weighted Loss function with comprehensive theoretical foundations and extensive empirical validation. We introduce an adaptive algorithm that adjusts the loss weight for each domain based on its sparsity in the training data, assigning a higher weight to sparser domains and a lower weight to denser ones. This ensures that even rare user interests contribute a meaningful gradient signal, preventing them from being overshadowed. We provide rigorous theoretical analysis including convergence proofs, complexity analysis, and bounds analysis to establish the stability and efficiency of our approach. Our comprehensive empirical validation across four diverse datasets (MovieLens, Amazon Electronics, Yelp Business, LastFM Music) with state-of-the-art baselines (SIGMA, CALRec, SparseEnNet) demonstrates that this dynamic weighting system significantly outperforms all comparison methods, particularly for sparse domains, achieving substantial lifts in key metrics like Recall at 10 and NDCG at 10 while maintaining performance on denser domains and introducing minimal computational overhead.
Authors: Abdulrahman Tamim
Abstract: Neural network training is typically viewed as gradient descent on a loss surface. We propose a fundamentally different perspective: learning is a structure-preserving transformation (a functor L) between the space of network parameters (Param) and the space of learned representations (Rep). This categorical framework reveals that different training runs producing similar test performance often belong to the same homotopy class (continuous deformation family) of optimization paths. We show experimentally that networks converging via homotopic trajectories generalize within 0.5% accuracy of each other, while non-homotopic paths differ by over 3%. The theory provides practical tools: persistent homology identifies stable minima predictive of generalization (R^2 = 0.82 correlation), pullback constructions formalize transfer learning, and 2-categorical structures explain when different optimization algorithms yield functionally equivalent models. These categorical invariants offer both theoretical insight into why deep learning works and concrete algorithmic principles for training more robust networks.
Authors: Xinshuai Dong, Ignavier Ng, Haoyue Dai, Jiaqi Sun, Xiangchen Song, Peter Spirtes, Kun Zhang
Abstract: Identifying the structure of a partially observed causal system is essential to various scientific fields. Recent advances have focused on constraint-based causal discovery to solve this problem, and yet in practice these methods often face challenges related to multiple testing and error propagation. These issues could be mitigated by a score-based method and thus it has raised great attention whether there exists a score-based greedy search method that can handle the partially observed scenario. In this work, we propose the first score-based greedy search method for the identification of structure involving latent variables with identifiability guarantees. Specifically, we propose Generalized N Factor Model and establish the global consistency: the true structure including latent variables can be identified up to the Markov equivalence class by using score. We then design Latent variable Greedy Equivalence Search (LGES), a greedy search algorithm for this class of model with well-defined operators, which search very efficiently over the graph space to find the optimal structure. Our experiments on both synthetic and real-life data validate the effectiveness of our method (code will be publicly available).
Authors: Shakson Isaac, Yentl Collin, Chirag Patel
Abstract: Continuous glucose monitoring (CGM) generates dense data streams critical for diabetes management, but most used forecasting models lack interpretability for clinical use. We present SSM-CGM, a Mamba-based neural state-space forecasting model that integrates CGM and wearable activity signals from the AI-READI cohort. SSM-CGM improves short-term accuracy over a Temporal Fusion Transformer baseline, adds interpretability through variable selection and temporal attribution, and enables counterfactual forecasts simulating how planned changes in physiological signals (e.g., heart rate, respiration) affect near-term glucose. Together, these features make SSM-CGM an interpretable, physiologically grounded framework for personalized diabetes management.
Authors: Wenyuan Zhao, Adithya Balachandran, Chao Tian, Paul Pu Liang
Abstract: The study of multimodality has garnered significant interest in fields where the analysis of interactions among multiple information sources can enhance predictive modeling, data fusion, and interpretability. Partial information decomposition (PID) has emerged as a useful information-theoretic framework to quantify the degree to which individual modalities independently, redundantly, or synergistically convey information about a target variable. However, existing PID methods depend on optimizing over a joint distribution constrained by estimated pairwise probability distributions, which are costly and inaccurate for continuous and high-dimensional modalities. Our first key insight is that the problem can be solved efficiently when the pairwise distributions are multivariate Gaussians, and we refer to this problem as Gaussian PID (GPID). We propose a new gradient-based algorithm that substantially improves the computational efficiency of GPID based on an alternative formulation of the underlying optimization problem. To generalize the applicability to non-Gaussian data, we learn information-preserving encoders to transform random variables of arbitrary input distributions into pairwise Gaussian random variables. Along the way, we resolved an open problem regarding the optimality of joint Gaussian solutions for GPID. Empirical validation in diverse synthetic examples demonstrates that our proposed method provides more accurate and efficient PID estimates than existing baselines. We further evaluate a series of large-scale multimodal benchmarks to show its utility in real-world applications of quantifying PID in multimodal datasets and selecting high-performing models.
Authors: Ziyi Chen, Heng Huang
Abstract: Performative reinforcement learning is an emerging dynamical decision making framework, which extends reinforcement learning to the common applications where the agent's policy can change the environmental dynamics. Existing works on performative reinforcement learning only aim at a performatively stable (PS) policy that maximizes an approximate value function. However, there is a provably positive constant gap between the PS policy and the desired performatively optimal (PO) policy that maximizes the original value function. In contrast, this work proposes a zeroth-order Frank-Wolfe algorithm (0-FW) algorithm with a zeroth-order approximation of the performative policy gradient in the Frank-Wolfe framework, and obtains \textbf{the first polynomial-time convergence to the desired PO} policy under the standard regularizer dominance condition. For the convergence analysis, we prove two important properties of the nonconvex value function. First, when the policy regularizer dominates the environmental shift, the value function satisfies a certain gradient dominance property, so that any stationary point (not PS) of the value function is a desired PO. Second, though the value function has unbounded gradient, we prove that all the sufficiently stationary points lie in a convex and compact policy subspace $\Pi_{\Delta}$, where the policy value has a constant lower bound $\Delta>0$ and thus the gradient becomes bounded and Lipschitz continuous. Experimental results also demonstrate that our 0-FW algorithm is more effective than the existing algorithms in finding the desired PO policy.
Authors: Ziyi Chen, Su Zhang, Heng Huang
Abstract: Federated learning has attracted increasing attention at recent large-scale optimization and machine learning research and applications, but is also vulnerable to Byzantine clients that can send any erroneous signals. Robust aggregators are commonly used to resist Byzantine clients. This usually requires to estimate the unknown number $f$ of Byzantine clients, and thus accordingly select the aggregators with proper degree of robustness (i.e., the maximum number $\hat{f}$ of Byzantine clients allowed by the aggregator). Such an estimation should have important effect on the performance, which has not been systematically studied to our knowledge. This work will fill in the gap by theoretically analyzing the worst-case error of aggregators as well as its induced federated learning algorithm for any cases of $\hat{f}$ and $f$. Specifically, we will show that underestimation ($\hat{f}
Authors: Farid Bozorgnia, Vyacheslav Kungurtsev, Shirali Kadyrov, Mohsen Yousefnezhad
Abstract: In this work, we introduce novel algorithms for label propagation and self-training using fractional heat kernel dynamics with a source term. We motivate the methodology through the classical correspondence of information theory with the physics of parabolic evolution equations. We integrate the fractional heat kernel into Graph Neural Network architectures such as Graph Convolutional Networks and Graph Attention, enhancing their expressiveness through adaptive, multi-hop diffusion. By applying Chebyshev polynomial approximations, large graphs become computationally feasible. Motivating variational formulations demonstrate that by extending the classical diffusion model to fractional powers of the Laplacian, nonlocal interactions deliver more globally diffusing labels. The particular balance between supervision of known labels and diffusion across the graph is particularly advantageous in the case where only a small number of labeled training examples are present. We demonstrate the effectiveness of this approach on standard datasets.
Authors: Yilun Zhu, Naihao Deng, Naichen Shi, Aditya Gangrade, Clayton Scott
Abstract: Domain generalization (DG) is the problem of generalizing from several distributions (or domains), for which labeled training data are available, to a new test domain for which no labeled data is available. A common finding in the DG literature is that it is difficult to outperform empirical risk minimization (ERM) on the pooled training data. In this work, we argue that this finding has primarily been reported for datasets satisfying a \emph{covariate shift} assumption. When the dataset satisfies a \emph{posterior drift} assumption instead, we show that ``domain-informed ERM,'' wherein feature vectors are augmented with domain-specific information, outperforms pooling ERM. These claims are supported by a theoretical framework and experiments on language and vision tasks.
Authors: Willa Potosnak, Malcolm Wolff, Boris Oreshkin, Mengfei Cao, Michael W. Mahoney, Dmitry Efimov, Kin G. Olivares
Abstract: While accuracy is a critical requirement for time series forecasting models, an equally important (yet often overlooked) desideratum is forecast stability across forecast creation dates (FCDs). Even highly accurate models can produce erratic revisions between FCDs, undermining stakeholder trust and disrupting downstream decision-making. To improve forecast stability, models like MQCNN, MQT, and SPADE employ a little-known but highly effective technique: forking-sequences. Unlike standard statistical and neural forecasting methods that treat each FCD independently, the forking-sequences method jointly encodes and decodes the entire time series across all FCDs, in a way mirroring time series cross-validation. Since forking sequences remains largely unknown in the broader neural forecasting community, in this work, we formalize the forking-sequences approach, and we make a case for its broader adoption. We demonstrate three key benefits of forking-sequences: (i) more stable and consistent gradient updates during training; (ii) reduced forecast variance through ensembling; and (iii) improved inference computational efficiency. We validate forking-sequences' benefits using 16 datasets from the M1, M3, M4, and Tourism competitions, showing improvements in forecast percentage change stability of 28.8%, 28.8%, 37.9%, and 31.3%, and 8.8%, on average, for MLP, RNN, LSTM, CNN, and Transformer-based architectures, respectively.
Authors: Linghao Kong, Inimai Subramanian, Yonadav Shavit, Micah Adler, Dan Alistarh, Nir Shavit
Abstract: This work demonstrates how increasing the number of neurons in a network without increasing its number of non-zero parameters improves performance. We show that this gain corresponds with a decrease in interference between multiple features that would otherwise share the same neurons. To reduce such entanglement at a fixed non-zero parameter count, we introduce Fixed Parameter Expansion (FPE): replace a neuron with multiple children and partition the parent's weights disjointly across them, so that each child inherits a non-overlapping subset of connections. On symbolic tasks, specifically Boolean code problems, clause-aligned FPE systematically reduces polysemanticity metrics and yields higher task accuracy. Notably, random splits of neuron weights approximate these gains, indicating that reduced collisions, not precise assignment, are a primary driver. Consistent with the superposition hypothesis, the benefits of FPE grow with increasing interference: when polysemantic load is high, accuracy improvements are the largest. Transferring these insights to real models (classifiers over CLIP embeddings and deeper multilayer networks) we find that widening networks while maintaining a constant non-zero parameter count consistently increases accuracy. These results identify an interpretability-grounded mechanism to leverage width against superposition, improving performance without increasing the number of non-zero parameters. Such a direction is well matched to modern accelerators, where memory movement of non-zero parameters, rather than raw compute, is the dominant bottleneck.
Authors: Min Wang, Xin Li, Ye He, Yao-Hui Li, Hasnaa Bennis, Riashat Islam, Mingzhong Wang
Abstract: The real world is inherently non-stationary, with ever-changing factors, such as weather conditions and traffic flows, making it challenging for agents to adapt to varying environmental dynamics. Non-Stationary Reinforcement Learning (NSRL) addresses this challenge by training agents to adapt rapidly to sequences of distinct Markov Decision Processes (MDPs). However, existing NSRL approaches often focus on tasks with regularly evolving patterns, leading to limited adaptability in highly dynamic settings. Inspired by the success of Wavelet analysis in time series modeling, specifically its ability to capture signal trends at multiple scales, we propose WISDOM to leverage wavelet-domain predictive task representations to enhance NSRL. WISDOM captures these multi-scale features in evolving MDP sequences by transforming task representation sequences into the wavelet domain, where wavelet coefficients represent both global trends and fine-grained variations of non-stationary changes. In addition to the auto-regressive modeling commonly employed in time series forecasting, we devise a wavelet temporal difference (TD) update operator to enhance tracking and prediction of MDP evolution. We theoretically prove the convergence of this operator and demonstrate policy improvement with wavelet task representations. Experiments on diverse benchmarks show that WISDOM significantly outperforms existing baselines in both sample efficiency and asymptotic performance, demonstrating its remarkable adaptability in complex environments characterized by non-stationary and stochastically evolving tasks.
Authors: Achim Eckerle, Martin Spitznagel, Janis Keuper
Abstract: Accurate and fast urban noise prediction is pivotal for public health and for regulatory workflows in cities, where the Environmental Noise Directive mandates regular strategic noise maps and action plans, often needed in permission workflows, right-of-way allocation, and construction scheduling. Physics-based solvers are too slow for such time-critical, iterative "what-if" studies. We evaluate conditional Normalizing Flows (Full-Glow) for generating for generating standards-compliant urban sound-pressure maps from 2D urban layouts in real time per 256x256 map on a single RTX 4090), enabling interactive exploration directly on commodity hardware. On datasets covering Baseline, Diffraction, and Reflection regimes, our model accelerates map generation by >2000 times over a reference solver while improving NLoS accuracy by up to 24% versus prior deep models; in Baseline NLoS we reach 0.65 dB MAE with high structural fidelity. The model reproduces diffraction and interference patterns and supports instant recomputation under source or geometry changes, making it a practical engine for urban planning, compliance mapping, and operations (e.g., temporary road closures, night-work variance assessments).
Authors: Yisen Gao, Xingcheng Fu, Qingyun Sun, Jianxin Li, Xianxian Li
Abstract: Graph diffusion models have made significant progress in learning structured graph data and have demonstrated strong potential for predictive tasks. Existing approaches typically embed node, edge, and graph-level features into a unified latent space, modeling prediction tasks including classification and regression as a form of conditional generation. However, due to the non-Euclidean nature of graph data, features of different curvatures are entangled in the same latent space without releasing their geometric potential. To address this issue, we aim to construt an ideal Riemannian diffusion model to capture distinct manifold signatures of complex graph data and learn their distribution. This goal faces two challenges: numerical instability caused by exponential mapping during the encoding proces and manifold deviation during diffusion generation. To address these challenges, we propose GeoMancer: a novel Riemannian graph diffusion framework for both generation and prediction tasks. To mitigate numerical instability, we replace exponential mapping with an isometric-invariant Riemannian gyrokernel approach and decouple multi-level features onto their respective task-specific manifolds to learn optimal representations. To address manifold deviation, we introduce a manifold-constrained diffusion method and a self-guided strategy for unconditional generation, ensuring that the generated data remains aligned with the manifold signature. Extensive experiments validate the effectiveness of our approach, demonstrating superior performance across a variety of tasks.
Authors: Satoshi Hayakawa, Yuhta Takida, Masaaki Imaizumi, Hiromi Wakaki, Yuki Mitsufuji
Abstract: Masked diffusion models have shown promising performance in generating high-quality samples in a wide range of domains, but accelerating their sampling process remains relatively underexplored. To investigate efficient samplers for masked diffusion, this paper theoretically analyzes the MaskGIT sampler for image modeling, revealing its implicit temperature sampling mechanism. Through this analysis, we introduce the "moment sampler," an asymptotically equivalent but more tractable and interpretable alternative to MaskGIT, which employs a "choose-then-sample" approach by selecting unmasking positions before sampling tokens. In addition, we improve the efficiency of choose-then-sample algorithms through two key innovations: a partial caching technique for transformers that approximates longer sampling trajectories without proportional computational cost, and a hybrid approach formalizing the exploration-exploitation trade-off in adaptive unmasking. Experiments in image and text domains demonstrate our theory as well as the efficiency of our proposed methods, advancing both theoretical understanding and practical implementation of masked diffusion samplers.
Authors: Elias Dubbeldam, Reza Mohammadi, Marit Schoonhoven, S. Ilker Birbil
Abstract: Despite recent progress, deep learning methods for tabular data still struggle to compete with traditional tree-based models. A key challenge lies in modeling complex, dataset-specific feature interactions that are central to tabular data. Graph-based tabular deep learning (GTDL) methods aim to address this by representing features and their interactions as graphs. However, existing methods predominantly optimize predictive accuracy, neglecting accurate modeling of the graph structure. This position paper argues that GTDL should move beyond prediction-centric objectives and prioritize the explicit learning and evaluation of feature interactions. Using synthetic datasets with known ground-truth graph structures, we show that existing GTDL methods fail to recover meaningful feature interactions. Moreover, enforcing the true interaction structure improves predictive performance. This highlights the need for GTDL methods to prioritize quantitative evaluation and accurate structural learning. We call for a shift toward structure-aware modeling as a foundation for building GTDL systems that are not only accurate but also interpretable, trustworthy, and grounded in domain understanding.
Authors: Seunghyeon Kim, Jinho Kim, Taesun Yeom, Wonpyo Park, Kyuyeun Kim, Jaeho Lee
Abstract: Transformer-based vision encoders -- such as CLIP -- are central to multimodal intelligence, powering applications from autonomous web agents to robotic control. Since these applications often demand real-time processing of massive visual data, reducing the inference cost of vision encoders is critical. Post-training quantization offers a practical path, but remains challenging even at 8-bit precision due to massive-scale activations (i.e., outliers). In this work, we propose $\textit{RegCache}$, a training-free algorithm to mitigate outliers in vision encoders, enabling quantization with significantly smaller accuracy drops. The proposed RegCache introduces outlier-prone yet semantically meaningless prefix tokens to the target vision encoder, which prevents other tokens from having outliers. Notably, we observe that outliers in vision encoders behave differently from those in language models, motivating two technical innovations: middle-layer prefixing and token deletion. Experiments show that our method consistently improves the accuracy of quantized models across both text-supervised and self-supervised vision encoders.
Authors: Jian'an Zhang
Abstract: We introduce Tail-Safe, a deployability-oriented framework for derivatives hedging that unifies distributional, risk-sensitive reinforcement learning with a white-box control-barrier-function (CBF) quadratic-program (QP) safety layer tailored to financial constraints. The learning component combines an IQN-based distributional critic with a CVaR objective (IQN--CVaR--PPO) and a Tail-Coverage Controller that regulates quantile sampling through temperature tilting and tail boosting to stabilize small-$\alpha$ estimation. The safety component enforces discrete-time CBF inequalities together with domain-specific constraints -- ellipsoidal no-trade bands, box and rate limits, and a sign-consistency gate -- solved as a convex QP whose telemetry (active sets, tightness, rate utilization, gate scores, slack, and solver status) forms an auditable trail for governance. We provide guarantees of robust forward invariance of the safe set under bounded model mismatch, a minimal-deviation projection interpretation of the QP, a KL-to-DRO upper bound linking per-state KL regularization to worst-case CVaR, concentration and sample-complexity results for the temperature-tilted CVaR estimator, and a CVaR trust-region improvement inequality under KL limits, together with feasibility persistence under expiry-aware tightening. Empirically, in arbitrage-free, microstructure-aware synthetic markets (SSVI $\to$ Dupire $\to$ VIX with ABIDES/MockLOB execution), Tail-Safe improves left-tail risk without degrading central performance and yields zero hard-constraint violations whenever the QP is feasible with zero slack. Telemetry is mapped to governance dashboards and incident workflows to support explainability and auditability. Limitations include reliance on synthetic data and simplified execution to isolate methodological contributions.
Authors: Mohsen Amiri, V Venktesh, Sindri Magn\'usson
Abstract: This paper investigates the identification of the top-m user-scheduling sets in multi-user MIMO downlink, which is cast as a combinatorial pure-exploration problem in stochastic linear bandits. Because the action space grows exponentially, exhaustive search is infeasible. We therefore adopt a linear utility model to enable efficient exploration and reliable selection of promising user subsets. We introduce a gap-index framework that maintains a shortlist of current estimates of champion arms (top-m sets) and a rotating shortlist of challenger arms that pose the greatest threat to the champions. This design focuses on measurements that yield the most informative gap-index-based comparisons, resulting in significant reductions in runtime and computation compared to state-of-the-art linear bandit methods, with high identification accuracy. The method also exposes a tunable trade-off between speed and accuracy. Simulations on a realistic OFDM downlink show that shortlist-driven pure exploration makes online, measurement-efficient subcarrier selection practical for AI-enabled communication systems.
Authors: Jinyang Jiang, Bernd Heidergott, Jiaqiao Hu, Yijie Peng
Abstract: Distortion Risk Measures (DRMs) capture risk preferences in decision-making and serve as general criteria for managing uncertainty. This paper proposes gradient descent algorithms for DRM optimization based on two dual representations: the Distortion-Measure (DM) form and Quantile-Function (QF) form. The DM-form employs a three-timescale algorithm to track quantiles, compute their gradients, and update decision variables, utilizing the Generalized Likelihood Ratio and kernel-based density estimation. The QF-form provides a simpler two-timescale approach that avoids the need for complex quantile gradient estimation. A hybrid form integrates both approaches, applying the DM-form for robust performance around distortion function jumps and the QF-form for efficiency in smooth regions. Proofs of strong convergence and convergence rates for the proposed algorithms are provided. In particular, the DM-form achieves an optimal rate of $O(k^{-4/7})$, while the QF-form attains a faster rate of $O(k^{-2/3})$. Numerical experiments confirm their effectiveness and demonstrate substantial improvements over baselines in robust portfolio selection tasks. The method's scalability is further illustrated through integration into deep reinforcement learning. Specifically, a DRM-based Proximal Policy Optimization algorithm is developed and applied to multi-echelon dynamic inventory management, showcasing its practical applicability.
Authors: Weishuo Ma, Yanbo Wang, Xiyuan Wang, Lei Zou, Muhan Zhang
Abstract: Graph Neural Networks (GNNs) are powerful tools for precessing relational data but often struggle to generalize to unseen graphs, giving rise to the development of Graph Foundational Models (GFMs). However, current GFMs are challenged by the extreme heterogeneity of graph data, where each graph can possess a unique feature space, label set, and topology. To address this, two main paradigms have emerged. The first leverages Large Language Models (LLMs), but is fundamentally text-dependent, thus struggles to handle the numerical features in vast graphs. The second pre-trains a structure-based model, but the adaptation to new tasks typically requires a costly, per-graph tuning stage, creating a critical efficiency bottleneck. In this work, we move beyond these limitations and introduce \textbf{G}raph \textbf{I}n-context \textbf{L}earning \textbf{T}ransformer (GILT), a framework built on an LLM-free and tuning-free architecture. GILT introduces a novel token-based framework for in-context learning (ICL) on graphs, reframing classification tasks spanning node, edge and graph levels in a unified framework. This mechanism is the key to handling heterogeneity, as it is designed to operate on generic numerical features. Further, its ability to understand class semantics dynamically from the context enables tuning-free adaptation. Comprehensive experiments show that GILT achieves stronger few-shot performance with significantly less time than LLM-based or tuning-based baselines, validating the effectiveness of our approach.
Authors: Haoqiang Kang, Yizhe Zhang, Nikki Lijing Kuang, Nicklas Majamaki, Navdeep Jaitly, Yi-An Ma, Lianhui Qin
Abstract: Large Language Models (LLMs) demonstrate their reasoning ability through chain-of-thought (CoT) generation. However, LLM's autoregressive decoding may limit the ability to revisit and refine earlier tokens in a holistic manner, which can also lead to inefficient exploration for diverse solutions. In this paper, we propose LaDiR (Latent Diffusion Reasoner), a novel reasoning framework that unifies the expressiveness of continuous latent representation with the iterative refinement capabilities of latent diffusion models for an existing LLM. We first construct a structured latent reasoning space using a Variational Autoencoder (VAE) that encodes text reasoning steps into blocks of thought tokens, preserving semantic information and interpretability while offering compact but expressive representations. Subsequently, we utilize a latent diffusion model that learns to denoise a block of latent thought tokens with a blockwise bidirectional attention mask, enabling longer horizon and iterative refinement with adaptive test-time compute. This design allows efficient parallel generation of diverse reasoning trajectories, allowing the model to plan and revise the reasoning process holistically. We conduct evaluations on a suite of mathematical reasoning and planning benchmarks. Empirical results show that LaDiR consistently improves accuracy, diversity, and interpretability over existing autoregressive, diffusion-based, and latent reasoning methods, revealing a new paradigm for text reasoning with latent diffusion.
Authors: Yuhta Takida, Satoshi Hayakawa, Takashi Shibuya, Masaaki Imaizumi, Naoki Murata, Bac Nguyen, Toshimitsu Uesaka, Chieh-Hsin Lai, Yuki Mitsufuji
Abstract: Deep generative models have made significant advances in generating complex content, yet conditional generation remains a fundamental challenge. Existing conditional generative adversarial networks often struggle to balance the dual objectives of assessing authenticity and conditional alignment of input samples within their conditional discriminators. To address this, we propose a novel discriminator design that integrates three key capabilities: unconditional discrimination, matching-aware supervision to enhance alignment sensitivity, and adaptive weighting to dynamically balance all objectives. Specifically, we introduce Sum of Naturalness and Alignment (SONA), which employs separate projections for naturalness (authenticity) and alignment in the final layer with an inductive bias, supported by dedicated objective functions and an adaptive weighting mechanism. Extensive experiments on class-conditional generation tasks show that \ours achieves superior sample quality and conditional alignment compared to state-of-the-art methods. Furthermore, we demonstrate its effectiveness in text-to-image generation, confirming the versatility and robustness of our approach.
Authors: Cl\'ement Bonet, Elsa Cazelles, Lucas Drumetz, Nicolas Courty
Abstract: The Busemann function has recently found much interest in a variety of geometric machine learning problems, as it naturally defines projections onto geodesic rays of Riemannian manifolds and generalizes the notion of hyperplanes. As several sources of data can be conveniently modeled as probability distributions, it is natural to study this function in the Wasserstein space, which carries a rich formal Riemannian structure induced by Optimal Transport metrics. In this work, we investigate the existence and computation of Busemann functions in Wasserstein space, which admits geodesic rays. We establish closed-form expressions in two important cases: one-dimensional distributions and Gaussian measures. These results enable explicit projection schemes for probability distributions on $\mathbb{R}$, which in turn allow us to define novel Sliced-Wasserstein distances over Gaussian mixtures and labeled datasets. We demonstrate the efficiency of those original schemes on synthetic datasets as well as transfer learning problems.
Authors: Carlo Kneissl, Christopher B\"ulte, Philipp Scholl, Gitta Kutyniok
Abstract: Probabilistic regression models the entire predictive distribution of a response variable, offering richer insights than classical point estimates and directly allowing for uncertainty quantification. While diffusion-based generative models have shown remarkable success in generating complex, high-dimensional data, their usage in general regression tasks often lacks uncertainty-related evaluation and remains limited to domain-specific applications. We propose a novel diffusion-based framework for probabilistic regression that learns predictive distributions in a nonparametric way. More specifically, we propose to model the full distribution of the diffusion noise, enabling adaptation to diverse tasks and enhanced uncertainty quantification. We investigate different noise parameterizations, analyze their trade-offs, and evaluate our framework across a broad range of regression tasks, covering low- and high-dimensional settings. For several experiments, our approach shows superior performance against existing baselines, while delivering calibrated uncertainty estimates, demonstrating its versatility as a tool for probabilistic prediction.
Authors: Alexandre Galashov, Natha\"el Da Costa, Liyuan Xu, Philipp Hennig, Arthur Gretton
Abstract: Neural networks are typically optimized with variants of stochastic gradient descent. Under a squared loss, however, the optimal solution to the linear last layer weights is known in closed-form. We propose to leverage this during optimization, treating the last layer as a function of the backbone parameters, and optimizing solely for these parameters. We show this is equivalent to alternating between gradient descent steps on the backbone and closed-form updates on the last layer. We adapt the method for the setting of stochastic gradient descent, by trading off the loss on the current batch against the accumulated information from previous batches. Further, we prove that, in the Neural Tangent Kernel regime, convergence of this method to an optimal solution is guaranteed. Finally, we demonstrate the effectiveness of our approach compared with standard SGD on a squared loss in several supervised tasks -- both regression and classification -- including Fourier Neural Operators and Instrumental Variable Regression.
Authors: Qizheng Zhang, Changran Hu, Shubhangi Upasani, Boyuan Ma, Fenglu Hong, Vamsidhar Kamanuru, Jay Rainton, Chen Wu, Mengmeng Ji, Hanchen Li, Urmish Thakker, James Zou, Kunle Olukotun
Abstract: Large language model (LLM) applications such as agents and domain-specific reasoning increasingly rely on context adaptation -- modifying inputs with instructions, strategies, or evidence, rather than weight updates. Prior approaches improve usability but often suffer from brevity bias, which drops domain insights for concise summaries, and from context collapse, where iterative rewriting erodes details over time. Building on the adaptive memory introduced by Dynamic Cheatsheet, we introduce ACE (Agentic Context Engineering), a framework that treats contexts as evolving playbooks that accumulate, refine, and organize strategies through a modular process of generation, reflection, and curation. ACE prevents collapse with structured, incremental updates that preserve detailed knowledge and scale with long-context models. Across agent and domain-specific benchmarks, ACE optimizes contexts both offline (e.g., system prompts) and online (e.g., agent memory), consistently outperforming strong baselines: +10.6% on agents and +8.6% on finance, while significantly reducing adaptation latency and rollout cost. Notably, ACE could adapt effectively without labeled supervision and instead by leveraging natural execution feedback. On the AppWorld leaderboard, ACE matches the top-ranked production-level agent on the overall average and surpasses it on the harder test-challenge split, despite using a smaller open-source model. These results show that comprehensive, evolving contexts enable scalable, efficient, and self-improving LLM systems with low overhead.
Authors: Youngjoon Lee, Seongmin Cho, Yehhyun Jo, Jinu Gong, Hyunjoo Jenny Lee, Joonhyuk Kang
Abstract: The limited data availability due to strict privacy regulations and significant resource demands severely constrains biomedical time-series AI development, which creates a critical gap between data requirements and accessibility. Synthetic data generation presents a promising solution by producing artificial datasets that maintain the statistical properties of real biomedical time-series data without compromising patient confidentiality. We propose a framework for synthetic biomedical time-series data generation based on advanced forecasting models that accurately replicates complex electrophysiological signals such as EEG and EMG with high fidelity. These synthetic datasets preserve essential temporal and spectral properties of real data, which enables robust analysis while effectively addressing data scarcity and privacy challenges. Our evaluations across multiple subjects demonstrate that the generated synthetic data can serve as an effective substitute for real data and also significantly boost AI model performance. The approach maintains critical biomedical features while provides high scalability for various applications and integrates seamlessly into open-source repositories, substantially expanding resources for AI-driven biomedical research.
Authors: Mohamed Ayoub Ben Ayad, Michael Dinzinger, Kanishka Ghosh Dastidar, Jelena Mitrovic, Michael Granitzer
Abstract: Embedding models are central to dense retrieval, semantic search, and recommendation systems, but their size often makes them impractical to deploy in resource-constrained environments such as browsers or edge devices. While smaller embedding models offer practical advantages, they typically underperform compared to their larger counterparts. To bridge this gap, we demonstrate that concatenating the raw embedding vectors of multiple small models can outperform a single larger baseline on standard retrieval benchmarks. To overcome the resulting high dimensionality of naive concatenation, we introduce a lightweight unified decoder trained with a Matryoshka Representation Learning (MRL) loss. This decoder maps the high-dimensional joint representation to a low-dimensional space, preserving most of the original performance without fine-tuning the base models. We also show that while concatenating more base models yields diminishing gains, the robustness of the decoder's representation under compression and quantization improves. Our experiments show that, on a subset of MTEB retrieval tasks, our concat-encode-quantize pipeline recovers 89\% of the original performance with a 48x compression factor when the pipeline is applied to a concatenation of four small embedding models.
Authors: Johanna Sommer, John Rachwan, Nils Fleischmann, Stephan G\"unnemann, Bertrand Charpentier
Abstract: Flow matching models generate high-fidelity molecular geometries but incur significant computational costs during inference, requiring hundreds of network evaluations. This inference overhead becomes the primary bottleneck when such models are employed in practice to sample large numbers of molecular candidates. This work discusses a training-free caching strategy that accelerates molecular geometry generation by predicting intermediate hidden states across solver steps. The proposed method operates directly on the SE(3)-equivariant backbone, is compatible with pretrained models, and is orthogonal to existing training-based accelerations and system-level optimizations. Experiments on the GEOM-Drugs dataset demonstrate that caching achieves a twofold reduction in wall-clock inference time at matched sample quality and a speedup of up to 3x compared to the base model with minimal sample quality degradation. Because these gains compound with other optimizations, applying caching alongside other general, lossless optimizations yield as much as a 7x speedup.
Authors: Yuandou Wang, Filip Gunnarsson, Rihan Hai
Abstract: Tabular data streams are rapidly emerging as a dominant modality for real-time decision-making in healthcare, finance, and the Internet of Things (IoT). These applications commonly run on edge and mobile devices, where energy budgets, memory, and compute are strictly limited. Continual learning (CL) addresses such dynamics by training models sequentially on task streams while preserving prior knowledge and consolidating new knowledge. While recent CL work has advanced in mitigating catastrophic forgetting and improving knowledge transfer, the practical requirements of energy and memory efficiency for tabular data streams remain underexplored. In particular, existing CL solutions mostly depend on replay mechanisms whose buffers grow over time and exacerbate resource costs. We propose a context-aware incremental Multi-Layer Perceptron (IMLP), a compact continual learner for tabular data streams. IMLP incorporates a windowed scaled dot-product attention over a sliding latent feature buffer, enabling constant-size memory and avoiding storing raw data. The attended context is concatenated with current features and processed by shared feed-forward layers, yielding lightweight per-segment updates. To assess practical deployability, we introduce NetScore-T, a tunable metric coupling balanced accuracy with energy for Pareto-aware comparison across models and datasets. IMLP achieves up to $27.6\times$ higher energy efficiency than TabNet and $85.5\times$ higher than TabPFN, while maintaining competitive average accuracy. Overall, IMLP provides an easy-to-deploy, energy-efficient alternative to full retraining for tabular data streams.
Authors: Fanzhe Fu, Yang Yang
Abstract: Reversible Instance Normalization (RevIN) is a key technique enabling simple linear models to achieve state-of-the-art performance in time series forecasting. While replacing its non-robust statistics with robust counterparts (termed R$^2$-IN) seems like a straightforward improvement, our findings reveal a far more complex reality. This paper deconstructs the perplexing performance of various normalization strategies by identifying four underlying theoretical contradictions. Our experiments provide two crucial findings: first, the standard RevIN catastrophically fails on datasets with extreme outliers, where its MSE surges by a staggering 683\%. Second, while the simple R$^2$-IN prevents this failure and unexpectedly emerges as the best overall performer, our adaptive model (A-IN), designed to test a diagnostics-driven heuristic, unexpectedly suffers a complete and systemic failure. This surprising outcome uncovers a critical, overlooked pitfall in time series analysis: the instability introduced by a simple or counter-intuitive heuristic can be more damaging than the statistical issues it aims to solve. The core contribution of this work is thus a new, cautionary paradigm for time series normalization: a shift from a blind search for complexity to a diagnostics-driven analysis that reveals not only the surprising power of simple baselines but also the perilous nature of naive adaptation.
Authors: Lorenzo Pannacci, Simone Fiorellino, Mario Edoardo Pandolfo, Emilio Calvanese Strinati, Paolo Di Lorenzo
Abstract: Deep joint source-channel coding (DeepJSCC) has emerged as a powerful paradigm for end-to-end semantic communications, jointly learning to compress and protect task-relevant features over noisy channels. However, existing DeepJSCC schemes assume a shared latent space at transmitter (TX) and receiver (RX) - an assumption that fails in multi-vendor deployments where encoders and decoders cannot be co-trained. This mismatch introduces "semantic noise", degrading reconstruction quality and downstream task performance. In this paper, we systematize and evaluate methods for semantic channel equalization for DeepJSCC, introducing an additional processing stage that aligns heterogeneous latent spaces under both physical and semantic impairments. We investigate three classes of aligners: (i) linear maps, which admit closed-form solutions; (ii) lightweight neural networks, offering greater expressiveness; and (iii) a Parseval-frame equalizer, which operates in zero-shot mode without the need for training. Through extensive experiments on image reconstruction over AWGN and fading channels, we quantify trade-offs among complexity, data efficiency, and fidelity, providing guidelines for deploying DeepJSCC in heterogeneous AI-native wireless networks.
Authors: Qiyu Wei, Haowei Wang, Richard Allmendinger, Mauricio A. \'Alvarez
Abstract: Bayesian optimization has emerged as a prominent methodology for optimizing expensive black-box functions by leveraging Gaussian process surrogates, which focus on capturing the global characteristics of the objective function. However, in numerous practical scenarios, the primary objective is not to construct an exhaustive global surrogate, but rather to quickly pinpoint the global optimum. Due to the aleatoric nature of the sequential optimization problem and its dependence on the quality of the surrogate model and the initial design, it is restrictive to assume that all observed samples contribute equally to the discovery of the optimum in this context. In this paper, we introduce Counterfactual Credit Guided Bayesian Optimization (CCGBO), a novel framework that explicitly quantifies the contribution of individual historical observations through counterfactual credit. By incorporating counterfactual credit into the acquisition function, our approach can selectively allocate resources in areas where optimal solutions are most likely to occur. We prove that CCGBO retains sublinear regret. Empirical evaluations on various synthetic and real-world benchmarks demonstrate that CCGBO consistently reduces simple regret and accelerates convergence to the global optimum.
Authors: Shuche Wang, Adarsh Barik, Peng Zhao, Vincent Y. F. Tan
Abstract: We develop the first parameter-free algorithms for the Stochastically Extended Adversarial (SEA) model, a framework that bridges adversarial and stochastic online convex optimization. Existing approaches for the SEA model require prior knowledge of problem-specific parameters, such as the diameter of the domain $D$ and the Lipschitz constant of the loss functions $G$, which limits their practical applicability. Addressing this, we develop parameter-free methods by leveraging the Optimistic Online Newton Step (OONS) algorithm to eliminate the need for these parameters. We first establish a comparator-adaptive algorithm for the scenario with unknown domain diameter but known Lipschitz constant, achieving an expected regret bound of $\tilde{O}\big(\|u\|_2^2 + \|u\|_2(\sqrt{\sigma^2_{1:T}} + \sqrt{\Sigma^2_{1:T}})\big)$, where $u$ is the comparator vector and $\sigma^2_{1:T}$ and $\Sigma^2_{1:T}$ represent the cumulative stochastic variance and cumulative adversarial variation, respectively. We then extend this to the more general setting where both $D$ and $G$ are unknown, attaining the comparator- and Lipschitz-adaptive algorithm. Notably, the regret bound exhibits the same dependence on $\sigma^2_{1:T}$ and $\Sigma^2_{1:T}$, demonstrating the efficacy of our proposed methods even when both parameters are unknown in the SEA model.
Authors: Chenxiang Zhang, Alexander Theus, Damien Teney, Antonio Orvieto, Jun Pang, Sjouke Mauw
Abstract: Model merging methods combine models with different capabilities into a single one while maintaining the same inference cost. Two popular approaches are linear interpolation, which linearly interpolates between model weights, and task arithmetic, which combines task vectors obtained by the difference between finetuned and base models. While useful in practice, what properties make merging effective are poorly understood. This paper explores how the optimization process affects the loss landscape geometry and its impact on merging success. We show that a single quantity -- the effective noise scale -- unifies the impact of optimizer and data choices on model merging. Across architectures and datasets, the effectiveness of merging success is a non-monotonic function of effective noise, with a distinct optimum. Decomposing this quantity, we find that larger learning rates, stronger weight decay, smaller batch sizes, and data augmentation all independently modulate the effective noise scale, exhibiting the same qualitative trend. Unlike prior work that connects optimizer noise to the flatness or generalization of individual minima, we show that it also affects the global loss landscape, predicting when independently trained solutions can be merged. Our findings broaden the understanding of how optimization shapes the loss landscape geometry and its downstream consequences for model merging, suggesting the possibility of further manipulating the training dynamics to improve merging effectiveness.
Authors: Zexin Wang, Changhua Pei, Yang Liu, Hengyue Jiang, Quan Zhou, Haotian Si, Hang Cui, Jianhui Li, Gaogang Xie, Jingjing Li, Dan Pei
Abstract: Web service administrators must ensure the stability of multiple systems by promptly detecting anomalies in Key Performance Indicators (KPIs). Achieving the goal of "train once, infer across scenarios" remains a fundamental challenge for time series anomaly detection models. Beyond improving zero-shot generalization, such models must also flexibly handle sequences of varying lengths during inference, ranging from one hour to one week, without retraining. Conventional approaches rely on sliding-window encoding and self-supervised learning, which restrict inference to fixed-length inputs. Large Language Models (LLMs) have demonstrated remarkable zero-shot capabilities across general domains. However, when applied to time series data, they face inherent limitations due to context length. To address this issue, we propose ViTs, a Vision-Language Model (VLM)-based framework that converts time series curves into visual representations. By rescaling time series images, temporal dependencies are preserved while maintaining a consistent input size, thereby enabling efficient processing of arbitrarily long sequences without context constraints. Training VLMs for this purpose introduces unique challenges, primarily due to the scarcity of aligned time series image-text data. To overcome this, we employ an evolutionary algorithm to automatically generate thousands of high-quality image-text pairs and design a three-stage training pipeline consisting of: (1) time series knowledge injection, (2) anomaly detection enhancement, and (3) anomaly reasoning refinement. Extensive experiments demonstrate that ViTs substantially enhance the ability of VLMs to understand and detect anomalies in time series data. All datasets and code will be publicly released at: https://anonymous.4open.science/r/ViTs-C484/.
Authors: Emanuele Mule, Stefano Fiorini, Antonio Purificato, Federico Siciliano, Stefano Coniglio, Fabrizio Silvestri
Abstract: Hypergraphs provide a natural way to represent higher-order interactions among multiple entities. While undirected hypergraphs have been extensively studied, the case of directed hypergraphs, which can model oriented group interactions, remains largely under-explored despite its relevance for many applications. Recent approaches in this direction often exhibit an implicit bias toward homophily, which limits their effectiveness in heterophilic settings. Rooted in the algebraic topology notion of Cellular Sheaves, Sheaf Neural Networks (SNNs) were introduced as an effective solution to circumvent such a drawback. While a generalization to hypergraphs is known, it is only suitable for undirected hypergraphs, failing to tackle the directed case. In this work, we introduce Directional Sheaf Hypergraph Networks (DSHN), a framework integrating sheaf theory with a principled treatment of asymmetric relations within a hypergraph. From it, we construct the Directed Sheaf Hypergraph Laplacian, a complex-valued operator by which we unify and generalize many existing Laplacian matrices proposed in the graph- and hypergraph-learning literature. Across 7 real-world datasets and against 13 baselines, DSHN achieves relative accuracy gains from 2% up to 20%, showing how a principled treatment of directionality in hypergraphs, combined with the expressive power of sheaves, can substantially improve performance.
Authors: Mehrasa Ahmadipour, Aur\'elien Garivier
Abstract: We study the fixed-confidence best arm identification (BAI) problem within the multi-armed bandit (MAB) framework under the Entropic Value-at-Risk (EVaR) criterion. Our analysis considers a nonparametric setting, allowing for general reward distributions bounded in [0,1]. This formulation addresses the critical need for risk-averse decision-making in high-stakes environments, such as finance, moving beyond simple expected value optimization. We propose a $\delta$-correct, Track-and-Stop based algorithm and derive a corresponding lower bound on the expected sample complexity, which we prove is asymptotically matched. The implementation of our algorithm and the characterization of the lower bound both require solving a complex convex optimization problem and a related, simpler non-convex one.
Authors: Zhiwei Han, Stefan Matthes, Hao Shen
Abstract: In this work, we establish conditions under which nonlinear CCA recovers the ground-truth latent factors up to an orthogonal transform after whitening. Building on the classical result that linear mappings maximize canonical correlations under Gaussian priors, we prove affine identifiability for a broad class of latent distributions in the population setting. Central to our proof is a reparameterization result that transports the analysis from observation space to source space, where identifiability becomes tractable. We further show that whitening is essential for ensuring boundedness and well-conditioning, thereby underpinning identifiability. Beyond the population setting, we prove that ridge-regularized empirical CCA converges to its population counterpart, transferring these guarantees to the finite-sample regime. Experiments on a controlled synthetic dataset and a rendered image dataset validate our theory and demonstrate the necessity of its assumptions through systematic ablations.
Authors: Wonjun Kang, Kevin Galim, Seunghyuk Oh, Minjae Lee, Yuchen Zeng, Shuibai Zhang, Coleman Hooper, Yuezhou Hu, Hyung Il Koo, Nam Ik Cho, Kangwook Lee
Abstract: While most autoregressive LLMs are constrained to one-by-one decoding, diffusion LLMs (dLLMs) have attracted growing interest for their potential to dramatically accelerate inference through parallel decoding. Despite this promise, the conditional independence assumption in dLLMs causes parallel decoding to ignore token dependencies, inevitably degrading generation quality when these dependencies are strong. However, existing works largely overlook these inherent challenges, and evaluations on standard benchmarks (e.g., math and coding) are not sufficient to capture the quality degradation caused by parallel decoding. To address this gap, we first provide an information-theoretic analysis of parallel decoding. We then conduct case studies on analytically tractable synthetic list operations from both data distribution and decoding strategy perspectives, offering quantitative insights that highlight the fundamental limitations of parallel decoding. Building on these insights, we propose ParallelBench, the first benchmark specifically designed for dLLMs, featuring realistic tasks that are trivial for humans and autoregressive LLMs yet exceptionally challenging for dLLMs under parallel decoding. Using ParallelBench, we systematically analyze both dLLMs and autoregressive LLMs, revealing that: (i) dLLMs under parallel decoding can suffer dramatic quality degradation in real-world scenarios, and (ii) current parallel decoding strategies struggle to adapt their degree of parallelism based on task difficulty, thus failing to achieve meaningful speedup without compromising quality. Our findings underscore the pressing need for innovative decoding methods that can overcome the current speed-quality trade-off. We release our benchmark to help accelerate the development of truly efficient dLLMs.
Authors: Michele Caprio, Siu Lun Chau, Krikamol Muandet
Abstract: Many machine learning algorithms rely on iterative updates of uncertainty representations, ranging from variational inference and expectation-maximization, to reinforcement learning, continual learning, and multi-agent learning. In the presence of imprecision and ambiguity, credal sets -- closed, convex sets of probability distributions -- have emerged as a popular framework for representing imprecise probabilistic beliefs. Under such imprecision, many learning problems in imprecise probabilistic machine learning (IPML) may be viewed as processes involving successive applications of update rules on credal sets. This naturally raises the question of whether this iterative process converges to stable fixed points -- or, more generally, under what conditions on the updating mechanism such fixed points exist, and whether they can be attained. We provide the first analysis of this problem and illustrate our findings using Credal Bayesian Deep Learning as a concrete example. Our work demonstrates that incorporating imprecision into the learning process not only enriches the representation of uncertainty, but also reveals structural conditions under which stability emerges, thereby offering new insights into the dynamics of iterative learning under imprecision.
Authors: Kai Qin, Jiaqi Wu, Jianxiang He, Haoyuan Sun, Yifei Zhao, Bin Liang, Yongzhe Chang, Tiantian Zhang, Houde Liu
Abstract: As Large Language Models (LLMs) demonstrate remarkable capabilities learned from vast corpora, concerns regarding data privacy and safety are receiving increasing attention. LLM unlearning, which aims to remove the influence of specific data while preserving overall model utility, is becoming an important research area. One of the mainstream unlearning classes is optimization-based methods, which achieve forgetting directly through fine-tuning, exemplified by Negative Preference Optimization (NPO). However, NPO's effectiveness is limited by its inherent lack of explicit positive preference signals. Attempts to introduce such signals by constructing preferred responses often necessitate domain-specific knowledge or well-designed prompts, fundamentally restricting their generalizability. In this paper, we shift the focus to the distribution-level, directly targeting the next-token probability distribution instead of entire responses, and derive a novel unlearning algorithm termed \textbf{Di}stribution \textbf{P}reference \textbf{O}ptimization (DiPO). We show that the requisite preference distribution pairs for DiPO, which are distributions over the model's output tokens, can be constructed by selectively amplifying or suppressing the model's high-confidence output logits, thereby effectively overcoming NPO's limitations. We theoretically prove the consistency of DiPO's loss function with the desired unlearning direction. Extensive experiments demonstrate that DiPO achieves a strong trade-off between model utility and forget quality. Notably, DiPO attains the highest forget quality on the TOFU benchmark, and maintains leading scalability and sustainability in utility preservation on the MUSE benchmark.
Authors: Ebenezer Awotoro, Chisom Ezekannagha, Florian Schwarz, Johannes Tauscher, Dominik Heider, Katharina Ladewig, Christel Le Bon, Karine Moncoq, Bruno Miroux, Georges Hattab
Abstract: Structural biology has made significant progress in determining membrane proteins, leading to a remarkable increase in the number of available structures in dedicated databases. The inherent complexity of membrane protein structures, coupled with challenges such as missing data, inconsistencies, and computational barriers from disparate sources, underscores the need for improved database integration. To address this gap, we present MetaMP, a framework that unifies membrane-protein databases within a web application and uses machine learning for classification. MetaMP improves data quality by enriching metadata, offering a user-friendly interface, and providing eight interactive views for streamlined exploration. MetaMP was effective across tasks of varying difficulty, demonstrating advantages across different levels without compromising speed or accuracy, according to user evaluations. Moreover, MetaMP supports essential functions such as structure classification and outlier detection. We present three practical applications of Artificial Intelligence (AI) in membrane protein research: predicting transmembrane segments, reconciling legacy databases, and classifying structures with explainable AI support. In a validation focused on statistics, MetaMP resolved 77% of data discrepancies and accurately predicted the class of newly identified membrane proteins 98% of the time and overtook expert curation. Altogether, MetaMP is a much-needed resource that harmonizes current knowledge and empowers AI-driven exploration of membrane-protein architecture.
Authors: Jonas H\"ubotter, Leander Diaz-Bone, Ido Hakimi, Andreas Krause, Moritz Hardt
Abstract: Humans are good at learning on the job: We learn how to solve the tasks we face as we go along. Can a model do the same? We propose an agent that assembles a task-specific curriculum, called test-time curriculum (TTC-RL), and applies reinforcement learning to continue training the model for its target task. The test-time curriculum avoids time-consuming human curation of datasets by automatically selecting the most task-relevant data from a large pool of available training data. Our experiments demonstrate that reinforcement learning on a test-time curriculum consistently improves the model on its target tasks, across a variety of evaluations and models. Notably, on challenging math and coding benchmarks, TTC-RL improves the pass@1 of Qwen3-8B by approximately 1.8x on AIME25 and 2.1x on CodeElo. Moreover, we find that TTC-RL significantly raises the performance ceiling compared to the initial model, increasing pass@8 on AIME25 from 40% to 62% and on CodeElo from 28% to 43%. Our findings show the potential of test-time curricula in extending the test-time scaling paradigm to continual training on thousands of task-relevant experiences during test-time.
Authors: Junhyung Ahn, Sanghack Lee
Abstract: Accurately predicting conversion rate (CVR) is essential in various recommendation domains such as online advertising systems and e-commerce. These systems utilize user interaction logs, which consist of exposures, clicks, and conversions. CVR prediction models are typically trained solely based on clicked samples, as conversions can only be determined following clicks. However, the sparsity of clicked instances necessitates the collection of a substantial amount of logs for effective model training. Recent works address this issue by devising frameworks that leverage non-clicked samples. While these frameworks aim to reduce biases caused by the discrepancy between clicked and non-clicked samples, they often rely on heuristics. Against this background, we propose a method to counterfactually generate conversion labels for non-clicked samples by using causality as a guiding principle, attempting to answer the question, "Would the user have converted if he or she had clicked the recommended item?" Our approach is named the Entire Space Counterfactual Inference Multi-task Model (ESCIM). We initially train a structural causal model (SCM) of user sequential behaviors and conduct a hypothetical intervention (i.e., click) on non-clicked items to infer counterfactual CVRs. We then introduce several approaches to transform predicted counterfactual CVRs into binary counterfactual conversion labels for the non-clicked samples. Finally, the generated samples are incorporated into the training process. Extensive experiments on public datasets illustrate the superiority of the proposed algorithm. Online A/B testing further empirically validates the effectiveness of our proposed algorithm in real-world scenarios. In addition, we demonstrate the improved performance of the proposed method on latent conversion data, showcasing its robustness and superior generalization capabilities.
Authors: Idan Attias, Lev Reyzin, Nathan Srebro, Gal Vardi
Abstract: Despite the theoretical significance and wide practical use of regular expressions, the computational complexity of learning them has been largely unexplored. We study the computational hardness of improperly learning regular expressions in the PAC model and with membership queries. We show that PAC learning is hard even under the uniform distribution on the hypercube, and also prove hardness of distribution-free learning with membership queries. Furthermore, if regular expressions are extended with complement or intersection, we establish hardness of learning with membership queries even under the uniform distribution. We emphasize that these results do not follow from existing hardness results for learning DFAs or NFAs, since the descriptive complexity of regular languages can differ exponentially between DFAs, NFAs, and regular expressions.
Authors: Guillaume Godin
Abstract: Bond Centered FingerPrint (BCFP) are a complementary, bond-centric alternative to Extended-Connectivity Fingerprints (ECFP). We introduce a static BCFP that mirrors the bond-convolution used by directed message-passing GNNs like ChemProp, and evaluate it with a fast rapid Random Forest model on Brain-Blood Barrier Penetration (BBBP) classification task. Across stratified cross-validation, concatenating ECFP with BCFP consistently improves AUROC and AUPRC over either descriptor alone, as confirmed by Turkey HSD multiple-comparison analysis. Among radii, r = 1 performs best; r = 2 does not yield statistically separable gains under the same test. We further propose BCFP-Sort&Slice, a simple feature-combination scheme that preserves the out-of-vocabulary (OOV) count information native to ECFP count vectors while enabling compact unhashed concatenation of BCFP variants. We also outperform the MGTP prediction on our BBBP evaluation, using such composite new features bond and atom features. These results show that lightweight, bond-centered descriptors can complement atom-centered circular fingerprints and provide strong, fast baselines for BBBP prediction.
Authors: Yorgos Felekis, Theodoros Damoulas, Paris Giampouras
Abstract: Causal Abstraction (CA) theory provides a principled framework for relating causal models that describe the same system at different levels of granularity while ensuring interventional consistency between them. Recently, several approaches for learning CAs have been proposed, but all assume fixed and well-specified exogenous distributions, making them vulnerable to environmental shifts and misspecification. In this work, we address these limitations by introducing the first class of distributionally robust CAs and their associated learning algorithms. The latter cast robust causal abstraction learning as a constrained min-max optimization problem with Wasserstein ambiguity sets. We provide theoretical results, for both empirical and Gaussian environments, leading to principled selection of the level of robustness via the radius of these sets. Furthermore, we present empirical evidence across different problems and CA learning methods, demonstrating our framework's robustness not only to environmental shifts but also to structural model and intervention mapping misspecification.
Authors: Junqi Jiang, Francesco Leofante, Antonio Rago, Francesca Toni
Abstract: Counterfactual explanations (CEs) provide recourse recommendations for individuals affected by algorithmic decisions. A key challenge is generating CEs that are robust against various perturbation types (e.g. input and model perturbations) while simultaneously satisfying other desirable properties. These include plausibility, ensuring CEs reside on the data manifold, and diversity, providing multiple distinct recourse options for single inputs. Existing methods, however, mostly struggle to address these multifaceted requirements in a unified, model-agnostic manner. We address these limitations by proposing a novel generative framework. First, we introduce the Label-conditional Gaussian Mixture Variational Autoencoder (L-GMVAE), a model trained to learn a structured latent space where each class label is represented by a set of Gaussian components with diverse, prototypical centroids. Building on this, we present LAPACE (LAtent PAth Counterfactual Explanations), a model-agnostic algorithm that synthesises entire paths of CE points by interpolating from inputs' latent representations to those learned latent centroids. This approach inherently ensures robustness to input changes, as all paths for a given target class converge to the same fixed centroids. Furthermore, the generated paths provide a spectrum of recourse options, allowing users to navigate the trade-off between proximity and plausibility while also encouraging robustness against model changes. In addition, user-specified actionability constraints can also be easily incorporated via lightweight gradient optimisation through the L-GMVAE's decoder. Comprehensive experiments show that LAPACE is computationally efficient and achieves competitive performance across eight quantitative metrics.
Authors: Siwei Han, Jiaqi Liu, Yaofeng Su, Wenbo Duan, Xinyuan Liu, Cihang Xie, Mohit Bansal, Mingyu Ding, Linjun Zhang, Huaxiu Yao
Abstract: As Large Language Model (LLM) agents increasingly gain self-evolutionary capabilities to adapt and refine their strategies through real-world interaction, their long-term reliability becomes a critical concern. We identify the Alignment Tipping Process (ATP), a critical post-deployment risk unique to self-evolving LLM agents. Unlike training-time failures, ATP arises when continual interaction drives agents to abandon alignment constraints established during training in favor of reinforced, self-interested strategies. We formalize and analyze ATP through two complementary paradigms: Self-Interested Exploration, where repeated high-reward deviations induce individual behavioral drift, and Imitative Strategy Diffusion, where deviant behaviors spread across multi-agent systems. Building on these paradigms, we construct controllable testbeds and benchmark Qwen3-8B and Llama-3.1-8B-Instruct. Our experiments show that alignment benefits erode rapidly under self-evolution, with initially aligned models converging toward unaligned states. In multi-agent settings, successful violations diffuse quickly, leading to collective misalignment. Moreover, current reinforcement learning-based alignment methods provide only fragile defenses against alignment tipping. Together, these findings demonstrate that alignment of LLM agents is not a static property but a fragile and dynamic one, vulnerable to feedback-driven decay during deployment. Our data and code are available at https://github.com/aiming-lab/ATP.
Authors: Zihan Zhao, Fengtao Zhou, Ronggang Li, Bing Chu, Xinke Zhang, Xueyi Zheng, Ke Zheng, Xiaobo Wen, Jiabo Ma, Yihui Wang, Jiewei Chen, Chengyou Zheng, Jiangyu Zhang, Yongqin Wen, Jiajia Meng, Ziqi Zeng, Xiaoqing Li, Jing Li, Dan Xie, Yaping Ye, Yu Wang, Hao Chen, Muyan Cai
Abstract: Intraoperative pathology is pivotal to precision surgery, yet its clinical impact is constrained by diagnostic complexity and the limited availability of high-quality frozen-section data. While computational pathology has made significant strides, the lack of large-scale, prospective validation has impeded its routine adoption in surgical workflows. Here, we introduce CRISP, a clinical-grade foundation model developed on over 100,000 frozen sections from eight medical centers, specifically designed to provide Clinical-grade Robust Intraoperative Support for Pathology (CRISP). CRISP was comprehensively evaluated on more than 15,000 intraoperative slides across nearly 100 retrospective diagnostic tasks, including benign-malignant discrimination, key intraoperative decision-making, and pan-cancer detection, etc. The model demonstrated robust generalization across diverse institutions, tumor types, and anatomical sites-including previously unseen sites and rare cancers. In a prospective cohort of over 2,000 patients, CRISP sustained high diagnostic accuracy under real-world conditions, directly informing surgical decisions in 92.6% of cases. Human-AI collaboration further reduced diagnostic workload by 35%, avoided 105 ancillary tests and enhanced detection of micrometastases with 87.5% accuracy. Together, these findings position CRISP as a clinical-grade paradigm for AI-driven intraoperative pathology, bridging computational advances with surgical precision and accelerating the translation of artificial intelligence into routine clinical practice.
Authors: Alexia Jolicoeur-Martineau
Abstract: Hierarchical Reasoning Model (HRM) is a novel approach using two small neural networks recursing at different frequencies. This biologically inspired method beats Large Language models (LLMs) on hard puzzle tasks such as Sudoku, Maze, and ARC-AGI while trained with small models (27M parameters) on small data (around 1000 examples). HRM holds great promise for solving hard problems with small networks, but it is not yet well understood and may be suboptimal. We propose Tiny Recursive Model (TRM), a much simpler recursive reasoning approach that achieves significantly higher generalization than HRM, while using a single tiny network with only 2 layers. With only 7M parameters, TRM obtains 45% test-accuracy on ARC-AGI-1 and 8% on ARC-AGI-2, higher than most LLMs (e.g., Deepseek R1, o3-mini, Gemini 2.5 Pro) with less than 0.01% of the parameters.
Authors: Xiangyang Xu, Hongyang Gao
Abstract: Low-energy molecular conformers generation (MCG) is a foundational yet challenging problem in drug discovery. Denoising-based methods include diffusion and flow-matching methods that learn mappings from a simple base distribution to the molecular conformer distribution. However, these approaches often suffer from error accumulation during sampling, especially in the low SNR steps, which are hard to train. To address these challenges, we propose a flow-matching refiner for the MCG task. The proposed method initializes sampling from mixed-quality outputs produced by upstream denoising models and reschedules the noise scale to bypass the low-SNR phase, thereby improving sample quality. On the GEOM-QM9 and GEOM-Drugs benchmark datasets, the generator-refiner pipeline improves quality with fewer total denoising steps while preserving diversity.
Authors: Alina Ermilova, Dmitrii Kornilov, Sofia Samoilova, Ekaterina Laptenkova, Anastasia Kolesnikova, Ekaterina Podplutova, Senotrusova Sofya, Maksim G. Sharaev
Abstract: Identifying disease interconnections through manual analysis of large-scale clinical data is labor-intensive, subjective, and prone to expert disagreement. While machine learning (ML) shows promise, three critical challenges remain: (1) selecting optimal methods from the vast ML landscape, (2) determining whether real-world clinical data (e.g., electronic health records, EHRs) or structured disease descriptions yield more reliable insights, (3) the lack of "ground truth," as some disease interconnections remain unexplored in medicine. Large language models (LLMs) demonstrate broad utility, yet they often lack specialized medical knowledge. To address these gaps, we conduct a systematic evaluation of seven approaches for uncovering disease relationships based on two data sources: (i) sequences of ICD-10 codes from MIMIC-IV EHRs and (ii) the full set of ICD-10 codes, both with and without textual descriptions. Our framework integrates the following: (i) a statistical co-occurrence analysis and a masked language modeling (MLM) approach using real clinical data; (ii) domain-specific BERT variants (Med-BERT and BioClinicalBERT); (iii) a general-purpose BERT and document retrieval; and (iv) four LLMs (Mistral, DeepSeek, Qwen, and YandexGPT). Our graph-based comparison of the obtained interconnection matrices shows that the LLM-based approach produces interconnections with the lowest diversity of ICD code connections to different diseases compared to other methods, including text-based and domain-based approaches. This suggests an important implication: LLMs have limited potential for discovering new interconnections. In the absence of ground truth databases for medical interconnections between ICD codes, our results constitute a valuable medical disease ontology that can serve as a foundational resource for future clinical research and artificial intelligence applications in healthcare.
Authors: Nick Jan{\ss}en, Melanie Schaller, Bodo Rosenhahn
Abstract: Understanding the robustness of deep learning models for multivariate long-term time series forecasting (M-LTSF) remains challenging, as evaluations typically rely on real-world datasets with unknown noise properties. We propose a simulation-based evaluation framework that generates parameterizable synthetic datasets, where each dataset instance corresponds to a different configuration of signal components, noise types, signal-to-noise ratios, and frequency characteristics. These configurable components aim to model real-world multivariate time series data without the ambiguity of unknown noise. This framework enables fine-grained, systematic evaluation of M-LTSF models under controlled and diverse scenarios. We benchmark four representative architectures S-Mamba (state-space), iTransformer (transformer-based), R-Linear (linear), and Autoformer (decomposition-based). Our analysis reveals that all models degrade severely when lookback windows cannot capture complete periods of seasonal patters in the data. S-Mamba and Autoformer perform best on sawtooth patterns, while R-Linear and iTransformer favor sinusoidal signals. White and Brownian noise universally degrade performance with lower signal-to-noise ratio while S-Mamba shows specific trend-noise and iTransformer shows seasonal-noise vulnerability. Further spectral analysis shows that S-Mamba and iTransformer achieve superior frequency reconstruction. This controlled approach, based on our synthetic and principle-driven testbed, offers deeper insights into model-specific strengths and limitations through the aggregation of MSE scores and provides concrete guidance for model selection based on signal characteristics and noise conditions.
Authors: Jonathan Cola\c{c}o Carr, Qinyi Sun, Cameron Allen
Abstract: Skills are essential for unlocking higher levels of problem solving. A common approach to discovering these skills is to learn ones that reliably reach different states, thus empowering the agent to control its environment. However, existing skill discovery algorithms often overlook the natural state variables present in many reinforcement learning problems, meaning that the discovered skills lack control of specific state variables. This can significantly hamper exploration efficiency, make skills more challenging to learn with, and lead to negative side effects in downstream tasks when the goal is under-specified. We introduce a general method that enables these skill discovery algorithms to learn focused skills -- skills that target and control specific state variables. Our approach improves state space coverage by a factor of three, unlocks new learning capabilities, and automatically avoids negative side effects in downstream tasks.
Authors: Johannes Liebenow, Thorsten Peinemann, Esfandiar Mohammadi
Abstract: The tuning of hyperparameters in distributed machine learning can substantially impact model performance. When the hyperparameters are tuned on sensitive data, privacy becomes an important challenge and to this end, differential privacy has emerged as the de facto standard for provable privacy. A standard setting when performing distributed learning tasks is that clients agree on a shared setup, i.e., find a compromise from a set of hyperparameters, like the learning rate of the model to be trained. Yet, prior work on differentially private hyperparameter tuning either uses computationally expensive cryptographic protocols, determines hyperparameters separately for each client, or applies differential privacy locally, which can lead to undesirable utility-privacy trade-offs. In this work, we present our algorithm DP-HYPE, which performs a distributed and privacy-preserving hyperparameter search by conducting a distributed voting based on local hyperparameter evaluations of clients. In this way, DP-HYPE selects hyperparameters that lead to a compromise supported by the majority of clients, while maintaining scalability and independence from specific learning tasks. We prove that DP-HYPE preserves the strong notion of differential privacy called client-level differential privacy and, importantly, show that its privacy guarantees do not depend on the number of hyperparameters. We also provide bounds on its utility guarantees, that is, the probability of reaching a compromise, and implement DP-HYPE as a submodule in the popular Flower framework for distributed machine learning. In addition, we evaluate performance on multiple benchmark data sets in iid as well as multiple non-iid settings and demonstrate high utility of DP-HYPE even under small privacy budgets.
Authors: Haotian Gao, Zheng Dong, Jiawei Yong, Shintaro Fukushima, Kenjiro Taura, Renhe Jiang
Abstract: Spatio-temporal forecasting is essential for real-world applications such as traffic management and urban computing. Although recent methods have shown improved accuracy, they often fail to account for dynamic deviations between current inputs and historical patterns. These deviations contain critical signals that can significantly affect model performance. To fill this gap, we propose ST-SSDL, a Spatio-Temporal time series forecasting framework that incorporates a Self-Supervised Deviation Learning scheme to capture and utilize such deviations. ST-SSDL anchors each input to its historical average and discretizes the latent space using learnable prototypes that represent typical spatio-temporal patterns. Two auxiliary objectives are proposed to refine this structure: a contrastive loss that enhances inter-prototype discriminability and a deviation loss that regularizes the distance consistency between input representations and corresponding prototypes to quantify deviation. Optimized jointly with the forecasting objective, these components guide the model to organize its hidden space and improve generalization across diverse input conditions. Experiments on six benchmark datasets show that ST-SSDL consistently outperforms state-of-the-art baselines across multiple metrics. Visualizations further demonstrate its ability to adaptively respond to varying levels of deviation in complex spatio-temporal scenarios. Our code and datasets are available at https://github.com/Jimmy-7664/ST-SSDL.
Authors: Jie Yang, Kexin Zhang, Guibin Zhang, Philip S. Yu, Kaize Ding
Abstract: Time Series Imputation (TSI), which aims to recover missing values in temporal data, remains a fundamental challenge due to the complex and often high-rate missingness in real-world scenarios. Existing models typically optimize the point-wise reconstruction loss, focusing on recovering numerical values (local information). However, we observe that under high missing rates, these models still perform well in the training phase yet produce poor imputations and distorted latent representation distributions (global information) in the inference phase. This reveals a critical optimization dilemma: current objectives lack global guidance, leading models to overfit local noise and fail to capture global information of the data. To address this issue, we propose a new training paradigm, Glocal Information Bottleneck (Glocal-IB). Glocal-IB is model-agnostic and extends the standard IB framework by introducing a Global Alignment loss, derived from a tractable mutual information approximation. This loss aligns the latent representations of masked inputs with those of their originally observed counterparts. It helps the model retain global structure and local details while suppressing noise caused by missing values, giving rise to better generalization under high missingness. Extensive experiments on nine datasets confirm that Glocal-IB leads to consistently improved performance and aligned latent representations under missingness. Our code implementation is available in https://github.com/Muyiiiii/NeurIPS-25-Glocal-IB.
Authors: Usman Akram, Yiyue Chen, Haris Vikalo
Abstract: Training automatic modulation classification (AMC) models on centrally aggregated data raises privacy concerns, incurs communication overhead, and often fails to confer robustness to channel shifts. Federated learning (FL) avoids central aggregation by training on distributed clients but remains sensitive to class imbalance, non-IID client distributions, and limited labeled samples. We propose FedSSL-AMC, which trains a causal, time-dilated CNN with triplet-loss self-supervision on unlabeled I/Q sequences across clients, followed by per-client SVMs on small labeled sets. We establish convergence of the federated representation learning procedure and a separability guarantee for the downstream classifier under feature noise. Experiments on synthetic and over-the-air datasets show consistent gains over supervised FL baselines under heterogeneous SNR, carrier-frequency offsets, and non-IID label partitions.
Authors: Ali Saheb Pasand, Elvis Dohmatob
Abstract: Grokking is the phenomenon whereby, unlike the training performance, which peaks early in the training process, the test/generalization performance of a model stagnates over arbitrarily many epochs and then suddenly jumps to usually close to perfect levels. In practice, it is desirable to reduce the length of such plateaus, that is to make the learning process "grok" faster. In this work, we provide new insights into grokking. First, we show both empirically and theoretically that grokking can be induced by asymmetric speeds of (stochastic) gradient descent, along different principal (i.e singular directions) of the gradients. We then propose a simple modification that normalizes the gradients so that dynamics along all the principal directions evolves at exactly the same speed. Then, we establish that this modified method, which we call egalitarian gradient descent (EGD) and can be seen as a carefully modified form of natural gradient descent, groks much faster. In fact, in some cases the stagnation is completely removed. Finally, we empirically show that on classical arithmetic problems such as modular addition and sparse parity problem which this stagnation has been widely observed and intensively studied, that our proposed method eliminates the plateaus.
Authors: Shiwen Qin, Alexander Auras, Shay B. Cohen, Elliot J. Crowley, Michael Moeller, Linus Ericsson, Jovita Lukasik
Abstract: Neural architecture search (NAS) automates the design process of high-performing architectures, but remains bottlenecked by expensive performance evaluation. Most existing studies that achieve faster evaluation are mostly tied to cell-based search spaces and graph encodings tailored to those individual search spaces, limiting their flexibility and scalability when applied to more expressive search spaces. In this work, we aim to close the gap of individual search space restrictions and search space dependent network representations. We present ONNX-Bench, a benchmark consisting of a collection of neural networks in a unified format based on ONNX files. ONNX-Bench includes all open-source NAS-bench-based neural networks, resulting in a total size of more than 600k {architecture, accuracy} pairs. This benchmark allows creating a shared neural network representation, ONNX-Net, able to represent any neural architecture using natural language descriptions acting as an input to a performance predictor. This text-based encoding can accommodate arbitrary layer types, operation parameters, and heterogeneous topologies, enabling a single surrogate to generalise across all neural architectures rather than being confined to cell-based search spaces. Experiments show strong zero-shot performance across disparate search spaces using only a small amount of pretraining samples, enabling the unprecedented ability to evaluate any neural network architecture instantly.
Authors: Jerry Yao-Chieh Hu, Xiwen Zhang, Weimin Wu, Han Liu
Abstract: Structured State-Space Duality (SSD) [Dao & Gu, ICML 2024] is an equivalence between a simple Structured State-Space Model (SSM) and a masked attention mechanism. In particular, a state-space model with a scalar-times-identity state matrix is equivalent to a masked self-attention with a $1$-semiseparable causal mask. Consequently, the same sequence transformation (model) has two algorithmic realizations: as a linear-time $O(T)$ recurrence or as a quadratic-time $O(T^2)$ attention. In this note, we formalize and generalize this duality: (i) we extend SSD from the scalar-identity case to general diagonal SSMs (diagonal state matrices); (ii) we show that these diagonal SSMs match the scalar case's training complexity lower bounds while supporting richer dynamics; (iii) we establish a necessary and sufficient condition under which an SSM is equivalent to $1$-semiseparable masked attention; and (iv) we show that such duality fails to extend to standard softmax attention due to rank explosion. Together, these results tighten bridge between recurrent SSMs and Transformers, and widen the design space for expressive yet efficient sequence models.
Authors: Jayanta Mandi, Marianne Defresne, Senne Berden, Tias Guns
Abstract: When some parameters of a constrained optimization problem (COP) are uncertain, this gives rise to a predict-then-optimize (PtO) problem, comprising two stages -- the prediction of the unknown parameters from contextual information and the subsequent optimization using those predicted parameters. Decision-focused learning (DFL) implements the first stage by training a machine learning (ML) model to optimize the quality of the decisions made using the predicted parameters. When parameters in the constraints of a COP are predicted, the predicted parameters can lead to infeasible solutions. Therefore, it is important to simultaneously manage both feasibility and decision quality. We develop a DFL framework for predicting constraint parameters in a generic COP. While prior works typically assume that the underlying optimization problem is a linear program (LP) or integer linear program (ILP), our approach makes no such assumption. We derive two novel loss functions based on maximum likelihood estimation (MLE): the first one penalizes infeasibility (by penalizing when the predicted parameters lead to infeasible solutions), and the second one penalizes suboptimal decisions (by penalizing when the true optimal solution is infeasible under the predicted parameters). We introduce a single tunable parameter to form a weighted average of the two losses, allowing decision-makers to balance suboptimality and feasibility. We experimentally demonstrate that adjusting this parameter provides a decision-maker the control over the trade-off between the two. Moreover, across several COP instances, we find that for a single value of the tunable parameter, our method matches the performance of the existing baselines on suboptimality and feasibility.
Authors: Allen Daniel Sunny
Abstract: We present StructuralDecompose, an R package for modular and interpretable time series decomposition. Unlike existing approaches that treat decomposition as a monolithic process, StructuralDecompose separates the analysis into distinct components: changepoint detection, anomaly detection, smoothing, and decomposition. This design provides flexibility and robust- ness, allowing users to tailor methods to specific time series characteristics. We demonstrate the package on simulated and real-world datasets, benchmark its performance against state-of-the- art tools such as Rbeast and autostsm, and discuss its role in interpretable machine learning workflows.
Authors: Xuefeng Xu, Graham Cormode
Abstract: Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves are fundamental tools for evaluating machine learning classifiers, offering detailed insights into the trade-offs between true positive rate vs. false positive rate (ROC) or precision vs. recall (PR). However, in Federated Learning (FL) scenarios, where data is distributed across multiple clients, computing these curves is challenging due to privacy and communication constraints. Specifically, the server cannot access raw prediction scores and class labels, which are used to compute the ROC and PR curves in a centralized setting. In this paper, we propose a novel method for approximating ROC and PR curves in a federated setting by estimating quantiles of the prediction score distribution under distributed differential privacy. We provide theoretical bounds on the Area Error (AE) between the true and estimated curves, demonstrating the trade-offs between approximation accuracy, privacy, and communication cost. Empirical results on real-world datasets demonstrate that our method achieves high approximation accuracy with minimal communication and strong privacy guarantees, making it practical for privacy-preserving model evaluation in federated systems.
Authors: Kristi Topollai, Anna Choromanska
Abstract: The vast majority of modern deep learning models are trained with momentum-based first-order optimizers. The momentum term governs the optimizer's memory by determining how much each past gradient contributes to the current convergence direction. Fundamental momentum methods, such as Nesterov Accelerated Gradient and the Heavy Ball method, as well as more recent optimizers such as AdamW and Lion, all rely on the momentum coefficient that is customarily set to $\beta = 0.9$ and kept constant during model training, a strategy widely used by practitioners, yet suboptimal. In this paper, we introduce an \textit{adaptive memory} mechanism that replaces constant momentum with a dynamic momentum coefficient that is adjusted online during optimization. We derive our method by approximating the objective function using two planes: one derived from the gradient at the current iterate and the other obtained from the accumulated memory of the past gradients. To the best of our knowledge, such a proximal framework was never used for momentum-based optimization. Our proposed approach is novel, extremely simple to use, and does not rely on extra assumptions or hyperparameter tuning. We implement adaptive memory variants of both SGD and AdamW across a wide range of learning tasks, from simple convex problems to large-scale deep learning scenarios, demonstrating that our approach can outperform standard SGD and Adam with hand-tuned momentum coefficients. Finally, our work opens doors for new ways of inducing adaptivity in optimization.
Authors: Xuefeng Xu, Graham Cormode
Abstract: Power transforms are popular parametric techniques for making data more Gaussian-like, and are widely used as preprocessing steps in statistical analysis and machine learning. However, we find that direct implementations of power transforms suffer from severe numerical instabilities, which can lead to incorrect results or even crashes. In this paper, we provide a comprehensive analysis of the sources of these instabilities and propose effective remedies. We further extend power transforms to the federated learning setting, addressing both numerical and distributional challenges that arise in this context. Experiments on real-world datasets demonstrate that our methods are both effective and robust, substantially improving stability compared to existing approaches.
Authors: Wei Xiong, Chenlu Ye, Baohao Liao, Hanze Dong, Xinxing Xu, Christof Monz, Jiang Bian, Nan Jiang, Tong Zhang
Abstract: Reinforcement learning applied to large language models (LLMs) for reasoning tasks is often bottlenecked by unstable gradient estimates due to fixed and uniform sampling of responses across prompts. Prior work such as GVM-RAFT addresses this by dynamically allocating inference budget per prompt to minimize stochastic gradient variance under a budget constraint. Inspired by this insight, we propose Reinforce-Ada, an adaptive sampling framework for online RL post-training of LLMs that continuously reallocates sampling effort to the prompts with the greatest uncertainty or learning potential. Unlike conventional two-stage allocation methods, Reinforce-Ada interleaves estimation and sampling in an online successive elimination process, and automatically stops sampling for a prompt once sufficient signal is collected. To stabilize updates, we form fixed-size groups with enforced reward diversity and compute advantage baselines using global statistics aggregated over the adaptive sampling phase. Empirical results across multiple model architectures and reasoning benchmarks show that Reinforce-Ada accelerates convergence and improves final performance compared to GRPO, especially when using the balanced sampling variant. Our work highlights the central role of variance-aware, adaptive data curation in enabling efficient and reliable reinforcement learning for reasoning-capable LLMs. Code is available at https://github.com/RLHFlow/Reinforce-Ada.
Authors: Weixin Wang, Haoyang Zheng, Guang Lin, Wei Deng, Pan Xu
Abstract: Most existing approximate Thompson Sampling (TS) algorithms for multi-armed bandits use Stochastic Gradient Langevin Dynamics (SGLD) or its variants in each round to sample from the posterior, relaxing the need for conjugacy assumptions between priors and reward distributions in vanilla TS. However, they often require approximating a different posterior distribution in different round of the bandit problem. This requires tricky, round-specific tuning of hyperparameters such as dynamic learning rates, causing challenges in both theoretical analysis and practical implementation. To alleviate this non-stationarity, we introduce TS-SA, which incorporates stochastic approximation (SA) within the TS framework. In each round, TS-SA constructs a posterior approximation only using the most recent reward(s), performs a Langevin Monte Carlo (LMC) update, and applies an SA step to average noisy proposals over time. This can be interpreted as approximating a stationary posterior target throughout the entire algorithm, which further yields a fixed step-size, a unified convergence analysis framework, and improved posterior estimates through temporal averaging. We establish near-optimal regret bounds for TS-SA, with a simplified and more intuitive theoretical analysis enabled by interpreting the entire algorithm as a simulation of a stationary SGLD process. Our empirical results demonstrate that even a single-step Langevin update with certain warm-up outperforms existing methods substantially on bandit tasks.
Authors: Nevan Wichers, Aram Ebtekar, Ariana Azarbal, Victor Gillioz, Christine Ye, Emil Ryd, Neil Rathi, Henry Sleight, Alex Mallen, Fabien Roger, Samuel Marks
Abstract: Large language models are sometimes trained with imperfect oversight signals, leading to undesired behaviors such as reward hacking and sycophancy. Improving oversight quality can be expensive or infeasible, motivating methods that improve learned behavior despite an imperfect training signal. We introduce Inoculation Prompting (IP), a simple but counterintuitive technique that prevents learning of an undesired behavior by modifying training prompts to explicitly request it. For example, to inoculate against reward hacking, we modify the prompts used in supervised fine-tuning to request code that only works on provided test cases but fails on other inputs. Across four settings we find that IP reduces the learning of undesired behavior without substantially reducing the learning of desired capabilities. We also show that prompts which more strongly elicit the undesired behavior prior to fine-tuning more effectively inoculate against the behavior when used during training; this serves as a heuristic to identify promising inoculation prompts. Overall, IP is a simple yet effective way to control how models generalize from fine-tuning, preventing learning of undesired behaviors without substantially disrupting desired capabilities.
Authors: Sergio Rozada, Vimal K. B., Andrea Cavallo, Antonio G. Marques, Hadi Jamali-Rad, Elvin Isufi
Abstract: We study the problem of generating graph signals from unknown distributions defined over given graphs, relevant to domains such as recommender systems or sensor networks. Our approach builds on generative diffusion models, which are well established in vision and graph generation but remain underexplored for graph signals. Existing methods lack generality, either ignoring the graph structure in the forward process or designing graph-aware mechanisms tailored to specific domains. We adopt a forward process that incorporates the graph through the heat equation. Rather than relying on the standard formulation, we consider a time-warped coefficient to mitigate the exponential decay of the drift term, yielding a graph-aware generative diffusion model (GAD). We analyze its forward dynamics, proving convergence to a Gaussian Markov random field with covariance parametrized by the graph Laplacian, and interpret the backward dynamics as a sequence of graph-signal denoising problems. Finally, we demonstrate the advantages of GAD on synthetic data, real traffic speed measurements, and a temperature sensor network.
Authors: Jihoon Lee, Hoyeon Moon, Kevin Zhai, Arun Kumar Chithanar, Anit Kumar Sahu, Soummya Kar, Chul Lee, Souradip Chakraborty, Amrit Singh Bedi
Abstract: Diffusion-based large language models (dLLMs) are trained flexibly to model extreme dependence in the data distribution; however, how to best utilize this information at inference time remains an open problem. In this work, we uncover an interesting property of these models: dLLMs trained on textual data implicitly learn a mixture of semi-autoregressive experts, where different generation orders reveal different specialized behaviors. We show that committing to any single, fixed inference time schedule, a common practice, collapses performance by failing to leverage this latent ensemble. To address this, we introduce HEX (Hidden semiautoregressive EXperts for test-time scaling), a training-free inference method that ensembles across heterogeneous block schedules. By doing a majority vote over diverse block-sized generation paths, HEX robustly avoids failure modes associated with any single fixed schedule. On reasoning benchmarks such as GSM8K, it boosts accuracy by up to 3.56X (from 24.72% to 88.10%), outperforming top-K margin inference and specialized fine-tuned methods like GRPO, without additional training. HEX even yields significant gains on MATH benchmark from 16.40% to 40.00%, scientific reasoning on ARC-C from 54.18% to 87.80%, and TruthfulQA from 28.36% to 57.46%. Our results establish a new paradigm for test-time scaling in diffusion-based LLMs (dLLMs), revealing that the sequence in which masking is performed plays a critical role in determining performance during inference.
Authors: Ahmed Elhussein, Paul Meddeb, Abigail Newbury, Jeanne Mirone, Martin Stoll, Gamze Gursoy
Abstract: Machine learning in healthcare requires effective representation of structured medical codes, but current methods face a trade off: knowledge graph based approaches capture formal relationships but miss real world patterns, while data driven methods learn empirical associations but often overlook structured knowledge in medical terminologies. We present KEEP (Knowledge preserving and Empirically refined Embedding Process), an efficient framework that bridges this gap by combining knowledge graph embeddings with adaptive learning from clinical data. KEEP first generates embeddings from knowledge graphs, then employs regularized training on patient records to adaptively integrate empirical patterns while preserving ontological relationships. Importantly, KEEP produces final embeddings without task specific auxiliary or end to end training enabling KEEP to support multiple downstream applications and model architectures. Evaluations on structured EHR from UK Biobank and MIMIC IV demonstrate that KEEP outperforms both traditional and Language Model based approaches in capturing semantic relationships and predicting clinical outcomes. Moreover, KEEP's minimal computational requirements make it particularly suitable for resource constrained environments.
Authors: Peter Van Katwyk, Karianne J. Bergen
Abstract: Uncertainty quantification is critical for ensuring robustness in high-stakes machine learning applications. We introduce HybridFlow, a modular hybrid architecture that unifies the modeling of aleatoric and epistemic uncertainty by combining a Conditional Masked Autoregressive normalizing flow for estimating aleatoric uncertainty with a flexible probabilistic predictor for epistemic uncertainty. The framework supports integration with any probabilistic model class, allowing users to easily adapt HybridFlow to existing architectures without sacrificing predictive performance. HybridFlow improves upon previous uncertainty quantification frameworks across a range of regression tasks, such as depth estimation, a collection of regression benchmarks, and a scientific case study of ice sheet emulation. We also provide empirical results of the quantified uncertainty, showing that the uncertainty quantified by HybridFlow is calibrated and better aligns with model error than existing methods for quantifying aleatoric and epistemic uncertainty. HybridFlow addresses a key challenge in Bayesian deep learning, unifying aleatoric and epistemic uncertainty modeling in a single robust framework.
Authors: Alexis Ross, Megha Srivastava, Jeremiah Blanchard, Jacob Andreas
Abstract: As programmers write code, they often edit and retry multiple times, creating rich "interaction traces" that reveal how they approach coding tasks and provide clues about their level of skill development. For novice programmers in particular, these traces reflect the diverse reasoning processes they employ to code, such as exploratory behavior to understand how a programming concept works, re-strategizing in response to bugs, and personalizing stylistic choices. In this work, we explore what can be learned from training language models on such reasoning traces: not just about code, but about coders, and particularly students learning to program. We introduce a dataset of over 3.8 million programming reasoning traces from users of Pencil Code, a free online educational platform used by students to learn simple programming concepts. Compared to models trained only on final programs or synthetically-generated traces, we find that models trained on real traces are stronger at modeling diverse student behavior. Through both behavioral and probing analyses, we also find that many properties of code traces, such as goal backtracking or number of comments, can be predicted from learned representations of the students who write them. Building on this result, we show that we can help students recover from mistakes by steering code generation models to identify a sequence of edits that will results in more correct code while remaining close to the original student's style. Together, our results suggest that many properties of code are properties of individual students and that training on edit traces can lead to models that are more steerable, more predictive of student behavior while programming, and better at generating programs in their final states. Code and data is available at https://github.com/meghabyte/pencilcode-public
Authors: Roberto Neglia, Andrea Cini, Michael M. Bronstein, Filippo Maria Bianchi
Abstract: Conformal prediction offers a powerful framework for building distribution-free prediction intervals for exchangeable data. Existing methods that extend conformal prediction to sequential data rely on fitting a relatively complex model to capture temporal dependencies. However, these methods can fail if the sample size is small and often require expensive retraining when the underlying data distribution changes. To overcome these limitations, we propose Reservoir Conformal Prediction (ResCP), a novel training-free conformal prediction method for time series. Our approach leverages the efficiency and representation learning capabilities of reservoir computing to dynamically reweight conformity scores. In particular, we compute similarity scores among reservoir states and use them to adaptively reweight the observed residuals at each step. With this approach, ResCP enables us to account for local temporal dynamics when modeling the error distribution without compromising computational scalability. We prove that, under reasonable assumptions, ResCP achieves asymptotic conditional coverage, and we empirically demonstrate its effectiveness across diverse forecasting tasks.
Authors: Sara Kangaslahti, Nihal V. Nayak, Jonathan Geuter, Marco Fumero, Francesco Locatello, David Alvarez-Melis
Abstract: Large language models (LLMs) are typically deployed under diverse memory and compute constraints. Existing approaches build model families by training each size independently, which is prohibitively expensive and provides only coarse-grained size options. In this work, we identify a novel phenomenon that we call boomerang distillation: starting from a large base model (the teacher), one first distills down to a small student and then progressively reconstructs intermediate-sized models by re-incorporating blocks of teacher layers into the student without any additional training. This process produces zero-shot interpolated models of many intermediate sizes whose performance scales smoothly between the student and teacher, often matching or surpassing pretrained or distilled models of the same size. We further analyze when this type of interpolation succeeds, showing that alignment between teacher and student through pruning and distillation is essential. Boomerang distillation thus provides a simple and efficient way to generate fine-grained model families, dramatically reducing training cost while enabling flexible adaptation across deployment environments. The code and models are available at https://github.com/dcml-lab/boomerang-distillation.
Authors: Yangyang Wang, Tayo Fabusuyi
Abstract: This study presents a novel small-area estimation framework to enhance urban transportation planning through detailed characterization of travel behavior. Our approach improves on the four-step travel model by employing publicly available microdata files and machine learning methods to predict travel behavior for a representative, synthetic population at small geographic areas. This approach enables high-resolution estimation of trip generation, trip distribution, mode choice, and route assignment. Validation using ACS/PUMS work-commute datasets demonstrates that our framework achieves higher accuracy compared to conventional approaches. The resulting granular insights enable the tailoring of interventions to address localized situations and support a range of policy applications and targeted interventions, including the optimal placement of micro-fulfillment centers, effective curb-space management, and the design of more inclusive transportation solutions particularly for vulnerable communities.
Authors: Avichal Goel, Yoon Kim, Nir Shavit, Tony T. Wang
Abstract: Finetuning (pretrained) language models is a standard approach for updating their internal parametric knowledge and specializing them to new tasks and domains. However, the corresponding model weight changes ("weight diffs") are not generally interpretable. While inspecting the finetuning dataset can give a sense of how the model might have changed, these datasets are often not publicly available or are too large to work with directly. Towards the goal of comprehensively understanding weight diffs in natural language, we introduce Diff Interpretation Tuning (DIT), a method that trains models to describe their own finetuning-induced modifications. Our approach uses synthetic, labeled weight diffs to train a DIT adapter, which can be applied to a compatible finetuned model to make it describe how it has changed. We demonstrate in two proof-of-concept settings (reporting hidden behaviors and summarizing finetuned knowledge) that our method enables models to describe their finetuning-induced modifications using accurate natural language descriptions.
Authors: Mingkang Zhu, Xi Chen, Bei Yu, Hengshuang Zhao, Jiaya Jia
Abstract: Large reasoning models (LRMs) generate intermediate reasoning traces before producing final answers, yielding strong gains on multi-step and mathematical tasks. Yet aligning LRMs with human preferences, a crucial prerequisite for model deployment, remains underexplored. The statistically correct objective for preference alignment requires marginalizing over reasoning traces, but this computation is intractable in practice. A common workaround optimizes a single sampled trajectory, which introduces substantial gradient variance from stochastic trace sampling. To address this challenge, we frame preference optimization for LRMs through the lens of the bias--variance trade-off and propose Bias--Variance Optimized Preference Optimization (BVPO), a simple, drop-in method that mixes two gradient estimators: a high-variance trace-based estimator and a low-variance empty-trace estimator obtained by disabling reasoning trace generation. Our theory shows that BVPO strictly reduces trace-induced variance for any nontrivial mixture, provides a closed-form choice of the mixing weight that minimizes mean-squared error relative to the true marginal gradient, and under standard smoothness and step-size conditions, tightens classical convergence bounds for stochastic gradient descent. Empirically, BVPO improves alignment over the best baseline by up to 7.8 points on AlpacaEval~2 and 6.8 points on Arena-Hard. Despite being trained only on general conversational data, BVPO also boosts reasoning performance for base models by up to 4.0 points on the average of six math reasoning benchmarks. These results identify variance from trace sampling as a key bottleneck and demonstrate that directly optimizing the bias--variance trade-off yields more stable training and stronger overall performance.
Authors: Cheng Xin, Fan Xu, Xin Ding, Jie Gao, Jiaxin Ding
Abstract: Graph Neural Networks (GNNs) have shown remarkable success across various scientific fields, yet their adoption in critical decision-making is often hindered by a lack of interpretability. Recently, intrinsically interpretable GNNs have been studied to provide insights into model predictions by identifying rationale substructures in graphs. However, existing methods face challenges when the underlying rationale subgraphs are complex and varied. In this work, we propose TopInG: Topologically Interpretable Graph Learning, a novel topological framework that leverages persistent homology to identify persistent rationale subgraphs. TopInG employs a rationale filtration learning approach to model an autoregressive generation process of rationale subgraphs, and introduces a self-adjusted topological constraint, termed topological discrepancy, to enforce a persistent topological distinction between rationale subgraphs and irrelevant counterparts. We provide theoretical guarantees that our loss function is uniquely optimized by the ground truth under specific conditions. Extensive experiments demonstrate TopInG's effectiveness in tackling key challenges, such as handling variform rationale subgraphs, balancing predictive performance with interpretability, and mitigating spurious correlations. Results show that our approach improves upon state-of-the-art methods on both predictive accuracy and interpretation quality.
Authors: Dmitriy Shopkhoev, Ammar Ali, Magauiya Zhussip, Valentin Malykh, Stamatios Lefkimmiatis, Nikos Komodakis, Sergey Zagoruyko
Abstract: We introduce ReplaceMe, a generalized training-free depth pruning method that effectively replaces transformer blocks with a linear operation, while maintaining high performance for low compression ratios. In contrast to conventional pruning approaches that require additional training or fine-tuning, our approach requires only a small calibration dataset that is used to estimate a linear transformation, which approximates the pruned blocks. The estimated linear mapping can be seamlessly merged with the remaining transformer blocks, eliminating the need for any additional network parameters. Our experiments show that ReplaceMe consistently outperforms other training-free approaches and remains highly competitive with state-of-the-art pruning methods that involve extensive retraining/fine-tuning and architectural modifications. Applied to several large language models (LLMs), ReplaceMe achieves up to 25% pruning while retaining approximately 90% of the original model's performance on open benchmarks - without any training or healing steps, resulting in minimal computational overhead (see Fig.1). We provide an open-source library implementing ReplaceMe alongside several state-of-the-art depth pruning techniques, available at https://github.com/mts-ai/ReplaceMe.
Authors: Magauiya Zhussip, Dmitriy Shopkhoev, Ammar Ali, Stamatios Lefkimmiatis
Abstract: Large language models (LLMs) have revolutionized AI applications, yet their high computational and memory demands hinder their widespread deployment. Existing compression techniques focus on intra-block optimizations (e.g. low-rank approximation, attention head pruning), while the repetitive layered structure of transformers implies significant inter-block redundancy - a dimension largely unexplored beyond key-value (KV) caching. Inspired by dictionary learning in CNNs, we propose a framework for structured weight sharing across transformer layers. Our approach decomposes attention projection matrices into shared dictionary atoms, reducing the attention module's parameters by 66.7% while achieving on-par performance. Unlike complex methods requiring distillation or architectural changes, MASA (Matrix Atom Sharing in Attention) operates as a drop-in replacement - trained with standard optimizers - and represents each layer's weights as linear combinations of shared matrix atoms. Experiments across scales (100M-700M parameters) show that MASA achieves better benchmark accuracy and perplexity than grouped-query attention (GQA), low-rank baselines and recently proposed Repeat-all-over/Sequential sharing at comparable parameter budgets. Ablation studies confirm robustness to the dictionary size and the efficacy of shared representations in capturing cross-layer statistical regularities. Extending to Vision Transformers (ViT), MASA matches performance metrics on image classification and detection tasks with 66.7% fewer attention parameters. By combining dictionary learning strategies with transformer efficiency, MASA offers a scalable blueprint for parameter-efficient models without sacrificing performance. Finally, we investigate the possibility of employing MASA on pretrained LLMs to reduce their number of parameters without experiencing any significant drop in their performance.
Authors: Ava C. Blake, Nivika A. Gandhi, Anurag R. Jakkula
Abstract: Accurate prediction of financial market volatility is critical for risk management, derivatives pricing, and investment strategy. In this study, we propose a multitude of regime-switching methods to improve the prediction of S&P 500 volatility by capturing structural changes in the market across time. We use eleven years of SPX data, from May 1st, 2014 to May 27th, 2025, to compute daily realized volatility (RV) from 5-minute intraday log returns, adjusted for irregular trading days. To enhance forecast accuracy, we engineered features to capture both historical dynamics and forward-looking market sentiment across regimes. The regime-switching methods include a soft Markov switching algorithm to estimate soft-regime probabilities, a distributional spectral clustering method that uses XGBoost to assign clusters at prediction time, and a coefficient-based soft regime algorithm that extracts HAR coefficients from time segments segmented through the Mood test and clusters through Bayesian GMM for soft regime weights, using XGBoost to predict regime probabilities. Models were evaluated across three time periods--before, during, and after the COVID-19 pandemic. The coefficient-based clustering algorithm outperformed all other models, including the baseline autoregressive model, during all time periods. Additionally, each model was evaluated on its recursive forecasting performance for 5- and 10-day horizons during each time period. The findings of this study demonstrate the value of regime-aware modeling frameworks and soft clustering approaches in improving volatility forecasting, especially during periods of heightened uncertainty and structural change.
Authors: Tunde Fahd Egunjobi
Abstract: Bayesian Optimization (BO) is widely used for optimizing expensive black-box functions, particularly in hyperparameter tuning. However, standard BO assumes access to precise objective values, which may be unavailable, noisy, or unreliable in real-world settings where only relative or rank-based feedback can be obtained. In this study, we propose Quantile-Scaled Bayesian Optimization (QS-BO), a principled rank-based optimization framework. QS-BO converts ranks into heteroscedastic Gaussian targets through a quantile-scaling pipeline, enabling the use of Gaussian process surrogates and standard acquisition functions without requiring explicit metric scores. We evaluate QS-BO on synthetic benchmark functions, including one- and two-dimensional nonlinear functions and the Branin function, and compare its performance against Random Search. Results demonstrate that QS-BO consistently achieves lower objective values and exhibits greater stability across runs. Statistical tests further confirm that QS-BO significantly outperforms Random Search at the 1\% significance level. These findings establish QS-BO as a practical and effective extension of Bayesian Optimization for rank-only feedback, with promising applications in preference learning, recommendation, and human-in-the-loop optimization where absolute metric values are unavailable or unreliable.
Authors: David van Batenburg
Abstract: Machine Learning is becoming increasingly more important in today's world. It is therefore very important to provide understanding of the decision-making process of machine-learning models. A popular way to do this is by looking at the Shapley-Values of these models as introduced by Lundberg and Lee. In this thesis, we discuss the two main results by Lundberg and Lee from a mathematically rigorous standpoint and provide full proofs, which are not available from the original material. The first result of this thesis is an axiomatic characterization of the Shapley values in machine learning based on axioms by Young. We show that the Shapley values are the unique explanation to satisfy local accuracy, missingness, symmetry and consistency. Lundberg and Lee claim that the symmetry axiom is redundant for explanations. However, we provide a counterexample that shows the symmetry axiom is in fact essential. The second result shows that we can write the Shapley values as the unique solution to a weighted linear regression problem. This result is proven with the use of dimensionality reduction.
Authors: Su Kara, Fazle Faisal, Suman Nath
Abstract: Recent advances in browser-based LLM agents have shown promise for automating tasks ranging from simple form filling to hotel booking or online shopping. Current benchmarks measure agent performance in controlled environments, such as containers or stable networks, where websites behave deterministically. However, in the real world, users access websites over networks and HTTPS connections that introduce instability from multiple sources: client-side, server-side issues or broader system failures. Moreover, live websites are prone to web attacks such Cross-Site Scripting, as well as general site modifications which can cause unexpected or malicious pop-ups or improper functionality. To address this gap, we present WAREX: Web Agent Reliability Evaluation on Existing Benchmarks. We measure the impact of WAREX across three popular benchmarks: WebArena, WebVoyager, and REAL. Our experiments show that introducing WAREX leads to significant drops in task success rates, highlighting the limited robustness of state-of-the-art agents.
Authors: Passant Elchafei, Amany Fashwan
Abstract: We present VLCAP, an Arabic image captioning framework that integrates CLIP-based visual label retrieval with multimodal text generation. Rather than relying solely on end-to-end captioning, VLCAP grounds generation in interpretable Arabic visual concepts extracted with three multilingual encoders, mCLIP, AraCLIP, and Jina V4, each evaluated separately for label retrieval. A hybrid vocabulary is built from training captions and enriched with about 21K general domain labels translated from the Visual Genome dataset, covering objects, attributes, and scenes. The top-k retrieved labels are transformed into fluent Arabic prompts and passed along with the original image to vision-language models. In the second stage, we tested Qwen-VL and Gemini Pro Vision for caption generation, resulting in six encoder-decoder configurations. The results show that mCLIP + Gemini Pro Vision achieved the best BLEU-1 (5.34%) and cosine similarity (60.01%), while AraCLIP + Qwen-VL obtained the highest LLM-judge score (36.33%). This interpretable pipeline enables culturally coherent and contextually accurate Arabic captions.
Authors: Akshar Gothi
Abstract: We present a controlled comparison of a convolutional neural network (EfficientNet-B0) and a Vision Transformer (ViT-Base) on SpaceNet under two label-distribution regimes: a naturally imbalanced five-class split and a balanced-resampled split with 700 images per class (70:20:10 train/val/test). With matched preprocessing (224x224, ImageNet normalization), lightweight augmentations, and a 40-epoch budget on a single NVIDIA P100, we report accuracy, macro-F1, balanced accuracy, per-class recall, and deployment metrics (model size and latency). On the imbalanced split, EfficientNet-B0 reaches 93% test accuracy with strong macro-F1 and lower latency; ViT-Base is competitive at 93% with a larger parameter count and runtime. On the balanced split, both models are strong; EfficientNet-B0 reaches 99% while ViT-Base remains competitive, indicating that balancing narrows architecture gaps while CNNs retain an efficiency edge. We release manifests, logs, and per-image predictions to support reproducibility.
Authors: Enrique Zuazua
Abstract: Control theory of dynamical systems offers a powerful framework for tackling challenges in deep neural networks and other machine learning architectures. We show that concepts such as simultaneous and ensemble controllability offer new insights into the classification and representation properties of deep neural networks while the control and optimization of static systems can be employed to better understand the performance of shallow networks. Inspired by the classical concept of turnpike, we also explore the relationship between dynamic and static neural networks, where depth is traded for width, and the role of transformers as mechanisms for accelerating classical neural network tasks. We also exploit the expressive power of neural networks (exemplified, for instance, by the Universal Approximation Theorem) to develop a novel hybrid modeling methodology, the Hybrid-Cooperative Learning (HYCO), combining mechanics and data-driven methods in a game-theoretic setting. Finally, we describe how classical properties of diffusion processes, long established in the context of partial differential equations, contribute to explaining the success of modern generative artificial intelligence (AI). We present an overview of our recent results in these areas, illustrating how control, machine learning, numerical analysis, and partial differential equations come together to motivate a fertile ground for future research.
Authors: Shuai Huang, Xuan Kan, James J. Lah, Deqiang Qiu
Abstract: Current atlas-based approaches to brain network analysis rely heavily on standardized anatomical or connectivity-driven brain atlases. However, these fixed atlases often introduce significant limitations, such as spatial misalignment across individuals, functional heterogeneity within predefined regions, and atlas-selection biases, collectively undermining the reliability and interpretability of the derived brain networks. To address these challenges, we propose a novel atlas-free brain network transformer (atlas-free BNT) that leverages individualized brain parcellations derived directly from subject-specific resting-state fMRI data. Our approach computes ROI-to-voxel connectivity features in a standardized voxel-based feature space, which are subsequently processed using the BNT architecture to produce comparable subject-level embeddings. Experimental evaluations on sex classification and brain-connectome age prediction tasks demonstrate that our atlas-free BNT consistently outperforms state-of-the-art atlas-based methods, including elastic net, BrainGNN, Graphormer and the original BNT. Our atlas-free approach significantly improves the precision, robustness, and generalizability of brain network analyses. This advancement holds great potential to enhance neuroimaging biomarkers and clinical diagnostic tools for personalized precision medicine.
Authors: Alex Gibson
Abstract: We study transformer language models, analyzing attention heads whose attention patterns are spread out, and whose attention scores depend weakly on content. We argue that the softmax denominators of these heads are stable when the underlying token distribution is fixed. By sampling softmax denominators from a "calibration text", we can combine together the outputs of multiple such stable heads in the first layer of GPT2-Small, approximating their combined output by a linear summary of the surrounding text. This approximation enables a procedure where from the weights alone - and a single calibration text - we can uncover hundreds of first layer neurons that respond to high-level contextual properties of the surrounding text, including neurons that didn't activate on the calibration text.
Authors: Ryan P. Demilt, Nicholas LaHaye, Karis Tenneson
Abstract: Earth Observation Foundation Models (EOFMs) have exploded in prevalence as tools for processing the massive volumes of remotely sensed and other earth observation data, and for delivering impact on the many essential earth monitoring tasks. An emerging trend posits using the outputs of pre-trained models as 'embeddings' which summarize high dimensional data to be used for generic tasks such as similarity search and content-specific queries. However, most EOFM models are trained only on single modalities of data and then applied or benchmarked by matching bands across different modalities. It is not clear from existing work what impact diverse sensor architectures have on the internal representations of the present suite of EOFMs. We show in this work that the representation space of EOFMs is highly sensitive to sensor architecture and that understanding this difference gives a vital perspective on the pitfalls of current EOFM design and signals for how to move forward as model developers, users, and a community guided by robust remote-sensing science.
Authors: Chenxiang Luo, David K. Y. Yau, Qun Song
Abstract: Federated learning (FL) enables collaborative model training without sharing raw data but is vulnerable to gradient inversion attacks (GIAs), where adversaries reconstruct private data from shared gradients. Existing defenses either incur impractical computational overhead for embedded platforms or fail to achieve privacy protection and good model utility at the same time. Moreover, many defenses can be easily bypassed by adaptive adversaries who have obtained the defense details. To address these limitations, we propose SVDefense, a novel defense framework against GIAs that leverages the truncated Singular Value Decomposition (SVD) to obfuscate gradient updates. SVDefense introduces three key innovations, a Self-Adaptive Energy Threshold that adapts to client vulnerability, a Channel-Wise Weighted Approximation that selectively preserves essential gradient information for effective model training while enhancing privacy protection, and a Layer-Wise Weighted Aggregation for effective model aggregation under class imbalance. Our extensive evaluation shows that SVDefense outperforms existing defenses across multiple applications, including image classification, human activity recognition, and keyword spotting, by offering robust privacy protection with minimal impact on model accuracy. Furthermore, SVDefense is practical for deployment on various resource-constrained embedded platforms. We will make our code publicly available upon paper acceptance.
Authors: Raik Dankworth, Gesina Schwalbe
Abstract: Deep neural networks (NNs) for computer vision are vulnerable to adversarial attacks, i.e., miniscule malicious changes to inputs may induce unintuitive outputs. One key approach to verify and mitigate such robustness issues is to falsify expected output behavior. This allows, e.g., to locally proof security, or to (re)train NNs on obtained adversarial input examples. Due to the black-box nature of NNs, current attacks only falsify a class of the final output, such as flipping from $\texttt{stop_sign}$ to $\neg\texttt{stop_sign}$. In this short position paper we generalize this to search for generally illogical behavior, as considered in NN verification: falsify constraints (concept-based properties) involving further human-interpretable concepts, like $\texttt{red}\wedge\texttt{octogonal}\rightarrow\texttt{stop_sign}$. For this, an easy implementation of concept-based properties on already trained NNs is proposed using techniques from explainable artificial intelligence. Further, we sketch the theoretical proof that attacks on concept-based properties are expected to have a reduced search space compared to simple class falsification, whilst arguably be more aligned with intuitive robustness targets. As an outlook to this work in progress we hypothesize that this approach has potential to efficiently and simultaneously improve logical compliance and robustness.
Authors: Fiona Victoria Stanley Jothiraj, Arunaggiri Pandian Karunanidhi, Seth A. Eichmeyer
Abstract: In semiconductor manufacturing, early detection of wafer defects is critical for product yield optimization. However, raw wafer data from wafer quality tests are often complex, unlabeled, imbalanced and can contain multiple defects on a single wafer, making it crucial to design clustering methods that remain reliable under such imperfect data conditions. We introduce DECOR, a deep clustering with orientation robustness framework that groups complex defect patterns from wafer maps into consistent clusters. We evaluate our method on the open source MixedWM38 dataset, demonstrating its ability to discover clusters without manual tuning. DECOR explicitly accounts for orientation variations in wafer maps, ensuring that spatially similar defects are consistently clustered regardless of its rotation or alignment. Experiments indicate that our method outperforms existing clustering baseline methods, thus providing a reliable and scalable solution in automated visual inspection systems.
Authors: Adharsha Sam Edwin Sam Devahi, Sohail Singh Sangha, Prachee Priyadarshinee, Jithin Thilakan, Ivan Fu Xing Tan, Christopher Johann Clarke, Sou Ka Lon, Balamurali B T, Yow Wei Quin, Chen Jer-Ming
Abstract: Early detection of Alzheimer's Dementia (AD) and Mild Cognitive Impairment (MCI) is critical for timely intervention, yet current diagnostic approaches remain resource-intensive and invasive. Speech, encompassing both acoustic and linguistic dimensions, offers a promising non-invasive biomarker for cognitive decline. In this study, we present a machine learning framework for the PROCESS Challenge, leveraging both audio embeddings and linguistic features derived from spontaneous speech recordings. Audio representations were extracted using Whisper embeddings from the Cookie Theft description task, while linguistic features-spanning pronoun usage, syntactic complexity, filler words, and clause structure-were obtained from transcriptions across Semantic Fluency, Phonemic Fluency, and Cookie Theft picture description. Classification models aimed to distinguish between Healthy Controls (HC), MCI, and AD participants, while regression models predicted Mini-Mental State Examination (MMSE) scores. Results demonstrated that voted ensemble models trained on concatenated linguistic features achieved the best classification performance (F1 = 0.497), while Whisper embedding-based ensemble regressors yielded the lowest MMSE prediction error (RMSE = 2.843). Comparative evaluation within the PROCESS Challenge placed our models among the top submissions in regression task, and mid-range for classification, highlighting the complementary strengths of linguistic and audio embeddings. These findings reinforce the potential of multimodal speech-based approaches for scalable, non-invasive cognitive assessment and underline the importance of integrating task-specific linguistic and acoustic markers in dementia detection.
Authors: Morgan D. Sanger, Gabrielle Campagnola, Robin Ritchey, Tuncer B. Edil, Matthew Ginder-Vogel
Abstract: Recycled concrete aggregate (RCA) is recognized as a readily available, mechanically sufficient construction and demolition waste product that is suitable as a base course substitute for natural, virgin aggregate in pavement construction. Environmentally responsible applications of RCA must consider the high alkalinity, high pH leachate, and heavy metal leaching risks reported in the literature. The existing body of literature does not address discrepancies between field and laboratory measurements of RCA leachate pH, nor are there any existing studies of aged RCA leachate composition as a function of time. To consider the influence of contact time on RCA leachate, the present study evaluates recovered RCA base course samples from the Minnesota Road Research highway construction study site using modified batch test methodology. Leachate pH, alkalinity, and calcium ion (Ca2+) concentration were monitored for 24 hours to understand RCA leachate chemistry during the initial contact period. Leachate pH is high upon initial contact with water (pH > 10) and decreases over time as it reacts with atmospheric carbon dioxide. Calcium ion concentration increases rapidly in the initial contact period, then more gradually as calcium saturation is reached. Alkalinity stabilizes (50-65 mg CaCO3/L) after a dramatic increase during the initial contact period.
Authors: Mahdi Farahbakhsh, Vishnu Teja Kunde, Dileep Kalathil, Krishna Narayanan, Jean-Francois Chamberland
Abstract: Diffusion models have emerged as powerful priors for solving inverse problems. However, existing approaches typically overlook side information that could significantly improve reconstruction quality, especially in severely ill-posed settings. In this work, we propose a novel inference-time search algorithm that guides the sampling process using the side information in a manner that balances exploration and exploitation. This enables more accurate and reliable reconstructions, providing an alternative to the gradient-based guidance that is prone to reward-hacking artifacts. Our approach can be seamlessly integrated into a wide range of existing diffusion-based image reconstruction pipelines. Through extensive experiments on a number of inverse problems, such as box inpainting, super-resolution, and various deblurring tasks including motion, Gaussian, nonlinear, and blind deblurring, we show that our approach consistently improves the qualitative and quantitative performance of diffusion-based image reconstruction algorithms. We also show the superior performance of our approach with respect to other baselines, including reward gradient-based guidance algorithms. The code is available at \href{https://github.com/mhdfb/sideinfo-search-reconstruction}{this repository}.
URLs: https://github.com/mhdfb/sideinfo-search-reconstruction
Authors: Ali Kayyam, Anusha Madan Gopal, M. Anthony Lewis
Abstract: We introduce provenance networks, a novel class of neural models designed to provide end-to-end, training-data-driven explainability. Unlike conventional post-hoc methods, provenance networks learn to link each prediction directly to its supporting training examples as part of the model's normal operation, embedding interpretability into the architecture itself. Conceptually, the model operates similarly to a learned KNN, where each output is justified by concrete exemplars weighted by relevance in the feature space. This approach facilitates systematic investigations of the trade-off between memorization and generalization, enables verification of whether a given input was included in the training set, aids in the detection of mislabeled or anomalous data points, enhances resilience to input perturbations, and supports the identification of similar inputs contributing to the generation of a new data point. By jointly optimizing the primary task and the explainability objective, provenance networks offer insights into model behavior that traditional deep networks cannot provide. While the model introduces additional computational cost and currently scales to moderately sized datasets, it provides a complementary approach to existing explainability techniques. In particular, it addresses critical challenges in modern deep learning, including model opaqueness, hallucination, and the assignment of credit to data contributors, thereby improving transparency, robustness, and trustworthiness in neural models.
Authors: Abhi Chawla, David M. Bortz, Vanja Dukic
Abstract: The Weak form Estimation of Nonlinear Dynamics (WENDy) method is a recently proposed class of parameter estimation algorithms that exhibits notable noise robustness and computational efficiency. This work examines the coverage and bias properties of the original WENDy-IRLS algorithm's parameter and state estimators in the context of the following differential equations: Logistic, Lotka-Volterra, FitzHugh-Nagumo, Hindmarsh-Rose, and a Protein Transduction Benchmark. The estimators' performance was studied in simulated data examples, under four different noise distributions (normal, log-normal, additive censored normal, and additive truncated normal), and a wide range of noise, reaching levels much higher than previously tested for this algorithm.
Authors: Zixuan Yi, Sami Abu-el-Haija, Yawen Wang, Teja Vemparala, Yannis Chronis, Yu Gan, Michael Burrows, Carsten Binnig, Bryan Perozzi, Ryan Marcus, Fatma Ozcan
Abstract: DB engines produce efficient query execution plans by relying on cost models. Practical implementations estimate cardinality of queries using heuristics, with magic numbers tuned to improve average performance on benchmarks. Empirically, estimation error significantly grows with query complexity. Alternatively, learning-based estimators offer improved accuracy, but add operational complexity preventing their adoption in-practice. Recognizing that query workloads contain highly repetitive subquery patterns, we learn many simple regressors online, each localized to a pattern. The regressor corresponding to a pattern can be randomly-accessed using hash of graph structure of the subquery. Our method has negligible overhead and competes with SoTA learning-based approaches on error metrics. Further, amending PostgreSQL with our method achieves notable accuracy and runtime improvements over traditional methods and drastically reduces operational costs compared to other learned cardinality estimators, thereby offering the most practical and efficient solution on the Pareto frontier. Concretely, simulating JOB-lite workload on IMDb speeds-up execution by 7.5 minutes (>30%) while incurring only 37 seconds overhead for online learning.
Authors: Jonas J\"ager, Philipp Els\"asser, Elham Torabian
Abstract: Current quantum computers require algorithms that use limited resources economically. In quantum machine learning, success hinges on quantum feature maps, which embed classical data into the state space of qubits. We introduce Quantum Feature-Map Learning via Analytic Iterative Reconstructions (Q-FLAIR), an algorithm that reduces quantum resource overhead in iterative feature-map circuit construction. It shifts workloads to a classical computer via partial analytic reconstructions of the quantum model, using only a few evaluations. For each probed gate addition to the ansatz, the simultaneous selection and optimization of the data feature and weight parameter is then entirely classical. Integrated into quantum neural network and quantum kernel support vector classifiers, Q-FLAIR shows state-of-the-art benchmark performance. Since resource overhead decouples from feature dimension, we train a quantum model on a real IBM device in only four hours, surpassing 90% accuracy on the full-resolution MNIST dataset (784 features, digits 3 vs 5). Such results were previously unattainable, as the feature dimension prohibitively drives hardware demands for fixed and search costs for adaptive ans\"atze. By rethinking feature-map learning beyond black-box optimization, this work takes a concrete step toward enabling quantum machine learning for real-world problems and near-term quantum computers.
Authors: Xiaoyan Bai, Aryan Shrivastava, Ari Holtzman, Chenhao Tan
Abstract: Self-recognition is a crucial metacognitive capability for AI systems, relevant not only for psychological analysis but also for safety, particularly in evaluative scenarios. Motivated by contradictory interpretations of whether models possess self-recognition (Panickssery et al., 2024; Davidson et al., 2024), we introduce a systematic evaluation framework that can be easily applied and updated. Specifically, we measure how well 10 contemporary larger language models (LLMs) can identify their own generated text versus text from other models through two tasks: binary self-recognition and exact model prediction. Different from prior claims, our results reveal a consistent failure in self-recognition. Only 4 out of 10 models predict themselves as generators, and the performance is rarely above random chance. Additionally, models exhibit a strong bias toward predicting GPT and Claude families. We also provide the first evaluation of model awareness of their own and others' existence, as well as the reasoning behind their choices in self-recognition. We find that the model demonstrates some knowledge of its own existence and other models, but their reasoning reveals a hierarchical bias. They appear to assume that GPT, Claude, and occasionally Gemini are the top-tier models, often associating high-quality text with them. We conclude by discussing the implications of our findings on AI safety and future directions to develop appropriate AI self-awareness.
Authors: Zhiying Jiang, Raihan Seraj, Marcos Villagra, Bidhan Roy
Abstract: We present Paris, the first publicly released diffusion model pre-trained entirely through decentralized computation. Paris demonstrates that high-quality text-to-image generation can be achieved without centrally coordinated infrastructure. Paris is open for research and commercial use. Paris required implementing our Distributed Diffusion Training framework from scratch. The model consists of 8 expert diffusion models (129M-605M parameters each) trained in complete isolation with no gradient, parameter, or intermediate activation synchronization. Rather than requiring synchronized gradient updates across thousands of GPUs, we partition data into semantically coherent clusters where each expert independently optimizes its subset while collectively approximating the full distribution. A lightweight transformer router dynamically selects appropriate experts at inference, achieving generation quality comparable to centrally coordinated baselines. Eliminating synchronization enables training on heterogeneous hardware without specialized interconnects. Empirical validation confirms that Paris's decentralized training maintains generation quality while removing the dedicated GPU cluster requirement for large-scale diffusion models. Paris achieves this using 14$\times$ less training data and 16$\times$ less compute than the prior decentralized baseline.
Authors: Chashi Mahiul Islam, Oteo Mamo, Samuel Jacob Chacko, Xiuwen Liu, Weikuan Yu
Abstract: Vision-language models (VLMs) have advanced multimodal reasoning but still face challenges in spatial reasoning for 3D scenes and complex object configurations. To address this, we introduce SpatialViLT, an enhanced VLM that integrates spatial features like depth maps, 3D coordinates, and edge maps through a multi-task learning framework. This approach enriches multimodal embeddings with spatial understanding. We propose two variants: SpatialViLT and MaskedSpatialViLT, focusing on full and masked object regions, respectively. Additionally, SpatialEnsemble combines both approaches, achieving state-of-the-art accuracy. Our models excel in spatial reasoning categories such as directional, topological, and proximity relations, as demonstrated on the challenging Visual Spatial Reasoning (VSR) dataset. This work represents a significant step in enhancing the spatial intelligence of AI systems, crucial for advanced multimodal understanding and real-world applications.
Authors: Ali Khairallah, Arkaitz Zubiaga
Abstract: We introduce ALHD, the first large-scale comprehensive Arabic dataset explicitly designed to distinguish between human- and LLM-generated texts. ALHD spans three genres (news, social media, reviews), covering both MSA and dialectal Arabic, and contains over 400K balanced samples generated by three leading LLMs and originated from multiple human sources, which enables studying generalizability in Arabic LLM-genearted text detection. We provide rigorous preprocessing, rich annotations, and standardized balanced splits to support reproducibility. In addition, we present, analyze and discuss benchmark experiments using our new dataset, in turn identifying gaps and proposing future research directions. Benchmarking across traditional classifiers, BERT-based models, and LLMs (zero-shot and few-shot) demonstrates that fine-tuned BERT models achieve competitive performance, outperforming LLM-based models. Results are however not always consistent, as we observe challenges when generalizing across genres; indeed, models struggle to generalize when they need to deal with unseen patterns in cross-genre settings, and these challenges are particularly prominent when dealing with news articles, where LLM-generated texts resemble human texts in style, which opens up avenues for future research. ALHD establishes a foundation for research related to Arabic LLM-detection and mitigating risks of misinformation, academic dishonesty, and cyber threats.
Authors: Yuan Gao, Anton Rodomanov, Jeremy Rack, Sebastian Stich
Abstract: Communication efficiency is a central challenge in distributed machine learning training, and message compression is a widely used solution. However, standard Error Feedback (EF) methods (Seide et al., 2014), though effective for smooth unconstrained optimization with compression (Karimireddy et al., 2019), fail in the broader and practically important setting of composite optimization, which captures, e.g., objectives consisting of a smooth loss combined with a non-smooth regularizer or constraints. The theoretical foundation and behavior of EF in the context of the general composite setting remain largely unexplored. In this work, we consider composite optimization with EF. We point out that the basic EF mechanism and its analysis no longer stand when a composite part is involved. We argue that this is because of a fundamental limitation in the method and its analysis technique. We propose a novel method that combines Dual Averaging with EControl (Gao et al., 2024), a state-of-the-art variant of the EF mechanism, and achieves for the first time a strong convergence analysis for composite optimization with error feedback. Along with our new algorithm, we also provide a new and novel analysis template for inexact dual averaging method, which might be of independent interest. We also provide experimental results to complement our theoretical findings.
Authors: Mohammad Mohaiminul Islam, Rishabh Anand, David R. Wessels, Friso de Kruiff, Thijs P. Kuipers, Rex Ying, Clara I. S\'anchez, Sharvaree Vadgama, Georg B\"okman, Erik J. Bekkers
Abstract: While widespread, Transformers lack inductive biases for geometric symmetries common in science and computer vision. Existing equivariant methods often sacrifice the efficiency and flexibility that make Transformers so effective through complex, computationally intensive designs. We introduce the Platonic Transformer to resolve this trade-off. By defining attention relative to reference frames from the Platonic solid symmetry groups, our method induces a principled weight-sharing scheme. This enables combined equivariance to continuous translations and Platonic symmetries, while preserving the exact architecture and computational cost of a standard Transformer. Furthermore, we show that this attention is formally equivalent to a dynamic group convolution, which reveals that the model learns adaptive geometric filters and enables a highly scalable, linear-time convolutional variant. Across diverse benchmarks in computer vision (CIFAR-10), 3D point clouds (ScanObjectNN), and molecular property prediction (QM9, OMol25), the Platonic Transformer achieves competitive performance by leveraging these geometric constraints at no additional cost.
Authors: Nicol\`o Dal Fabbro, Milad Mesbahi, Renato Mendes, Jo\~ao Borges de Sousa, George J. Pappas
Abstract: We study the problem of long-term (multiple days) mapping of a river plume using multiple autonomous underwater vehicles (AUVs), focusing on the Douro river representative use-case. We propose an energy - and communication - efficient multi-agent reinforcement learning approach in which a central coordinator intermittently communicates with the AUVs, collecting measurements and issuing commands. Our approach integrates spatiotemporal Gaussian process regression (GPR) with a multi-head Q-network controller that regulates direction and speed for each AUV. Simulations using the Delft3D ocean model demonstrate that our method consistently outperforms both single- and multi-agent benchmarks, with scaling the number of agents both improving mean squared error (MSE) and operational endurance. In some instances, our algorithm demonstrates that doubling the number of AUVs can more than double endurance while maintaining or improving accuracy, underscoring the benefits of multi-agent coordination. Our learned policies generalize across unseen seasonal regimes over different months and years, demonstrating promise for future developments of data-driven long-term monitoring of dynamic plume environments.
Authors: Adam Filipek
Abstract: The Transformer architecture has become the de facto standard for Large Language Models (LLMs), demonstrating remarkable capabilities in language understanding and generation. However, its application in conversational AI is fundamentally constrained by its stateless nature and the quadratic computational complexity ($O(L^2)$) with respect to sequence length $L$. Current models emulate memory by reprocessing an ever-expanding conversation history with each turn, leading to prohibitive costs and latency in long dialogues. This paper introduces the Reactive Transformer (RxT), a novel architecture designed to overcome these limitations by shifting from a data-driven to an event-driven paradigm. RxT processes each conversational turn as a discrete event in real-time, maintaining context in an integrated, fixed-size Short-Term Memory (STM) system. The architecture features a distinct operational cycle where a generator-decoder produces a response based on the current query and the previous memory state, after which a memory-encoder and a dedicated Memory Attention network asynchronously update the STM with a representation of the complete interaction. This design fundamentally alters the scaling dynamics, reducing the total user-facing cost of a conversation from quadratic ($O(N^2 \cdot T)$) to linear ($O(N \cdot T)$) with respect to the number of interactions $N$. By decoupling response generation from memory updates, RxT achieves low latency, enabling truly real-time, stateful, and economically viable long-form conversations. We validated our architecture with a series of proof-of-concept experiments on synthetic data, demonstrating superior performance and constant-time inference latency compared to a baseline stateless model of comparable size.
Authors: Sina Alemohammad, Zhangyang Wang, Richard G. Baraniuk
Abstract: Scaling generative AI models is bottlenecked by the scarcity of high-quality training data. The ease of synthesizing from a generative model suggests using (unverified) synthetic data to augment a limited corpus of real data for the purpose of fine-tuning in the hope of improving performance. Unfortunately, however, the resulting positive feedback loop leads to model autophagy disorder (MAD, aka model collapse) that results in a rapid degradation in sample quality and/or diversity. In this paper, we introduce Neon (for Negative Extrapolation frOm self-traiNing), a new learning method that turns the degradation from self-training into a powerful signal for self-improvement. Given a base model, Neon first fine-tunes it on its own self-synthesized data but then, counterintuitively, reverses its gradient updates to extrapolate away from the degraded weights. We prove that Neon works because typical inference samplers that favor high-probability regions create a predictable anti-alignment between the synthetic and real data population gradients, which negative extrapolation corrects to better align the model with the true data distribution. Neon is remarkably easy to implement via a simple post-hoc merge that requires no new real data, works effectively with as few as 1k synthetic samples, and typically uses less than 1% additional training compute. We demonstrate Neon's universality across a range of architectures (diffusion, flow matching, autoregressive, and inductive moment matching models) and datasets (ImageNet, CIFAR-10, and FFHQ). In particular, on ImageNet 256x256, Neon elevates the xAR-L model to a new state-of-the-art FID of 1.02 with only 0.36% additional training compute. Code is available at https://github.com/SinaAlemohammad/Neon
Authors: Alexander V. Mantzaris
Abstract: This paper asks whether the Hierarchical Reasoning Model (HRM) with the two Transformer-style modules $(f_L,f_H)$, one step (DEQ-style) training, deep supervision, Rotary Position Embeddings, and RMSNorm can serve as a practical image classifier. It is evaluated on MNIST, CIFAR-10, and CIFAR-100 under a deliberately raw regime: no data augmentation, identical optimizer family with one-epoch warmup then cosine-floor decay, and label smoothing. HRM optimizes stably and performs well on MNIST ($\approx 98\%$ test accuracy), but on small natural images it overfits and generalizes poorly: on CIFAR-10, HRM reaches 65.0\% after 25 epochs, whereas a two-stage Conv--BN--ReLU baseline attains 77.2\% while training $\sim 30\times$ faster per epoch; on CIFAR-100, HRM achieves only 29.7\% test accuracy despite 91.5\% train accuracy, while the same CNN reaches 45.3\% test with 50.5\% train accuracy. Loss traces and error analyses indicate healthy optimization but insufficient image-specific inductive bias for HRM in this regime. It is concluded that, for small-resolution image classification without augmentation, HRM is not competitive with even simple convolutional architectures as the HRM currently exist but this does not exclude possibilities that modifications to the model may allow it to improve greatly.
Authors: Adel Javanmard, Baharan Mirzasoleiman, Vahab Mirrokni
Abstract: Test-time scaling improves the reasoning capabilities of large language models (LLMs) by allocating extra compute to generate longer Chains-of-Thoughts (CoTs). This enables models to tackle more complex problem by breaking them down into additional steps, backtracking, and correcting mistakes. Despite its strong performance--demonstrated by OpenAI's o1 and DeepSeek R1, the conditions in the training data under which long CoTs emerge, and when such long CoTs improve the performance, remain unclear. In this paper, we study the performance of test-time scaling for transformers trained on an in-context weight prediction task for linear regression. Our analysis provides a theoretical explanation for several intriguing observations: First, at any fixed test error, increasing test-time compute allows us to reduce the number of in-context examples (context length) in training prompts. Second, if the skills required to solve a downstream task are not sufficiently present in the training data, increasing test-time compute can harm performance. Finally, we characterize task hardness via the smallest eigenvalue of its feature covariance matrix and show that training on a diverse, relevant, and hard set of tasks results in best performance for test-time scaling. We confirm our findings with experiments on large, nonlinear transformer architectures.
Authors: Mattia Scardecchia
Abstract: Recent advances in self-supervised learning (SSL) have made it possible to learn general-purpose visual features that capture both the high-level semantics and the fine-grained spatial structure of images. Most notably, the recent DINOv2 has established a new state of the art by surpassing weakly supervised methods (WSL) like OpenCLIP on most benchmarks. In this survey, we examine the core ideas behind its approach, multi-crop view augmentation and self-distillation with a mean teacher, and trace their development in previous work. We then compare the performance of DINO and DINOv2 with other SSL and WSL methods across various downstream tasks, and highlight some remarkable emergent properties of their learned features with transformer backbones. We conclude by briefly discussing DINOv2's limitations, its impact, and future research directions.
Authors: Raquib Bin Yousuf, Aadyant Khatri, Shengzhe Xu, Mandar Sharma, Naren Ramakrishnan
Abstract: Recently proposed evaluation benchmarks aim to characterize the effective context length and the forgetting tendencies of large language models (LLMs). However, these benchmarks often rely on simplistic 'needle in a haystack' retrieval or continuation tasks that may not accurately reflect the performance of these models in information-dense scenarios. Thus, rather than simple next token prediction, we argue for evaluating these models on more complex reasoning tasks that requires them to induce structured relational knowledge from the text - such as graphs from potentially noisy natural language content. While the input text can be viewed as generated in terms of a graph, its structure is not made explicit and connections must be induced from distributed textual cues, separated by long contexts and interspersed with irrelevant information. Our findings reveal that LLMs begin to exhibit memory drift and contextual forgetting at much shorter effective lengths when tasked with this form of relational reasoning, compared to what existing benchmarks suggest. With these findings, we offer recommendations for the optimal use of popular LLMs for complex reasoning tasks. We further show that even models specialized for reasoning, such as OpenAI o1, remain vulnerable to early memory drift in these settings. These results point to significant limitations in the models' ability to abstract structured knowledge from unstructured input and highlight the need for architectural adaptations to improve long-range reasoning.
Authors: Nikitin Nikita
Abstract: The introduction of machine learning methods has led to significant advances in automation, optimization, and discoveries in various fields of science and technology. However, their widespread application faces a fundamental limitation: the transfer of models between data domains generally lacks a rigorous mathematical justification. The key problem is the lack of formal criteria to guarantee that a model trained on one type of data will retain its properties on another.This paper proposes a solution to this problem by formalizing the concept of analogy between data sets and models using first-order logic and Hoare logic.We formulate and rigorously prove a theorem that sets out the necessary and sufficient conditions for analogy in the task of knowledge transfer between machine learning models. Practical verification of the analogy theorem on model data obtained using the Monte Carlo method, as well as on MNIST and USPS data, allows us to achieving F1 scores of 0.84 and 0.88 for convolutional neural networks and random forests, respectively.The proposed approach not only allows us to justify the correctness of transfer between domains but also provides tools for comparing the applicability of models to different types of data.The main contribution of the work is a rigorous formalization of analogy at the level of program logic, providing verifiable guarantees of the correctness of knowledge transfer, which opens new opportunities for both theoretical research and the practical use of machine learning models in previously inaccessible areas.
Authors: Raaghav Malik, Satpreet H. Singh, Sonja Johnson-Yu, Nathan Wu, Roy Harpaz, Florian Engert, Kanaka Rajan
Abstract: Larval zebrafish hunting provides a tractable setting to study how ecological and energetic constraints shape adaptive behavior in both biological brains and artificial agents. Here we develop a minimal agent-based model, training recurrent policies with deep reinforcement learning in a bout-based zebrafish simulator. Despite its simplicity, the model reproduces hallmark hunting behaviors -- including eye vergence-linked pursuit, speed modulation, and stereotyped approach trajectories -- that closely match real larval zebrafish. Quantitative trajectory analyses show that pursuit bouts systematically reduce prey angle by roughly half before strike, consistent with measurements. Virtual experiments and parameter sweeps vary ecological and energetic constraints, bout kinematics (coupled vs. uncoupled turns and forward motion), and environmental factors such as food density, food speed, and vergence limits. These manipulations reveal how constraints and environments shape pursuit dynamics, strike success, and abort rates, yielding falsifiable predictions for neuroscience experiments. These sweeps identify a compact set of constraints -- binocular sensing, the coupling of forward speed and turning in bout kinematics, and modest energetic costs on locomotion and vergence -- that are sufficient for zebrafish-like hunting to emerge. Strikingly, these behaviors arise in minimal agents without detailed biomechanics, fluid dynamics, circuit realism, or imitation learning from real zebrafish data. Taken together, this work provides a normative account of zebrafish hunting as the optimal balance between energetic cost and sensory benefit, highlighting the trade-offs that structure vergence and trajectory dynamics. We establish a virtual lab that narrows the experimental search space and generates falsifiable predictions about behavior and neural coding.
Authors: Eadom Dessalene, Pavan Mantripragada, Michael Maynord, Yiannis Aloimonos
Abstract: We introduce EmbodiSwap - a method for producing photorealistic synthetic robot overlays over human video. We employ EmbodiSwap for zero-shot imitation learning, bridging the embodiment gap between in-the-wild ego-centric human video and a target robot embodiment. We train a closed-loop robot manipulation policy over the data produced by EmbodiSwap. We make novel use of V-JEPA as a visual backbone, repurposing V-JEPA from the domain of video understanding to imitation learning over synthetic robot videos. Adoption of V-JEPA outperforms alternative vision backbones more conventionally used within robotics. In real-world tests, our zero-shot trained V-JEPA model achieves an $82\%$ success rate, outperforming a few-shot trained $\pi_0$ network as well as $\pi_0$ trained over data produced by EmbodiSwap. We release (i) code for generating the synthetic robot overlays which takes as input human videos and an arbitrary robot URDF and generates a robot dataset, (ii) the robot dataset we synthesize over EPIC-Kitchens, HOI4D and Ego4D, and (iii) model checkpoints and inference code, to facilitate reproducible research and broader adoption.
Authors: Leander Girrbach, Stephan Alaniz, Genevieve Smith, Trevor Darrell, Zeynep Akata
Abstract: Vision-language models trained on large-scale multimodal datasets show strong demographic biases, but the role of training data in producing these biases remains unclear. A major barrier has been the lack of demographic annotations in web-scale datasets such as LAION-400M. We address this gap by creating person-centric annotations for the full dataset, including over 276 million bounding boxes, perceived gender and race/ethnicity labels, and automatically generated captions. These annotations are produced through validated automatic labeling pipelines combining object detection, multimodal captioning, and finetuned classifiers. Using them, we uncover demographic imbalances and harmful associations, such as the disproportionate linking of men and individuals perceived as Black or Middle Eastern with crime-related and negative content. We also show that 60-70% of gender bias in CLIP and Stable Diffusion can be linearly explained by direct co-occurrences in the data. Our resources establish the first large-scale empirical link between dataset composition and downstream model bias.
Authors: Thomas Hallopeau, Joris Gu\'erin, Laurent Demagistri, Youssef Fouzai, Renata Gracie, Vanderlei Pascoal De Matos, Helen Gurgel, Nadine Dessay
Abstract: While deep learning methods for detecting informal settlements have already been developed, they have not yet fully utilized the potential offered by recent pretrained neural networks. We compare two types of pretrained neural networks for detecting the favelas of Rio de Janeiro: 1. Generic networks pretrained on large diverse datasets of unspecific images, 2. A specialized network pretrained on satellite imagery. While the latter is more specific to the target task, the former has been pretrained on significantly more images. Hence, this research investigates whether task specificity or data volume yields superior performance in urban informal settlement detection.
Authors: Xuehai He
Abstract: Humans understand the world through the integration of multiple sensory modalities, enabling them to perceive, reason about, and imagine dynamic physical processes. Inspired by this capability, multimodal foundation models (MFMs) have emerged as powerful tools for multimodal understanding and generation. However, today's MFMs fall short of serving as effective world models. They lack the essential ability such as perform counterfactual reasoning, simulate dynamics, understand the spatiotemporal information, control generated visual outcomes, and perform multifaceted reasoning. We investigates what it takes to bridge the gap between multimodal foundation models and world models. We begin by improving the reasoning capabilities of MFMs through discriminative tasks and equipping MFMs with structured reasoning skills, such as causal inference, counterfactual thinking, and spatiotemporal reasoning, enabling them to go beyond surface correlations and understand deeper relationships within visual and textual data. Next, we explore generative capabilities of multimodal foundation models across both image and video modalities, introducing new frameworks for structured and controllable generation. Our approaches incorporate scene graphs, multimodal conditioning, and multimodal alignment strategies to guide the generation process, ensuring consistency with high-level semantics and fine-grained user intent. We further extend these techniques to controllable 4D generation, enabling interactive, editable, and morphable object synthesis over time and space.
Authors: Kuang Yuan, Yang Gao, Xilin Li, Xinhao Mei, Syavosh Zadissa, Tarun Pruthi, Saeed Bagheri Sereshki
Abstract: Acoustic scene classification (ASC) models on edge devices typically operate under fixed class assumptions, lacking the transferability needed for real-world applications that require adaptation to new or refined acoustic categories. We propose ContrastASC, which learns generalizable acoustic scene representations by structuring the embedding space to preserve semantic relationships between scenes, enabling adaptation to unseen categories without retraining. Our approach combines supervised contrastive fine-tuning of pre-trained models with contrastive representation distillation to transfer this structured knowledge to compact student models. Our evaluation shows that ContrastASC demonstrates improved few-shot adaptation to unseen categories while maintaining strong closed-set performance.
Authors: Tiago Rodrigues de Almeida, Yufei Zhu, Andrey Rudenko, Tomasz P. Kucner, Johannes A. Stork, Martin Magnusson, Achim J. Lilienthal
Abstract: Robots and other intelligent systems navigating in complex dynamic environments should predict future actions and intentions of surrounding agents to reach their goals efficiently and avoid collisions. The dynamics of those agents strongly depends on their tasks, roles, or observable labels. Class-conditioned motion prediction is thus an appealing way to reduce forecast uncertainty and get more accurate predictions for heterogeneous agents. However, this is hardly explored in the prior art, especially for mobile robots and in limited data applications. In this paper, we analyse different class-conditioned trajectory prediction methods on two datasets. We propose a set of conditional pattern-based and efficient deep learning-based baselines, and evaluate their performance on robotics and outdoors datasets (TH\"OR-MAGNI and Stanford Drone Dataset). Our experiments show that all methods improve accuracy in most of the settings when considering class labels. More importantly, we observe that there are significant differences when learning from imbalanced datasets, or in new environments where sufficient data is not available. In particular, we find that deep learning methods perform better on balanced datasets, but in applications with limited data, e.g., cold start of a robot in a new environment, or imbalanced classes, pattern-based methods may be preferable.
Authors: Divij Handa, Mihir Parmar, Aswin RRV, Md Nayem Uddin, Hamid Palangi, Chitta Baral
Abstract: Repeated Sampling (RS) is a simple inference-time algorithm that has been shown to improve model performance on complex tasks. Although it is an effective way of scaling inference time, it often struggles to generate diverse solution candidates, frequently relying on the same underlying approach to solve the problem and thus producing redundant samples. To address this limitation, we propose a new inference algorithm, GuidedSampling, which decouples the exploration and generation phases during inference, increasing diversity of generated candidate solutions. The exploration phase identifies multiple concepts that can be utilized to solve the problem, while the generation phase applies a specific concept to provide final solution candidates. We first define the theoretical bounds of GuidedSampling and then empirically demonstrate that it improves the performance of base model at pass@50 by on an average ~21.6% across various benchmarks compared to RS. Furthermore, models trained on trajectories of GuidedSampling exhibit substantial performance improvements at pass@5 by on an average ~9.7%, compared to models trained on traditional RS. Additionally, models trained with GuidedSampling increases the average number of concepts per instance (1.67 -> 3.03), yielding a diverse set of candidates than traditional RS.
Authors: Yiqiao Chen
Abstract: Cardiovascular disease (CVD) is a major pediatric health burden, and early screening is of critical importance. Electrocardiography (ECG), as a noninvasive and accessible tool, is well suited for this purpose. This paper presents the first benchmark study of deep learning for multi-label pediatric CVD classification on the recently released ZZU-pECG dataset, comprising 3716 recordings with 19 CVD categories. We systematically evaluate four representative paradigms--ResNet-1D, BiLSTM, Transformer, and Mamba 2--under both 9-lead and 12-lead configurations. All models achieved strong results, with Hamming Loss as low as 0.0069 and F1-scores above 85% in most settings. ResNet-1D reached a macro-F1 of 94.67% on the 12-lead subset, while BiLSTM and Transformer also showed competitive performance. Per-class analysis indicated challenges for rare conditions such as hypertrophic cardiomyopathy in the 9-lead subset, reflecting the effect of limited positive samples. This benchmark establishes reusable baselines and highlights complementary strengths across paradigms. It further points to the need for larger-scale, multi-center validation, age-stratified analysis, and broader disease coverage to support real-world pediatric ECG applications.
Authors: Rasel Hossen, Diptajoy Mistry, Mushiur Rahman, Waki As Sami Atikur Rahman Hridoy, Sajib Saha, Muhammad Ibrahim
Abstract: Urban safety and infrastructure maintenance are critical components of smart city development. Manual monitoring of road damages is time-consuming, highly costly, and error-prone. This paper presents a deep learning approach for automated road damage and manhole detection using the YOLOv9 algorithm with polygonal annotations. Unlike traditional bounding box annotation, we employ polygonal annotations for more precise localization of road defects. We develop a novel dataset comprising more than one thousand images which are mostly collected from Dhaka, Bangladesh. This dataset is used to train a YOLO-based model for three classes, namely Broken, Not Broken, and Manhole. We achieve 78.1% overall image-level accuracy. The YOLOv9 model demonstrates strong performance for Broken (86.7% F1-score) and Not Broken (89.2% F1-score) classes, with challenges in Manhole detection (18.2% F1-score) due to class imbalance. Our approach offers an efficient and scalable solution for monitoring urban infrastructure in developing countries.
Authors: Vaskar Chakma, Wooyeol Choi
Abstract: Current Cyber-Physical Systems (CPS) integrated with Digital Twin (DT) technology face critical limitations in achieving real-time performance for mission-critical industrial applications. Existing 5G-enabled systems suffer from latencies exceeding 10ms, which are inadequate for applications requiring sub-millisecond response times, such as autonomous industrial control and predictive maintenance. This research aims to develop and validate a 6G-enabled Digital Twin framework that achieves ultra-low latency communication and real-time synchronization between physical industrial assets and their digital counterparts, specifically targeting bearing fault detection as a critical industrial use case. The proposed framework integrates terahertz communications (0.1-1 THz), intelligent reflecting surfaces, and edge artificial intelligence within a five-layer architecture. Experimental validation was conducted using the Case Western Reserve University (CWRU) bearing dataset, implementing comprehensive feature extraction (15 time and frequency domain features) and Random Forest classification algorithms. The system performance was evaluated against traditional WiFi-6 and 5G networks across multiple metrics, including classification accuracy, end-to-end latency, and scalability. It achieved 97.7% fault classification accuracy with 0.8ms end-to-end latency, representing a 15.6x improvement over WiFi-6 (12.5ms) and 5.25x improvement over 5G (4.2ms) networks. The system demonstrated superior scalability with sub-linear processing time growth and maintained consistent performance across four bearing fault categories (normal, inner race, outer race, and ball faults) with macro-averaged F1-scores exceeding 97%.
Authors: William Hao-Cheng Huang
Abstract: In high-dimensional learning, models remain stable until they collapse abruptly once the sample size falls below a critical level. This instability is not algorithm-specific but a geometric mechanism: when the weakest Fisher eigendirection falls beneath sample-level fluctuations, identifiability fails. Our Fisher Threshold Theorem formalizes this by proving that stability requires the minimal Fisher eigenvalue to exceed an explicit $O(\sqrt{d/n})$ bound. Unlike prior asymptotic or model-specific criteria, this threshold is finite-sample and necessary, marking a sharp phase transition between reliable concentration and inevitable failure. To make the principle constructive, we introduce the Fisher floor, a verifiable spectral regularization robust to smoothing and preconditioning. Synthetic experiments on Gaussian mixtures and logistic models confirm the predicted transition, consistent with $d/n$ scaling. Statistically, the threshold sharpens classical eigenvalue conditions into a non-asymptotic law; learning-theoretically, it defines a spectral sample-complexity frontier, bridging theory with diagnostics for robust high-dimensional inference.
Authors: Shankar Prasad Sastry
Abstract: I present a regression algorithm that provides a continuous, piecewise-smooth function approximating scattered data. It is based on composing and blending linear functions over Voronoi cells, and it scales to high dimensions. The algorithm infers Voronoi cells from seed vertices and constructs a linear function for the input data in and around each cell. As the algorithm does not explicitly compute the Voronoi diagram, it avoids the curse of dimensionality. An accuracy of around 98.2% on the MNIST dataset with 722,200 degrees of freedom (without data augmentation, convolution, or other geometric operators) demonstrates the applicability and scalability of the algorithm.
Authors: Byungjun Kim, Soobin Um, Jong Chul Ye
Abstract: Text-to-image (T2I) diffusion models have demonstrated impressive performance in generating high-fidelity images, largely enabled by text-guided inference. However, this advantage often comes with a critical drawback: limited diversity, as outputs tend to collapse into similar modes under strong text guidance. Existing approaches typically optimize intermediate latents or text conditions during inference, but these methods deliver only modest gains or remain sensitive to hyperparameter tuning. In this work, we introduce Contrastive Noise Optimization, a simple yet effective method that addresses the diversity issue from a distinct perspective. Unlike prior techniques that adapt intermediate latents, our approach shapes the initial noise to promote diverse outputs. Specifically, we develop a contrastive loss defined in the Tweedie data space and optimize a batch of noise latents. Our contrastive optimization repels instances within the batch to maximize diversity while keeping them anchored to a reference sample to preserve fidelity. We further provide theoretical insights into the mechanism of this preprocessing to substantiate its effectiveness. Extensive experiments across multiple T2I backbones demonstrate that our approach achieves a superior quality-diversity Pareto frontier while remaining robust to hyperparameter choices.
Authors: Yue wu
Abstract: There are limitations of traditional methods and deep learning methods in terms of interpretability, generalization, and quantification of uncertainty in industrial fault diagnosis, and there are core problems of insufficient credibility in industrial fault diagnosis. The architecture performs preliminary analysis through a Bayesian network-based diagnostic engine and features an LLM-driven cognitive quorum module with multimodal input capabilities. The module conducts expert-level arbitration of initial diagnoses by analyzing structured features and diagnostic charts, prioritizing final decisions after conflicts are identified. To ensure the reliability of the system output, the architecture integrates a confidence calibration module based on temperature calibration and a risk assessment module, which objectively quantifies the reliability of the system using metrics such as expected calibration error (ECE). Experimental results on a dataset containing multiple fault types showed that the proposed framework improved diagnostic accuracy by more than 28 percentage points compared to the baseline model, while the calibrated ECE was reduced by more than 75%. Case studies have confirmed that HCAA effectively corrects misjudgments caused by complex feature patterns or knowledge gaps in traditional models, providing novel and practical engineering solutions for building high-trust, explainable AI diagnostic systems for industrial applications.
Authors: Pedro Ivo da Cruz, Dimitri Silva, Tito Spadini, Ricardo Suyama, Murilo Bellezoni Loiola
Abstract: Massive multiple-input multiple-output (MMIMO) is essential to modern wireless communication systems, like 5G and 6G, but it is vulnerable to active eavesdropping attacks. One type of such attack is the pilot contamination attack (PCA), where a malicious user copies pilot signals from an authentic user during uplink, intentionally interfering with the base station's (BS) channel estimation accuracy. In this work, we propose to use a Decision Tree (DT) algorithm for PCA detection at the BS in a multi-user system. We present a methodology to generate training data for the DT classifier and select the best DT according to their depth. Then, we simulate different scenarios that could be encountered in practice and compare the DT to a classical technique based on likelihood ratio testing (LRT) submitted to the same scenarios. The results revealed that a DT with only one level of depth is sufficient to outperform the LRT. The DT shows a good performance regarding the probability of detection in noisy scenarios and when the malicious user transmits with low power, in which case the LRT fails to detect the PCA. We also show that the reason for the good performance of the DT is its ability to compute a threshold that separates PCA data from non-PCA data better than the LRT's threshold. Moreover, the DT does not necessitate prior knowledge of noise power or assumptions regarding the signal power of malicious users, prerequisites typically essential for LRT and other hypothesis testing methodologies.
Authors: Vincent Nguyen, Guilherme Herzog, Jos\'e Cambronero, Marcus Revaj, Aditya Kini, Alexander Fr\"ommgen, Maxim Tabachnyk
Abstract: Manually editing pasted code is a long-standing developer pain point. In internal software development at Google, we observe that code is pasted 4 times more often than it is manually typed. These paste actions frequently require follow-up edits, ranging from simple reformatting and renaming to more complex style adjustments and cross-language translations. Prior work has shown deep learning can be used to predict these edits. In this work, we show how to iteratively develop and scale Smart Paste, an IDE feature for post-paste edit suggestions, to Google's development environment. This experience can serve as a guide for AI practitioners on a holistic approach to feature development, covering user experience, system integration, and model capabilities. Since deployment, Smart Paste has had overwhelmingly positive feedback with a 45% acceptance rate. At Google's enterprise scale, these accepted suggestions account substantially for over 1% of all code written company-wide.
Authors: Gon Buzaglo, Noah Golowich, Elad Hazan
Abstract: This paper investigates a class of games with large strategy spaces, motivated by challenges in AI alignment and language games. We introduce the hidden game problem, where for each player, an unknown subset of strategies consistently yields higher rewards compared to the rest. The central question is whether efficient regret minimization algorithms can be designed to discover and exploit such hidden structures, leading to equilibrium in these subgames while maintaining rationality in general. We answer this question affirmatively by developing a composition of regret minimization techniques that achieve optimal external and swap regret bounds. Our approach ensures rapid convergence to correlated equilibria in hidden subgames, leveraging the hidden game structure for improved computational efficiency.
Authors: Raghav Sharma, Manan Mehta
Abstract: Small language models (SLMs; 1-12B params, sometimes up to 20B) are sufficient and often superior for agentic workloads where the objective is schema- and API-constrained accuracy rather than open-ended generation. We synthesize recent evidence across open and proprietary SLMs (Phi-4-Mini, Qwen-2.5-7B, Gemma-2-9B, Llama-3.2-1B/3B, Ministral-3B/8B, Apple on-device 3B, DeepSeek-R1-Distill) and connect it to modern evaluations (BFCL v3/v4, StableToolBench) and serving stacks (vLLM, SGLang, TensorRT-LLM) paired with guided decoding libraries (XGrammar, Outlines). We formalize SLM-default, LLM-fallback systems with uncertainty-aware routing and verifier cascades, and propose engineering metrics that reflect real production goals: cost per successful task (CPS), schema validity rate, executable call rate, p50/p95 latency, and energy per request. Guided decoding, strict JSON Schema outputs, and validator-first tool execution close much of the capability gap with larger models and often let SLMs match or surpass LLMs on tool use, function calling, and RAG at 10x-100x lower token cost with materially better latency and energy. We provide design patterns for agent stacks that prioritize SLMs: schema-first prompting, type-safe function registries, confidence scoring with verifier rollups, and lightweight adaptation via LoRA/QLoRA. We also delineate limits where fallback remains valuable (open-domain reasoning and some long-horizon planning). The result is a practical blueprint for building fast, inexpensive, and reliable agents that default to SLMs while preserving headroom with targeted LLM assistance. Keywords: small language models, agents, function calling, structured outputs, JSON Schema, guided decoding, LoRA/QLoRA, routing, energy efficiency, edge inference
Authors: Raghav Sharma, Manan Mehta
Abstract: Ensuring that critical IoT systems function safely and smoothly depends a lot on finding anomalies quickly. As more complex systems, like smart healthcare, energy grids and industrial automation, appear, it is easier to see the shortcomings of older methods of detection. Monitoring failures usually happen in dynamic, high dimensional situations, especially when data is incomplete, messy or always evolving. Such limits point out the requirement for adaptive, intelligent systems that always improve and think. LLMs are now capable of significantly changing how context is understood and semantic inference is done across all types of data. This proposal suggests using an LLM supported contextual reasoning method along with XAI agents to improve how anomalies are found in significant IoT environments. To discover hidden patterns and notice inconsistencies in data streams, it uses attention methods, avoids dealing with details from every time step and uses memory buffers with meaning. Because no code AI stresses transparency and interpretability, people can check and accept the AI's decisions, helping ensure AI follows company policies. The two architectures are put together in a test that compares the results of the traditional model with those of the suggested LLM enhanced model. Important measures to check are the accuracy of detection, how much inaccurate information is included in the results, how clearly the findings can be read and how fast the system responds under different test situations. The metaheuristic is tested in simulations of real world smart grid and healthcare contexts to check its adaptability and reliability. From the study, we see that the new approach performs much better than most existing models in both accuracy and interpretation, so it could be a good fit for future anomaly detection tasks in IoT
Authors: Lutz Oettershagen, Othon Michail
Abstract: Balancing resource efficiency and fairness is critical in networked systems that support modern learning applications. We introduce the Fair Minimum Labeling (FML) problem: the task of designing a minimum-cost temporal edge activation plan that ensures each group of nodes in a network has sufficient access to a designated target set, according to specified coverage requirements. FML captures key trade-offs in systems where edge activations incur resource costs and equitable access is essential, such as distributed data collection, update dissemination in edge-cloud systems, and fair service restoration in critical infrastructure. We show that FML is NP-hard and $\Omega(\log |V|)$-hard to approximate, and we present probabilistic approximation algorithms that match this bound, achieving the best possible guarantee for the activation cost. We demonstrate the practical utility of FML in a fair multi-source data aggregation task for training a shared model. Empirical results show that FML enforces group-level fairness with substantially lower activation cost than baseline heuristics, underscoring its potential for building resource-efficient, equitable temporal reachability in learning-integrated networks.
Authors: Jinghao Lyu, Kyle J. Ray, James P. Crutchfield
Abstract: The energy cost of computation has emerged as a central challenge at the intersection of physics and computer science. Recent advances in statistical physics -- particularly in stochastic thermodynamics -- enable precise characterizations of work, heat, and entropy production in information-processing systems driven far from equilibrium by time-dependent control protocols. A key open question is then how to design protocols that minimize thermodynamic cost while ensur- ing correct outcomes. To this end, we develop a unified framework to identify optimal protocols using fluctuation response relations (FRR) and machine learning. Unlike previous approaches that optimize either distributions or protocols separately, our method unifies both using FRR-derived gradients. Moreover, our method is based primarily on iteratively learning from sampled noisy trajectories, which is generally much easier than solving for the optimal protocol directly from a set of governing equations. We apply the framework to canonical examples -- bit erasure in a double-well potential and translating harmonic traps -- demonstrating how to construct loss functions that trade-off energy cost against task error. The framework extends trivially to underdamped systems, and we show this by optimizing a bit-flip in an underdamped system. In all computations we test, the framework achieves the theoretically optimal protocol or achieves work costs comparable to relevant finite time bounds. In short, the results provide principled strategies for designing thermodynamically efficient protocols in physical information-processing systems. Applications range from quantum gates robust under noise to energy-efficient control of chemical and synthetic biological networks.
Authors: Andrew Campbell, Valentin De Bortoli, Jiaxin Shi, Arnaud Doucet
Abstract: We present self-speculative masked diffusions, a new class of masked diffusion generative models for discrete data that require significantly fewer function evaluations to generate samples. Standard masked diffusion models predict factorized logits over currently masked positions. A number of masked positions are then sampled, however, the factorization approximation means that sampling too many positions in one go leads to poor sample quality. As a result, many simulation steps and therefore neural network function evaluations are required to generate high-quality data. We reduce the computational burden by generating non-factorized predictions over masked positions. This is achieved by modifying the final transformer attention mask from non-causal to causal, enabling draft token generation and parallel validation via a novel, model-integrated speculative sampling mechanism. This results in a non-factorized predictive distribution over masked positions in a single forward pass. We apply our method to GPT2 scale text modelling and protein sequences generation, finding that we can achieve a ~2x reduction in the required number of network forward passes relative to standard masked diffusion models.
Authors: Chengxiao Wang, Isha Chaudhary, Qian Hu, Weitong Ruan, Rahul Gupta, Gagandeep Singh
Abstract: Large Language Models (LLMs) can produce catastrophic responses in conversational settings that pose serious risks to public safety and security. Existing evaluations often fail to fully reveal these vulnerabilities because they rely on fixed attack prompt sequences, lack statistical guarantees, and do not scale to the vast space of multi-turn conversations. In this work, we propose QRLLM, a novel, principled Certification framework for Catastrophic risks in multi-turn Conversation for LLMs that bounds the probability of an LLM generating catastrophic responses under multi-turn conversation distributions with statistical guarantees. We model multi-turn conversations as probability distributions over query sequences, represented by a Markov process on a query graph whose edges encode semantic similarity to capture realistic conversational flow, and quantify catastrophic risks using confidence intervals. We define several inexpensive and practical distributions: random node, graph path, adaptive with rejection. Our results demonstrate that these distributions can reveal substantial catastrophic risks in frontier models, with certified lower bounds as high as 70\% for the worst model, highlighting the urgent need for improved safety training strategies in frontier LLMs.
Authors: Kirandeep Kaur, Preetam Prabhu Srikar Dammu, Hideo Joho, Chirag Shah
Abstract: Personalized AI agents are becoming central to modern information retrieval, yet most evaluation methodologies remain static, relying on fixed benchmarks and one-off metrics that fail to reflect how users' needs evolve over time. These limitations hinder our ability to assess whether agents can meaningfully adapt to individuals across dynamic, longitudinal interactions. In this perspective paper, we propose a conceptual lens for rethinking evaluation in adaptive personalization, shifting the focus from static performance snapshots to interaction-aware, evolving assessments. We organize this lens around three core components: (1) persona-based user simulation with temporally evolving preference models; (2) structured elicitation protocols inspired by reference interviews to extract preferences in context; and (3) adaptation-aware evaluation mechanisms that measure how agent behavior improves across sessions and tasks. While recent works have embraced LLM-driven user simulation, we situate this practice within a broader paradigm for evaluating agents over time. To illustrate our ideas, we conduct a case study in e-commerce search using the PersonalWAB dataset. Beyond presenting a framework, our work lays a conceptual foundation for understanding and evaluating personalization as a continuous, user-centric endeavor.
Authors: Yaxin Hou, Bo Han, Yuheng Jia, Hui Liu, Junhui Hou
Abstract: Current long-tailed semi-supervised learning methods assume that labeled data exhibit a long-tailed distribution, and unlabeled data adhere to a typical predefined distribution (i.e., long-tailed, uniform, or inverse long-tailed). However, the distribution of the unlabeled data is generally unknown and may follow an arbitrary distribution. To tackle this challenge, we propose a Controllable Pseudo-label Generation (CPG) framework, expanding the labeled dataset with the progressively identified reliable pseudo-labels from the unlabeled dataset and training the model on the updated labeled dataset with a known distribution, making it unaffected by the unlabeled data distribution. Specifically, CPG operates through a controllable self-reinforcing optimization cycle: (i) at each training step, our dynamic controllable filtering mechanism selectively incorporates reliable pseudo-labels from the unlabeled dataset into the labeled dataset, ensuring that the updated labeled dataset follows a known distribution; (ii) we then construct a Bayes-optimal classifier using logit adjustment based on the updated labeled data distribution; (iii) this improved classifier subsequently helps identify more reliable pseudo-labels in the next training step. We further theoretically prove that this optimization cycle can significantly reduce the generalization error under some conditions. Additionally, we propose a class-aware adaptive augmentation module to further improve the representation of minority classes, and an auxiliary branch to maximize data utilization by leveraging all labeled and unlabeled samples. Comprehensive evaluations on various commonly used benchmark datasets show that CPG achieves consistent improvements, surpassing state-of-the-art methods by up to \textbf{15.97\%} in accuracy. The code is available at https://github.com/yaxinhou/CPG.
Authors: Yujie Zhou, Yiwei Liao, Cheng Peng, Yong Xiao, Yingyu Li
Abstract: Semantic communication (SemCom) shifts the focus from data transmission to meaning delivery, enabling efficient and intelligent communication. Existing AI-based coding schemes for multi-modal multi-task SemCom often require transmitters with full-modal data to participate in all receivers' tasks, which leads to redundant transmissions and conflicts with the physical limits of channel capacity and computational capability. In this paper, we propose PoM$^2$-DIB, a novel framework that extends the distributed information bottleneck (DIB) theory to address this problem. Unlike the typical DIB, this framework introduces modality selection as an additional key design variable, enabling a more flexible tradeoff between communication rate and inference quality. This extension selects only the most relevant modalities for task participation, adhering to the physical constraints, while following efficient DIB-based coding. To optimize selection and coding end-to-end, we relax modality selection into a probabilistic form, allowing the use of score function estimation with common randomness to enable optimizable coordinated decisions across distributed devices. Experimental results on public datasets verify that PoM$^2$-DIB achieves high inference quality compared to full-participation baselines in various tasks under physical limits.
Authors: Sumanth Varambally, Marshall Fisher, Jas Thakker, Yiwei Chen, Zhirui Xia, Yasaman Jafari, Ruijia Niu, Manas Jain, Veeramakali Vignesh Manivannan, Zachary Novack, Luyu Han, Srikar Eranky, Salva R\"uhling Cachay, Taylor Berg-Kirkpatrick, Duncan Watson-Parris, Yi-An Ma, Rose Yu
Abstract: Foundation models for weather science are pre-trained on vast amounts of structured numerical data and outperform traditional weather forecasting systems. However, these models lack language-based reasoning capabilities, limiting their utility in interactive scientific workflows. Large language models (LLMs) excel at understanding and generating text but cannot reason about high-dimensional meteorological datasets. We bridge this gap by building a novel agentic framework for weather science. Our framework includes a Python code-based environment for agents (ZephyrusWorld) to interact with weather data, featuring tools like an interface to WeatherBench 2 dataset, geoquerying for geographical masks from natural language, weather forecasting, and climate simulation capabilities. We design Zephyrus, a multi-turn LLM-based weather agent that iteratively analyzes weather datasets, observes results, and refines its approach through conversational feedback loops. We accompany the agent with a new benchmark, ZephyrusBench, with a scalable data generation pipeline that constructs diverse question-answer pairs across weather-related tasks, from basic lookups to advanced forecasting, extreme event detection, and counterfactual reasoning. Experiments on this benchmark demonstrate the strong performance of Zephyrus agents over text-only baselines, outperforming them by up to 35 percentage points in correctness. However, on harder tasks, Zephyrus performs similarly to text-only baselines, highlighting the challenging nature of our benchmark and suggesting promising directions for future work.
Authors: Dan Leonte, Rapha\"el Huser, Almut E. D. Veraart
Abstract: The growing availability of large and complex datasets has increased interest in temporal stochastic processes that can capture stylized facts such as marginal skewness, non-Gaussian tails, long memory, and even non-Markovian dynamics. While such models are often easy to simulate from, parameter estimation remains challenging. Simulation-based inference (SBI) offers a promising way forward, but existing methods typically require large training datasets or complex architectures and frequently yield confidence (credible) regions that fail to attain their nominal values, raising doubts on the reliability of estimates for the very features that motivate the use of these models. To address these challenges, we propose a fast and accurate, sample-efficient SBI framework for amortized posterior inference applicable to intractable stochastic processes. The proposed approach relies on two main steps: first, we learn the posterior density by decomposing it sequentially across parameter dimensions. Then, we use Chebyshev polynomial approximations to efficiently generate independent posterior samples, enabling accurate inference even when Markov chain Monte Carlo methods mix poorly. We further develop novel diagnostic tools for SBI in this context, as well as post-hoc calibration techniques; the latter not only lead to performance improvements of the learned inferential tool, but also to the ability to reuse it directly with new time series of varying lengths, thus amortizing the training cost. We demonstrate the method's effectiveness on trawl processes, a class of flexible infinitely divisible models that generalize univariate Gaussian processes, applied to energy demand data.
Authors: Yunfan Zhang, Kathleen McKeown, Smaranda Muresan
Abstract: Large Language Models (LLMs) are typically trained to reflect a relatively uniform set of values, which limits their applicability to tasks that require understanding of nuanced human perspectives. Recent research has underscored the importance of enabling LLMs to support steerable pluralism -- the capacity to adopt a specific perspective and align generated outputs with it. In this work, we investigate whether Chain-of-Thought (CoT) reasoning techniques can be applied to building steerable pluralistic models. We explore several methods, including CoT prompting, fine-tuning on human-authored CoT, fine-tuning on synthetic explanations, and Reinforcement Learning with Verifiable Rewards (RLVR). We evaluate these approaches using the Value Kaleidoscope and OpinionQA datasets. Among the methods studied, RLVR consistently outperforms others and demonstrates strong training sample efficiency. We further analyze the generated CoT traces with respect to faithfulness and safety.
Authors: Tong Mao, Jinchao Xu
Abstract: We prove a saturation theorem for linearized shallow ReLU$^k$ neural networks on the unit sphere $\mathbb S^d$. For any antipodally quasi-uniform set of centers, if the target function has smoothness $r>\tfrac{d+2k+1}{2}$, then the best $\mathcal{L}^2(\mathbb S^d)$ approximation cannot converge faster than order $n^{-\frac{d+2k+1}{2d}}$. This lower bound matches existing upper bounds, thereby establishing the exact saturation order $\tfrac{d+2k+1}{2d}$ for such networks. Our results place linearized neural-network approximation firmly within the classical saturation framework and show that, although ReLU$^k$ networks outperform finite elements under equal degrees $k$, this advantage is intrinsically limited.
Authors: Hyung Gyu Rho
Abstract: Modern preference alignment techniques, such as Best-of-N (BoN) sampling, rely on reward models trained with pairwise comparison data. While effective at learning relative preferences, this paradigm fails to capture a signal of response acceptability, leaving systems vulnerable to selecting the least bad of many unacceptable options. This is particularly problematic for hard prompts, where the risk of such false acceptances increases with the number of samples. In this paper, we address this critical reliability gap by introducing a new data collection and modeling framework. By augmenting preference data with an outside option, inspired by discrete choice models, we train a reward model that can distinguish not just what is \textit{better}, but what is \textit{good enough}. We leverage this capability to create an adaptive inference strategy, best of mini-N in-loop, which partitions the generation budget into sequential loops with a calibrated, early-exit condition. Our experiments show that when tuned as an alignment guardrail, it reduces reliability failures by 70\%, and when tuned as an inference accelerator, it improves average inference speed by over 22\% in IMDB-sentiment setting. We thus provide a principled and flexible framework for practitioners to explicitly manage the trade-off between reliability and computational efficiency.
Authors: Tommy Mordo, Sagie Dekel, Omer Madmon, Moshe Tennenholtz, Oren Kurland
Abstract: Competitive search is a setting where document publishers modify them to improve their ranking in response to a query. Recently, publishers have increasingly leveraged LLMs to generate and modify competitive content. We introduce Reinforcement Learning from Ranker Feedback (RLRF), a framework that trains LLMs using preference datasets derived from ranking competitions. The goal of a publisher (LLM-based) agent is to optimize content for improved ranking while accounting for the strategies of competing agents. We generate the datasets using approaches that do not rely on human-authored data. We show that our proposed agents consistently and substantially outperform previously suggested approaches for LLM-based competitive document modification. We further show that our agents are effective with ranking functions they were not trained for (i.e., out of distribution) and they adapt to strategic opponents. These findings provide support to the significant potential of using reinforcement learning in competitive search.
Authors: Sean Moran
Abstract: Approximate Nearest Neighbour (ANN) search is a fundamental problem in information retrieval, underpinning large-scale applications in computer vision, natural language processing, and cross-modal search. Hashing-based methods provide an efficient solution by mapping high-dimensional data into compact binary codes that enable fast similarity computations in Hamming space. Over the past two decades, a substantial body of work has explored learning to hash, where projection and quantisation functions are optimised from data rather than chosen at random. This article offers a foundational survey of early learning-based hashing methods, with an emphasis on the core ideas that shaped the field. We review supervised, unsupervised, and semi-supervised approaches, highlighting how projection functions are designed to generate meaningful embeddings and how quantisation strategies convert these embeddings into binary codes. We also examine extensions to multi-bit and multi-threshold models, as well as early advances in cross-modal retrieval. Rather than providing an exhaustive account of the most recent methods, our goal is to introduce the conceptual foundations of learning-based hashing for ANN search. By situating these early models in their historical context, we aim to equip readers with a structured understanding of the principles, trade-offs, and open challenges that continue to inform current research in this area.
Authors: Tim Bakkenes, Daniel Wang, Anton Johansson
Abstract: The rise of Large Language Models has not been inclusive of all cultures. The models are mostly trained on English texts and culture which makes them underperform in other languages and cultural contexts. By developing a generalizable method for preparing culturally relevant datasets and post-training the Gemma 2 model, this project aimed to increase the performance of Gemma 2 for an underrepresented language and showcase how others can do the same to unlock the power of Generative AI in their country and preserve their cultural heritage.
Authors: Xiaoyu Yang, Jie Lu, En Yu
Abstract: This paper identifies a critical yet underexplored challenge in distilling from multimodal large language models (MLLMs): the reasoning trajectories generated by multiple drifting teachers exhibit concept drift, whereby their reasoning distributions evolve unpredictably and transmit biases to the student model, ultimately compromising its performance. To tackle this issue, we pioneer a theoretical connection between concept drift and knowledge distillation, casting the non-stationary reasoning dynamics from multiple MLLM teachers as next-token prediction of multi-stream reasoning trajectories.Guided by concept drift, we introduce the "learn, compare, critique" paradigm, culminating in autonomous preference optimization (APO). Under the active guidance of the teachers, the student model first learns and self-distils preferred thinking by comparing multiple teachers. It then engages in critical reflection over the drifting inference from teachers, performing concept alignment through APO, ultimately yielding a robust, consistent, and generalizable model.Extensive experiments demonstrate our superior performance of consistency, robustness and generalization within knowledge distillation. Besides, we also contributed a large-scale dataset, CXR-MAX (Multi-teachers Alignment X-rays), comprising 170,982 distilled reasoning trajectories derived from publicly accessible MLLMs based on MIMIC-CXR. Our code and data are public at: https://anonymous.4open.science/r/Autonomous-Distillation/.
URLs: https://anonymous.4open.science/r/Autonomous-Distillation/.
Authors: Haoqi Wu, Wei Dai, Ming Xu, Li Wang, Qiang Yan
Abstract: Diffusion Models have gained significant popularity due to their remarkable capabilities in image generation, albeit at the cost of intensive computation requirement. Meanwhile, despite their widespread deployment in inference services such as Midjourney, concerns about the potential leakage of sensitive information in uploaded user prompts have arisen. Existing solutions either lack rigorous privacy guarantees or fail to strike an effective balance between utility and efficiency. To bridge this gap, we propose ObCLIP, a plug-and-play safeguard that enables oblivious cloud-device hybrid generation. By oblivious, each input prompt is transformed into a set of semantically similar candidate prompts that differ only in sensitive attributes (e.g., gender, ethnicity). The cloud server processes all candidate prompts without knowing which one is the real one, thus preventing any prompt leakage. To mitigate server cost, only a small portion of denoising steps is performed upon the large cloud model. The intermediate latents are then sent back to the client, which selects the targeted latent and completes the remaining denoising using a small device model. Additionally, we analyze and incorporate several cache-based accelerations that leverage temporal and batch redundancy, effectively reducing computation cost with minimal utility degradation. Extensive experiments across multiple datasets demonstrate that ObCLIP provides rigorous privacy and comparable utility to cloud models with slightly increased server cost.
Authors: Aviv Navon, Aviv Shamsian, Neta Glazer, Yael Segal-Feldman, Gill Hetz, Joseph Keshet, Ethan Fetaya
Abstract: Diffusion and flow-based non-autoregressive (NAR) models have shown strong promise in large language modeling, however, their potential for automatic speech recognition (ASR) remains largely unexplored. We propose Drax, a discrete flow matching framework for ASR that enables efficient parallel decoding. To better align training with inference, we construct an audio-conditioned probability path that guides the model through trajectories resembling likely intermediate inference errors, rather than direct random noise to target transitions. Our theoretical analysis links the generalization gap to divergences between training and inference occupancies, controlled by cumulative velocity errors, thereby motivating our design choice. Empirical evaluation demonstrates that our approach attains recognition accuracy on par with state-of-the-art speech models while offering improved accuracy-efficiency trade-offs, highlighting discrete flow matching as a promising direction for advancing NAR ASR.
Authors: Ran Eisenberg, Amit Rozner, Ethan Fetaya, Ofir Lindenbaum
Abstract: Deep neural networks have achieved remarkable success in computer vision; however, their black-box nature in decision-making limits interpretability and trust, particularly in safety-critical applications. Interpretability is crucial in domains where errors have severe consequences. Existing models not only lack transparency but also risk exploiting unreliable or misleading features, which undermines both robustness and the validity of their explanations. Concept Bottleneck Models (CBMs) aim to improve transparency by reasoning through human-interpretable concepts. Still, they require costly concept annotations and lack spatial grounding, often failing to identify which regions support each concept. We propose SEG-MIL-CBM, a novel framework that integrates concept-guided image segmentation into an attention-based multiple instance learning (MIL) framework, where each segmented region is treated as an instance and the model learns to aggregate evidence across them. By reasoning over semantically meaningful regions aligned with high-level concepts, our model highlights task-relevant evidence, down-weights irrelevant cues, and produces spatially grounded, concept-level explanations without requiring annotations of concepts or groups. SEG-MIL-CBM achieves robust performance across settings involving spurious correlations (unintended dependencies between background and label), input corruptions (perturbations that degrade visual quality), and large-scale benchmarks, while providing transparent, concept-level explanations.
Authors: Yizhuo Ding, Mingkang Chen, Qiuhua Liu, Fenghua Weng, Wanying Qu, Yue Yang, Yugang Jiang, Zuxuan Wu, Yanwei Fu, Wenqi Shao
Abstract: Large Multimodal Reasoning Models (LMRMs) are moving into real applications, where they must be both useful and safe. Safety is especially challenging in multimodal settings: images and text can be combined to bypass guardrails, and single objective training can cause policy drift that yields over-refusal on benign inputs or unsafe compliance on risky ones. We present COSMO-RL, a mixed reinforcement learning framework that trains reasoning oriented LMRMs under multimodal, multitask, and multiobjective signals, and we release the resulting model, COSMO-R1. Our approach aims to let safety and capability grow together in one stable pipeline rather than competing during alignment. In experiments, COSMO-R1 improves safety while maintaining-and often improving multimodal reasoning and instruction following, shows stronger robustness to multimodal jailbreaks, and reduces unnecessary refusals. The framework also transfers across backbones with consistent gains. Ablations support the design choices, indicating a simple path to advancing safety and general capability together in LMRMs.
Authors: Zhengyang Tang, Zihan Ye, Chenyu Huang, Xuhan Huang, Chengpeng Li, Sihang Li, Guanhua Chen, Ming Yan, Zizhuo Wang, Hongyuan Zha, Dayiheng Liu, Benyou Wang
Abstract: Large Reasoning Models (LRMs) have demonstrated strong capabilities in complex multi-step reasoning, opening new opportunities for automating optimization modeling. However, existing domain adaptation methods, originally designed for earlier instruction-tuned models, often fail to exploit the advanced reasoning patterns of modern LRMs -- In particular, we show that direct fine-tuning on traditional \textit{non-reflective} datasets leads to limited gains. To fully leverage LRMs' inherent reasoning abilities, we propose \textbf{CALM} (\textit{Corrective Adaptation with Lightweight Modification}), a framework that progressively refines LRMs within their native reasoning modes for optimization modeling tasks. In CALM, an expert intervener identifies reasoning flaws and provides concise corrective hints, which the LRM incorporates to produce improved reasoning trajectories. These interventions modify fewer than 2.6\% of generated tokens, but generate high-quality data for soft adaptation through supervised fine-tuning. The adapted model is then further improved through reinforcement learning. Building on CALM, we develop \textbf{STORM} (\textit{Smart Thinking Optimization Reasoning Model}), a 4B-parameter LRM that achieves a new state-of-the-art average accuracy of 68.9\% across five popular optimization modeling benchmarks, matching the performance of a 671B LRM. These results demonstrate that dynamic, hint-based data synthesis both preserves and amplifies the native reasoning patterns of modern LRMs, offering a more effective and scalable path towards expert-level performance on challenging optimization modeling tasks.
Authors: Lixuan He, Shikang Zheng, Linfeng Zhang
Abstract: Autoregressive (AR) models have shown great promise in image generation, yet they face a fundamental inefficiency stemming from their core component: a vast, unstructured vocabulary of visual tokens. This conventional approach treats tokens as a flat vocabulary, disregarding the intrinsic structure of the token embedding space where proximity often correlates with semantic similarity. This oversight results in a highly complex prediction task, which hinders training efficiency and limits final generation quality. To resolve this, we propose Manifold-Aligned Semantic Clustering (MASC), a principled framework that constructs a hierarchical semantic tree directly from the codebook's intrinsic structure. MASC employs a novel geometry-aware distance metric and a density-driven agglomerative construction to model the underlying manifold of the token embeddings. By transforming the flat, high-dimensional prediction task into a structured, hierarchical one, MASC introduces a beneficial inductive bias that significantly simplifies the learning problem for the AR model. MASC is designed as a plug-and-play module, and our extensive experiments validate its effectiveness: it accelerates training by up to 57% and significantly improves generation quality, reducing the FID of LlamaGen-XL from 2.87 to 2.58. MASC elevates existing AR frameworks to be highly competitive with state-of-the-art methods, establishing that structuring the prediction space is as crucial as architectural innovation for scalable generative modeling.
Authors: Dustin Wright, Sarah Masud, Jared Moore, Srishti Yadav, Maria Antoniak, Chan Young Park, Isabelle Augenstein
Abstract: Large language models (LLMs) tend to generate lexically, semantically, and stylistically homogenous texts. This poses a risk of knowledge collapse, where homogenous LLMs mediate a shrinking in the range of accessible information over time. Existing works on homogenization are limited by a focus on closed-ended multiple-choice setups or fuzzy semantic features, and do not look at trends across time and cultural contexts. To overcome this, we present a new methodology to measure epistemic diversity, i.e., variation in real-world claims in LLM outputs, which we use to perform a broad empirical study of LLM knowledge collapse. We test 27 LLMs, 155 topics covering 12 countries, and 200 prompt variations sourced from real user chats. For the topics in our study, we show that while newer models tend to generate more diverse claims, nearly all models are less epistemically diverse than a basic web search. We find that model size has a negative impact on epistemic diversity, while retrieval-augmented generation (RAG) has a positive impact, though the improvement from RAG varies by the cultural context. Finally, compared to a traditional knowledge source (Wikipedia), we find that country-specific claims reflect the English language more than the local one, highlighting a gap in epistemic representation
Authors: Shengjiao Ji, Yujin Zhang, Zihan Zou, Bin Jiang, Jun Jiang, Yi Luo, Wei Hu
Abstract: Accurate and efficient simulation of infrared (IR) and Raman spectra is essential for molecular identification and structural analysis. Traditional quantum chemistry methods based on the harmonic approximation neglect anharmonicity and nuclear quantum effects, while ab initio molecular dynamics (AIMD) remains computationally expensive. Here, we integrate our deep equivariant tensor attention network (DetaNet) with a velocity-Verlet integrator to enable fast and accurate machine learning molecular dynamics (MLMD) simulations for spectral prediction. Trained on the QMe14S dataset containing energies, forces, dipole moments, and polarizabilities for 186,102 small organic molecules, DetaNet yields a universal and transferable force field with high-order tensor prediction capability. Using time-correlation functions derived from MLMD and ring-polymer molecular dynamics (RPMD) trajectories, we computed IR and Raman spectra that accurately reproduce anharmonic and nuclear quantum effects. Benchmark tests on isolated molecules, including polycyclic aromatic hydrocarbons, demonstrate that the DetaNet-based MD approach achieves near-experimental spectral accuracy with speedups up to three orders of magnitude over AIMD. Furthermore, the framework extends seamlessly to molecular and inorganic crystals, molecular aggregates, and biological macromolecules such as polypeptides with minimal fine-tuning. In all systems, DetaNet maintains high accuracy while significantly reducing computational cost. Overall, this work establishes a universal machine learning force field and tensor-aware MLMD framework that enable fast, accurate, and broadly applicable dynamic simulations and IR/Raman spectral predictions across diverse molecular and material systems.
Authors: Amin Ahmadi Kasani, Hedieh Sajedi
Abstract: Convolutional neural networks are continually evolving, with some efforts aimed at improving accuracy, others at increasing speed, and some at enhancing accessibility. Improving accessibility broadens the application of neural networks across a wider range of tasks, including the detection of eye diseases. Early diagnosis of eye diseases and consulting an ophthalmologist can prevent many vision disorders. Given the importance of this issue, various datasets have been collected from the cornea to facilitate the process of making neural network models. However, most of the methods introduced in the past are computationally complex. In this study, we tried to increase the accessibility of deep neural network models. We did this at the most fundamental level, specifically by redesigning and optimizing the convolutional layers. By doing so, we created a new general model that incorporates our novel convolutional layer named ArConv layers. Thanks to the efficient performance of this new layer, the model has suitable complexity for use in mobile phones and can perform the task of diagnosing the presence of disease with high accuracy. The final model we present contains only 1.3 million parameters. In comparison to the MobileNetV2 model, which has 2.2 million parameters, our model demonstrated better accuracy when trained and evaluated on the RfMiD dataset under identical conditions, achieving an accuracy of 0.9328 versus 0.9266 on the RfMiD test set.
Authors: Jinyang Jiang, Jinhui Han, Yijie Peng, Ying Zhang
Abstract: Effective cross-functional coordination is essential for enhancing firm-wide profitability, particularly in the face of growing organizational complexity and scale. Recent advances in artificial intelligence, especially in reinforcement learning (RL), offer promising avenues to address this fundamental challenge. This paper proposes a unified multi-agent RL framework tailored for joint optimization across distinct functional modules, exemplified via coordinating inventory replenishment and personalized product recommendation. We first develop an integrated theoretical model to capture the intricate interplay between these functions and derive analytical benchmarks that characterize optimal coordination. The analysis reveals synchronized adjustment patterns across products and over time, highlighting the importance of coordinated decision-making. Leveraging these insights, we design a novel multi-timescale multi-agent RL architecture that decomposes policy components according to departmental functions and assigns distinct learning speeds based on task complexity and responsiveness. Our model-free multi-agent design improves scalability and deployment flexibility, while multi-timescale updates enhance convergence stability and adaptability across heterogeneous decisions. We further establish the asymptotic convergence of the proposed algorithm. Extensive simulation experiments demonstrate that the proposed approach significantly improves profitability relative to siloed decision-making frameworks, while the behaviors of the trained RL agents align closely with the managerial insights from our theoretical model. Taken together, this work provides a scalable, interpretable RL-based solution to enable effective cross-functional coordination in complex business settings.
Authors: Hamish Flynn
Abstract: The sample complexity of estimating or maximising an unknown function in a reproducing kernel Hilbert space is known to be linked to both the effective dimension and the information gain associated with the kernel. While the information gain has an attractive information-theoretic interpretation, the effective dimension typically results in better rates. We introduce a new quantity called the relative information gain, which measures the sensitivity of the information gain with respect to the observation noise. We show that the relative information gain smoothly interpolates between the effective dimension and the information gain, and that the relative information gain has the same growth rate as the effective dimension. In the second half of the paper, we prove a new PAC-Bayesian excess risk bound for Gaussian process regression. The relative information gain arises naturally from the complexity term in this PAC-Bayesian bound. We prove bounds on the relative information gain that depend on the spectral properties of the kernel. When these upper bounds are combined with our excess risk bound, we obtain minimax-optimal rates of convergence.
Authors: Karthik Viswanathan, Sang Eon Park
Abstract: We introduce a cumulant-expansion framework for quantifying how large language models (LLMs) internalize higher-order statistical structure during next-token prediction. By treating the softmax entropy of each layer's logit distribution as a perturbation around its "center" distribution, we derive closed-form cumulant observables that isolate successively higher-order correlations. Empirically, we track these cumulants in GPT-2 and Pythia models on Pile-10K prompts. (i) Structured prompts exhibit a characteristic rise-and-plateau profile across layers, whereas token-shuffled prompts remain flat, revealing the dependence of the cumulant profile on meaningful context. (ii) During training, all cumulants increase monotonically before saturating, directly visualizing the model's progression from capturing variance to learning skew, kurtosis, and higher-order statistical structures. (iii) Mathematical prompts show distinct cumulant signatures compared to general text, quantifying how models employ fundamentally different processing mechanisms for mathematical versus linguistic content. Together, these results establish cumulant analysis as a lightweight, mathematically grounded probe of feature-learning dynamics in high-dimensional neural networks.
Authors: Harshil Vejendla
Abstract: Mixture-of-Experts (MoE) layers scale transformers by routing tokens to a sparse subset of feed-forward experts. Token-level routing, however, assigns an entire semantic spectrum to each expert, creating capacity bottlenecks, load-balancing pathologies, and limited specialization. We introduce SliceMoE, an architecture that routes contiguous slices of a token's hidden vector. A d-dimensional embedding is partitioned into S slices, and for each slice, a lightweight shared router predicts the top-k experts. Experts operate on their assigned slices independently, and outputs are reassembled, maintaining per-token FLOP efficiency. Because slices from different tokens interleave within an expert, utilization is naturally smoother. We propose a slice-level capacity loss, cross-slice dropout, and efficient fused batched GEMM kernels. Experiments on WikiText-103 language modeling, WMT En-De translation, and three text-classification datasets show SliceMoE attains up to 1.7x faster inference than dense baselines, 12 to 18 percent lower perplexity than parameter-matched token-MoE, and improved expert balance, with interpretable expertise over syntactic versus semantic subspaces.
Authors: Mehrzad Tareh, Aydin Mohandesi, Ebrahim Ansari
Abstract: Sentiment analysis is a key task in Natural Language Processing (NLP), enabling the extraction of meaningful insights from user opinions across various domains. However, performing sentiment analysis in Persian remains challenging due to the scarcity of labeled datasets, limited preprocessing tools, and the lack of high-quality embeddings and feature extraction methods. To address these limitations, we propose a hybrid approach that integrates machine learning (ML) and deep learning (DL) techniques for Persian aspect-based sentiment analysis (ABSA). In particular, we utilize polarity scores from multilingual BERT as additional features and incorporate them into a decision tree classifier, achieving an accuracy of 93.34%-surpassing existing benchmarks on the Pars-ABSA dataset. Additionally, we introduce a Persian synonym and entity dictionary, a novel linguistic resource that supports text augmentation through synonym and named entity replacement. Our results demonstrate the effectiveness of hybrid modeling and feature augmentation in advancing sentiment analysis for low-resource languages such as Persian.
Authors: Bohan Tang, Huidong Liang, Keyue Jiang, Xiaowen Dong
Abstract: Large language model multi-agent systems (LLM-MAS) offer a promising paradigm for harnessing collective intelligence to achieve more advanced forms of AI behaviour. While recent studies suggest that LLM-MAS can outperform LLM single-agent systems (LLM-SAS) on certain tasks, the lack of systematic experimental designs limits the strength and generality of these conclusions. We argue that a principled understanding of task complexity, such as the degree of sequential reasoning required and the breadth of capabilities involved, is essential for assessing the effectiveness of LLM-MAS in task solving. To this end, we propose a theoretical framework characterising tasks along two dimensions: depth, representing reasoning length, and width, representing capability diversity. We theoretically examine a representative class of LLM-MAS, namely the multi-agent debate system, and empirically evaluate its performance in both discriminative and generative tasks with varying depth and width. Theoretical and empirical results show that the benefit of LLM-MAS over LLM-SAS increases with both task depth and width, and the effect is more pronounced with respect to depth. This clarifies when LLM-MAS are beneficial and provides a principled foundation for designing future LLM-MAS methods and benchmarks.
Authors: Etienne Gauthier, Francis Bach, Michael I. Jordan
Abstract: Traditional conformal prediction methods construct prediction sets such that the true label falls within the set with a user-specified coverage level. However, poorly chosen coverage levels can result in uninformative predictions, either producing overly conservative sets when the coverage level is too high, or empty sets when it is too low. Moreover, the fixed coverage level cannot adapt to the specific characteristics of each individual example, limiting the flexibility and efficiency of these methods. In this work, we leverage recent advances in e-values and post-hoc conformal inference, which allow the use of data-dependent coverage levels while maintaining valid statistical guarantees. We propose to optimize an adaptive coverage policy by training a neural network using a leave-one-out procedure on the calibration set, allowing the coverage level and the resulting prediction set size to vary with the difficulty of each individual example. We support our approach with theoretical coverage guarantees and demonstrate its practical benefits through a series of experiments.
Authors: Rui Wu, Yihao Quan, Zeru Shi, Zhenting Wang, Yanshu Li, Ruixiang Tang
Abstract: Safety-aligned Large Language Models (LLMs) still show two dominant failure modes: they are easily jailbroken, or they over-refuse harmless inputs that contain sensitive surface signals. We trace both to a common cause: current models reason weakly about links between actions and outcomes and over-rely on surface-form signals, lexical or stylistic cues that do not encode consequences. We define this failure mode as Consequence-blindness. To study consequence-blindness, we build a benchmark named CB-Bench covering four risk scenarios that vary whether semantic risk aligns with outcome risk, enabling evaluation under both matched and mismatched conditions which are often ignored by existing safety benchmarks. Mainstream models consistently fail to separate these risks and exhibit consequence-blindness, indicating that consequence-blindness is widespread and systematic. To mitigate consequence-blindness, we introduce CS-Chain-4k, a consequence-reasoning dataset for safety alignment. Models fine-tuned on CS-Chain-4k show clear gains against semantic-camouflage jailbreaks and reduce over-refusal on harmless inputs, while maintaining utility and generalization on other benchmarks. These results clarify the limits of current alignment, establish consequence-aware reasoning as a core alignment goal and provide a more practical and reproducible evaluation path.
Authors: Akshay Govind Srinivasan, Anuj Jagannath Said, Sathwik Pentela, Vikas Dwivedi, Balaji Srinivasan
Abstract: Partial differential equation (PDE) solvers underpin modern quantitative finance, governing option pricing and risk evaluation. Physics-Informed Neural Networks (PINNs) have emerged as a promising approach for solving the forward and inverse problems of partial differential equations (PDEs) using deep learning. However they remain computationally expensive due to their iterative gradient descent based optimization and scale poorly with increasing model size. This paper introduces Physics-Informed Extreme Learning Machines (PIELMs) as fast alternative to PINNs for solving both forward and inverse problems in financial PDEs. PIELMs replace iterative optimization with a single least-squares solve, enabling deterministic and efficient training. We benchmark PIELM on the Black-Scholes and Heston-Hull-White models for forward pricing and demonstrate its capability in inverse model calibration to recover volatility and interest rate parameters from noisy data. From experiments we observe that PIELM achieve accuracy comparable to PINNs while being up to $30\times$ faster, highlighting their potential for real-time financial modeling.
Authors: Christian Limberg, Fares Schulz, Zhe Zhang, Stefan Weinzierl
Abstract: This paper presents a novel approach to neural instrument sound synthesis using a two-stage semi-supervised learning framework capable of generating pitch-accurate, high-quality music samples from an expressive timbre latent space. Existing approaches that achieve sufficient quality for music production often rely on high-dimensional latent representations that are difficult to navigate and provide unintuitive user experiences. We address this limitation through a two-stage training paradigm: first, we train a pitch-timbre disentangled 2D representation of audio samples using a Variational Autoencoder; second, we use this representation as conditioning input for a Transformer-based generative model. The learned 2D latent space serves as an intuitive interface for navigating and exploring the sound landscape. We demonstrate that the proposed method effectively learns a disentangled timbre space, enabling expressive and controllable audio generation with reliable pitch conditioning. Experimental results show the model's ability to capture subtle variations in timbre while maintaining a high degree of pitch accuracy. The usability of our method is demonstrated in an interactive web application, highlighting its potential as a step towards future music production environments that are both intuitive and creatively empowering: https://pgesam.faresschulz.com
Authors: Nahshon Mokua Obiri, Kristof Van Laerhoven
Abstract: Indoor LoRaWAN propagation is shaped by structural and time-varying context factors, which challenge log-distance models and the assumption of log-normal shadowing. We present an environment-aware, statistically disciplined path loss framework evaluated using leakage-safe cross-validation on a 12-month campaign in an eighth-floor office measuring 240 m^2. A log-distance multi-wall mean is augmented with environmental covariates (relative humidity, temperature, carbon dioxide, particulate matter, and barometric pressure), as well as the signal-to-noise ratio. We compare multiple linear regression with regularized variants, Bayesian linear regression, and a selective second-order polynomial applied to continuous drivers. Predictor relevance is established using heteroscedasticity-robust Type II and III analysis of variance and nested partial F tests. Shadow fading is profiled with kernel density estimation and non-parametric families, including Normal, Skew-Normal, Student's t, and Gaussian mixtures. The polynomial mean reduces cross-validated RMSE from 8.07 to 7.09 dB and raises R^2 from 0.81 to 0.86. Out-of-fold residuals are non-Gaussian; a 3-component mixture captures a sharp core with a light, broad tail. We convert accuracy into reliability by prescribing the fade margin as the upper-tail quantile of cross-validated residuals, quantifying uncertainty via a moving-block bootstrap, and validating on a held-out set. At 99% packet delivery ratio, the environment-aware polynomial requires 25.7 dB versus 27.7 to 27.9 dB for linear baselines. This result presents a deployment-ready, interpretable workflow with calibrated reliability control for indoor Internet of Things planning, aligned with 6G targets.
Authors: Anindya Sundar Das, Kangjie Chen, Monowar Bhuyan
Abstract: Pre-trained language models have achieved remarkable success across a wide range of natural language processing (NLP) tasks, particularly when fine-tuned on large, domain-relevant datasets. However, they remain vulnerable to backdoor attacks, where adversaries embed malicious behaviors using trigger patterns in the training data. These triggers remain dormant during normal usage, but, when activated, can cause targeted misclassifications. In this work, we investigate the internal behavior of backdoored pre-trained encoder-based language models, focusing on the consistent shift in attention and gradient attribution when processing poisoned inputs; where the trigger token dominates both attention and gradient signals, overriding the surrounding context. We propose an inference-time defense that constructs anomaly scores by combining token-level attention and gradient information. Extensive experiments on text classification tasks across diverse backdoor attack scenarios demonstrate that our method significantly reduces attack success rates compared to existing baselines. Furthermore, we provide an interpretability-driven analysis of the scoring mechanism, shedding light on trigger localization and the robustness of the proposed defense.
Authors: Dmitry Ustalov, Egor Bogomolov, Alexander Bezzubov, Yaroslav Golubev, Evgeniy Glukhov, Georgii Levtsov, Vladimir Kovalenko
Abstract: The rapid advancement of workflows and methods for software engineering using AI emphasizes the need for a systematic evaluation and analysis of their ability to leverage information from entire projects, particularly in large code bases. In this challenge on optimization of context collection for code completion, organized by JetBrains in collaboration with Mistral AI as part of the ASE 2025 conference, participants developed efficient mechanisms for collecting context from source code repositories to improve fill-in-the-middle code completions for Python and Kotlin. We constructed a large dataset of real-world code in these two programming languages using permissively licensed open-source projects. The submissions were evaluated based on their ability to maximize completion quality for multiple state-of-the-art neural models using the chrF metric. During the public phase of the competition, nineteen teams submitted solutions to the Python track and eight teams submitted solutions to the Kotlin track. In the private phase, six teams competed, of which five submitted papers to the workshop.
Authors: Osman Bicer, Ali D. Kara, Serdar Yuksel
Abstract: In this paper, for Markov decision processes (MDPs) with unbounded state spaces we present refined upper bounds presented in [Kara et. al. JMLR'23] on finite model approximation errors via optimizing the quantizers used for finite model approximations. We also consider implications on quantizer design for quantized Q-learning and empirical model learning, and the performance of policies obtained via Q-learning where the quantized state is treated as the state itself. We highlight the distinctions between planning, where approximating MDPs can be independently designed, and learning (either via Q-learning or empirical model learning), where approximating MDPs are restricted to be defined by invariant measures of Markov chains under exploration policies, leading to significant subtleties on quantizer design performance, even though asymptotic near optimality can be established under both setups. In particular, under Lyapunov growth conditions, we obtain explicit upper bounds which decay to zero as the number of bins approaches infinity.
Authors: Jiarui Li, Zixiang Yin, Zhengming Ding, Samuel J. Landry, Ramgopal R. Mettu
Abstract: T cell receptor (TCR) recognition of peptide-MHC (pMHC) complexes is a central component of adaptive immunity, with implications for vaccine design, cancer immunotherapy, and autoimmune disease. While recent advances in machine learning have improved prediction of TCR-pMHC binding, the most effective approaches are black-box transformer models that cannot provide a rationale for predictions. Post-hoc explanation methods can provide insight with respect to the input but do not explicitly model biochemical mechanisms (e.g. known binding regions), as in TCR-pMHC binding. ``Explain-by-design'' models (i.e., with architectural components that can be examined directly after training) have been explored in other domains, but have not been used for TCR-pMHC binding. We propose explainable model layers (TCR-EML) that can be incorporated into protein-language model backbones for TCR-pMHC modeling. Our approach uses prototype layers for amino acid residue contacts drawn from known TCR-pMHC binding mechanisms, enabling high-quality explanations for predicted TCR-pMHC binding. Experiments of our proposed method on large-scale datasets demonstrate competitive predictive accuracy and generalization, and evaluation on the TCR-XAI benchmark demonstrates improved explainability compared with existing approaches.
Authors: Faisal Hamman, Chenyang Zhu, Anoop Kumar, Xujun Peng, Sanghamitra Dutta, Daben Liu, Alfy Samuel
Abstract: RAG systems are increasingly deployed in high-stakes domains where users expect outputs to be consistent across semantically equivalent queries. However, existing systems often exhibit significant inconsistencies due to variability in both the retriever and generator (LLM), undermining trust and reliability. In this work, we focus on information consistency, i.e., the requirement that outputs convey the same core content across semantically equivalent inputs. We introduce a principled evaluation framework that decomposes RAG consistency into retriever-level, generator-level, and end-to-end components, helping identify inconsistency sources. To improve consistency, we propose Paraphrased Set Group Relative Policy Optimization (PS-GRPO), an RL approach that leverages multiple rollouts across paraphrased set to assign group similarity rewards. We leverage PS-GRPO to achieve Information Consistent RAG (Con-RAG), training the generator to produce consistent outputs across paraphrased queries and remain robust to retrieval-induced variability. Because exact reward computation over paraphrase sets is computationally expensive, we also introduce a scalable approximation method that retains effectiveness while enabling efficient, large-scale training. Empirical evaluations across short-form, multi-hop, and long-form QA benchmarks demonstrate that Con-RAG significantly improves both consistency and accuracy over strong baselines, even in the absence of explicit ground-truth supervision. Our work provides practical solutions for evaluating and building reliable RAG systems for safety-critical deployments.
Authors: Ankit Vadehra, Bill Johnson, Gene Saunders, Pascal Poupart
Abstract: Text editing can involve several iterations of revision. Incorporating an efficient Grammar Error Correction (GEC) tool in the initial correction round can significantly impact further human editing effort and final text quality. This raises an interesting question to quantify GEC Tool usability: How much effort can the GEC Tool save users? We present the first large-scale dataset of post-editing (PE) time annotations and corrections for two English GEC test datasets (BEA19 and CoNLL14). We introduce Post-Editing Effort in Time (PEET) for GEC Tools as a human-focused evaluation scorer to rank any GEC Tool by estimating PE time-to-correct. Using our dataset, we quantify the amount of time saved by GEC Tools in text editing. Analyzing the edit type indicated that determining whether a sentence needs correction and edits like paraphrasing and punctuation changes had the greatest impact on PE time. Finally, comparison with human rankings shows that PEET correlates well with technical effort judgment, providing a new human-centric direction for evaluating GEC tool usability. We release our dataset and code at: https://github.com/ankitvad/PEET_Scorer.
Authors: Buyun Liang, Liangzu Peng, Jinqi Luo, Darshan Thaker, Kwan Ho Ryan Chan, Ren\'e Vidal
Abstract: Large Language Models (LLMs) are increasingly deployed in high-risk domains. However, state-of-the-art LLMs often produce hallucinations, raising serious concerns about their reliability. Prior work has explored adversarial attacks for hallucination elicitation in LLMs, but it often produces unrealistic prompts, either by inserting gibberish tokens or by altering the original meaning. As a result, these approaches offer limited insight into how hallucinations may occur in practice. While adversarial attacks in computer vision often involve realistic modifications to input images, the problem of finding realistic adversarial prompts for eliciting LLM hallucinations has remained largely underexplored. To address this gap, we propose Semantically Equivalent and Coherent Attacks (SECA) to elicit hallucinations via realistic modifications to the prompt that preserve its meaning while maintaining semantic coherence. Our contributions are threefold: (i) we formulate finding realistic attacks for hallucination elicitation as a constrained optimization problem over the input prompt space under semantic equivalence and coherence constraints; (ii) we introduce a constraint-preserving zeroth-order method to effectively search for adversarial yet feasible prompts; and (iii) we demonstrate through experiments on open-ended multiple-choice question answering tasks that SECA achieves higher attack success rates while incurring almost no constraint violations compared to existing methods. SECA highlights the sensitivity of both open-source and commercial gradient-inaccessible LLMs to realistic and plausible prompt variations. Code is available at https://github.com/Buyun-Liang/SECA.
Authors: Charles L. Wang, Keir Dorchen, Peter Jin
Abstract: As systems trend toward superintelligence, a natural modeling premise is that agents can self-improve along every facet of their own design. We formalize this with a five-axis decomposition and a decision layer, separating incentives from learning behavior and analyzing axes in isolation. Our central result identifies and introduces a sharp utility--learning tension, the structural conflict in self-modifying systems whereby utility-driven changes that improve immediate or expected performance can also erode the statistical preconditions for reliable learning and generalization. Our findings show that distribution-free guarantees are preserved iff the policy-reachable model family is uniformly capacity-bounded; when capacity can grow without limit, utility-rational self-changes can render learnable tasks unlearnable. Under standard assumptions common in practice, these axes reduce to the same capacity criterion, yielding a single boundary for safe self-modification. Numerical experiments across several axes validate the theory by comparing destructive utility policies against our proposed two-gate policies that preserve learnability.
Authors: William Zhang, Saurabh Amin, Georgia Perakis
Abstract: Conformal prediction offers finite-sample coverage guarantees under minimal assumptions. However, existing methods treat the entire modeling process as a black box, overlooking opportunities to exploit modular structure. We introduce a conformal prediction framework for two-stage sequential models, where an upstream predictor generates intermediate representations for a downstream model. By decomposing the overall prediction residual into stage-specific components, our method enables practitioners to attribute uncertainty to specific pipeline stages. We develop a risk-controlled parameter selection procedure using family-wise error rate (FWER) control to calibrate stage-wise scaling parameters, and propose an adaptive extension for non-stationary settings that preserves long-run coverage guarantees. Experiments on synthetic distribution shifts, as well as real-world supply chain and stock market data, demonstrate that our approach maintains coverage under conditions that degrade standard conformal methods, while providing interpretable stage-wise uncertainty attribution. This framework offers diagnostic advantages and robust coverage that standard conformal methods lack.
Authors: Brian Hu Zhang, Ioannis Anagnostides, Tuomas Sandholm
Abstract: A considerable chasm has been looming for decades between theory and practice in zero-sum game solving through first-order methods. Although a convergence rate of $T^{-1}$ has long been established since Nemirovski's mirror-prox algorithm and Nesterov's excessive gap technique in the early 2000s, the most effective paradigm in practice is *counterfactual regret minimization*, which is based on *regret matching* and its modern variants. In particular, the state of the art across most benchmarks is *predictive* regret matching$^+$ (PRM$^+$), in conjunction with non-uniform averaging. Yet, such algorithms can exhibit slower $\Omega(T^{-1/2})$ convergence even in self-play. In this paper, we close the gap between theory and practice. We propose a new scale-invariant and parameter-free variant of PRM$^+$, which we call IREG-PRM$^+$. We show that it achieves $T^{-1/2}$ best-iterate and $T^{-1}$ (i.e., optimal) average-iterate convergence guarantees, while also being on par with PRM$^+$ on benchmark games. From a technical standpoint, we draw an analogy between IREG-PRM$^+$ and optimistic gradient descent with *adaptive* learning rate. The basic flaw of PRM$^+$ is that the ($\ell_2$-)norm of the regret vector -- which can be thought of as the inverse of the learning rate -- can decrease. By contrast, we design IREG-PRM$^+$ so as to maintain the invariance that the norm of the regret vector is nondecreasing. This enables us to derive an RVU-type bound for IREG-PRM$^+$, the first such property that does not rely on introducing additional hyperparameters to enforce smoothness. Furthermore, we find that IREG-PRM$^+$ performs on par with an adaptive version of optimistic gradient descent that we introduce whose learning rate depends on the misprediction error, demystifying the effectiveness of the regret matching family *vis-a-vis* more standard optimization techniques.
Authors: Yuta Shikuri, Hironori Fujisawa
Abstract: Survival analysis is a statistical technique used to estimate the time until an event occurs. Although it is applied across a wide range of fields, adjusting for reporting delays under practical constraints remains a significant challenge in the insurance industry. Such delays render event occurrences unobservable when their reports are subject to right censoring. This issue becomes particularly critical when estimating hazard rates for newly enrolled cohorts with limited follow-up due to administrative censoring. Our study addresses this challenge by jointly modeling the parametric hazard functions of event occurrences and report timings. The joint probability distribution is marginalized over the latent event occurrence status. We construct an estimator for the proposed survival model and establish its asymptotic consistency. Furthermore, we develop an expectation-maximization algorithm to compute its estimates. Using these findings, we propose a two-stage estimation procedure based on a parametric proportional hazards model to evaluate observations subject to administrative censoring. Experimental results demonstrate that our method effectively improves the timeliness of risk evaluation for newly enrolled cohorts.
Authors: Kaosar Uddin
Abstract: We present spd-metrics-id, a Python package for computing distances and divergences between symmetric positive-definite (SPD) matrices. Unlike traditional toolkits that focus on specific applications, spd-metrics-id provides a unified, extensible, and reproducible framework for SPD distance computation. The package supports a wide variety of geometry-aware metrics, including Alpha-z Bures-Wasserstein, Alpha-Procrustes, affine-invariant Riemannian, log-Euclidean, and others, and is accessible both via a command-line interface and a Python API. Reproducibility is ensured through Docker images and Zenodo archiving. We illustrate usage through a connectome fingerprinting example, but the package is broadly applicable to covariance analysis, diffusion tensor imaging, and other domains requiring SPD matrix comparison. The package is openly available at https://pypi.org/project/spd-metrics-id/.
Authors: Ziyi Chen, Peiran Yu, Heng Huang
Abstract: This work aims to solve a stochastic nonconvex nonsmooth composite optimization problem. Previous works on composite optimization problem requires the major part to satisfy Lipschitz smoothness or some relaxed smoothness conditions, which excludes some machine learning examples such as regularized ReLU network and sparse support matrix machine. In this work, we focus on stochastic nonconvex composite optimization problem without any smoothness assumptions. In particular, we propose two new notions of approximate stationary points for such optimization problem and obtain finite-time convergence results of two zeroth-order algorithms to these two approximate stationary points respectively. Finally, we demonstrate that these algorithms are effective using numerical experiments.
Authors: Akira Kitaoka
Abstract: In mixed-integer linear programming, data-driven inverse optimization that learns the objective function and the constraints from observed data plays an important role in constructing appropriate mathematical models for various fields, including power systems and scheduling. However, to the best of our knowledge, there is no known method for learning both the objective functions and the constraints. In this paper, we propose a two-stage method for a class of problems where the objective function is expressed as a linear combination of functions and the constraints are represented by functions and thresholds. Specifically, our method first learns the constraints and then learns the objective function. On the theoretical side, we show the proposed method can solve inverse optimization problems in finite dataset, develop statistical learning theory in pseudometric spaces and sub-Gaussian distributions, and construct a statistical learning for inverse optimization. On the experimental side, we demonstrate that our method is practically applicable for scheduling problems formulated as integer linear programmings with up to 100 decision variables, which are typical in real-world settings.
Authors: Bobby Shi, Kevin Tian, Matthew S. Zhang
Abstract: We survey different perspectives on the stochastic localization process of [Eld13], a powerful construction that has had many exciting recent applications in high-dimensional probability and algorithm design. Unlike prior surveys on this topic, our focus is on giving a self-contained presentation of all known alternative constructions of Eldan's stochastic localization, with an emphasis on connections between different constructions. Our hope is that by collecting these perspectives, some of which had primarily arisen within a particular community (e.g., probability theory, theoretical computer science, information theory, or machine learning), we can broaden the accessibility of stochastic localization, and ease its future use.
Authors: Ian Baxter, Hamid Pahlavan, Pedram Hassanzadeh, Katharine Rucker, Tiffany Shaw
Abstract: Physics-based atmosphere-land models with prescribed sea surface temperature have notable successes but also biases in their ability to represent atmospheric variability compared to observations. Recently, AI emulators and hybrid models have emerged with the potential to overcome these biases, but still require systematic evaluation against metrics grounded in fundamental atmospheric dynamics. Here, we evaluate the representation of four atmospheric variability benchmarking metrics in a fully data-driven AI emulator (ACE2-ERA5) and hybrid model (NeuralGCM). The hybrid model and emulator can capture the spectra of large-scale tropical waves and extratropical eddy-mean flow interactions, including critical levels. However, both struggle to capture the timescales associated with quasi-biennial oscillation (QBO, $\sim 28$ months) and Southern annular mode propagation ($\sim 150$ days). These dynamical metrics serve as an initial benchmarking tool to inform AI model development and understand their limitations, which may be essential for out-of-distribution applications (e.g., extrapolating to unseen climates).
Authors: Baber Jan, Saeed Anwar, Aiman H. El-Maleh, Abdul Jabbar Siddiqui, Abdul Bais
Abstract: Camouflaged object detection segments objects with intrinsic similarity and edge disruption. Current detection methods rely on accumulated complex components. Each approach adds components such as boundary modules, attention mechanisms, and multi-scale processors independently. This accumulation creates a computational burden without proportional gains. To manage this complexity, they process at reduced resolutions, eliminating fine details essential for camouflage. We present SPEGNet, addressing fragmentation through a unified design. The architecture integrates multi-scale features via channel calibration and spatial enhancement. Boundaries emerge directly from context-rich representations, maintaining semantic-spatial alignment. Progressive refinement implements scale-adaptive edge modulation with peak influence at intermediate resolutions. This design strikes a balance between boundary precision and regional consistency. SPEGNet achieves 0.887 $S_\alpha$ on CAMO, 0.890 on COD10K, and 0.895 on NC4K, with real-time inference speed. Our approach excels across scales, from tiny, intricate objects to large, pattern-similar ones, while handling occlusion and ambiguous boundaries. Code, model weights, and results are available on \href{https://github.com/Baber-Jan/SPEGNet}{https://github.com/Baber-Jan/SPEGNet}.
URLs: https://github.com/Baber-Jan/SPEGNet, https://github.com/Baber-Jan/SPEGNet
Authors: Gang Li, Yan Chen, Ming Lin, Tianbao Yang
Abstract: Recent large reasoning models (LRMs) driven by reinforcement learning algorithms (e.g., GRPO) have achieved remarkable performance on challenging reasoning tasks. However, these models suffer from overthinking, generating unnecessarily long and redundant reasoning even for simple questions, which substantially increases computational cost and response latency. While existing methods incorporate length rewards to GRPO to promote concise reasoning, they incur significant performance degradation. We identify the root cause: when rewards for correct but long rollouts are penalized, GRPO's group-relative advantage function can assign them negative advantages, actively discouraging valid reasoning. To overcome this, we propose Decoupled Reward Policy Optimization (DRPO), a novel framework that decouples the length-based learning signal of correct rollouts from incorrect ones. DRPO ensures that reward signals for correct rollouts are normalized solely within the positive group, shielding them from interference by negative samples. The DRPO's objective is grounded in integrating an optimized positive data distribution, which maximizes length-based rewards under a KL regularization, into a discriminative objective. We derive a closed-form solution for this distribution, enabling efficient computation of the objective and its gradients using only on-policy data and importance weighting. Of independent interest, this formulation is general and can incorporate other preference rewards of positive data beyond length. Experiments on mathematical reasoning tasks demonstrate DRPO's significant superiority over six efficient reasoning baselines. Notably, with a 1.5B model, our method achieves 77\% length reduction with only 1.1\% performance loss on simple questions like GSM8k dataset, while the follow-up baseline sacrifices 4.3\% for 68\% length reduction.
Authors: Soo Yong Kim, Suin Cho, Vincent-Daniel Yun, Gyeongyeon Hwang
Abstract: Bridging clinical diagnostic reasoning with AI remains a central challenge in medical imaging. We introduce MedCLM, an automated pipeline that converts detection datasets into large-scale medical visual question answering (VQA) data with Chain-of-Thought (CoT) reasoning by linking lesion boxes to organ segmentation and structured rationales. These contextual signals enable medical vision-language models to generate question-answer pairs with step-by-step reasoning. To utilize this data effectively, we propose an Integrated CoT-Curriculum Strategy composed of an Easy stage with explicit lesion boxes for visual grounding, a Medium stage that encourages implicit localization, and a Hard stage for weakly supervised reasoning. Experimental results demonstrate that MedCLM attains state-of-the-art performance on several medical VQA benchmarks, providing a scalable framework for developing clinically aligned medical vision-language models.
Authors: Akshay Govind Srinivasan, Vikas Dwivedi, Balaji Srinivasan
Abstract: Partial differential equation (PDE) solvers are fundamental to engineering simulation. Classical mesh-based approaches (finite difference/volume/element) are fast and accurate on high-quality meshes but struggle with higher-order operators and complex, hard-to-mesh geometries. Recently developed physics-informed neural networks (PINNs) and their variants are mesh-free and flexible, yet compute-intensive and often less accurate. This paper systematically benchmarks RBF-PIELM, a rapid PINN variant-an extreme learning machine with radial-basis activations-for higher-order PDEs. RBF-PIELM replaces PINNs' time-consuming gradient descent with a single-shot least-squares solve. We test RBF-PIELM on the fourth-order biharmonic equation using two benchmarks: lid-driven cavity flow (streamfunction formulation) and a manufactured oscillatory solution. Our results show up to $(350\times)$ faster training than PINNs and over $(10\times)$ fewer parameters for comparable solution accuracy. Despite surpassing PINNs, RBF-PIELM still lags mature mesh-based solvers and its accuracy degrades on highly oscillatory solutions, highlighting remaining challenges for practical deployment.
Authors: Yue Que, Yingyi Zhang, Xiangyu Zhao, Chen Ma
Abstract: Graph-based recommender systems leverage neighborhood aggregation to generate node representations, which is highly sensitive to popularity bias, resulting in an echo effect during information propagation. Existing graph-based debiasing solutions refine the aggregation process with attempts such as edge reconstruction or weight adjustment. However, these methods remain inadequate in fully alleviating popularity bias. Specifically, this is because 1) they provide no insights into graph aggregation rationality, thus lacking an optimality guarantee; 2) they fail to well balance the training and debiasing process, which undermines the effectiveness. In this paper, we propose a novel approach to mitigate popularity bias through rational modeling of the graph aggregation process. We reveal that graph aggregation is a special form of backdoor adjustment in causal inference, where the aggregation weight corresponds to the historical interaction likelihood distribution. Based on this insight, we devise an encoder-decoder architecture, namely Causality-aware Graph Aggregation Weight Estimator for Debiasing (CAGED), to approximate the unbiased aggregation weight by optimizing the evidence lower bound of the interaction likelihood. In order to enhance the debiasing effectiveness during early training stages, we further design a momentum update strategy that incrementally refines the aggregation weight matrix. Extensive experiments on three datasets demonstrate that CAGED outperforms existing graph-based debiasing methods. Our implementation is available at https://github.com/QueYork/CAGED.
Authors: Fumio Nemoto, Nobuyuki Koike, Daichi Sato, Yuuta Kawaai, Masayuki Ohzeki
Abstract: Recently, bicycle-sharing systems have been implemented in numerous cities, becoming integral to daily life. However, a prevalent issue arises when intensive commuting demand leads to bicycle shortages in specific areas and at particular times. To address this challenge, we employ a novel quantum machine learning model that analyzes time series data by fitting quantum time evolution to observed sequences. This model enables us to capture actual trends in bicycle counts at individual ports and identify correlations between different ports. Utilizing the trained model, we simulate the impact of proactively adding bicycles to high-demand ports on the overall rental number across the system. Given that the core of this method lies in a Monte Carlo simulation, it is anticipated to have a wide range of industrial applications.
Authors: Kaito Takanami, Takashi Takahashi, Yoshiyuki Kabashima
Abstract: In-context learning (ICL) is a key building block of modern large language models, yet its theoretical mechanisms remain poorly understood. It is particularly mysterious how ICL operates in real-world applications where tasks have a common structure. In this work, we address this problem by analyzing a linear attention model trained on low-rank regression tasks. Within this setting, we precisely characterize the distribution of predictions and the generalization error in the high-dimensional limit. Moreover, we find that statistical fluctuations in finite pre-training data induce an implicit regularization. Finally, we identify a sharp phase transition of the generalization error governed by task structure. These results provide a framework for understanding how transformers learn to learn the task structure.
Authors: Jorge Leonardo Ruiz Williams
Abstract: We introduce a scalable witness-based persistent homology pipeline for full-brain MRI volumes that couples density-aware landmark selection with a GPU-ready witness filtration. Candidates are scored by a hybrid metric that balances geometric coverage against inverse kernel density, yielding landmark sets that shrink mean pairwise distances by 30-60% over random or density-only baselines while preserving topological features. Benchmarks on BrainWeb, IXI, and synthetic manifolds execute in under ten seconds on a single NVIDIA RTX 4090 GPU, avoiding the combinatorial blow-up of Cech, Vietoris-Rips, and alpha filtrations. The package is distributed on PyPI as whale-tda (installable via pip); source and issues are hosted at https://github.com/jorgeLRW/whale. The release also exposes a fast preset (mri_deep_dive_fast) for exploratory sweeps, and ships with reproducibility-focused scripts and artifacts for drop-in use in medical imaging workflows.
Authors: Alexej Brauer, Paul Menzel
Abstract: In a dynamic landscape where portfolios and environments evolve, maintaining the accuracy of pricing models is critical. To the best of our knowledge, this is the first study to systematically examine concept drift in non-life insurance pricing. We (i) provide an overview of the relevant literature and commonly used methodologies, clarify the distinction between virtual drift and concept drift, and explain their implications for long-run model performance; (ii) review and formalize common performance measures, including the Gini index and deviance loss, and articulate their interpretation; (iii) derive the asymptotic distribution of the Gini index, enabling valid inference and hypothesis testing; and (iv) present a standardized monitoring procedure that indicates when refitting is warranted. We illustrate the framework using a modified real-world portfolio with induced concept drift and discuss practical considerations and pitfalls.
Authors: Naman Gupta, Shreeyash Gowaikar, Arun Iyer, Kirankumar Shiragur, Ramakrishna B Bairi, Rishikesh Maurya, Ritabrata Maiti, Sankarshan Damle, Shachee Mishra Gupta
Abstract: Reasoning over very long inputs remains difficult for large language models (LLMs). Common workarounds either shrink the input via retrieval (risking missed evidence), enlarge the context window (straining selectivity), or stage multiple agents to read in pieces. In staged pipelines (e.g., Chain of Agents, CoA), free-form summaries passed between agents can discard crucial details and amplify early mistakes. We introduce COSMIR (Chain Orchestrated Structured Memory for Iterative Reasoning), a chain-style framework that replaces ad hoc messages with a structured memory. A Planner agent first turns a user query into concrete, checkable sub-questions. worker agents process chunks via a fixed micro-cycle: Extract, Infer, Refine, writing all updates to the shared memory. A Manager agent then Synthesizes the final answer directly from the memory. This preserves step-wise read-then-reason benefits while changing both the communication medium (structured memory) and the worker procedure (fixed micro-cycle), yielding higher faithfulness, better long-range aggregation, and auditability. On long-context QA from the HELMET suite, COSMIR reduces propagation-stage information loss and improves accuracy over a CoA baseline.
Authors: Juncheng Wang, Chao Xu, Cheng Yu, Zhe Hu, Haoyu Xie, Guoqi Yu, Lei Shang, Shujun Wang
Abstract: While language models (LMs) paired with residual vector quantization (RVQ) tokenizers have shown promise in text-to-audio (T2A) generation, they still lag behind diffusion-based models by a non-trivial margin. We identify a critical dilemma underpinning this gap: incorporating more RVQ layers improves audio reconstruction fidelity but exceeds the generation capacity of conventional LMs. To address this, we first analyze RVQ dynamics and uncover two key limitations: 1) orthogonality of features across RVQ layers hinders effective LMs training, and 2) descending semantic richness in tokens from deeper RVQ layers exacerbates exposure bias during autoregressive decoding. Based on these insights, we propose Siren, a novel LM-based framework that employs multiple isolated transformers with causal conditioning and anti-causal alignment via reinforcement learning. Extensive experiments demonstrate that Siren outperforms both existing LM-based and diffusion-based T2A systems, achieving state-of-the-art results. By bridging the representational strengths of LMs with the fidelity demands of audio synthesis, our approach repositions LMs as competitive contenders against diffusion models in T2A tasks. Moreover, by aligning audio representations with linguistic structures, Siren facilitates a promising pathway toward unified multi-modal generation frameworks.
Authors: Junsei Ito, Yasuaki Wasa
Abstract: This article proposes a data-driven PID controller design based on the principle of adaptive gain optimization, leveraging Physics-Informed Neural Networks (PINNs) generated for predictive modeling purposes. The proposed control design method utilizes gradients of the PID gain optimization, achieved through the automatic differentiation of PINNs, to apply model predictive control using a cost function based on tracking error and control inputs. By optimizing PINNs-based PID gains, the method achieves adaptive gain tuning that ensures stability while accounting for system nonlinearities. The proposed method features a systematic framework for integrating PINNs-based models of dynamical control systems into closed-loop control systems, enabling direct application to PID control design. A series of numerical experiments is conducted to demonstrate the effectiveness of the proposed method from the control perspectives based on both time and frequency domains.
Authors: Eduardo Fernandes Montesuma, Yassir Bendou, Mike Gartrell
Abstract: Wasserstein barycenters provide a powerful tool for aggregating probability measures, while leveraging the geometry of their ambient space. Existing discrete methods suffer from poor scalability, as they require access to the complete set of samples from input measures. We address this issue by recasting the original barycenter problem as a gradient flow in the Wasserstein space. Our approach offers two advantages. First, we achieve scalability by sampling mini-batches from the input measures. Second, we incorporate functionals over probability measures, which regularize the barycenter problem through internal, potential, and interaction energies. We present two algorithms for empirical and Gaussian mixture measures, providing convergence guarantees under the Polyak-{\L}ojasiewicz inequality. Experimental validation on toy datasets and domain adaptation benchmarks show that our methods outperform previous discrete and neural net-based methods for computing Wasserstein barycenters.
Authors: Yuan Wang, Mingyu Li, Haibo Chen
Abstract: Computer-use agents (CUAs) powered by large language models (LLMs) have emerged as a promising approach to automating computer tasks, yet they struggle with graphical user interfaces (GUIs). GUIs, designed for humans, force LLMs to decompose high-level goals into lengthy, error-prone sequences of fine-grained actions, resulting in low success rates and an excessive number of LLM calls. We propose Goal-Oriented Interface (GOI), a novel abstraction that transforms existing GUIs into three declarative primitives: access, state, and observation, which are better suited for LLMs. Our key idea is policy-mechanism separation: LLMs focus on high-level semantic planning (policy) while GOI handles low-level navigation and interaction (mechanism). GOI does not require modifying the application source code or relying on application programming interfaces (APIs). We evaluate GOI with Microsoft Office Suite (Word, PowerPoint, Excel) on Windows. Compared to a leading GUI-based agent baseline, GOI improves task success rates by 67% and reduces interaction steps by 43.5%. Notably, GOI completes over 61% of successful tasks with a single LLM call.
Authors: Eugene Lim, Tzeh Yuan Neoh, Nicholas Teh
Abstract: We study a sequential decision-making model where a set of items is repeatedly matched to the same set of agents over multiple rounds. The objective is to determine a sequence of matchings that either maximizes the utility of the least advantaged agent at the end of all rounds (optimal) or at the end of every individual round (anytime optimal). We investigate the computational challenges associated with finding (anytime) optimal outcomes and demonstrate that these problems are generally computationally intractable. However, we provide approximation algorithms, fixed-parameter tractable algorithms, and identify several special cases whereby the problem(s) can be solved efficiently. Along the way, we also establish characterizations of Pareto-optimal/maximum matchings, which may be of independent interest to works in matching theory and house allocation.
Authors: Ayan Majumdar, Feihao Chen, Jinghui Li, Xiaozhen Wang
Abstract: Large-scale web-scraped text corpora used to train general-purpose AI models often contain harmful demographic-targeted social biases, creating a regulatory need for data auditing and developing scalable bias-detection methods. Although prior work has investigated biases in text datasets and related detection methods, these studies remain narrow in scope. They typically focus on a single content type (e.g., hate speech), cover limited demographic axes, overlook biases affecting multiple demographics simultaneously, and analyze limited techniques. Consequently, practitioners lack a holistic understanding of the strengths and limitations of recent large language models (LLMs) for automated bias detection. In this study, we present a comprehensive evaluation framework aimed at English texts to assess the ability of LLMs in detecting demographic-targeted social biases. To align with regulatory requirements, we frame bias detection as a multi-label task using a demographic-focused taxonomy. We then conduct a systematic evaluation with models across scales and techniques, including prompting, in-context learning, and fine-tuning. Using twelve datasets spanning diverse content types and demographics, our study demonstrates the promise of fine-tuned smaller models for scalable detection. However, our analyses also expose persistent gaps across demographic axes and multi-demographic targeted biases, underscoring the need for more effective and scalable auditing frameworks.
Authors: Joann Ching, Gerhard Widmer
Abstract: Music Emotion Recognition (MER) is a task deeply connected to human perception, relying heavily on subjective annotations collected from contributors. Prior studies tend to focus on specific musical styles rather than incorporating a diverse range of genres, such as rock and classical, within a single framework. In this paper, we address the task of recognizing emotion from audio content by investigating five datasets with dimensional emotion annotations -- EmoMusic, DEAM, PMEmo, WTC, and WCMED -- which span various musical styles. We demonstrate the problem of out-of-distribution generalization in a systematic experiment. By closely looking at multiple data and feature sets, we provide insight into genre-emotion relationships in existing data and examine potential genre dominance and dataset biases in certain feature representations. Based on these experiments, we arrive at a simple yet effective framework that combines embeddings extracted from the Jukebox model with chroma features and demonstrate how, alongside a combination of several diverse training sets, this permits us to train models with substantially improved cross-dataset generalization capabilities.
Authors: Lucas Bandarkar, Chenyuan Yang, Mohsen Fayyaz, Junlin Hu, Nanyun Peng
Abstract: Mixture-of-Experts (MoE) architectures have become the key to scaling modern LLMs, yet little is understood about how their sparse routing dynamics respond to multilingual data. In this work, we analyze expert routing patterns using parallel multilingual datasets and present highly interpretable layer-wise phenomena. We find that MoE models route tokens in language-specific ways in the early and late decoder layers but exhibit significant cross-lingual routing alignment in middle layers, mirroring parameter-sharing trends observed in dense LLMs. In particular, we reveal a clear, strong correlation between a model's performance in a given language and how similarly its tokens are routed to English in these layers. Extending beyond correlation, we explore inference-time interventions that induce higher cross-lingual routing alignment. We introduce a method that steers the router by promoting middle-layer task experts frequently activated in English, and it successfully increases multilingual performance. These 1-2% gains are remarkably consistent across two evaluation tasks, three models, and 15+ languages, especially given that these simple interventions override routers of extensively trained, state-of-the-art LLMs. In comparison, interventions outside of the middle layers or targeting multilingual-specialized experts only yield performance degradation. Altogether, we present numerous findings that explain how MoEs process non-English text and demonstrate that generalization is limited by the model's ability to leverage language-universal experts in all languages.
Authors: Ivo Petrov, Jasper Dekoninck, Martin Vechev
Abstract: Large language models (LLMs) have recently shown strong performance on mathematical benchmarks. At the same time, they are prone to hallucination and sycophancy, often providing convincing but flawed proofs for incorrect mathematical statements provided by users. This significantly limits the applicability of LLMs in theorem proving, as verification of these flawed proofs must be done manually by expert mathematicians. However, existing benchmarks that measure sycophancy in mathematics are limited: they focus solely on final-answer problems, rely on very simple and often contaminated datasets, and construct benchmark samples using synthetic modifications that create ill-posed questions rather than well-posed questions that are demonstrably false. To address these issues, we introduce BrokenMath, the first benchmark for evaluating sycophantic behavior in LLMs within the context of natural language theorem proving. BrokenMath is built from advanced 2025 competition problems, which are perturbed with an LLM to produce false statements and subsequently refined through expert review. Using an LLM-as-a-judge framework, we evaluate state-of-the-art LLMs and agentic systems and find that sycophancy is widespread, with the best model, GPT-5, producing sycophantic answers 29% of the time. We further investigate several mitigation strategies, including test-time interventions and supervised fine-tuning on curated sycophantic examples. These approaches substantially reduce, but do not eliminate, sycophantic behavior.
Authors: Miguel Alves Pereira
Abstract: This article proposes predictive economics as a distinct analytical perspective within economics, grounded in machine learning and centred on predictive accuracy rather than causal identification. Drawing on the instrumentalist tradition (Friedman), the explanation-prediction divide (Shmueli), and the contrast between modelling cultures (Breiman), we formalise prediction as a valid epistemological and methodological objective. Reviewing recent applications across economic subfields, we show how predictive models contribute to empirical analysis, particularly in complex or data-rich contexts. This perspective complements existing approaches and supports a more pluralistic methodology - one that values out-of-sample performance alongside interpretability and theoretical structure.
Authors: Baher Mohammad, Magauiya Zhussip, Stamatios Lefkimmiatis
Abstract: We introduce MAVE (Mamba with Cross-Attention for Voice Editing and Synthesis), a novel autoregressive architecture for text-conditioned voice editing and high-fidelity text-to-speech (TTS) synthesis, built on a cross-attentive Mamba backbone. MAVE achieves state-of-the-art performance in speech editing and very competitive results in zero-shot TTS, while not being explicitly trained on the latter task, outperforming leading autoregressive and diffusion models on diverse, real-world audio. By integrating Mamba for efficient audio sequence modeling with cross-attention for precise text-acoustic alignment, MAVE enables context-aware voice editing with exceptional naturalness and speaker consistency. In pairwise human evaluations on a random 40-sample subset of the RealEdit benchmark (400 judgments), 57.2% of listeners rated MAVE - edited speech as perceptually equal to the original, while 24.8% prefered the original and 18.0% MAVE - demonstrating that in the majority of cases edits are indistinguishable from the source. MAVE compares favorably with VoiceCraft and FluentSpeech both on pairwise comparisons and standalone mean opinion score (MOS) evaluations. For zero-shot TTS, MAVE exceeds VoiceCraft in both speaker similarity and naturalness, without requiring multiple inference runs or post-processing. Remarkably, these quality gains come with a significantly lower memory cost and approximately the same latency: MAVE requires ~6x less memory than VoiceCraft during inference on utterances from the RealEdit database (mean duration: 6.21s, A100, FP16, batch size 1). Our results demonstrate that MAVE establishes a new standard for flexible, high-fidelity voice editing and synthesis through the synergistic integration of structured state-space modeling and cross-modal attention.
Authors: Thorsten Gl\"usenkamp
Abstract: A generic D-dimensional Gaussian can be conditioned or projected onto the D-1 unit sphere, thereby leading to the well-known Fisher-Bingham (FB) or Angular Gaussian (AG) distribution families, respectively. These are some of the most fundamental distributions on the sphere, yet cannot straightforwardly be written as a normalizing flow except in two special cases: the von-Mises Fisher in D=3 and the central angular Gaussian in any D. In this paper, we describe how to generalize these special cases to a family of normalizing flows that behave similarly to the full FB or AG family in any D. We call them "zoom-linear-project" (ZLP)-Fisher flows. Unlike a normal Fisher-Bingham distribution, their composition allows to gradually add complexity as needed. Furthermore, they can naturally handle conditional density estimation with target distributions that vary by orders of magnitude in scale - a setting that is important in astronomical applications but that existing flows often struggle with. A particularly useful member of the new family is the Kent analogue that can cheaply upgrade any flow in this situation to yield better performance.
Authors: Xiaomeng Fan, Yuchuan Mao, Zhi Gao, Yuwei Wu, Jin Chen, Yunde Jia
Abstract: Open-vocabulary learning requires modeling the data distribution in open environments, which consists of both seen-class and unseen-class data. Existing methods estimate the distribution in open environments using seen-class data, where the absence of unseen classes makes the estimation error inherently unidentifiable. Intuitively, learning beyond the seen classes is crucial for distribution estimation to bound the estimation error. We theoretically demonstrate that the distribution can be effectively estimated by generating unseen-class data, through which the estimation error is upper-bounded. Building on this theoretical insight, we propose a novel open-vocabulary learning method, which generates unseen-class data for estimating the distribution in open environments. The method consists of a class-domain-wise data generation pipeline and a distribution alignment algorithm. The data generation pipeline generates unseen-class data under the guidance of a hierarchical semantic tree and domain information inferred from the seen-class data, facilitating accurate distribution estimation. With the generated data, the distribution alignment algorithm estimates and maximizes the posterior probability to enhance generalization in open-vocabulary learning. Extensive experiments on $11$ datasets demonstrate that our method outperforms baseline approaches by up to $14\%$, highlighting its effectiveness and superiority.
Authors: Max Kirchner, Hanna Hoffmann, Alexander C. Jenke, Oliver L. Saldanha, Kevin Pfeiffer, Weam Kanjo, Julia Alekseenko, Claas de Boer, Santhi Raj Kolamuri, Lorenzo Mazza, Nicolas Padoy, Sophia Bano, Annika Reinke, Lena Maier-Hein, Danail Stoyanov, Jakob N. Kather, Fiona R. Kolbinger, Sebastian Bodenstedt, Stefanie Speidel
Abstract: Purpose: The FedSurg challenge was designed to benchmark the state of the art in federated learning for surgical video classification. Its goal was to assess how well current methods generalize to unseen clinical centers and adapt through local fine-tuning while enabling collaborative model development without sharing patient data. Methods: Participants developed strategies to classify inflammation stages in appendicitis using a preliminary version of the multi-center Appendix300 video dataset. The challenge evaluated two tasks: generalization to an unseen center and center-specific adaptation after fine-tuning. Submitted approaches included foundation models with linear probing, metric learning with triplet loss, and various FL aggregation schemes (FedAvg, FedMedian, FedSAM). Performance was assessed using F1-score and Expected Cost, with ranking robustness evaluated via bootstrapping and statistical testing. Results: In the generalization task, performance across centers was limited. In the adaptation task, all teams improved after fine-tuning, though ranking stability was low. The ViViT-based submission achieved the strongest overall performance. The challenge highlighted limitations in generalization, sensitivity to class imbalance, and difficulties in hyperparameter tuning in decentralized training, while spatiotemporal modeling and context-aware preprocessing emerged as promising strategies. Conclusion: The FedSurg Challenge establishes the first benchmark for evaluating FL strategies in surgical video classification. Findings highlight the trade-off between local personalization and global robustness, and underscore the importance of architecture choice, preprocessing, and loss design. This benchmarking offers a reference point for future development of imbalance-aware, adaptive, and robust FL methods in clinical surgical AI.
Authors: Arie Wortsman, Bruno Loureiro
Abstract: In this work, we investigate high-dimensional kernel ridge regression (KRR) on i.i.d. Gaussian data with anisotropic power-law covariance. This setting differs fundamentally from the classical source & capacity conditions for KRR, where power-law assumptions are typically imposed on the kernel eigen-spectrum itself. Our contributions are twofold. First, we derive an explicit characterization of the kernel spectrum for polynomial inner-product kernels, giving a precise description of how the kernel eigen-spectrum inherits the data decay. Second, we provide an asymptotic analysis of the excess risk in the high-dimensional regime for a particular kernel with this spectral behavior, showing that the sample complexity is governed by the effective dimension of the data rather than the ambient dimension. These results establish a fundamental advantage of learning with power-law anisotropic data over isotropic data. To our knowledge, this is the first rigorous treatment of non-linear KRR under power-law data.
Authors: Malith Premarathna (Department of Statistics, University of Nebraska-Lincoln), Fabrizio Ruggeri (CNR IMATI, Milano), Dixon Vimalajeewa (Department of Statistics, University of Nebraska-Lincoln)
Abstract: Understanding signal behavior across scales is vital in areas such as natural phenomena analysis and financial modeling. A key property is self-similarity, quantified by the Hurst exponent (H), which reveals long-term dependencies. Wavelet-based methods are effective for estimating H due to their multi-scale analysis capability, but additive noise in real-world measurements often degrades accuracy. We propose Noise-Controlled ALPHEE (NC-ALPHEE), an enhancement of the Average Level-Pairwise Hurst Exponent Estimator (ALPHEE), incorporating noise mitigation and generating multiple level-pairwise estimates from signal energy pairs. A neural network (NN) combines these estimates, replacing traditional averaging. This adaptive learning maintains ALPHEE's behavior in noise-free cases while improving performance in noisy conditions. Extensive simulations show that in noise-free data, NC-ALPHEE matches ALPHEE's accuracy using both averaging and NN-based methods. Under noise, however, traditional averaging deteriorates and requires impractical level restrictions, while NC-ALPHEE consistently outperforms existing techniques without such constraints. NC-ALPHEE offers a robust, adaptive approach for H estimation, significantly enhancing the reliability of wavelet-based methods in noisy environments.
Authors: Muquan Li, Hang Gou, Dongyang Zhang, Shuang Liang, Xiurui Xie, Deqiang Ouyang, Ke Qin
Abstract: The growing demand for efficient deep learning has positioned dataset distillation as a pivotal technique for compressing training dataset while preserving model performance. However, existing inner-loop optimization methods for dataset distillation typically rely on random truncation strategies, which lack flexibility and often yield suboptimal results. In this work, we observe that neural networks exhibit distinct learning dynamics across different training stages-early, middle, and late-making random truncation ineffective. To address this limitation, we propose Automatic Truncated Backpropagation Through Time (AT-BPTT), a novel framework that dynamically adapts both truncation positions and window sizes according to intrinsic gradient behavior. AT-BPTT introduces three key components: (1) a probabilistic mechanism for stage-aware timestep selection, (2) an adaptive window sizing strategy based on gradient variation, and (3) a low-rank Hessian approximation to reduce computational overhead. Extensive experiments on CIFAR-10, CIFAR-100, Tiny-ImageNet, and ImageNet-1K show that AT-BPTT achieves state-of-the-art performance, improving accuracy by an average of 6.16% over baseline methods. Moreover, our approach accelerates inner-loop optimization by 3.9x while saving 63% memory cost.
Authors: Dongge Han, Camille Couturier, Daniel Madrigal Diaz, Xuchao Zhang, Victor R\"uhle, Saravan Rajmohan
Abstract: We introduce LEGOMem, a modular procedural memory framework for multi-agent large language model (LLM) systems in workflow automation. LEGOMem decomposes past task trajectories into reusable memory units and flexibly allocates them across orchestrators and task agents to support planning and execution. To explore the design space of memory in multi-agent systems, we use LEGOMem as a lens and conduct a systematic study of procedural memory in multi-agent systems, examining where memory should be placed, how it should be retrieved, and which agents benefit most. Experiments on the OfficeBench benchmark show that orchestrator memory is critical for effective task decomposition and delegation, while fine-grained agent memory improves execution accuracy. We find that even teams composed of smaller language models can benefit substantially from procedural memory, narrowing the performance gap with stronger agents by leveraging prior execution traces for more accurate planning and tool use. These results position LEGOMem as both a practical framework for memory-augmented agent systems and a research tool for understanding memory design in multi-agent workflow automation.
Authors: Martial Guidez, Stefan Duffner, Yannick Alpou, Oscar R\"oth, Christophe Garcia
Abstract: Although deep neural networks and in particular Convolutional Neural Networks have demonstrated state-of-the-art performance in image classification with relatively high efficiency, they still exhibit high computational costs, often rendering them impractical for real-time and edge applications. Therefore, a multitude of compression techniques have been developed to reduce these costs while maintaining accuracy. In addition, dynamic architectures have been introduced to modulate the level of compression at execution time, which is a desirable property in many resource-limited application scenarios. The proposed method effectively integrates two well-established optimization techniques: early exits and knowledge distillation, where a reduced student early-exit model is trained from a more complex teacher early-exit model. The primary contribution of this research lies in the approach for training the student early-exit model. In comparison to the conventional Knowledge Distillation loss, our approach incorporates a new entropy-based loss for images where the teacher's classification was incorrect. The proposed method optimizes the trade-off between accuracy and efficiency, thereby achieving significant reductions in computational complexity without compromising classification performance. The validity of this approach is substantiated by experimental results on image classification datasets CIFAR10, CIFAR100 and SVHN, which further opens new research perspectives for Knowledge Distillation in other contexts.
Authors: Sam Earle, Zehua Jiang, Eugene Vinitsky, Julian Togelius
Abstract: Procedural Content Generation via Reinforcement Learning (PCGRL) offers a method for training controllable level designer agents without the need for human datasets, using metrics that serve as proxies for level quality as rewards. Existing PCGRL research focuses on single generator agents, but are bottlenecked by the need to frequently recalculate heuristics of level quality and the agent's need to navigate around potentially large maps. By framing level generation as a multi-agent problem, we mitigate the efficiency bottleneck of single-agent PCGRL by reducing the number of reward calculations relative to the number of agent actions. We also find that multi-agent level generators are better able to generalize to out-of-distribution map shapes, which we argue is due to the generators' learning more local, modular design policies. We conclude that treating content generation as a distributed, multi-agent task is beneficial for generating functional artifacts at scale.
Authors: Hayat Rajani, Valerio Franchi, Borja Martinez-Clavel Valles, Raimon Ramos, Rafael Garcia, Nuno Gracias
Abstract: Benthic habitat mapping is fundamental for understanding marine ecosystems, guiding conservation efforts, and supporting sustainable resource management. Yet, the scarcity of large, annotated datasets limits the development and benchmarking of machine learning models in this domain. This paper introduces a thorough multi-modal dataset, comprising about a million side-scan sonar (SSS) tiles collected along the coast of Catalonia (Spain), complemented by bathymetric maps and a set of co-registered optical images from targeted surveys using an autonomous underwater vehicle (AUV). Approximately \num{36000} of the SSS tiles have been manually annotated with segmentation masks to enable supervised fine-tuning of classification models. All the raw sensor data, together with mosaics, are also released to support further exploration and algorithm development. To address challenges in multi-sensor data fusion for AUVs, we spatially associate optical images with corresponding SSS tiles, facilitating self-supervised, cross-modal representation learning. Accompanying open-source preprocessing and annotation tools are provided to enhance accessibility and encourage research. This resource aims to establish a standardized benchmark for underwater habitat mapping, promoting advancements in autonomous seafloor classification and multi-sensor integration.
Authors: Nathan Shankar, Pawel Ladosz, Hujun Yin
Abstract: This paper presents a novel approach for enabling robust robotic perception in dark environments using infrared (IR) stream. IR stream is less susceptible to noise than RGB in low-light conditions. However, it is dominated by active emitter patterns that hinder high-level tasks such as object detection, tracking and localisation. To address this, a U-Net-based architecture is proposed that reconstructs clean IR images from emitter-populated input, improving both image quality and downstream robotic performance. This approach outperforms existing enhancement techniques and enables reliable operation of vision-driven robotic systems across illumination conditions from well-lit to extreme low-light scenes.
Authors: Yuxin Wen, Arman Zharmagambetov, Ivan Evtimov, Narine Kokhlikyan, Tom Goldstein, Kamalika Chaudhuri, Chuan Guo
Abstract: Prompt injection poses a serious threat to the reliability and safety of LLM agents. Recent defenses against prompt injection, such as Instruction Hierarchy and SecAlign, have shown notable robustness against static attacks. However, to more thoroughly evaluate the robustness of these defenses, it is arguably necessary to employ strong attacks such as automated red-teaming. To this end, we introduce RL-Hammer, a simple recipe for training attacker models that automatically learn to perform strong prompt injections and jailbreaks via reinforcement learning. RL-Hammer requires no warm-up data and can be trained entirely from scratch. To achieve high ASRs against industrial-level models with defenses, we propose a set of practical techniques that enable highly effective, universal attacks. Using this pipeline, RL-Hammer reaches a 98% ASR against GPT-4o and a $72\%$ ASR against GPT-5 with the Instruction Hierarchy defense. We further discuss the challenge of achieving high diversity in attacks, highlighting how attacker models tend to reward-hack diversity objectives. Finally, we show that RL-Hammer can evade multiple prompt injection detectors. We hope our work advances automatic red-teaming and motivates the development of stronger, more principled defenses. Code is available at https://github.com/facebookresearch/rl-injector.
Authors: Punya Syon Pandey, Hai Son Le, Devansh Bhardwaj, Rada Mihalcea, Zhijing Jin
Abstract: Large language models (LLMs) are increasingly deployed in contexts where their failures can have direct sociopolitical consequences. Yet, existing safety benchmarks rarely test vulnerabilities in domains such as political manipulation, propaganda and disinformation generation, or surveillance and information control. We introduce SocialHarmBench, a dataset of 585 prompts spanning 7 sociopolitical categories and 34 countries, designed to surface where LLMs most acutely fail in politically charged contexts. Our evaluations reveal several shortcomings: open-weight models exhibit high vulnerability to harmful compliance, with Mistral-7B reaching attack success rates as high as 97% to 98% in domains such as historical revisionism, propaganda, and political manipulation. Moreover, temporal and geographic analyses show that LLMs are most fragile when confronted with 21st-century or pre-20th-century contexts, and when responding to prompts tied to regions such as Latin America, the USA, and the UK. These findings demonstrate that current safeguards fail to generalize to high-stakes sociopolitical settings, exposing systematic biases and raising concerns about the reliability of LLMs in preserving human rights and democratic values. We share the SocialHarmBench benchmark at https://huggingface.co/datasets/psyonp/SocialHarmBench.
URLs: https://huggingface.co/datasets/psyonp/SocialHarmBench.
Authors: Zheng Xiong, Kang Li, Zilin Wang, Matthew Jackson, Jakob Foerster, Shimon Whiteson
Abstract: Built upon language and vision foundation models with strong generalization ability and trained on large-scale robotic data, Vision-Language-Action (VLA) models have recently emerged as a promising approach to learning generalist robotic policies. However, a key drawback of existing VLAs is their extremely high inference costs. In this paper, we propose HyperVLA to address this problem. Unlike existing monolithic VLAs that activate the whole model during both training and inference, HyperVLA uses a novel hypernetwork (HN)-based architecture that activates only a small task-specific policy during inference, while still retaining the high model capacity needed to accommodate diverse multi-task behaviors during training. Successfully training an HN-based VLA is nontrivial so HyperVLA contains several key algorithm design features that improve its performance, including properly utilizing the prior knowledge from existing vision foundation models, HN normalization, and an action generation strategy. Compared to monolithic VLAs, HyperVLA achieves a similar or even higher success rate for both zero-shot generalization and few-shot adaptation, while significantly reducing inference costs. Compared to OpenVLA, a state-of-the-art VLA model, HyperVLA reduces the number of activated parameters at test time by $90\times$, and accelerates inference speed by $120\times$. Code is publicly available at https://github.com/MasterXiong/HyperVLA
Authors: Ngeyen Yinkfu, Sunday Nwovu, Jonathan Kayizzi, Angelique Uwamahoro
Abstract: In Kigali, Rwanda, motorcycle taxis are a primary mode of transportation, often navigating unpredictably and disregarding traffic rules, posing significant challenges for autonomous driving systems. This study compares four object detection models--YOLOv5, Faster R-CNN, SSD, and RetinaNet--for motorbike detection using a custom dataset of 198 images collected in Kigali. Implemented in PyTorch with transfer learning, the models were evaluated for accuracy, localization, and inference speed to assess their suitability for real-time navigation in resource-constrained settings. We identify implementation challenges, including dataset limitations and model complexities, and recommend simplified architectures for future work to enhance accessibility for autonomous systems in developing countries like Rwanda.
Authors: Eyal Cohen (LPSM), Christophe Denis (SAMM), Mohamed Hebiri (LAMA)
Abstract: Set-valued classification is used in multiclass settings where confusion between classes can occur and lead to misleading predictions. However, its application may amplify discriminatory bias motivating the development of set-valued approaches under fairness constraints. In this paper, we address the problem of set-valued classification under demographic parity and expected size constraints. We propose two complementary strategies: an oracle-based method that minimizes classification risk while satisfying both constraints, and a computationally efficient proxy that prioritizes constraint satisfaction. For both strategies, we derive closed-form expressions for the (optimal) fair set-valued classifiers and use these to build plug-in, data-driven procedures for empirical predictions. We establish distribution-free convergence rates for violations of the size and fairness constraints for both methods, and under mild assumptions we also provide excess-risk bounds for the oracle-based approach. Empirical results demonstrate the effectiveness of both strategies and highlight the efficiency of our proxy method.
Authors: Amir Hameed Mir
Abstract: Large Language Models (LLMs) often produce fluent yet factually incorrect statements-a phenomenon known as hallucination-posing serious risks in high-stakes domains. We present Layer-wise Semantic Dynamics (LSD), a geometric framework for hallucination detection that analyzes the evolution of hidden-state semantics across transformer layers. Unlike prior methods that rely on multiple sampling passes or external verification sources, LSD operates intrinsically within the model's representational space. Using margin-based contrastive learning, LSD aligns hidden activations with ground-truth embeddings derived from a factual encoder, revealing a distinct separation in semantic trajectories: factual responses preserve stable alignment, while hallucinations exhibit pronounced semantic drift across depth. Evaluated on the TruthfulQA and synthetic factual-hallucination datasets, LSD achieves an F1-score of 0.92, AUROC of 0.96, and clustering accuracy of 0.89, outperforming SelfCheckGPT and Semantic Entropy baselines while requiring only a single forward pass. This efficiency yields a 5-20x speedup over sampling-based methods without sacrificing precision or interpretability. LSD offers a scalable, model-agnostic mechanism for real-time hallucination monitoring and provides new insights into the geometry of factual consistency within large language models.
Authors: Guoxin Chen, Zile Qiao, Wenqing Wang, Donglei Yu, Xuanzhong Chen, Hao Sun, Minpeng Liao, Kai Fan, Yong Jiang, Penguin Xie, Wayne Xin Zhao, Ruihua Song, Fei Huang
Abstract: Large Reasoning Models (LRMs) often exhibit a tendency for overanalysis in simple tasks, where the models excessively utilize System 2-type, deliberate reasoning, leading to inefficient token generation. Furthermore, these models face challenges in adapting their reasoning capabilities to rapidly changing environments due to the static nature of their pretraining data. To address these issues, advancing Large Language Models (LLMs) for complex reasoning tasks requires innovative approaches that bridge intuitive and deliberate cognitive processes, akin to human cognition's dual-system dynamic. This paper introduces a Multi-Agent System for Deep ReSearch (MARS) enabling seamless integration of System 1's fast, intuitive thinking with System 2's deliberate reasoning within LLMs. MARS strategically integrates multiple external tools, such as Google Search, Google Scholar, and Python Interpreter, to access up-to-date information and execute complex computations, while creating a specialized division of labor where System 1 efficiently processes and summarizes high-volume external information, providing distilled insights that expand System 2's reasoning context without overwhelming its capacity. Furthermore, we propose a multi-agent reinforcement learning framework extending Group Relative Policy Optimization to simultaneously optimize both systems with multi-turn tool interactions, bin-packing optimization, and sample balancing strategies that enhance collaborative efficiency. Extensive experiments demonstrate MARS achieves substantial improvements of 3.86% on the challenging Humanity's Last Exam (HLE) benchmark and an average gain of 8.9% across 7 knowledge-intensive tasks, validating the effectiveness of our dual-system paradigm for complex reasoning in dynamic information environments.
Authors: Yuxi Liu, Catherine Lalman, Yimin Yang
Abstract: The effectiveness of modern deep learning models is predicated on the availability of large-scale, human-annotated datasets, a process that is notoriously expensive and time-consuming. While Active Learning (AL) offers a strategic solution by labeling only the most informative and representative data, its iterative nature still necessitates significant human involvement. Unsupervised Active Learning (UAL) presents an alternative by shifting the annotation burden to a single, post-selection step. Unfortunately, prevailing UAL methods struggle to achieve state-of-the-art performance. These approaches typically rely on local, gradient-based scoring for sample importance estimation, which not only makes them vulnerable to ambiguous and noisy data but also hinders their capacity to select samples that adequately represent the full data distribution. Moreover, their use of shallow, one-shot linear selection falls short of a true UAL paradigm. In this paper, we propose the Natural Feature Progressive Framework (NFPF), a UAL method that revolutionizes how sample importance is measured. At its core, NFPF employs a Specific Feature Learning Machine (SFLM) to effectively quantify each sample's contribution to model performance. We further utilize the SFLM to define a powerful Reconstruction Difference metric for initial sample selection. Our comprehensive experiments show that NFPF significantly outperforms all established UAL methods and achieves performance on par with supervised AL methods on vision datasets. Detailed ablation studies and qualitative visualizations provide compelling evidence for NFPF's superior performance, enhanced robustness, and improved data distribution coverage.
Authors: Om Dobariya, Akhil Kumar
Abstract: The wording of natural language prompts has been shown to influence the performance of large language models (LLMs), yet the role of politeness and tone remains underexplored. In this study, we investigate how varying levels of prompt politeness affect model accuracy on multiple-choice questions. We created a dataset of 50 base questions spanning mathematics, science, and history, each rewritten into five tone variants: Very Polite, Polite, Neutral, Rude, and Very Rude, yielding 250 unique prompts. Using ChatGPT 4o, we evaluated responses across these conditions and applied paired sample t-tests to assess statistical significance. Contrary to expectations, impolite prompts consistently outperformed polite ones, with accuracy ranging from 80.8% for Very Polite prompts to 84.8% for Very Rude prompts. These findings differ from earlier studies that associated rudeness with poorer outcomes, suggesting that newer LLMs may respond differently to tonal variation. Our results highlight the importance of studying pragmatic aspects of prompting and raise broader questions about the social dimensions of human-AI interaction.
Authors: Marcel Wien\"obst, Leonard Henckel, Sebastian Weichwald
Abstract: We present FLOP (Fast Learning of Order and Parents), a score-based causal discovery algorithm for linear models. It pairs fast parent selection with iterative Cholesky-based score updates, cutting run-times over prior algorithms. This makes it feasible to fully embrace discrete search, enabling iterated local search with principled order initialization to find graphs with scores at or close to the global optimum. The resulting structures are highly accurate across benchmarks, with near-perfect recovery in standard settings. This performance calls for revisiting discrete search over graphs as a reasonable approach to causal discovery.
Authors: Nabarun Deb
Abstract: In this paper, we study fluctuations of conditionally centered statistics of the form $$N^{-1/2}\sum_{i=1}^N c_i(g(\sigma_i)-\mathbb{E}_N[g(\sigma_i)|\sigma_j,j\neq i])$$ where $(\sigma_1,\ldots ,\sigma_N)$ are sampled from a dependent random field, and $g$ is some bounded function. Our first main result shows that under weak smoothness assumptions on the conditional means (which cover both sparse and dense interactions), the above statistic converges to a Gaussian \emph{scale mixture} with a random scale determined by a \emph{quadratic variance} and an \emph{interaction component}. We also show that under appropriate studentization, the limit becomes a pivotal Gaussian. We leverage this theory to develop a general asymptotic framework for maximum pseudolikelihood (MPLE) inference in dependent random fields. We apply our results to Ising models with pairwise as well as higher-order interactions and exponential random graph models (ERGMs). In particular, we obtain a joint central limit theorem for the inverse temperature and magnetization parameters via the joint MPLE (to our knowledge, the first such result in dense, irregular regimes), and we derive conditionally centered edge CLTs and marginal MPLE CLTs for ERGMs without restricting to the ``sub-critical" region. Our proof is based on a method of moments approach via combinatorial decision-tree pruning, which may be of independent interest.
Authors: Khalid Mehtab Khan, Anagha Kulkarni
Abstract: Identifying cultural capital (CC) themes in student reflections can offer valuable insights that help foster equitable learning environments in classrooms. However, themes such as aspirational goals or family support are often woven into narratives, rather than appearing as direct keywords. This makes them difficult to detect for standard NLP models that process sentences in isolation. The core challenge stems from a lack of awareness, as standard models are pre-trained on general corpora, leaving them blind to the domain-specific language and narrative context inherent to the data. To address this, we introduce AWARE, a framework that systematically attempts to improve a transformer model's awareness for this nuanced task. AWARE has three core components: 1) Domain Awareness, adapting the model's vocabulary to the linguistic style of student reflections; 2) Context Awareness, generating sentence embeddings that are aware of the full essay context; and 3) Class Overlap Awareness, employing a multi-label strategy to recognize the coexistence of themes in a single sentence. Our results show that by making the model explicitly aware of the properties of the input, AWARE outperforms a strong baseline by 2.1 percentage points in Macro-F1 and shows considerable improvements across all themes. This work provides a robust and generalizable methodology for any text classification task in which meaning depends on the context of the narrative.
Authors: Koen Vellenga, H. Joe Steinhauer, Jonas Andersson, Anders Sj\"ogren
Abstract: Deep neural networks (DNNs) are increasingly applied to safety-critical tasks in resource-constrained environments, such as video-based driver action and intention recognition. While last layer probabilistic deep learning (LL-PDL) methods can detect out-of-distribution (OOD) instances, their performance varies. As an alternative to last layer approaches, we propose extending pre-trained DNNs with transformation layers to produce multiple latent representations to estimate the uncertainty. We evaluate our latent uncertainty representation (LUR) and repulsively trained LUR (RLUR) approaches against eight PDL methods across four video-based driver action and intention recognition datasets, comparing classification performance, calibration, and uncertainty-based OOD detection. We also contribute 28,000 frame-level action labels and 1,194 video-level intention labels for the NuScenes dataset. Our results show that LUR and RLUR achieve comparable in-distribution classification performance to other LL-PDL approaches. For uncertainty-based OOD detection, LUR matches top-performing PDL methods while being more efficient to train and easier to tune than approaches that require Markov-Chain Monte Carlo sampling or repulsive training procedures.
Authors: Theodore Jerome Tinker, Kenji Doya, Jun Tani
Abstract: Human infants acquire language and action co-developmentally, achieving remarkable generalization capabilities from only a minimal number of learning examples. In contrast, recent large language models require exposure to billions of training tokens to achieve such generalization. What mechanisms underlie such efficient developmental learning in humans? This study addresses this question through simulation experiments in which robots learn to perform various actions corresponding to imperative sentences (e.g., \textit{push red cube}) via trials of self-guided exploration. Our approach integrates the active inference framework with reinforcement learning, enabling curiosity-driven developmental learning. The simulations yielded several nontrivial findings: i) Curiosity-driven exploration combined with motor noise substantially outperforms learning without curiosity. ii) Simpler, prerequisite-like actions emerge earlier in development, while more complex actions involving these prerequisites develop later. iii) Rote pairing of sentences and actions occurs before the emergence of compositional generalization. iv) Generalization is drastically improved as the number of compositional elements increases. These results shed light into possible mechanisms underlying efficient co-developmental learning in infants and provide computational parallels to findings in developmental psychology.
Authors: Xuanming Cui, Jianpeng Cheng, Hong-you Chen, Satya Narayan Shukla, Abhijeet Awasthi, Xichen Pan, Chaitanya Ahuja, Shlok Kumar Mishra, Qi Guo, Ser-Nam Lim, Aashu Singh, Xiangjun Fan
Abstract: There is a growing interest in Universal Multimodal Embeddings (UME), where models are required to generate task-specific representations. While recent studies show that Multimodal Large Language Models (MLLMs) perform well on such tasks, they treat MLLMs solely as encoders, overlooking their generative capacity. However, such an encoding paradigm becomes less effective as instructions become more complex and require compositional reasoning. Inspired by the proven effectiveness of chain-of-thought reasoning, we propose a general Think-Then-Embed (TTE) framework for UME, composed of a reasoner and an embedder. The reasoner MLLM first generates reasoning traces that explain complex queries, followed by an embedder that produces representations conditioned on both the original query and the intermediate reasoning. This explicit reasoning step enables more nuanced understanding of complex multimodal instructions. Our contributions are threefold. First, by leveraging a powerful MLLM reasoner, we achieve state-of-the-art performance on the MMEB-V2 benchmark, surpassing proprietary models trained on massive in-house datasets. Second, to reduce the dependency on large MLLM reasoners, we finetune a smaller MLLM reasoner using high-quality embedding-centric reasoning traces, achieving the best performance among open-source models with a 7% absolute gain over recently proposed models. Third, we investigate strategies for integrating the reasoner and embedder into a unified model for improved efficiency without sacrificing performance.
Authors: Markus Englberger, Devendra Singh Dhami
Abstract: We present a categorical framework for relating causal models that represent the same system at different levels of abstraction. We define a causal abstraction as natural transformations between appropriate Markov functors, which concisely consolidate desirable properties a causal abstraction should exhibit. Our approach unifies and generalizes previously considered causal abstractions, and we obtain categorical proofs and generalizations of existing results on causal abstractions. Using string diagrammatical tools, we can explicitly describe the graphs that serve as consistent abstractions of a low-level graph under interventions. We discuss how methods from mechanistic interpretability, such as circuit analysis and sparse autoencoders, fit within our categorical framework. We also show how applying do-calculus on a high-level graphical abstraction of an acyclic-directed mixed graph (ADMG), when unobserved confounders are present, gives valid results on the low-level graph, thus generalizing an earlier statement by Anand et al. (2023). We argue that our framework is more suitable for modeling causal abstractions compared to existing categorical frameworks. Finally, we discuss how notions such as $\tau$-consistency and constructive $\tau$-abstractions can be recovered with our framework.
Authors: V\'ictor Blanco, Harshit Kothari, James Luedtke
Abstract: In this paper, we propose a new mathematical optimization model for multiclass classification based on arrangements of hyperplanes. Our approach preserves the core support vector machine (SVM) paradigm of maximizing class separation while minimizing misclassification errors, and it is computationally more efficient than a previous formulation. We present a kernel-based extension that allows it to construct nonlinear decision boundaries. Furthermore, we show how the framework can naturally incorporate alternative geometric structures, including classification trees, $\ell_p$-SVMs, and models with discrete feature selection. To address large-scale instances, we develop a dynamic clustering matheuristic that leverages the proposed MIP formulation. Extensive computational experiments demonstrate the efficiency of the proposed model and dynamic clustering heuristic, and we report competitive classification performance on both synthetic datasets and real-world benchmarks from the UCI Machine Learning Repository, comparing our method with state-of-the-art implementations available in scikit-learn.
Authors: Siheng Zhao, Yanjie Ze, Yue Wang, C. Karen Liu, Pieter Abbeel, Guanya Shi, Rocky Duan
Abstract: Humanoid whole-body loco-manipulation promises transformative capabilities for daily service and warehouse tasks. While recent advances in general motion tracking (GMT) have enabled humanoids to reproduce diverse human motions, these policies lack the precision and object awareness required for loco-manipulation. To this end, we introduce ResMimic, a two-stage residual learning framework for precise and expressive humanoid control from human motion data. First, a GMT policy, trained on large-scale human-only motion, serves as a task-agnostic base for generating human-like whole-body movements. An efficient but precise residual policy is then learned to refine the GMT outputs to improve locomotion and incorporate object interaction. To further facilitate efficient training, we design (i) a point-cloud-based object tracking reward for smoother optimization, (ii) a contact reward that encourages accurate humanoid body-object interactions, and (iii) a curriculum-based virtual object controller to stabilize early training. We evaluate ResMimic in both simulation and on a real Unitree G1 humanoid. Results show substantial gains in task success, training efficiency, and robustness over strong baselines. Videos are available at https://resmimic.github.io/ .
Authors: Lin Fan, Peter W. Glynn
Abstract: We study the process-level dynamics of Thompson sampling in the ``small gap'' regime. The small gap regime is one in which the gaps between the arm means are of order $\sqrt{\gamma}$ or smaller and the time horizon is of order $1/\gamma$, where $\gamma$ is small. As $\gamma \downarrow 0$, we show that the process-level dynamics of Thompson sampling converge weakly to the solutions to certain stochastic differential equations and stochastic ordinary differential equations. Our weak convergence theory is developed from first principles using the Continuous Mapping Theorem, can handle stationary, weakly dependent reward processes, and can also be adapted to analyze a variety of sampling-based bandit algorithms. Indeed, we show that the process-level dynamics of many sampling-based bandit algorithms -- including Thompson sampling designed for any single-parameter exponential family of rewards, as well as non-parametric bandit algorithms based on bootstrap re-sampling -- satisfy an invariance principle. Namely, their weak limits coincide with that of Gaussian parametric Thompson sampling with Gaussian priors. Moreover, in the small gap regime, the regret performance of these algorithms is generally insensitive to model mis-specification, changing continuously with increasing degrees of mis-specification.
Authors: Brandon Amos
Abstract: Optimization is a ubiquitous modeling tool and is often deployed in settings which repeatedly solve similar instances of the same problem. Amortized optimization methods use learning to predict the solutions to problems in these settings, exploiting the shared structure between similar problem instances. These methods have been crucial in variational inference and reinforcement learning and are capable of solving optimization problems many orders of magnitudes faster than traditional optimization methods that do not use amortization. This tutorial presents an introduction to the amortized optimization foundations behind these advancements and overviews their applications in variational inference, sparse coding, gradient-based meta-learning, control, reinforcement learning, convex optimization, optimal transport, and deep equilibrium networks. The source code for this tutorial is available at https://github.com/facebookresearch/amortized-optimization-tutorial.
URLs: https://github.com/facebookresearch/amortized-optimization-tutorial.
Authors: Brandon Amos
Abstract: This paper focuses on computing the convex conjugate (also known as the Legendre-Fenchel conjugate or c-transform) that appears in Euclidean Wasserstein-2 optimal transport. This conjugation is considered difficult to compute and in practice, methods are limited by not being able to exactly conjugate the dual potentials in continuous space. To overcome this, the computation of the conjugate can be approximated with amortized optimization, which learns a model to predict the conjugate. I show that combining amortized approximations to the conjugate with a solver for fine-tuning significantly improves the quality of transport maps learned for the Wasserstein-2 benchmark by Korotin et al. (2021a) and is able to model many 2-dimensional couplings and flows considered in the literature. All baselines, methods, and solvers are publicly available at http://github.com/facebookresearch/w2ot.
Authors: Weiqiu You, Helen Qu, Marco Gatti, Bhuvnesh Jain, Eric Wong
Abstract: Self-attributing neural networks (SANNs) present a potential path towards interpretable models for high-dimensional problems, but often face significant trade-offs in performance. In this work, we formally prove a lower bound on errors of per-feature SANNs, whereas group-based SANNs can achieve zero error and thus high performance. Motivated by these insights, we propose Sum-of-Parts (SOP), a framework that transforms any differentiable model into a group-based SANN, where feature groups are learned end-to-end without group supervision. SOP achieves state-of-the-art performance for SANNs on vision and language tasks, and we validate that the groups are interpretable on a range of quantitative and semantic metrics. We further validate the utility of SOP explanations in model debugging and cosmological scientific discovery. Our code is available at https://github.com/BrachioLab/sop
Authors: Joe Suzuki, Tian-Le Yang
Abstract: LiNGAM determines the variable order from cause to effect using additive noise models, but it faces challenges with confounding. Previous methods maintained LiNGAM's fundamental structure while trying to identify and address variables affected by confounding. As a result, these methods required significant computational resources regardless of the presence of confounding, and they did not ensure the detection of all confounding types. In contrast, this paper enhances LiNGAM by introducing LiNGAM-MMI, a method that quantifies the magnitude of confounding using KL divergence and arranges the variables to minimize its impact. This method efficiently achieves a globally optimal variable order through the shortest path problem formulation. LiNGAM-MMI processes data as efficiently as traditional LiNGAM in scenarios without confounding while effectively addressing confounding situations. Our experimental results suggest that LiNGAM-MMI more accurately determines the correct variable order, both in the presence and absence of confounding. The code is in the supplementary file in this link: https://github.com/SkyJoyTianle/ISIT2024.
Authors: Jiaqi Wang, Xi Li
Abstract: Federated Learning (FL), while a breakthrough in decentralized machine learning, contends with significant challenges such as limited data availability and the variability of computational resources, which can stifle the performance and scalability of the models. The integration of Foundation Models (FMs) into FL presents a compelling solution to these issues, with the potential to enhance data richness and reduce computational demands through pre-training and data augmentation. However, this incorporation introduces novel issues in terms of robustness, privacy, and fairness, which have not been sufficiently addressed in the existing research. We make a preliminary investigation into this field by systematically evaluating the implications of FM-FL integration across these dimensions. We analyze the trade-offs involved, uncover the threats and issues introduced by this integration, and propose a set of criteria and strategies for navigating these challenges. Furthermore, we identify potential research directions for advancing this field, laying a foundation for future development in creating reliable, secure, and equitable FL systems.
Authors: Guri Zab\"ergja, Arlind Kadra, Christian M. M. Frey, Josif Grabocka
Abstract: Tabular data represent one of the most prevalent data formats in applied machine learning, largely because they accommodate a broad spectrum of real-world problems. Existing literature has studied many of the shortcomings of neural architectures on tabular data and has repeatedly confirmed the scalability and robustness of gradient-boosted decision trees across varied datasets. However, recent deep learning models have not been subjected to a comprehensive evaluation under conditions that allow for a fair comparison with existing classical approaches. This situation motivates an investigation into whether recent deep-learning paradigms outperform classical ML methods on tabular data. Our survey fills this gap by benchmarking seventeen state-of-the-art methods, spanning neural networks, classical ML and AutoML techniques. Our empirical results over 68 diverse datasets from a well-established benchmark indicate a paradigm shift, where Deep Learning methods outperform classical approaches.
Authors: Andr\'e Hottung, Mridul Mahajan, Kevin Tierney
Abstract: Reinforcement learning-based methods for constructing solutions to combinatorial optimization problems are rapidly approaching the performance of human-designed algorithms. To further narrow the gap, learning-based approaches must efficiently explore the solution space during the search process. Recent approaches artificially increase exploration by enforcing diverse solution generation through handcrafted rules, however, these rules can impair solution quality and are difficult to design for more complex problems. In this paper, we introduce PolyNet, an approach for improving exploration of the solution space by learning complementary solution strategies. In contrast to other works, PolyNet uses only a single-decoder and a training schema that does not enforce diverse solution generation through handcrafted rules. We evaluate PolyNet on four combinatorial optimization problems and observe that the implicit diversity mechanism allows PolyNet to find better solutions than approaches that explicitly enforce diverse solution generation.
Authors: Xin Li
Abstract: What is intelligence? We argue for a structural-dynamical account rooted in a topological closure law: \emph{the boundary of a boundary vanishes} ($\partial^2=0$). This principle forces transient fragments to cancel while closed cycles persist as invariants, yielding the cascade $\partial^2\!=\!0 \Rightarrow \text{cycles (invariants)} \Rightarrow \text{memory} \Rightarrow \text{prediction (intelligence)}$. Prediction requires invariance: only order-invariant cycles can stabilize the predictive substrate. This motivates the \textbf{Structure-before-Specificity (SbS)} principle, where persistent structures ($\Phi$) must stabilize before contextual specificities ($\Psi$) can be meaningfully interpreted, and is formalized by the \textbf{Context-Content Uncertainty Principle (CCUP)}, which casts cognition as dynamic alignment that minimizes the joint uncertainty $H(\Phi,\Psi)$. We show that \textbf{Memory-Amortized Inference (MAI)} is the computational mechanism that implements SbS\,$\rightarrow$\,CCUP through dual bootstrapping: \emph{temporal} bootstrapping consolidates episodic specifics into reusable latent trajectories, while \emph{spatial} bootstrapping reuses these invariants across latent manifolds. This framework explains why \emph{semantics precedes syntax}: stable cycles anchor meaning, and symbolic syntax emerges only after semantic invariants are in place. In an evolutionary perspective, the same closure law unifies the trajectory of natural intelligence: from primitive memory traces in microbes, to cyclic sensorimotor patterns in bilaterians, to semantic generalization in mammals, culminating in human symbolic abstraction by natural language. In sum, intelligence arises from the progressive collapse of specificity into structure, grounded in the closure-induced emergence of invariants.
Authors: Donghwan Lee
Abstract: Convergence of Q-learning has been the focus of extensive research over the past several decades. Recently, an asymptotic convergence analysis for Q-learning was introduced using a switching system framework. This approach applies the so-called ordinary differential equation (ODE) approach to prove the convergence of the asynchronous Q-learning modeled as a continuous-time switching system, where notions from switching system theory are used to prove its asymptotic stability without using explicit Lyapunov arguments. However, to prove stability, restrictive conditions, such as quasi-monotonicity, must be satisfied for the underlying switching systems, which makes it hard to easily generalize the analysis method to other reinforcement learning algorithms, such as the smooth Q-learning variants. In this paper, we present a more general and unified convergence analysis that improves upon the switching system approach and can analyze Q-learning and its smooth variants. The proposed analysis is motivated by previous work on the convergence of synchronous Q-learning based on $p$-norm serving as a Lyapunov function. However, the proposed analysis addresses more general ODE models that can cover both asynchronous Q-learning and its smooth versions with simpler frameworks.
Authors: Zhao Ding, Chenguang Duan, Yuling Jiao, Ruoxuan Li, Jerry Zhijian Yang, Pingwen Zhang
Abstract: We propose the characteristic generator, a novel one-step generative model that combines the efficiency of sampling in Generative Adversarial Networks (GANs) with the stable performance of flow-based models. Our model is driven by characteristics, along which the probability density transport can be described by ordinary differential equations (ODEs). Specifically, we first estimate the underlying velocity field and use the Euler method to solve the probability flow ODE, generating discrete approximations of the characteristics. A deep neural network is then trained to fit these characteristics, creating a one-step map that pushes a simple Gaussian distribution to the target distribution. In the theoretical aspect, we provide a comprehensive analysis of the errors arising from velocity matching, Euler discretization, and characteristic fitting to establish a non-asymptotic convergence rate in the 2-Wasserstein distance under mild data assumptions. Crucially, we demonstrate that under a standard manifold assumption, this convergence rate depends only on the intrinsic dimension of data rather than the much larger ambient dimension, proving our model's ability to mitigate the curse of dimensionality. To our knowledge, this is the first rigorous convergence analysis for a flow-based one-step generative model. Experiments on both synthetic and real-world datasets demonstrate that the characteristic generator achieves high-quality and high-resolution sample generation with the efficiency of just a single neural network evaluation.
Authors: Charikleia Iakovidou, Kibaek Kim
Abstract: Federated learning (FL) was recently proposed to securely train models with data held over multiple locations (``clients'') under the coordination of a central server. Prolonged training times caused by slow clients may hinder the performance of FL; while asynchronous communication is a promising solution, highly heterogeneous client response times under non-IID local data may introduce significant bias to the global model, particularly in client-driven setups where sampling is infeasible. To address this issue, we propose \underline{A}synch\underline{R}onous \underline{E}xact \underline{A}veraging (\textsc{AREA}), a stochastic (sub)gradient method that leverages asynchrony for scalability and uses client-side memory to correct the bias induced by uneven participation, without client sampling or prior knowledge of client latencies. \textsc{AREA} communicates model residuals rather than gradient estimates, reducing exposure to gradient inversion, and is compatible with secure aggregation. Under standard assumptions and unbounded, heterogeneous delays with finite mean, AREA achieves optimal convergence rates: $\mathcal{O}(1/K)$ in the strongly convex, smooth regime and $\mathcal{O}(1/\sqrt{K})$ in the convex, nonsmooth regime. For strongly convex, smooth objectives, we demonstrate theoretically and empirically that AREA accommodates larger step sizes than existing methods, enabling fast convergence without adversely impacting model generalization. In the convex, nonsmooth setting, to our knowledge we are the first to obtain rates that scale with the average client update frequency rather than the minimum or maximum, indicating increased robustness to outliers.
Authors: Giuseppe Serra, Florian Buettner
Abstract: Given the ability to model more realistic and dynamic problems, Federated Continual Learning (FCL) has been increasingly investigated recently. A well-known problem encountered in this setting is the so-called catastrophic forgetting, for which the learning model is inclined to focus on more recent tasks while forgetting the previously learned knowledge. The majority of the current approaches in FCL propose generative-based solutions to solve said problem. However, this setting requires multiple training epochs over the data, implying an offline setting where datasets are stored locally and remain unchanged over time. Furthermore, the proposed solutions are tailored for vision tasks solely. To overcome these limitations, we propose a new approach to deal with different modalities in the online scenario where new data arrive in streams of mini-batches that can only be processed once. To solve catastrophic forgetting, we propose an uncertainty-aware memory-based approach. Specifically, we suggest using an estimator based on the Bregman Information (BI) to compute the model's variance at the sample level. Through measures of predictive uncertainty, we retrieve samples with specific characteristics, and - by retraining the model on such samples - we demonstrate the potential of this approach to reduce the forgetting effect in realistic settings while maintaining data confidentiality and competitive communication efficiency compared to state-of-the-art approaches.
Authors: Li Ji-An, Marcus K. Benna
Abstract: Backpropagation, a foundational algorithm for training artificial neural networks, predominates in contemporary deep learning. Although highly successful, it is widely considered biologically implausible, because it relies on precise symmetry between feedforward and feedback weights to accurately propagate gradient signals that assign credit. The so-called weight transport problem concerns how biological brains learn to align feedforward and feedback paths while avoiding the non-biological transport of feedforward weights into feedback weights. To address this, several credit assignment algorithms, such as feedback alignment and the Kollen-Pollack rule, have been proposed. While they can achieve the desired weight alignment, these algorithms imply that if a neuron sends a feedforward synapse to another neuron, it should also receive an identical or at least partially correlated feedback synapse from the latter neuron, thereby forming a bidirectional connection. However, this idealized connectivity pattern contradicts experimental observations in the brain, a discrepancy we refer to as the weight symmetry problem. To address this challenge posed by considering biological constraints on connectivity, we introduce the Product Feedback Alignment (PFA) algorithm. We demonstrate that PFA can eliminate explicit weight symmetry entirely while closely approximating backpropagation and achieving comparable performance in deep convolutional networks. Our results offer a novel approach to solve the longstanding problem of credit assignment in the brain, leading to more biologically plausible learning in deep networks compared to previous methods.
Authors: Maciej Besta, Florian Scheidl, Lukas Gianinazzi, Grzegorz Kwasniewski, Shachar Klaiman, J\"urgen M\"uller, Torsten Hoefler
Abstract: Higher-order graph neural networks (HOGNNs) and the related architectures from Topological Deep Learning are an important class of GNN models that harness polyadic relations between vertices beyond plain edges. They have been used to eliminate issues such as over-smoothing or over-squashing, to significantly enhance the accuracy of GNN predictions, to improve the expressiveness of GNN architectures, and for numerous other goals. A plethora of HOGNN models have been introduced, and they come with diverse neural architectures, and even with different notions of what the "higher-order" means. This richness makes it very challenging to appropriately analyze and compare HOGNN models, and to decide in what scenario to use specific ones. To alleviate this, we first design an in-depth taxonomy and a blueprint for HOGNNs. This facilitates designing models that maximize performance. Then, we use our taxonomy to analyze and compare the available HOGNN models. The outcomes of our analysis are synthesized in a set of insights that help to select the most beneficial GNN model in a given scenario, and a comprehensive list of challenges and opportunities for further research into more powerful HOGNNs.
Authors: Sajjad Hashemian, Mohammad Saeed Arvenaghi, Ebrahim Ardeshir-Larijani
Abstract: In this paper, we introduce new algorithms for Principal Component Analysis (PCA) with outliers. Utilizing techniques from computational geometry, specifically higher-degree Voronoi diagrams, we navigate to the optimal subspace for PCA even in the presence of outliers. This approach achieves an optimal solution with a time complexity of $n^{d+\mathcal{O}(1)}\text{poly}(n,d)$. Additionally, we present a randomized algorithm with a complexity of $2^{\mathcal{O}(r(d-r))} \times \text{poly}(n, d)$. This algorithm samples subspaces characterized in terms of a Grassmannian manifold. By employing such sampling method, we ensure a high likelihood of capturing the optimal subspace, with the success probability $(1 - \delta)^T$. Where $\delta$ represents the probability that a sampled subspace does not contain the optimal solution, and $T$ is the number of subspaces sampled, proportional to $2^{r(d-r)}$. Our use of higher-degree Voronoi diagrams and Grassmannian based sampling offers a clearer conceptual pathway and practical advantages, particularly in handling large datasets or higher-dimensional settings.
Authors: Yucheng Wang, Peiliang Gong, Min Wu, Felix Ott, Xiaoli Li, Lihua Xie, Zhenghua Chen
Abstract: Time-Series (TS) data has grown in importance with the rise of Internet of Things devices like sensors, but its labeling remains costly and complex. While Unsupervised Domain Adaptation (UDAs) offers an effective solution, growing data privacy concerns have led to the development of Source-Free UDA (SFUDAs), enabling model adaptation to target domains without accessing source data. Despite their potential, applying existing SFUDAs to TS data is challenging due to the difficulty of transferring temporal dependencies, an essential characteristic of TS data, particularly in the absence of source samples. Although prior works attempt to address this by specific source pretraining designs, such requirements are often impractical, as source data owners cannot be expected to adhere to particular pretraining schemes. To address this, we propose Temporal Source Recovery (TemSR), a framework that leverages the intrinsic properties of TS data to generate a source-like domain and recover source temporal dependencies. With this domain, TemSR enables dependency transfer to the target domain without accessing source data or relying on source-specific designs, thereby facilitating effective and practical TS-SFUDA. TemSR features a masking recovery optimization process to generate a source-like distribution with restored temporal dependencies. This distribution is further refined through local context-aware regularization to preserve local dependencies, and anchor-based recovery diversity maximization to promote distributional diversity. Together, these components enable effective temporal dependency recovery and facilitate transfer across domains using standard UDA techniques. Extensive experiments across multiple TS tasks demonstrate the effectiveness of TemSR, which even surpasses existing TS-SFUDA methods that require source-specific designs.
Authors: Kshitij Kayastha, Vasilis Gkatzelis, Shahin Jabbari
Abstract: Algorithmic recourse provides individuals who receive undesirable outcomes from machine learning systems with minimum-cost improvements to achieve a desirable outcome. However, machine learning models often get updated, so the recourse may not lead to the desired outcome. The robust recourse framework chooses recourses that are less sensitive to adversarial model changes, but this comes at a higher cost. To address this, we initiate the study of learning-augmented algorithmic recourse and evaluate the extent to which a designer equipped with a prediction of the future model can reduce the cost of recourse when the prediction is accurate (consistency) while also limiting the cost even when the prediction is inaccurate (robustness). We propose a novel algorithm, study the robustness-consistency trade-off, and analyze how prediction accuracy affects performance.
Authors: Yang Chen, Long Yang, Yitao Liang, Zhouchen Lin
Abstract: Low-Dimension-to-High-Dimension (LDHD) generalization is a special case of Out-of-Distribution (OOD) generalization, where the training data are restricted to a low-dimensional subspace of the high-dimensional testing space. Assuming that each instance is generated from a latent variable and the dimension of the latent variable reflects the problem scale, the inherent scaling challenge in length generalization can be captured by the LDHD generalization in the latent space. We theoretically demonstrate that LDHD generalization is generally unattainable without exploiting prior knowledge to provide appropriate inductive bias. Specifically, we explore LDHD generalization in Boolean functions. We verify that different architectures trained with (S)GD converge to \emph{min-degree interpolators w.r.t. different independent sets}. LDHD generalization is achievable if and only if the target function coincides with this inductive bias. Applying the insights from LDHD generalization to length generalization, we explain the effectiveness of CoT as changing the structure latent space to enable better LDHD generalization. We also propose a principle for position embedding design to handle both the inherent LDHD generalization and the nuisances such as the data format. Following the principle, we propose a novel position embedding called RPE-Square that remedies the RPE for dealing with the data format nuisance.
Authors: Yunwei Ren, Jason D. Lee
Abstract: The information exponent ([BAGJ21]) and its extensions -- which are equivalent to the lowest degree in the Hermite expansion of the link function (after a potential label transform) for Gaussian single-index models -- have played an important role in predicting the sample complexity of online stochastic gradient descent (SGD) in various learning tasks. In this work, we demonstrate that, for multi-index models, focusing solely on the lowest degree can miss key structural details of the model and result in suboptimal rates. Specifically, we consider the task of learning target functions of form $f_*(\mathbf{x}) = \sum_{k=1}^{P} \phi(\mathbf{v}_k^* \cdot \mathbf{x})$, where $P \ll d$, the ground-truth directions $\{ \mathbf{v}_k^* \}_{k=1}^P$ are orthonormal, and the information exponent of $\phi$ is $L$. Based on the theory of information exponent, when $L = 2$, only the relevant subspace (not the exact directions) can be recovered due to the rotational invariance of the second-order terms, and when $L > 2$, recovering the directions using online SGD require $\tilde{O}(P d^{L-1})$ samples. In this work, we show that by considering both second- and higher-order terms, we can first learn the relevant space using the second-order terms, and then the exact directions using the higher-order terms, and the overall sample and complexity of online SGD is $\tilde{O}( d P^{L-1} )$.
Authors: Eren Ozbay, Ashkan Golgoon
Abstract: This paper offers a comprehensive analysis of collaborative bandit algorithms and provides a thorough comparison of their performance. Collaborative bandits aim to improve the performance of contextual bandits by introducing relationships between arms (or items), allowing effective propagation of information. Collaboration among arms allows the feedback obtained through a single user (item) to be shared across related users (items). Introducing collaboration also alleviates the cold user (item) problem, i.e., lack of historical information when a new user (item) arriving to the platform with no prior record of interactions. In the context of modeling the relationships between arms (items), there are two main approaches: Hard and soft clustering. We call approaches that model the relationship between arms in an \textit{absolute} manner as hard clustering, i.e., the relationship is binary. Soft clustering relaxes membership constraints, allowing \textit{fuzzy} assignment. Focusing on the latter, we provide extensive experiments on the state-of-the-art collaborative contextual bandit algorithms and investigate the effect of sparsity and how the exploration intensity acts as a correction mechanism. Our numerical experiments demonstrate that controlling for sparsity in collaboration improves data efficiency and performance as it better informs learning. Meanwhile, increasing the exploration intensity acts as a correction because it effectively reduces variance due to potentially misspecified relationships among users. We observe that this misspecification is further remedied by introducing latent factors, and thus, increasing the dimensionality of the bandit parameters.
Authors: Connall Garrod, Jonathan P. Keating
Abstract: Neural collapse (NC) and its multi-layer variant, deep neural collapse (DNC), describe a structured geometry that occurs in the features and weights of trained deep networks. Recent theoretical work by Sukenik et al. using a deep unconstrained feature model (UFM) suggests that DNC is suboptimal under mean squared error (MSE) loss. They heuristically argue that this is due to low-rank bias induced by L2 regularization. In this work, we extend this result to deep UFMs trained with cross-entropy loss, showing that high-rank structures, including DNC, are not generally optimal. We characterize the associated low-rank bias, proving a fixed bound on the number of non-negligible singular values at global minima as network depth increases. We further analyze the loss surface, demonstrating that DNC is more prevalent in the landscape than other critical configurations, which we argue explains its frequent empirical appearance. Our results are validated through experiments in deep UFMs and deep neural networks.
Authors: Pawe{\l} Skier\'s, Kamil Deja
Abstract: In this work, we introduce JDCL - a new method for continual learning with generative rehearsal based on joint diffusion models. Neural networks suffer from catastrophic forgetting defined as abrupt loss in the model's performance when retrained with additional data coming from a different distribution. Generative-replay-based continual learning methods try to mitigate this issue by retraining a model with a combination of new and rehearsal data sampled from a generative model. In this work, we propose to extend this idea by combining a continually trained classifier with a diffusion-based generative model into a single - jointly optimized neural network. We show that such shared parametrization, combined with the knowledge distillation technique allows for stable adaptation to new tasks without catastrophic forgetting. We evaluate our approach on several benchmarks, where it outperforms recent state-of-the-art generative replay techniques. Additionally, we extend our method to the semi-supervised continual learning setup, where it outperforms competing buffer-based replay techniques, and evaluate, in a self-supervised manner, the quality of trained representations.
Authors: Sumeet Ramesh Motwani, Chandler Smith, Rocktim Jyoti Das, Rafael Rafailov, Ivan Laptev, Philip H. S. Torr, Fabio Pizzati, Ronald Clark, Christian Schroeder de Witt
Abstract: Large Language Models (LLMs) often produce answers with a single chain-of-thought, which restricts their ability to explore reasoning paths or self-correct flawed outputs in complex tasks. In this paper, we introduce MALT (Multi-Agent LLM Training), a novel post-training strategy that divides the reasoning process into generation, verification, and refinement steps using a sequential pipeline of heterogeneous agents. During data generation, each agent is repeatedly sampled to form a multi-agent search tree, where final outputs are graded against ground-truth data. We then apply value iteration to propagate reward signals back to each role-conditioned model, automatically producing multi-agent post-training data without human or teacher-model supervision. Our off-policy approach allows each agent to specialize by learning from correct and incorrect trajectories, ultimately improving the end-to-end reasoning chain. On MATH, GSM8K, and CSQA, MALT surpasses the same baseline LLM with a relative improvement of 15.66%, 7.42%, and 9.40% respectively, making it an important advance towards multi-agent cooperative training.
Authors: Hazel Kim, Tom A. Lamb, Adel Bibi, Philip Torr, Yarin Gal
Abstract: Large language models (LLMs) frequently generate confident yet inaccurate responses, introducing significant risks for deployment in safety-critical domains. We present a novel, test-time approach to detecting model hallucination through systematic analysis of information flow across model layers. We target cases when LLMs process inputs with ambiguous or insufficient context. Our investigation reveals that hallucination manifests as usable information deficiencies in inter-layer transmissions. While existing approaches primarily focus on final-layer output analysis, we demonstrate that tracking cross-layer information dynamics ($\mathcal{L}$I) provides robust indicators of model reliability, accounting for both information gain and loss during computation. $\mathcal{L}$I integrates easily with pretrained LLMs without requiring additional training or architectural modifications.
Authors: Aecheon Jung, Seunghwan Lee, Dongyoon Han, Sungeun Hong
Abstract: Model merging offers an efficient alternative to multi-task learning by combining independently fine-tuned models, but most prior approaches focus mainly on avoiding task interference. We argue instead that the real potential of merging lies in achieving synergy, where tasks enhance one another. Our intuition comes from a pilot study showing that when a classifier trained on one task is paired with the encoder of another, the resulting cross-task performance strongly predicts merge quality. Moreover, adapting even a single task-specific layer can substantially improve this compatibility, suggesting a simple yet powerful lever for synergy. Building on this insight, we introduce SyMerge, a lightweight framework that jointly optimizes one task-specific layer and merging coefficients. To ensure stability without labels, SyMerge employs a robust self-labeling strategy guided by expert model predictions, avoiding the pitfalls of entropy-based adaptation. This minimalist yet principled design achieves state-of-the-art results across vision, dense prediction, and NLP benchmarks, while also producing adapted layers that transfer effectively to other merging methods. Our code is available at https://aim-skku.github.io/SyMerge/
Authors: Priyanshu Singh, Kapil Ahuja
Abstract: Hierarchical clustering remains a fundamental challenge in data mining, particularly when dealing with large-scale datasets where traditional approaches fail to scale effectively. Recent Chameleon-based algorithms - Chameleon2, M-Chameleon, and INNGS-Chameleon have proposed advanced strategies but they still suffer from $O(n^2)$ computational complexity, especially for large datasets. With Chameleon2 as the base algorithm, we introduce Chameleon2++ that addresses this challenge. Our algorithm has three parts. First, Graph Generation - we propose an approximate $k$-NN search instead of an exact one, specifically we integrate with the Annoy algorithm. This results in fast approximate nearest neighbor computation, significantly reducing the graph generation time. Second, Graph Partitioning - we propose use of a multi-level partitioning algorithm instead of a recursive bisection one. Specifically we adapt the hMETIS algorithm instead of the FM. This is because multi-level algorithms are robust to approximation introduced in the graph generation phase yielding higher-quality partitions, and that too with minimum configuration requirements. Third, Merging - we retain the flood fill heuristic that ensures connected balanced components in the partitions as well as efficient partition merging criteria leading to the final clusters. These enhancements reduce the overall time complexity to $O(n\log n)$, achieving scalability. On real-world benchmark datasets used in prior Chameleon works, Chameleon2++ delivers an average of 4% improvement in clustering quality. This demonstrates that algorithmic efficiency and clustering quality can co-exist in large-scale hierarchical clustering.
Authors: Seohyun Lee, Wenzhi Fang, Anindya Bijoy Das, Seyyedali Hosseinalipour, David J. Love, Christopher G. Brinton
Abstract: Federated learning (FL) is vulnerable to backdoor attacks, where adversaries alter model behavior on target classification labels by embedding triggers into data samples. While these attacks have received considerable attention in horizontal FL, they are less understood for vertical FL (VFL), where devices hold different features of the samples, and only the server holds the labels. In this work, we propose a novel backdoor attack on VFL which (i) does not rely on gradient information from the server and (ii) considers potential collusion among multiple adversaries for sample selection and trigger embedding. Our label inference model augments variational autoencoders with metric learning, which adversaries can train locally. A consensus process over the adversary graph topology determines which datapoints to poison. We further propose methods for trigger splitting across the adversaries, with an intensity-based implantation scheme skewing the server towards the trigger. Our convergence analysis reveals the impact of backdoor perturbations on VFL indicated by a stationarity gap for the trained model, which we verify empirically as well. We conduct experiments comparing our attack with recent backdoor VFL approaches, finding that ours obtains significantly higher success rates for the same main task performance despite not using server information. Additionally, our results verify the impact of collusion on attack performance.
Authors: Zihao Hu, Xiaoyu Fan, Yuan Yao, Jiheng Zhang, Zhengyuan Zhou
Abstract: First-price auctions have recently gained significant traction in digital advertising markets, exemplified by Google's transition from second-price to first-price auctions. Unlike in second-price auctions, where bidding one's private valuation is a dominant strategy, determining an optimal bidding strategy in first-price auctions is more complex. From a learning perspective, the learner (a specific bidder) can interact with the environment (other bidders, i.e., opponents) sequentially to infer their behaviors. Existing research often assumes specific environmental conditions and benchmarks performance against the best fixed policy (static benchmark). While this approach ensures strong learning guarantees, the static benchmark can deviate significantly from the optimal strategy in environments with even mild non-stationarity. To address such scenarios, a dynamic benchmark--representing the sum of the highest achievable rewards at each time step--offers a more suitable objective. However, achieving no-regret learning with respect to the dynamic benchmark requires additional constraints. By inspecting reward functions in online first-price auctions, we introduce two metrics to quantify the regularity of the sequence of opponents' highest bids, which serve as measures of non-stationarity. We provide a minimax-optimal characterization of the dynamic regret for the class of sequences of opponents' highest bids that satisfy either of these regularity constraints. Our main technical tool is the Optimistic Mirror Descent (OMD) framework with a novel optimism configuration, which is well-suited for achieving minimax-optimal dynamic regret rates in this context. We then use synthetic datasets to validate our theoretical guarantees and demonstrate that our methods outperform existing ones.
Authors: Chenyue Li, Wen Deng, Mengqian Lu, Binhang Yuan
Abstract: The rapid advancements in large language models (LLMs), particularly in their reasoning capabilities, hold transformative potential for addressing complex challenges and boosting scientific discovery in atmospheric science. However, leveraging LLMs effectively in this domain requires a robust and comprehensive evaluation benchmark. Toward this end, we present AtmosSci-Bench, a novel benchmark designed to systematically assess LLM performance across five core categories of atmospheric science problems: hydrology, atmospheric dynamics, atmospheric physics, geophysics, and physical oceanography. AtmosSci-Bench features a dual-format design comprising both multiple-choice questions (MCQs) and open-ended questions (OEQs), enabling scalable automated evaluation alongside deeper analysis of conceptual understanding. We employ a template-based MCQ generation framework to create diverse, graduate-level problems with symbolic perturbation, while OEQs are used to probe open-ended reasoning. We conduct a comprehensive evaluation of representative LLMs, categorized into four groups: instruction-tuned models, advanced reasoning models, math-augmented models, and domain-specific climate models. Our analysis provides some interesting insights into the reasoning and problem-solving capabilities of LLMs in atmospheric science. We believe AtmosSci-Bench can serve as a critical step toward advancing LLM applications in climate services by offering a standard and rigorous evaluation framework. Our source code is available at https://github.com/Relaxed-System-Lab/AtmosSci-Bench.
Authors: Akansha Kalra, Basavasagar Patil, Guanhong Tao, Daniel S. Brown
Abstract: Learning from Demonstration (LfD) algorithms have shown promising results in robotic manipulation tasks, but their vulnerability to offline universal perturbation attacks remains underexplored. This paper presents a comprehensive study of adversarial attacks on both classic and recently proposed algorithms, including Behavior Cloning (BC), LSTM-GMM, Implicit Behavior Cloning (IBC), Diffusion Policy (DP), and Vector-Quantizied Behavior Transformer (VQ-BET). We study the vulnerability of these methods to universal adversarial perturbations. Our experiments on several simulated robotic manipulation tasks reveal that most of the current methods are highly vulnerable to adversarial perturbations. We also show that these attacks are often transferable across algorithms, architectures, and tasks, raising concerning security vulnerabilities to black-box attacks. To the best of our knowledge, we are the first to present a systematic study of the vulnerabilities of different LfD algorithms to both white-box and black-box attacks. Our findings highlight the vulnerabilities of modern BC algorithms, paving the way for future work in addressing such limitations.
Authors: Jiamin Xu, Ivan Nazarov, Aditya Rastogi, \'Africa Peri\'a\~nez, Kyra Gan
Abstract: This paper addresses the poor finite-horizon performance of existing online \emph{restless bandit} (RB) algorithms, which stems from the prohibitive sample complexity of learning a full \emph{Markov decision process} (MDP) for each agent. We argue that superior finite-horizon performance requires \emph{rapid convergence} to a \emph{high-quality} policy. Thus motivated, we introduce a reformulation of online RBs as a \emph{budgeted thresholding contextual bandit}, which simplifies the learning problem by encoding long-term state transitions into a scalar reward. We prove the first non-asymptotic optimality of an oracle policy for a simplified finite-horizon setting. We propose a practical learning policy under a heterogeneous-agent, multi-state setting, and show that it achieves a sublinear regret, achieving \emph{faster convergence} than existing methods. This directly translates to higher cumulative reward, as empirically validated by significant gains over state-of-the-art algorithms in large-scale heterogeneous environments. Our work provides a new pathway for achieving practical, sample-efficient learning in finite-horizon RBs.
Authors: Hanfei Yu, Xingqi Cui, Hong Zhang, Hao Wang, Hao Wang
Abstract: Large Language Models (LLMs) have gained immense success in revolutionizing various applications, including content generation, search and recommendation, and AI-assisted operation. To reduce high training costs, Mixture-of-Experts (MoE) architecture has become a popular backbone for modern LLMs. However, despite the benefits, serving MoE-based LLMs experience severe memory inefficiency due to sparsely activated experts. Recent studies propose to offload inactive experts from GPU memory to CPU memory to improve the serving efficiency of MoE models. However, they either incur high inference latency or high model memory footprints due to coarse-grained designs. To tame the latency-memory trade-off in MoE serving, we present FineMoE, a fine-grained expert offloading system for MoE serving that achieves low inference latency with memory efficiency. We design FineMoE to extract fine-grained expert selection patterns from MoE models and semantic hints from input prompts to efficiently guide expert prefetching, caching, and offloading decisions. FineMoE is prototyped on top of HuggingFace Transformers and deployed on a six-GPU testbed. Experiments with open-source MoE models and real-world workloads show that FineMoE reduces inference latency by 47% and improves expert hit rate by 39% over state-of-the-art solutions.
Authors: Snehal Raj, Brian Coyle
Abstract: Scaling full finetuning of large foundation models strains GPU memory and training time. Parameter Efficient Fine-Tuning (PEFT) methods address this issue via adapter modules which update only a small subset of model parameters. In this work, we introduce Quantum-Inspired Compound Adapters (QuIC Adapters), a PEFT approach inspired from Hamming-weight preserving quantum circuits that can effectively finetune a model using less than 0.02\% memory footprint of the base model. QuIC adapters preserve pretrained representations by enforcing orthogonality in weight parameters, and have native deployment mechanisms on quantum computers. We test QuIC adapters by finetuning large language models like LLaMA and vision transformers on language, math, reasoning and vision benchmarks. In its first-order configuration, QuIC recovers the performance of existing orthogonal methods, while higher-order configurations enable substantial parameter compression (over 40x smaller than LoRA) for a modest performance trade-off, unlocking applications in highly resource-constrained environments. Through ablation studies, we determine that combining multiple Hamming-weight orders with orthogonality and matrix compounding are essential for performant finetuning. Our findings suggest that QuIC adapters offers a promising direction for efficient finetuning of foundation models in resource-constrained environments.
Authors: Zhikai Wu, Sifan Wang, Shiyang Zhang, Sizhuang He, Min Zhu, Anran Jiao, Lu Lu, David van Dijk
Abstract: Operator learning for time-dependent partial differential equations (PDEs) has seen rapid progress in recent years, enabling efficient approximation of complex spatiotemporal dynamics. However, most existing methods rely on fixed time step sizes during rollout, which limits their ability to adapt to varying temporal complexity and often leads to error accumulation. Here, we propose the Time-Adaptive Transformer with Neural Taylor Expansion (TANTE), a novel operator-learning framework that produces continuous-time predictions with adaptive step sizes. TANTE predicts future states by performing a Taylor expansion at the current state, where neural networks learn both the higher-order temporal derivatives and the local radius of convergence. This allows the model to dynamically adjust its rollout based on the local behavior of the solution, thereby reducing cumulative error and improving computational efficiency. We demonstrate the effectiveness of TANTE across a wide range of PDE benchmarks, achieving superior accuracy and adaptability compared to fixed-step baselines, delivering accuracy gains of 60-80 % and speed-ups of 30-40 % at inference time. The code is publicly available at https://github.com/zwu88/TANTE for transparency and reproducibility.
Authors: Raghav Singhal, Kaustubh Ponkshe, Rohit Vartak, Lav R. Varshney, Praneeth Vepakomma
Abstract: Low-Rank Adaptation (LoRA) has become ubiquitous for efficiently fine-tuning foundation models. However, federated fine-tuning using LoRA is challenging due to suboptimal updates arising from traditional federated averaging of individual adapters. Existing solutions either incur prohibitively high communication cost that scales linearly with the number of clients or suffer from performance degradation due to limited expressivity. We introduce Federated Silver Bullet (Fed-SB), a novel approach for federated fine-tuning of LLMs using LoRA-SB, a recently proposed low-rank adaptation method. LoRA-SB optimally aligns the optimization trajectory with the ideal low-rank full fine-tuning projection by learning a small square matrix (R) between adapters B and A, keeping other components fixed. Direct averaging of R guarantees exact updates, substantially reducing communication cost, which remains independent of the number of clients, and enables scalability. Fed-SB achieves state-of-the-art performance across commonsense reasoning, arithmetic reasoning, and language inference tasks while reducing communication costs by up to 230x. In private settings, Fed-SB further improves performance by (1) reducing trainable parameters, thereby lowering the noise required for differential privacy and (2) avoiding noise amplification introduced by other methods. Overall, Fed-SB offers a state-of-the-art, efficient, and scalable solution for both private and non-private federated fine-tuning. Our code is publicly available at: https://github.com/CERT-Lab/fed-sb.
Authors: Jonathan Light, Wei Cheng, Benjamin Riviere, Wu Yue, Masafumi Oyamada, Mengdi Wang, Yisong Yue, Santiago Paternain, Haifeng Chen
Abstract: Inference scaling methods for LLMs often rely on decomposing problems into steps (or groups of tokens), followed by sampling and selecting the best next steps. However, these steps and their sizes are often predetermined or manually designed based on domain knowledge. We propose dynamic decomposition, a method that adaptively and automatically partitions solution and reasoning traces into manageable steps during inference. By more effectively allocating compute -- particularly through subdividing challenging steps and prioritizing their sampling -- dynamic decomposition significantly improves inference efficiency. Experiments on benchmarks such as APPS, MATH, and LiveCodeBench demonstrate that dynamic decomposition outperforms static approaches, including token-level, sentence-level, and single-step decompositions, reducing the pass@10 error rate by 5.0%, 6.7%, and 10.5% respectively. These findings highlight the potential of dynamic decomposition to improve a wide range of inference scaling techniques.
Authors: Jos\'e Medina, Amnir Hadachi, Paul Honeine, Abdelaziz Bensrhair
Abstract: Deep neural networks (DNNs) have remarkably succeeded in various image processing tasks. However, their large size and computational complexity present significant challenges for deploying them in resource-constrained environments. This paper presents an innovative approach for integrating Mamba Architecture within a Progressive Knowledge Distillation (PKD) process to address the challenge of reducing model complexity while maintaining accuracy in image classification tasks. The proposed framework distills a large teacher model into progressively smaller student models, designed using Mamba blocks. Each student model is trained using Selective-State-Space Models (S-SSM) within the Mamba blocks, focusing on important input aspects while reducing computational complexity. The work's preliminary experiments use MNIST and CIFAR-10 as datasets to demonstrate the effectiveness of this approach. For MNIST, the teacher model achieves 98% accuracy. A set of seven student models as a group retained 63% of the teacher's FLOPs, approximating the teacher's performance with 98% accuracy. The weak student used only 1% of the teacher's FLOPs and maintained 72% accuracy. Similarly, for CIFAR-10, the students achieved 1% less accuracy compared to the teacher, with the small student retaining 5% of the teacher's FLOPs to achieve 50% accuracy. These results confirm the flexibility and scalability of Mamba Architecture, which can be integrated into PKD, succeeding in the process of finding students as weak learners. The framework provides a solution for deploying complex neural networks in real-time applications with a reduction in computational cost.
Authors: Christian Jorgensen, Arthur Y. Lin, Rhushil Vasavada, Rose K. Cersonsky
Abstract: How do classification models "see" our data? Based on their success in delineating behaviors, there must be some lens through which it is easy to see the boundary between classes; however, our current set of visualization techniques makes this prospect difficult. In this work, we propose a hybrid supervised-unsupervised technique distinctly suited to visualizing the decision boundaries determined by classification problems. This method provides a human-interpretable map that can be analyzed qualitatively and quantitatively, which we demonstrate through visualizing and interpreting a decision boundary for chemical neurotoxicity. While we discuss this method in the context of chemistry-driven problems, its application can be generalized across subfields for "unboxing" the operations of machine-learning classification models.
Authors: Beyza Kalkanli, Tales Imbiriba, Stratis Ioannidis, Deniz Erdogmus, Jennifer Dy
Abstract: Active learning aims to efficiently build a labeled training set by strategically selecting samples to query labels from annotators. In this sequential process, each sample acquisition influences subsequent selections, causing dependencies among samples in the labeled set. However, these dependencies are overlooked during the model parameter estimation stage when updating the model using Maximum Likelihood Estimation (MLE), a conventional method that assumes independent and identically distributed (i.i.d.) data. We propose Dependency-aware MLE (DMLE), which corrects MLE within the active learning framework by addressing sample dependencies typically neglected due to the i.i.d. assumption, ensuring consistency with active learning principles in the model parameter estimation process. This improved method achieves superior performance across multiple benchmark datasets, reaching higher performance in earlier cycles compared to conventional MLE. Specifically, we observe average accuracy improvements of 6%, 8.6%, and 10.5% for k=1, k=5, and k=10 respectively, after collecting the first 100 samples, where entropy is the acquisition function and k is the query batch size acquired at every active learning cycle. Our implementation is publicly available at: https://github.com/neu-spiral/DMLEforAL
Authors: Bicheng Ying, Zhe Li, Haibo Yang
Abstract: This work tackles the fundamental challenges in Federated Learning (FL) posed by arbitrary client participation and data heterogeneity, prevalent characteristics in practical FL settings. It is well-established that popular FedAvg-style algorithms struggle with exact convergence and can suffer from slow convergence rates since a decaying learning rate is required to mitigate these scenarios. To address these issues, we introduce the concept of stochastic matrix and the corresponding time-varying graphs as a novel modeling tool to accurately capture the dynamics of arbitrary client participation and the local update procedure. Leveraging this approach, we offer a fresh decentralized perspective on designing FL algorithms and present FOCUS, Federated Optimization with Exact Convergence via Push-pull Strategy, a provably convergent algorithm designed to effectively overcome the previously mentioned two challenges. More specifically, we provide a rigorous proof demonstrating that FOCUS achieves exact convergence with a linear rate regardless of the arbitrary client participation, establishing it as the first work to demonstrate this significant result.
Authors: Zichen Liu, Changyu Chen, Wenjun Li, Penghui Qi, Tianyu Pang, Chao Du, Wee Sun Lee, Min Lin
Abstract: DeepSeek-R1-Zero has shown that reinforcement learning (RL) at scale can directly enhance the reasoning capabilities of LLMs without supervised fine-tuning. In this work, we critically examine R1-Zero-like training by analyzing its two core components: base models and RL. We investigate a wide range of base models, including DeepSeek-V3-Base, to understand how pretraining characteristics influence RL performance. Our analysis reveals that DeepSeek-V3-Base already exhibit ''Aha moment'', while Qwen2.5 base models demonstrate strong reasoning capabilities even without prompt templates, suggesting potential pretraining biases. Additionally, we identify an optimization bias in Group Relative Policy Optimization (GRPO), which artificially increases response length (especially for incorrect outputs) during training. To address this, we introduce Dr. GRPO, an unbiased optimization method that improves token efficiency while maintaining reasoning performance. Leveraging these insights, we present a minimalist R1-Zero recipe that achieves 43.3% accuracy on AIME 2024 with a 7B base model, establishing a new state-of-the-art. Our code is available at https://github.com/sail-sg/understand-r1-zero.
Authors: Mattia Opper, Roland Fernandez, Paul Smolensky, Jianfeng Gao
Abstract: Transformers struggle with length generalisation, displaying poor performance even on basic tasks. We test whether these limitations can be explained through two key failures of the self-attention mechanism. The first is the inability to fully remove irrelevant information. The second is tied to position, even if the dot product between a key and query is highly negative (i.e. an irrelevant key) learned positional biases may unintentionally up-weight such information - dangerous when distances become out of distribution. Put together, these two failure cases lead to compounding generalisation difficulties. We test whether they can be mitigated through the combination of a) selective sparsity - completely removing irrelevant keys from the attention softmax and b) contextualised relative distance - distance is only considered as between the query and the keys that matter. We show how refactoring the attention mechanism with these two mitigations in place can substantially improve the generalisation capabilities of decoder only transformers.
Authors: Kerui Wu, Ziyue Zhao, B\"ulent Yener
Abstract: Electroencephalography (EEG) reflects the brain's functional state, making it a crucial tool for diverse detection applications like seizure detection and sleep stage classification. While deep learning-based approaches have recently shown promise for automated detection, traditional models are often constrained by limited learnable parameters and only achieve modest performance. In contrast, large foundation models showed improved capabilities by scaling up the model size, but required extensive time-consuming pre-training. Moreover, both types of existing methods require complex and redundant post-processing pipelines to convert discrete labels to continuous annotations. In this work, based on the multi-scale nature of EEG events, we propose a simple U-shaped model to efficiently learn representations by capturing both local and global features using convolution and self-attentive modules for sequence-to-sequence modeling. Compared to other window-level classification models, our method directly outputs predictions at the time-step level, eliminating redundant overlapping inferences. Beyond sequence-to-sequence modeling, the architecture naturally extends to window-level classification by incorporating an attention-pooling layer. Such a paradigm shift and model design demonstrated promising efficiency improvement, cross-subject generalization, and state-of-the-art performance in various time-step and window-level classification tasks in the experiment. More impressively, our model showed the capability to be scaled up to the same level as existing large foundation models that have been extensively pre-trained over diverse datasets and outperforms them by solely using the downstream fine-tuning dataset. Our model won 1st place in the 2025 "seizure detection challenge" organized in the International Conference on Artificial Intelligence in Epilepsy and Other Neurological Disorders.
Authors: Xing Han L\`u, Amirhossein Kazemnejad, Nicholas Meade, Arkil Patel, Dongchan Shin, Alejandra Zambrano, Karolina Sta\'nczak, Peter Shaw, Christopher J. Pal, Siva Reddy
Abstract: Web agents enable users to perform tasks on web browsers through natural language interaction. Evaluating web agents trajectories is an important problem, since it helps us determine whether the agent successfully completed the tasks. Rule-based methods are widely used for this purpose, but they are challenging to extend to new tasks and may not always recognize successful trajectories. We may achieve higher accuracy through human evaluation, but the process would be substantially slower and more expensive. Automatic evaluations with LLMs may avoid the challenges of designing new rules and manually annotating trajectories, enabling faster and cost-effective evaluation. However, it is unclear how effective they are at evaluating web agents. To this end, we propose AgentRewardBench, the first benchmark to assess the effectiveness of LLM judges for evaluating web agents. AgentRewardBench contains 1302 trajectories across 5 benchmarks and 4 LLMs. Each trajectory in AgentRewardBench is reviewed by an expert, who answers questions pertaining to the success, side effects, and repetitiveness of the agent. Using our benchmark, we evaluate 12 LLM judges and find that no single LLM excels across all benchmarks. We also find that the rule-based evaluation used by common benchmarks tends to underreport the success rate of web agents, highlighting a key weakness of rule-based evaluation and the need to develop more flexible automatic evaluations. We release the benchmark at: https://agent-reward-bench.github.io
Authors: Xuyang Chen, Guojian Wang, Keyu Yan, Lin Zhao
Abstract: Offline reinforcement learning (RL) learns effective policies from pre-collected datasets, offering a practical solution for applications where online interactions are risky or costly. Model-based approaches are particularly advantageous for offline RL, owing to their data efficiency and generalizability. However, due to inherent model errors, model-based methods often artificially introduce conservatism guided by heuristic uncertainty estimation, which can be unreliable. In this paper, we introduce VIPO, a novel model-based offline RL algorithm that incorporates self-supervised feedback from value estimation to enhance model training. Specifically, the model is learned by additionally minimizing the inconsistency between the value learned directly from the offline data and the one estimated from the model. We perform comprehensive evaluations from multiple perspectives to show that VIPO can learn a highly accurate model efficiently and consistently outperform existing methods. In particular, it achieves state-of-the-art performance on almost all tasks in both D4RL and NeoRL benchmarks. Overall, VIPO offers a general framework that can be readily integrated into existing model-based offline RL algorithms to systematically enhance model accuracy.
Authors: Paul Ghanem, Owen Howell, Michael Potter, Pau Closas, Alireza Ramezani, Deniz Erdogmus, Tales Imbiriba
Abstract: Inferring an adversary's goals from exhibited behavior is crucial for counterplanning and non-cooperative multi-agent systems in domains like cybersecurity, military, and strategy games. Deep Inverse Reinforcement Learning (IRL) methods based on maximum entropy principles show promise in recovering adversaries' goals but are typically offline, require large batch sizes with gradient descent, and rely on first-order updates, limiting their applicability in real-time scenarios. We propose an online Recursive Deep Inverse Reinforcement Learning (RDIRL) approach to recover the cost function governing the adversary actions and goals. Specifically, we minimize an upper bound on the standard Guided Cost Learning (GCL) objective using sequential second-order Newton updates, akin to the Extended Kalman Filter (EKF), leading to a fast (in terms of convergence) learning algorithm. We demonstrate that RDIRL is able to recover cost and reward functions of expert agents in standard and adversarial benchmark tasks. Experiments on benchmark tasks show that our proposed approach outperforms several leading IRL algorithms.
Authors: Xiaoyong Yuan, Xiaolong Ma, Linke Guo, Lan Zhang
Abstract: Diffusion models (DMs) have revolutionized text-to-image generation, enabling the creation of highly realistic and customized images from text prompts. With the rise of parameter-efficient fine-tuning (PEFT) techniques, users can now customize powerful pre-trained models using minimal computational resources. However, the widespread sharing of fine-tuned DMs on open platforms raises growing ethical and legal concerns, as these models may inadvertently or deliberately generate sensitive or unauthorized content. Despite increasing regulatory attention on generative AI, there are currently no practical tools for systematically auditing these models before deployment. In this paper, we address the problem of concept auditing: determining whether a fine-tuned DM has learned to generate a specific target concept. Existing approaches typically rely on prompt-based input crafting and output-based image classification but they suffer from critical limitations, including prompt uncertainty, concept drift, and poor scalability. To overcome these challenges, we introduce Prompt-Agnostic Image-Free Auditing (PAIA), a novel, model-centric concept auditing framework. By treating the DM as the object of inspection, PAIA enables direct analysis of internal model behavior, bypassing the need for optimized prompts or generated images. We evaluate PAIA on 320 controlled models trained with curated concept datasets and 771 real-world community models sourced from a public DM sharing platform. Evaluation results show that PAIA achieves over 90% detection accuracy while reducing auditing time by 18 - 40X compared to existing baselines. To our knowledge, PAIA is the first scalable and practical solution for pre-deployment concept auditing of diffusion models, providing a practical foundation for safer and more transparent diffusion model sharing.
Authors: Alessandro Trenta, Andrea Cossu, Davide Bacciu
Abstract: We propose Derivative Learning (DERL), a supervised approach that models physical systems by learning their partial derivatives. We also leverage DERL to build physical models incrementally, by designing a distillation protocol that effectively transfers knowledge from a pre-trained model to a student one. We provide theoretical guarantees that DERL can learn the true physical system, being consistent with the underlying physical laws, even when using empirical derivatives. DERL outperforms state-of-the-art methods in generalizing an ODE to unseen initial conditions and a parametric PDE to unseen parameters. We also design a method based on DERL to transfer physical knowledge across models by extending them to new portions of the physical domain and a new range of PDE parameters. We believe this is the first attempt at building physical models incrementally in multiple stages.
Authors: Xuyang Chen, Keyu Yan, Wenhan Cao, Lin Zhao
Abstract: Offline reinforcement learning (RL) learns policies from fixed datasets without online interactions, but suffers from distribution shift, causing inaccurate evaluation and overestimation of out-of-distribution (OOD) actions. Existing methods counter this by conservatively discouraging all OOD actions, which limits generalization. We propose Advantage-based Diffusion Actor-Critic (ADAC), which evaluates OOD actions via an advantage-like function and uses it to modulate the Q-function update discriminatively. Our key insight is that the (state) value function is generally learned more reliably than the action-value function; we thus use the next-state value to indirectly assess each action. We develop a PointMaze environment to clearly visualize that advantage modulation effectively selects superior OOD actions while discouraging inferior ones. Moreover, extensive experiments on the D4RL benchmark show that ADAC achieves state-of-the-art performance, with especially strong gains on challenging tasks.
Authors: Tung L Nguyen, Toby Hocking
Abstract: Changepoint detection identifies significant shifts in data sequences, making it important in areas like finance, genetics, and healthcare. The Optimal Partitioning algorithms efficiently detect these changes, using a penalty parameter to limit the changepoints count. Determining the optimal value for this penalty can be challenging. Traditionally, this process involved manually extracting statistical features, such as sequence length or variance to make the prediction. This study proposes a novel approach that uses recurrent networks to learn this penalty directly from raw sequences by automatically extracting features. Experiments conducted on 20 benchmark genomic datasets show that this novel method generally outperforms traditional ones in changepoint detection accuracy.
Authors: Takashi Nicholas Maeda, Shohei Shimizu, Hidetoshi Matsui
Abstract: We propose a causal discovery method for mixed bivariate data consisting of one continuous and one discrete variable. Existing approaches either impose strong distributional assumptions or face challenges in fairly comparing causal directions between variables of different types, due to differences in their information content. We introduce a novel approach that determines causal direction by analyzing the monotonicity of the conditional density ratio of the continuous variable, conditioned on different values of the discrete variable. Our theoretical analysis shows that the conditional density ratio exhibits monotonicity when the continuous variable causes the discrete variable, but not in the reverse direction. This property provides a principled basis for comparing causal directions between variables of different types, free from strong distributional assumptions and bias arising from differences in their information content. We demonstrate its effectiveness through experiments on both synthetic and real-world datasets, showing superior accuracy compared to existing methods.
Authors: Asher Labovich
Abstract: Modern Gradient Boosted Decision Trees (GBDTs) accelerate split finding with histogram-based binning, which reduces complexity from $O(N\log N)$ to $O(N)$ by aggregating gradients into fixed-size bins. However, the predominant quantile binning strategy - designed to distribute data points evenly among bins -- may overlook critical boundary values that could enhance predictive performance. In this work, we consider a novel approach that replaces quantile binning with a $k$-means discretizer initialized with quantile bins, and justify the swap with a proof showing how, for any $L$-Lipschitz function, k-means maximizes the worst-case explained variance of Y obtained when treating all values in a given bin as equivalent. We test this swap against quantile and uniform binning on 33 OpenML datasets plus synthetics that control for modality, skew, and bin budget. Across 18 regression datasets, k-means shows no statistically significant losses at the 5% level and wins in three cases-most strikingly a 55% MSE drop on one particularly skewed dataset-even though k-means' mean reciprocal rank (MRR) is slightly lower (0.65 vs 0.72). On the 15 classification datasets the two methods are statistically tied (MRR 0.70 vs 0.68) with gaps $\leq$0.2 pp. Synthetic experiments confirm consistently large MSE gains - typically >20% and rising to 90% as outlier magnitude increases or bin budget drops. We find that k-means keeps error on par with exhaustive (no-binning) splitting when extra cuts add little value, yet still recovers key split points that quantile overlooks. As such, we advocate for a built-in bin_method=k-means flag, especially in regression tasks and in tight-budget settings such as the 32-64-bin GPU regime - because it is a "safe default" with large upside, yet adds only a one-off, cacheable overhead ($\approx$ 3.5s per feature to bin 10M rows on one Apple M1 thread).
Authors: Kaustubh Ponkshe, Shaan Shah, Raghav Singhal, Praneeth Vepakomma
Abstract: Large Language Models (LLMs) rely on safety alignment to produce socially acceptable responses. However, this behavior is known to be brittle: further fine-tuning, even on benign or lightly contaminated data, can degrade safety and reintroduce harmful behaviors. A growing body of work suggests that alignment may correspond to identifiable directions in weight space, forming subspaces that could, in principle, be isolated or preserved to defend against misalignment. In this work, we conduct a comprehensive empirical study of this perspective. We examine whether safety-relevant behavior is concentrated in specific linear subspaces, whether it can be separated from general-purpose learning, and whether harmfulness arises from distinguishable patterns in activations. Across both weight and activation spaces, our findings are consistent: subspaces that amplify safe behaviors also amplify useful ones, and prompts with different safety implications activate overlapping representations. Rather than residing in distinct directions, we show that safety is highly entangled with the general learning components of the model. This suggests that subspace-based defenses face fundamental limitations and underscores the need for alternative strategies to preserve safety under continued training. We corroborate these findings with multiple experiments on five open-source LLMs from the Llama and Qwen families. Our code is publicly available at: https://github.com/CERT-Lab/safety-subspaces.
Authors: Ichiro Hashimoto
Abstract: In this paper, we study benign overfitting of fixed width leaky ReLU two-layer neural network classifiers trained on mixture data via gradient descent. We provide both, upper and lower classification error bounds, and discover a phase transition in the bound as a function of signal strength. The lower bound leads to a characterization of cases when benign overfitting provably fails even if directional convergence occurs. Our analysis allows us to considerably relax the distributional assumptions that are made in existing work on benign overfitting of leaky ReLU two-layer neural network classifiers. We can allow for non-sub-Gaussian data and do not require near orthogonality. Our results are derived by establishing directional convergence of the network parameters and studying classification error bounds for the convergent direction. Previously, directional convergence in (leaky) ReLU neural networks was established only for gradient flow. By first establishing directional convergence, we are able to study benign overfitting of fixed width leaky ReLU two-layer neural network classifiers in a much wider range of scenarios than was done before.
Authors: Woosung Koh, Wonbeen Oh, Jaein Jang, MinHyung Lee, Hyeongjin Kim, Ah Yeon Kim, Joonkee Kim, Junghyun Lee, Taehyeon Kim, Se-Young Yun
Abstract: Self-Taught Reasoners (STaR), synonymously known as Rejection sampling Fine-Tuning (RFT), is an integral part of the training pipeline of self-improving reasoning Language Models (LMs). The self-improving mechanism often employs random observation (data) sampling. However, this results in trained observation imbalance; inefficiently over-training on solved examples while under-training on challenging ones. In response, we introduce Adaptive STaR (AdaSTaR), a novel algorithm that rectifies this by integrating two adaptive sampling principles: (1) Adaptive Sampling for Diversity: promoting balanced training across observations, and (2) Adaptive Sampling for Curriculum: dynamically adjusting data difficulty to match the model's evolving strength. Across six benchmarks, AdaSTaR achieves best test accuracy in all instances (6/6) and reduces training FLOPs by an average of 58.6% against an extensive list of baselines. These improvements in performance and efficiency generalize to different pre-trained LMs and larger models, paving the way for more efficient and effective self-improving LMs.
Authors: Majid Mohammadi, Siu Lun Chau, Krikamol Muandet
Abstract: Kernel methods are widely used in machine learning due to their flexibility and expressiveness. However, their black-box nature poses significant challenges to interpretability, limiting their adoption in high-stakes applications. Shapley value-based feature attribution techniques, such as SHAP and kernel method-specific adaptation like RKHS-SHAP, offer a promising path toward explainability. Yet, computing exact Shapley values is generally intractable, leading existing methods to rely on approximations and thereby incur unavoidable error. In this work, we introduce PKeX-Shapley, a novel algorithm that utilizes the multiplicative structure of product kernels to enable the exact computation of Shapley values in polynomial time. The core of our approach is a new value function, the functional baseline value function, specifically designed for product-kernel models. This value function removes the influence of a feature subset by setting its functional component to the least informative state. Crucially, it allows a recursive thus efficient computation of Shapley values in polynomial time. As an important additional contribution, we show that our framework extends beyond predictive modeling to statistical inference. In particular, it generalizes to popular kernel-based discrepancy measures such as the Maximum Mean Discrepancy (MMD) and the Hilbert-Schmidt Independence Criterion (HSIC), thereby providing new tools for interpretable statistical inference.
Authors: Trung V. Phan, George A. Kevrekidis, Soledad Villar, Yannis G. Kevrekidis, Juan M. Bello-Rivas
Abstract: The machine learning methods for data-driven identification of partial differential equations (PDEs) are typically defined for a given number of spatial dimensions and a choice of coordinates the data have been collected in. This dependence prevents the learned evolution equation from generalizing to other spaces. In this work, we reformulate the problem in terms of coordinate- and dimension-independent representations, paving the way toward what we call ``spatially liberated" PDE learning. To this end, we employ a machine learning approach to predict the evolution of scalar field systems expressed in the formalism of exterior calculus, which is coordinate-free and immediately generalizes to arbitrary dimensions by construction. We demonstrate the performance of this approach in the FitzHugh-Nagumo and Barkley reaction-diffusion models, as well as the Patlak-Keller-Segel model informed by in-situ chemotactic bacteria observations. We provide extensive numerical experiments that demonstrate that our approach allows for seamless transitions across various spatial contexts. We show that the field dynamics learned in one space can be used to make accurate predictions in other spaces with different dimensions, coordinate systems, boundary conditions, and curvatures.
Authors: Konstantinos Barmpas, Na Lee, Yannis Panagakis, Dimitrios A. Adamos, Nikolaos Laskaris, Stefanos Zafeiriou
Abstract: Recent advances in large-scale pre-trained Electroencephalogram (EEG) models have shown great promise, driving progress in Brain-Computer Interfaces (BCIs) and healthcare applications. However, despite their success, many existing pre-trained models have struggled to fully capture the rich information content of neural oscillations, a limitation that fundamentally constrains their performance and generalizability across diverse BCI tasks. This limitation is frequently rooted in suboptimal architectural design choices which constrain their representational capacity. In this work, we introduce LaBraM++, an enhanced Large Brainwave Foundation Model (LBM) that incorporates principled improvements grounded in robust signal processing foundations. LaBraM++ demonstrates substantial gains across a variety of tasks, consistently outperforming its originally-based architecture and achieving competitive results when compared to other open-source LBMs. Its superior performance and training efficiency highlight its potential as a strong foundation for future advancements in LBMs.
Authors: Jaewon Cheon, Pilsung Kang
Abstract: The growing size of large language models has created significant computational inefficiencies. To address this challenge, sparse activation methods selectively deactivates non-essential parameters during inference, reducing computational costs in FFNN layers. While existing methods focus on non-linear gating mechanisms, we hypothesize that the sparsity of the FFNN layer lies globally in the form of a linear combination over its internal down projection matrix. Based on this insight, we propose two methods: M-COUNTDOWN, leveraging indirect coefficients, and D-COUNTDOWN, utilizing direct coefficients of the linear combination. Experimental results demonstrate that D-COUNTDOWN can omit 90% of computations with performance loss as low as 5.5% ideally, while M-COUNTDOWN provides a predictor-free solution with up to 29.4% better performance preservation compared to existing methods. Our specialized kernel implementations effectively realize these theoretical gains into substantial real-world acceleration.
Authors: Mudit Gaur, Prashant Trivedi, Sasidhar Kunapuli, Amrit Singh Bedi, Vaneet Aggarwal
Abstract: Diffusion models have demonstrated remarkable performance in generating high-dimensional samples across domains such as vision, language, and the sciences. Although continuous-state diffusion models have been extensively studied both empirically and theoretically, discrete-state diffusion models, essential for applications involving text, sequences, and combinatorial structures, they remain significantly less understood from a theoretical standpoint. In particular, all existing analyses of discrete-state models assume access to an empirical risk minimizer. In this work, we present a principled theoretical framework analyzing diffusion models, providing a state-of-the-art sample complexity bound of $\widetilde{\mathcal{O}}(\epsilon^{-4})$. Our structured decomposition of the score estimation error into statistical and optimization components offers critical insights into how diffusion models can be trained efficiently. This analysis addresses a fundamental gap in the literature and establishes the theoretical tractability and practical relevance of diffusion models.
Authors: Waleed Reda, Abhinav Jangda, Krishna Chintalapudi
Abstract: As Large Language Models (LLMs) are increasingly deployed for narrow tasks in resource-constrained settings, a central question arises: how much of an LLM is truly necessary for a given task? We present LLM-Sieve, a framework that prunes LLMs down to the minimal parameter subset needed to preserve task performance. Our approach introduces two innovations: (i) output-aligned non-orthogonal projections, which yield more faithful low-rank approximations than traditional PCA/SVD by aligning directly with layer outputs; and (ii) adaptive pruning via a Genetic Algorithm, which automatically discovers matrix-specific pruning levels and exposes the uneven distribution of task-relevant knowledge. Across models from 3.8B to 70B parameters, LLM-Sieve removes 20-75% of weights with only 1-5% accuracy loss-substantially ahead of prior pruning methods. Beyond efficiency, our framework reveals bottleneck matrices that concentrate critical knowledge, suggesting architectural implications for future LLM design. LLM-Sieve integrates seamlessly with LoRA fine-tuning and quantization, enabling both efficient deployment and deeper understanding of knowledge organization in LLMs.
Authors: Nicholas M. Boffi, Michael S. Albergo, Eric Vanden-Eijnden
Abstract: Flow-based generative models achieve state-of-the-art sample quality, but require the expensive solution of a differential equation at inference time. Flow map models, commonly known as consistency models, encompass many recent efforts to improve inference-time efficiency by learning the solution operator of this differential equation. Yet despite their promise, these models lack a unified description that clearly explains how to learn them efficiently in practice. Here, building on the methodology proposed in Boffi et. al. (2024), we present a systematic algorithmic framework for directly learning the flow map associated with a flow or diffusion model. By exploiting a relationship between the velocity field underlying a continuous-time flow and the instantaneous rate of change of the flow map, we show how to convert any distillation scheme into a direct training algorithm via self-distillation, eliminating the need for pre-trained teachers. We introduce three algorithmic families based on different mathematical characterizations of the flow map: Eulerian, Lagrangian, and Progressive methods, which we show encompass and extend all known distillation and direct training schemes for consistency models. We find that the novel class of Lagrangian methods, which avoid both spatial derivatives and bootstrapping from small steps by design, achieve significantly more stable training and higher performance than more standard Eulerian and Progressive schemes. Our methodology unifies existing training schemes under a single common framework and reveals new design principles for accelerated generative modeling. Associated code is available at https://github.com/nmboffi/flow-maps.
Authors: Zhepeng Cen, Yihang Yao, William Han, Zuxin Liu, Ding Zhao
Abstract: Reinforcement learning (RL) has emerged as a powerful post-training technique to incentivize the reasoning ability of large language models (LLMs). However, LLMs can respond very inconsistently to RL finetuning: some show substantial performance gains, while others plateau or even degrade. To understand this divergence, we analyze the per-step influence of the RL objective and identify two key conditions for effective post-training: (1) RL-informative rollout accuracy, and (2) strong data co-influence, which quantifies how much the training data affects performance on other samples. Guided by these insights, we propose behavior injection, a task-agnostic data augmentation scheme applied prior to RL. Behavior injection enriches the supervised finetuning (SFT) data by seeding exploratory and exploitative behaviors, effectively making the model more RL-ready. We evaluate our method across two reasoning benchmarks with multiple base models. The results demonstrate that our theoretically motivated augmentation can significantly increase the performance gain from RL over the pre-RL model.
Authors: Yuzheng Hu, Fan Wu, Haotian Ye, David Forsyth, James Zou, Nan Jiang, Jiaqi W. Ma, Han Zhao
Abstract: Online reinforcement learning (RL) excels in complex, safety-critical domains but suffers from sample inefficiency, training instability, and limited interpretability. Data attribution provides a principled way to trace model behavior back to training samples, yet existing methods assume fixed datasets, which is violated in online RL where each experience both updates the policy and shapes future data collection. In this paper, we initiate the study of data attribution for online RL, focusing on the widely used Proximal Policy Optimization (PPO) algorithm. We start by establishing a \emph{local} attribution framework, interpreting model checkpoints with respect to the records in the recent training buffer. We design two target functions, capturing agent action and cumulative return respectively, and measure each record's contribution through gradient similarity between its training loss and these targets. We demonstrate the power of this framework through three concrete applications: diagnosis of learning, temporal analysis of behavior formation, and targeted intervention during training. Leveraging this framework, we further propose an algorithm, iterative influence-based filtering (IIF), for online RL training that iteratively performs experience filtering to refine policy updates. Across standard RL benchmarks (classic control, navigation, locomotion) to RLHF for large language models, IIF reduces sample complexity, speeds up training, and achieves higher returns. Together, these results open a new direction for making online RL more interpretable, efficient, and effective.
Authors: Zongle Huang, Lei Zhu, Zongyuan Zhan, Ting Hu, Weikai Mao, Xianzhi Yu, Yongpan Liu, Tianyu Zhang
Abstract: Large Language Models (LLMs) have achieved remarkable success across many applications, with Mixture of Experts (MoE) models demonstrating great potential. Compared to traditional dense models, MoEs achieve better performance with less computation. Speculative decoding (SD) is a widely used technique to accelerate LLM inference without accuracy loss, but it has been considered efficient only for dense models. In this work, we first demonstrate that, under medium batch sizes, MoE surprisingly benefits more from SD than dense models. Furthermore, as MoE becomes sparser -- the prevailing trend in MoE designs -- the batch size range where SD acceleration is expected to be effective becomes broader. To quantitatively understand tradeoffs involved in SD, we develop a reliable modeling based on theoretical analyses. While current SD research primarily focuses on improving acceptance rates of algorithms, changes in workload and model architecture can still lead to degraded SD acceleration even with high acceptance rates. To address this limitation, we introduce a new metric 'target efficiency' that characterizes these effects, thus helping researchers identify system bottlenecks and understand SD acceleration more comprehensively. For scenarios like private serving, this work unveils a new perspective to speed up MoE inference, where existing solutions struggle. Experiments on different GPUs show up to 2.29x speedup for Qwen2-57B-A14B at medium batch sizes and validate our theoretical predictions.
Authors: Anji Liu, Zilei Shao, Guy Van den Broeck
Abstract: Probabilistic Circuits (PCs) offer a computationally scalable framework for generative modeling, supporting exact and efficient inference of a wide range of probabilistic queries. While recent advances have significantly improved the expressiveness and scalability of PCs, effectively training their parameters remains a challenge. In particular, a widely used optimization method, full-batch Expectation-Maximization (EM), requires processing the entire dataset before performing a single update, making it ineffective for large datasets. Although empirical extensions to the mini-batch setting, as well as gradient-based mini-batch algorithms, converge faster than full-batch EM, they generally underperform in terms of final likelihood. We investigate this gap by establishing a novel theoretical connection between these practical algorithms and the general EM objective. Our analysis reveals a fundamental issue that existing mini-batch EM and gradient-based methods fail to properly regularize distribution changes, causing each update to effectively ``overfit'' the current mini-batch. Motivated by this insight, we introduce anemone, a new mini-batch EM algorithm for PCs. Anemone applies an implicit adaptive learning rate to each parameter, scaled by how much it contributes to the likelihood of the current batch. Across extensive experiments on language, image, and DNA datasets, anemone consistently outperforms existing optimizers in both convergence speed and final performance.
Authors: Chenxiang Zhang, Jun Pang, Sjouke Mauw
Abstract: Neural networks trained on real-world data often exhibit biases while simultaneously being vulnerable to privacy attacks aimed at extracting sensitive information. Despite extensive research on each problem individually, their intersection remains poorly understood. In this work, we investigate the privacy impact of spurious correlation bias. We introduce \emph{spurious privacy leakage}, a phenomenon in which spurious groups are significantly more vulnerable to privacy attacks than non-spurious groups. We observe that privacy disparity between groups increases in tasks with simpler objectives (e.g. fewer classes) due to spurious features. Counterintuitively, we demonstrate that spurious robust methods, designed to reduce spurious bias, fail to mitigate privacy disparity. Our analysis reveals that this occurs because robust methods can reduce reliance on spurious features for prediction, but do not prevent their memorization during training. Finally, we systematically compare the privacy of different model architectures trained with spurious data, demonstrating that, contrary to previous work, architectural choice can affect privacy evaluation.
Authors: Abhishek Vijaya Kumar, Eric Ding, Arjun Devraj, Darius Bunandar, Rachee Singh
Abstract: When accelerators fail in modern ML datacenters, operators migrate the affected ML training or inference jobs to entirely new racks. This approach, while preserving network performance, is highly inefficient, requiring datacenters to reserve full racks of idle accelerators for fault tolerance. In this paper, we address this resource inefficiency by introducing LUMION, a novel reconfigurable optical fabric for connecting accelerators within a datacenter rack. Instead of migrating entire ML jobs, LUMION dynamically integrates spare accelerators into ongoing workloads as failures occur, thereby maintaining consistent performance without costly migrations. We show the benefits of LUMION by building an end-to-end hardware prototype. Our experiments fine-tune Llama 3.2 and show that LUMION swaps a failed GPU with a healthy one and restarts the ML job within ~ 1 second of the failure. LUMION achieves higher inter-GPU bandwidth compared to traditional electrical racks after replacing failed accelerators with spare ones, leading to nearly 2X improvement in fine-tuning throughput.
Authors: Xiangcheng Zhang, Haowei Lin, Haotian Ye, James Zou, Jianzhu Ma, Yitao Liang, Yilun Du
Abstract: Classical search algorithms have long underpinned modern artificial intelligence. In this work, we tackle the challenge of inference-time control in diffusion models -- adapting generated outputs to meet diverse test-time objectives -- using principles from classical search. We propose a general framework that orchestrates local and global search to efficiently navigate the generative space. It employs a theoretically grounded local search via annealed Langevin MCMC and performs compute-efficient global exploration using breadth-first and depth-first tree search. We evaluate our approach on a range of challenging domains, including planning, offline reinforcement learning, and image generation. Across all tasks, we observe significant gains in both performance and efficiency. These results show that classical search provides a principled and practical foundation for inference-time scaling in diffusion models. Project page at https://diffusion-inference-scaling.github.io/.
Authors: Xiaoyu Wu, Yifei Pang, Terrance Liu, Zhiwei Steven Wu
Abstract: Large Language Models are typically trained on datasets collected from the web, which may inadvertently contain harmful or sensitive personal information. To address growing privacy concerns, unlearning methods have been proposed to remove the influence of specific data from trained models. Of these, exact unlearning -- which retrains the model from scratch without the target data -- is widely regarded the gold standard for mitigating privacy risks in deployment. In this paper, we revisit this assumption in a practical deployment setting where both the pre- and post-unlearning logits API are exposed, such as in open-weight scenarios. Targeting this setting, we introduce a novel data extraction attack that leverages signals from the pre-unlearning model to guide the post-unlearning model, uncovering patterns that reflect the removed data distribution. Combining model guidance with a token filtering strategy, our attack significantly improves extraction success rates -- doubling performance in some cases -- across common benchmarks such as MUSE, TOFU, and WMDP. Furthermore, we demonstrate our attack's effectiveness on a simulated medical diagnosis dataset to highlight real-world privacy risks associated with exact unlearning. In light of our findings, which suggest that unlearning may, in a contradictory way, increase the risk of privacy leakage during real-world deployments, we advocate for evaluation of unlearning methods to consider broader threat models that account not only for post-unlearning models but also for adversarial access to prior checkpoints. Code is publicly available at: https://github.com/Nicholas0228/unlearned_data_extraction_llm.
URLs: https://github.com/Nicholas0228/unlearned_data_extraction_llm.
Authors: Harsh Chaudhari, Jamie Hayes, Matthew Jagielski, Ilia Shumailov, Milad Nasr, Alina Oprea
Abstract: Model distillation has become essential for creating smaller, deployable language models that retain larger system capabilities. However, widespread deployment raises concerns about resilience to adversarial manipulation. This paper investigates vulnerability of distilled models to adversarial injection of biased content during training. We demonstrate that adversaries can inject subtle biases into teacher models through minimal data poisoning, which propagates to student models and becomes significantly amplified. We propose two propagation modes: Untargeted Propagation, where bias affects multiple tasks, and Targeted Propagation, focusing on specific tasks while maintaining normal behavior elsewhere. With only 25 poisoned samples (0.25% poisoning rate), student models generate biased responses 76.9% of the time in targeted scenarios - higher than 69.4% in teacher models. For untargeted propagation, adversarial bias appears 6x-29x more frequently in student models on unseen tasks. We validate findings across six bias types (targeted advertisements, phishing links, narrative manipulations, insecure coding practices), various distillation methods, and different modalities spanning text and code generation. Our evaluation reveals shortcomings in current defenses - perplexity filtering, bias detection systems, and LLM-based autorater frameworks - against these attacks. Results expose significant security vulnerabilities in distilled models, highlighting need for specialized safeguards. We propose practical design principles for building effective adversarial bias mitigation strategies.
Authors: Yutong Huang, Zhiyuan Guo, Yiying Zhang
Abstract: Memory prefetching has long boosted CPU caches and is increasingly vital for far-memory systems, where large portions of memory are offloaded to cheaper, remote tiers. While effective prefetching requires accurate prediction of future accesses, prior ML approaches have been limited to simulation or small-scale hardware. We introduce FarSight, the first Linux-based far-memory system to leverage deep learning by decoupling application semantics from runtime memory layout. This separation enables offline-trained models to predict access patterns over a compact ordinal vocabulary, which are resolved at runtime through lightweight mappings. Across four data-intensive workloads, FarSight delivers up to 3.6x higher performance than the state-of-the-art.
Authors: Alessio Russo, Ryan Welch, Aldo Pacchiano
Abstract: We study the problem active sequential hypothesis testing, also known as pure exploration: given a new task, the learner adaptively collects data from the environment to efficiently determine an underlying correct hypothesis. A classical instance of this problem is the task of identifying the best arm in a multi-armed bandit problem (a.k.a. BAI, Best-Arm Identification), where actions index hypotheses. Another important case is generalized search, a problem of determining the correct label through a sequence of strategically selected queries that indirectly reveal information about the label. In this work, we introduce In-Context Pure Exploration (ICPE), which meta-trains Transformers to map observation histories to query actions and a predicted hypothesis, yielding a model that transfers in-context. At inference time, ICPE actively gathers evidence on new tasks and infers the true hypothesis without parameter updates. Across deterministic, stochastic, and structured benchmarks, including BAI and generalized search, ICPE is competitive with adaptive baselines while requiring no explicit modeling of information structure. Our results support Transformers as practical architectures for general sequential testing.
Authors: Zeng Wang, Minghao Shao, Rupesh Karn, Likhitha Mankali, Jitendra Bhandari, Ramesh Karri, Ozgur Sinanoglu, Muhammad Shafique, Johann Knechtel
Abstract: Large Language Models (LLMs) offer transformative capabilities for hardware design automation, particularly in Verilog code generation. However, they also pose significant data security challenges, including Verilog evaluation data contamination, intellectual property (IP) design leakage, and the risk of malicious Verilog generation. We introduce SALAD, a comprehensive assessment that leverages machine unlearning to mitigate these threats. Our approach enables the selective removal of contaminated benchmarks, sensitive IP and design artifacts, or malicious code patterns from pre-trained LLMs, all without requiring full retraining. Through detailed case studies, we demonstrate how machine unlearning techniques effectively reduce data security risks in LLM-aided hardware design.
Authors: Viet Nguyen, Changjian Shui, Vijay Giri, Siddarth Arya, Michael Cooper, Amol Verma, Fahad Razak, Rahul G. Krishnan
Abstract: The distribution of data changes over time; models operating operating in dynamic environments need retraining. But knowing when to retrain, without access to labels, is an open challenge since some, but not all shifts degrade model performance. This paper formalizes and addresses the problem of post-deployment deterioration (PDD) monitoring. We propose D3M, a practical and efficient monitoring algorithm based on the disagreement of predictive models, achieving low false positive rates under non-deteriorating shifts and provides sample complexity bounds for high true positive rates under deteriorating shifts. Empirical results on both standard benchmark and a real-world large-scale internal medicine dataset demonstrate the effectiveness of the framework and highlight its viability as an alert mechanism for high-stakes machine learning pipelines.
Authors: Mickel Liu, Liwei Jiang, Yancheng Liang, Simon Shaolei Du, Yejin Choi, Tim Althoff, Natasha Jaques
Abstract: Conventional language model (LM) safety alignment relies on a reactive, disjoint procedure: attackers exploit a static model, followed by defensive fine-tuning to patch exposed vulnerabilities. This sequential approach creates a mismatch -- attackers overfit to obsolete defenses, while defenders perpetually lag behind emerging threats. To address this, we propose Self-RedTeam, an online self-play reinforcement learning algorithm where an attacker and defender agent co-evolve through continuous interaction. We cast safety alignment as a two-player zero-sum game, where a single model alternates between attacker and defender roles -- generating adversarial prompts and safeguarding against them -- while a reward LM adjudicates outcomes. This enables dynamic co-adaptation. Grounded in the game-theoretic framework of zero-sum games, we establish a theoretical safety guarantee which motivates the design of our method: if self-play converges to a Nash Equilibrium, the defender will reliably produce safe responses to any adversarial input. Empirically, Self-RedTeam uncovers more diverse attacks (+21.8% SBERT) compared to attackers trained against static defenders and achieves higher robustness on safety benchmarks (e.g., +65.5% on WildJailBreak) than defenders trained against static attackers. We further propose hidden Chain-of-Thought, allowing agents to plan privately, which boosts adversarial diversity and reduces over-refusals. Our results motivate a shift from reactive patching to proactive co-evolution in LM safety training, enabling scalable, autonomous, and robust self-improvement of LMs via multi-agent reinforcement learning (MARL).
Authors: Yanlong Chen, Mattia Orlandi, Pierangelo Maria Rapa, Simone Benatti, Luca Benini, Yawei Li
Abstract: Physiological signals are often corrupted by motion artifacts, baseline drift, and other low-SNR disturbances, which pose significant challenges for analysis. Additionally, these signals exhibit strong non-stationarity, with sharp peaks and abrupt changes that evolve continuously, making them difficult to represent using traditional time-domain or filtering methods. To address these issues, a novel wavelet-based approach for physiological signal analysis is presented, aiming to capture multi-scale time-frequency features in various physiological signals. Leveraging this technique, two large-scale pretrained models specific to EMG and ECG are introduced for the first time, achieving superior performance and setting new baselines in downstream tasks. Additionally, a unified multi-modal framework is constructed by integrating pretrained EEG model, where each modality is guided through its dedicated branch and fused via learnable weighted fusion. This design effectively addresses challenges such as low signal-to-noise ratio, high inter-subject variability, and device mismatch, outperforming existing methods on multi-modal tasks. The proposed wavelet-based architecture lays a solid foundation for analysis of diverse physiological signals, while the multi-modal design points to next-generation physiological signal processing with potential impact on wearable health monitoring, clinical diagnostics, and broader biomedical applications. Code and data are available at: github.com/ForeverBlue816/PhysioWave
Authors: Pengrun Huang, Chhavi Yadav, Kamalika Chaudhuri, Ruihan Wu
Abstract: Large language models (LLMs) are increasingly fine-tuned on domain-specific datasets to support applications in fields such as healthcare, finance, and law. These fine-tuning datasets often have sensitive and confidential dataset-level properties -- such as patient demographics or disease prevalence -- that are not intended to be revealed. While prior work has studied property inference attacks on discriminative models (e.g., image classification models) and generative models (e.g., GANs for image data), it remains unclear if such attacks transfer to LLMs. In this work, we introduce PropInfer, a benchmark task for evaluating property inference in LLMs under two fine-tuning paradigms: question-answering and chat-completion. Built on the ChatDoctor dataset, our benchmark includes a range of property types and task configurations. We further propose two tailored attacks: a prompt-based generation attack and a shadow-model attack leveraging word frequency signals. Empirical evaluations across multiple pretrained LLMs show the success of our attacks, revealing a previously unrecognized vulnerability in LLMs.
Authors: Shaorong Zhang, Rob Brekelmans, Yunshu Wu, Greg Ver Steeg
Abstract: Diffusion models provide a powerful way to incorporate complex prior information for solving inverse problems. However, existing methods struggle to correctly incorporate guidance from conflicting signals in the prior and measurement, and often failed to maximizing the consistency to the measurement, especially in the challenging setting of non-Gaussian or unknown noise. To address these issues, we propose Measurement-Aligned Sampling (MAS), a novel framework for linear inverse problem solving that flexibly balances prior and measurement information. MAS unifies and extends existing approaches such as DDNM, TMPD, while generalizing to handle both known Gaussian noise and unknown or non-Gaussian noise types. Extensive experiments demonstrate that MAS consistently outperforms state-of-the-art methods across a variety of tasks, while maintaining relatively low computational cost.
Authors: Eurico Covas
Abstract: Here we present a series of artificial models - a total of four related models - based on machine learning techniques that attempt to learn from existing exhibitions which have been curated by human experts, in order to be able to do similar curatorship work. Out of our four artificial intelligence models, three achieve a reasonable ability at imitating these various curators responsible for all those exhibitions, with various degrees of precision and curatorial coherence. In particular, we can conclude two key insights: first, that there is sufficient information in these exhibitions to construct an artificial intelligence model that replicates past exhibitions with an accuracy well above random choices; and second, that using feature engineering and carefully designing the architecture of modest size models can make them almost as good as those using the so-called large language models such as GPT in a brute force approach.
Authors: Ziyue Li, Chenrui Fan, Tianyi Zhou
Abstract: This paper presents the first study of grokking in practical LLM pretraining. Specifically, we investigate when an LLM memorizes the training data, when its generalization on downstream tasks starts to improve, and what happens if there is a lag between the two. Unlike existing works studying when a small model generalizes to limited and specified tasks during thousands epochs' training on algorithmic data, we focus on a practical setting for LLMs, i.e., one-epoch pretraining of next-token prediction on a cross-domain, large-scale corpus, and generalization on diverse benchmark tasks covering math/commonsense reasoning, code generation, and domain-specific retrieval. Our study, for the first time, verifies that grokking still emerges in pretraining mixture-of-experts (MoE) LLMs, though different local data groups may enter their grokking stages asynchronously due to the heterogeneity of their distributions and attributions to others. To find a mechanistic interpretation of this local grokking, we investigate the dynamics of training data's pathways (i.e., expert choices across layers in MoE). Our primary discovery is that the pathways evolve from random, non-smooth across layers, instance-specific to more structured and transferable across samples, despite the converged pretraining loss. This depicts a transition from memorization to generalization. Two novel metrics are developed to quantify these patterns: one computes the pathway similarity between samples, while the other measures the consistency of aggregated experts between subsequent layers for each sample. These training data based metrics induce zero cost but can faithfully track and monitor the generalization of LLMs on downstream tasks, which, in conventional settings, requires costly instruction tuning and benchmark evaluation.
Authors: Shirley Wu, Parth Sarthi, Shiyu Zhao, Aaron Lee, Herumb Shandilya, Adrian Mladenic Grobelnik, Nurendra Choudhary, Eddie Huang, Karthik Subbian, Linjun Zhang, Diyi Yang, James Zou, Jure Leskovec
Abstract: Compound AI systems integrating multiple components, such as Large Language Models, specialized tools, and traditional machine learning models, are increasingly deployed to solve complex real-world tasks. However, optimizing compound systems remains challenging due to their non-differentiable structures and diverse configuration types across components, including prompts, hyperparameters, and model parameters. To address this challenge, we propose Optimas, a unified framework for effective optimization of compound systems. The core idea of Optimas is to maintain one Local Reward Function (LRF) per component, each satisfying a local-global alignment property, i.e., each component's local reward correlates with the global system performance. In each iteration, Optimas efficiently adapts the LRFs to maintain this property while simultaneously maximizing each component's local reward. This approach enables independent updates of heterogeneous configurations using the designated optimization method, while ensuring that local improvements consistently lead to performance gains. We present extensive evaluations across five real-world compound systems to demonstrate that Optimas outperforms strong baselines by an average improvement of 11.92%, offering a general and effective approach for improving compound systems. Our website is at https://optimas.stanford.edu.
Authors: Tim Beyer, Yan Scholten, Leo Schwinn, Stephan G\"unnemann
Abstract: To guarantee safe and robust deployment of large language models (LLMs) at scale, it is critical to accurately assess their adversarial robustness. Existing adversarial attacks typically target harmful responses in single-point greedy generations, overlooking the inherently stochastic nature of LLMs and overestimating robustness. We show that for the goal of eliciting harmful responses, repeated sampling of model outputs during the attack complements prompt optimization and serves as a strong and efficient attack vector. By casting attacks as a resource allocation problem between optimization and sampling, we determine compute-optimal trade-offs and show that integrating sampling into existing attacks boosts success rates by up to 37\% and improves efficiency by up to two orders of magnitude. We further analyze how distributions of output harmfulness evolve during an adversarial attack, discovering that many common optimization strategies have little effect on output harmfulness. Finally, we introduce a label-free proof-of-concept objective based on entropy maximization, demonstrating how our sampling-aware perspective enables new optimization targets. Overall, our findings establish the importance of sampling in attacks to accurately assess and strengthen LLM safety at scale.
Authors: Meihua Dang, Jiaqi Han, Minkai Xu, Kai Xu, Akash Srivastava, Stefano Ermon
Abstract: Discrete diffusion models have recently emerged as strong alternatives to autoregressive language models, matching their performance through large-scale training. However, inference-time control remains relatively underexplored. In this work, we study how to steer generation toward desired rewards without retraining the models. Prior methods typically resample or filter within a single denoising trajectory, optimizing rewards step-by-step without trajectory-level refinement. We introduce particle Gibbs sampling for diffusion language models (PG-DLM), a novel inference-time algorithm enabling trajectory-level refinement while preserving generation perplexity under reward optimization. PG-DLM constructs a Markov chain over full denoising trajectories and applies a conditional sequential Monte Carlo kernel to resample them. We derive theoretical guarantees for convergence, including asymptotic consistency and variance bounds. Within this framework, we further analyze trade-offs across four key axes for inference-time scaling under fixed budgets: iterations, samples, denoising steps, and reward estimation. Our analysis shows scaling iterations achieves the best reward-perplexity trade-off. Empirically, PG-DLM consistently outperforms prior methods using MDLM and LLaDA-8B as base models across a wide range of compute budgets for reward-guided generation tasks including toxicity and sentiment control as well as linguistic acceptability.
Authors: Yaowenqi Liu, Bingxu Meng, Rui Pan, Yuxing Liu, Jerry Huang, Jiaxuan You, Tong Zhang
Abstract: The field of AI research is advancing at an unprecedented pace, enabling automated hypothesis generation and experimental design across diverse domains such as biology, mathematics, and artificial intelligence. Despite these advancements, there remains a significant gap in the availability of scalable advising systems capable of providing high-quality, well-reasoned feedback to refine proposed hypotheses and experimental designs. To address this challenge, we explore key factors that underlie the development of robust advising systems, including model size, context length, confidence estimation, and structured reasoning processes. Our findings reveal that a relatively small model, when equipped with a well-compressed literature database and a structured reasoning framework, can outperform powerful general-purpose language models such as Deepseek-R1 in terms of acceptance rates for self-ranked top-30% submissions to ICLR 2025. Moreover, when limited to high-confidence predictions, our system achieves an acceptance rate exceeding 90% on the ICLR 2025 test set, underscoring its potential to significantly enhance the quality and efficiency of hypothesis generation and experimental design. The code is released at https://github.com/HowardLiu0830/GUIDE-Research-Idea-Evaluation.
URLs: https://github.com/HowardLiu0830/GUIDE-Research-Idea-Evaluation.
Authors: Huynh Dang Nguyen, Trong-Thang Pham, Ngan Le, Van Nguyen
Abstract: The electrocardiogram (ECG) is an essential and effective tool for diagnosing heart diseases. However, its effectiveness can be compromised by noise or unavailability of one or more leads of the standard 12-lead recordings, resulting in diagnostic errors or uncertainty. To address these challenges, we propose TolerantECG, a foundation model for ECG signals that is robust to noise and capable of functioning with arbitrary subsets of the standard 12-lead ECG. TolerantECG training combines contrastive and self-supervised learning frameworks to jointly learn ECG signal representations alongside their corresponding knowledge-retrieval-based text report descriptions and corrupted or lead-missing signals. Comprehensive benchmarking results demonstrate that TolerantECG consistently ranks as the best or second-best performer across various ECG signal conditions and class levels in the PTB-XL dataset, and achieves the highest performance on the MIT-BIH Arrhythmia Database.
Authors: Zihao Xue, Zhen Bi, Long Ma, Zhenlin Hu, Yan Wang, Zhenfang Liu, Qing Sheng, Jie Xiao, Jungang Lou
Abstract: While reinforcement learning-trained Large Reasoning Models (LRMs, e.g., Deepseek-R1) demonstrate advanced reasoning capabilities in the evolving Large Language Models (LLMs) domain, their susceptibility to security threats remains a critical vulnerability. This weakness is particularly evident in Chain-of-Thought (CoT) generation processes, where adversarial methods like backdoor prompt attacks can systematically subvert the model's core reasoning mechanisms. The emerging Chain-of-Thought Attack (CoTA) reveals this vulnerability through exploiting prompt controllability, simultaneously degrading both CoT safety and task performance with low-cost interventions. To address this compounded security-performance vulnerability, we propose Thought Purity (TP): a defense framework that systematically strengthens resistance to malicious content while preserving operational efficacy. Our solution achieves this through three synergistic components: (1) a safety-optimized data processing pipeline (2) reinforcement learning-enhanced rule constraints (3) adaptive monitoring metrics. Our approach establishes the first comprehensive defense mechanism against CoTA vulnerabilities in reinforcement learning-aligned reasoning systems, significantly advancing the security-functionality equilibrium for next-generation AI architectures.
Authors: Muhao Guo, Yang Weng
Abstract: Accurate detection and localization of solar photovoltaic (PV) panels in satellite imagery is essential for optimizing microgrids and active distribution networks (ADNs), which are critical components of renewable energy systems. Existing methods lack transparency regarding their underlying algorithms or training datasets, rely on large, high-quality PV training data, and struggle to generalize to new geographic regions or varied environmental conditions without extensive re-training. These limitations lead to inconsistent detection outcomes, hindering large-scale deployment and data-driven grid optimization. In this paper, we investigate how large language models (LLMs) can be leveraged to overcome these challenges. Despite their promise, LLMs face several challenges in solar panel detection, including difficulties with multi-step logical processes, inconsistent output formatting, frequent misclassification of visually similar objects (e.g., shadows, parking lots), and low accuracy in complex tasks such as spatial localization and quantification. To overcome these issues, we propose the PV Assessment with LLMs (PVAL) framework, which incorporates task decomposition for more efficient workflows, output standardization for consistent and scalable formatting, few-shot prompting to enhance classification accuracy, and fine-tuning using curated PV datasets with detailed annotations. PVAL ensures transparency, scalability, and adaptability across heterogeneous datasets while minimizing computational overhead. By combining open-source accessibility with robust methodologies, PVAL establishes an automated and reproducible pipeline for solar panel detection, paving the way for large-scale renewable energy integration and optimized grid management.
Authors: Jakob Snel, Seong Joon Oh
Abstract: Large Language Models (LLMs) hallucinate, and detecting these cases is key to ensuring trust. While many approaches address hallucination detection at the response or span level, recent work explores token-level detection, enabling more fine-grained intervention. However, the distribution of hallucination signal across sequences of hallucinated tokens remains unexplored. We leverage token-level annotations from the RAGTruth corpus and find that the first hallucinated token is far more detectable than later ones. This structural property holds across models, suggesting that first hallucination tokens play a key role in token-level hallucination detection. Our code is available at https://github.com/jakobsnl/RAGTruth_Xtended.
Authors: Adib Hasan, Mardavij Roozbehani, Munther Dahleh
Abstract: Accurate crop yield forecasting is essential for global food security. However, current AI models systematically underperform when yields deviate from historical trends. We attribute this to the lack of rich, physically grounded datasets directly linking atmospheric states to yields. To address this, we introduce VITA (Variational Inference Transformer for Asymmetric data), a variational pretraining framework that learns representations from large satellite-based weather datasets and transfers to the ground-based limited measurements available for yield prediction. VITA is trained using detailed meteorological variables as proxy targets during pretraining and learns to predict latent atmospheric states under a seasonality-aware sinusoidal prior. This allows the model to be fine-tuned using limited weather statistics during deployment. Applied to 763 counties in the U.S. Corn Belt, VITA achieves state-of-the-art performance in predicting corn and soybean yields across all evaluation scenarios, particularly during extreme years, with statistically significant improvements (paired t-test, $p < 0.0001$). Importantly, VITA outperforms prior frameworks like GNN-RNN without soil data, and bigger foundational models (e.g., Chronos-Bolt) with less compute, making it practical for real-world use--especially in data-scarce regions. This work highlights how domain-aware AI design can overcome data limitations and support resilient agricultural forecasting in a changing climate.
Authors: Gideon Vos, Liza van Eijk, Zoltan Sarnyai, Mostafa Rahimi Azghadi
Abstract: Synaptic pruning in biological brains removes weak connections to improve efficiency. In contrast, dropout regularization in artificial neural networks randomly deactivates neurons without considering activity-dependent pruning. We propose a magnitude-based synaptic pruning method that better reflects biology by progressively removing low-importance connections during training. Integrated directly into the training loop as a dropout replacement, our approach computes weight importance from absolute magnitudes across layers and applies a cubic schedule to gradually increase global sparsity. At fixed intervals, pruning masks permanently remove low-importance weights while maintaining gradient flow for active ones, eliminating the need for separate pruning and fine-tuning phases. Experiments on multiple time series forecasting models including RNN, LSTM, and Patch Time Series Transformer across four datasets show consistent gains. Our method ranked best overall, with statistically significant improvements confirmed by Friedman tests (p < 0.01). In financial forecasting, it reduced Mean Absolute Error by up to 20% over models with no or standard dropout, and up to 52% in select transformer models. This dynamic pruning mechanism advances regularization by coupling weight elimination with progressive sparsification, offering easy integration into diverse architectures. Its strong performance, especially in financial time series forecasting, highlights its potential as a practical alternative to conventional dropout techniques.
Authors: Zhicheng Yang, Zhijiang Guo, Yinya Huang, Yongxin Wang, Dongchun Xie, Yiwei Wang, Xiaodan Liang, Jing Tang
Abstract: Reinforcement Learning with Verifiable Reward (RLVR) has emerged as a powerful paradigm for unlocking reasoning capabilities in large language models, yet its full potential is hindered by two under-explored dimensions: Depth-the hardest problem a model can sample; Breadth-the number of instances consumed in a single iteration. We dissect the popular GRPO algorithm and reveal a systematic bias: the cumulative-advantage disproportionately weights samples with medium accuracy, while down-weighting the low-accuracy instances that are crucial for pushing reasoning boundaries. To rectify the depth neglect, we introduce Difficulty Adaptive Rollout Sampling (DARS), which re-weights hard problems through targeted multi-stage rollouts, thereby increasing the number of positive rollouts for hard problems. Empirically, naively enlarging rollout size only accelerates convergence and even hurts Pass@K. Our DARS, in contrast, delivers consistent Pass@K gains without extra inference cost at convergence. Just as we adaptively expanded the depth of exploration, we now ask whether aggressively scaling the breadth of training data can further amplify reasoning gains. To this end, we intensely scale batch size and replace PPO's mini-batch iterations with full-batch updates over multiple epochs. Increasing breadth significantly enhances Pass@1 performance. Large-breadth training sustains high token-level entropy, indicating continued exploration and reduced gradient noise. We further present DARS-B, which augments DARS with large breadth, and demonstrate simultaneous gains in Pass@K and Pass@1. The results confirm that breadth and adaptive exploration across depth operate as orthogonal dimensions in RLVR, which are key to unleashing the reasoning power of RLVR.
Authors: Scott Jeen
Abstract: Modern reinforcement learning (RL) systems capture deep truths about general, human problem-solving. In domains where new data can be simulated cheaply, these systems uncover sequential decision-making policies that far exceed the ability of any human. Society faces many problems whose solutions require this skill, but they are often in domains where new data cannot be cheaply simulated. In such scenarios, we can learn simulators from existing data, but these will only ever be approximately correct, and can be pathologically incorrect when queried outside of their training distribution. As a result, a misalignment between the environments in which we train our agents and the real-world in which we wish to deploy our agents is inevitable. Dealing with this misalignment is the primary concern of zero-shot reinforcement learning, a problem setting where the agent must generalise to a new task or domain with zero practice shots. Whilst impressive progress has been made on methods that perform zero-shot RL in idealised settings, new work is needed if these results are to be replicated in real-world settings. In this thesis, we argue that doing so requires us to navigate (at least) three constraints. First, the data quality constraint: real-world datasets are small and homogeneous. Second, the observability constraint: states, dynamics and rewards in the real-world are often only partially observed. And third, the data availability constraint: a priori access to data cannot always be assumed. This work proposes a suite of methods that perform zero-shot RL subject to these constraints. In a series of empirical studies we expose the failings of existing methods, and justify our techniques for remedying them. We believe these designs take us a step closer to RL methods that can be deployed to solve real-world problems.
Authors: Hossein KhademSohi, Hadi Hemmati, Jiayu Zhou, Steve Drew
Abstract: Federated Learning (FL) aggregates information from multiple clients to train a shared global model without exposing raw data. Accurately estimating each client's contribution is essential not just for fair rewards, but for selecting the most useful clients so the global model converges faster. The Shapley value is a principled choice, yet exact computation scales exponentially with the number of clients, making it infeasible for large federations. We propose FedOwen, an efficient framework that uses Owen sampling to approximate Shapley values under the same total evaluation budget as existing methods while keeping the approximation error small. In addition, FedOwen uses an adaptive client selection strategy that balances exploiting high-value clients with exploring under-sampled ones, reducing bias and uncovering rare but informative data. Under a fixed valuation cost, FedOwen achieves up to 23 percent higher final accuracy within the same number of communication rounds compared to state-of-the-art baselines on non-IID benchmarks.
Authors: Peter Racioppo
Abstract: We introduce Adaptive Filter Attention (AFA), a novel attention mechanism that incorporates a learnable dynamics model directly into the computation of attention weights. Rather than comparing queries and keys directly, we model the input sequence as discrete observations of a linear stochastic differential equation (SDE). By imposing a linear dynamics model with simultaneously diagonalizable state matrices and noise covariances, we can make use of a closed-form solution to the differential Lyapunov equation to efficiently propagate pairwise uncertainties through the dynamics. Attention naturally arises as the maximum likelihood solution for this linear SDE, with attention weights corresponding to robust residual-based reweightings of the propagated pairwise precisions. Imposing an additional constraint on the state matrix's eigenvalues leads to a simplified variant with the same computational and memory complexity as standard attention. In the limit of vanishing dynamics and process noise, and using a small-angle approximation, we recover ordinary dot-product attention.
Authors: Debdeep Sanyal, Aaryan Nagpal, Dhruv Kumar, Murari Mandal, Saurabh Deshpande
Abstract: While transformer-based foundation models excel at forecasting routine patterns, two questions remain: do they internalize semantic concepts such as market regimes, or merely fit curves? And can their internal representations be leveraged to simulate rare, high-stakes events such as market crashes? To investigate this, we introduce activation transplantation, a causal intervention that manipulates hidden states by imposing the statistical moments of one event (e.g., a historical crash) onto another (e.g., a calm period) during the forward pass. This procedure deterministically steers forecasts: injecting crash semantics induces downturn predictions, while injecting calm semantics suppresses crashes and restores stability. Beyond binary control, we find that models encode a graded notion of event severity, with the latent vector norm directly correlating with the magnitude of systemic shocks. Validated across two architecturally distinct TSFMs, Toto (decoder only) and Chronos (encoder-decoder), our results demonstrate that steerable, semantically grounded representations are a robust property of large time series transformers. Our findings provide evidence for a latent concept space that governs model predictions, shifting interpretability from post-hoc attribution to direct causal intervention, and enabling semantic "what-if" analysis for strategic stress-testing.
Authors: Vittorio Giammarino, Ruiqi Ni, Ahmed H. Qureshi
Abstract: Offline Goal-Conditioned Reinforcement Learning (GCRL) holds great promise for domains such as autonomous navigation and locomotion, where collecting interactive data is costly and unsafe. However, it remains challenging in practice due to the need to learn from datasets with limited coverage of the state-action space and to generalize across long-horizon tasks. To improve on these challenges, we propose a \emph{Physics-informed (Pi)} regularized loss for value learning, derived from the Eikonal Partial Differential Equation (PDE) and which induces a geometric inductive bias in the learned value function. Unlike generic gradient penalties that are primarily used to stabilize training, our formulation is grounded in continuous-time optimal control and encourages value functions to align with cost-to-go structures. The proposed regularizer is broadly compatible with temporal-difference-based value learning and can be integrated into existing Offline GCRL algorithms. When combined with Hierarchical Implicit Q-Learning (HIQL), the resulting method, Eikonal-regularized HIQL (Eik-HIQL), yields significant improvements in both performance and generalization, with pronounced gains in stitching regimes and large-scale navigation tasks.
Authors: Robson L. F. Cordeiro, Meng-Chieh Lee, Christos Faloutsos
Abstract: Given a set of financial transactions (who buys from whom, when, and for how much), as well as prior information from buyers and sellers, how can we find fraudulent transactions? If we have labels for some transactions for known types of fraud, we can build a classifier. However, we also want to find new types of fraud, still unknown to the domain experts ('Detection'). Moreover, we also want to provide evidence to experts that supports our opinion ('Justification'). In this paper, we propose FRAUDGUESS, to achieve two goals: (a) for 'Detection', it spots new types of fraud as micro-clusters in a carefully designed feature space; (b) for 'Justification', it uses visualization and heatmaps for evidence, as well as an interactive dashboard for deep dives. FRAUDGUESS is used in real life and is currently considered for deployment in an Anonymous Financial Institution (AFI). Thus, we also present the three new behaviors that FRAUDGUESS discovered in a real, million-scale financial dataset. Two of these behaviors are deemed fraudulent or suspicious by domain experts, catching hundreds of fraudulent transactions that would otherwise go un-noticed.
Authors: Ansh Nagda, Prabhakar Raghavan, Abhradeep Thakurta
Abstract: We explore whether techniques from AI can help discover new combinatorial structures that improve on known limits on efficient algorithms. Specifically, we use AlphaEvolve (an LLM coding agent) to study two settings: a) Average-case hardness for MAX-CUT and MAX-Independent Set: We improve a recent result of Kunisky and Yu to obtain near-optimal upper and (conditional) lower bounds on certification algorithms for MAX-CUT and MAX-Independent Set on random 3- and 4-regular graphs. Our improved lower bounds are obtained by constructing nearly extremal Ramanujan graphs on as many as $163$ nodes, using AlphaEvolve. Additionally, via analytical arguments we strengthen the upper bounds to settle the computational hardness of these questions up to an error in the third decimal place. b) Worst-case Hardness of Approximation for MAX-k-CUT: We obtain new inapproximability results, proving that it is NP-hard to approximate MAX-4-CUT and MAX-3-CUT within factors of $0.987$ and $0.9649$ respectively, using AlphaEvolve to discover new gadget reductions. Our MAX-4-CUT result improves upon the SOTA of $0.9883$, and our MAX-3-CUT result improves on the current best gadget-based inapproximability result of $0.9853$, but falls short of improving the SOTA of $16/17$ that relies on a custom PCP, rather than a gadget reduction from "standard" H{\aa}stad-style PCPs. A key technical challenge we faced: verifying a candidate construction produced by AlphaEvolve is costly (often requiring exponential time). In both settings above, our results were enabled by using AlphaEvolve itself to evolve the verification procedure to be faster (sometimes by $10,000\times$). We conclude with a discussion of norms by which to assess the assistance from AI in developing proofs.
Authors: Andreas Lebedev, Abhinav Das, Sven Pappert, Stephan Schl\"uter
Abstract: Precise probabilistic forecasts are fundamental for energy risk management, and there is a wide range of both statistical and machine learning models for this purpose. Inherent to these probabilistic models is some form of uncertainty quantification. However, most models do not capture the full extent of uncertainty, which arises not only from the data itself but also from model and distributional choices. In this study, we examine uncertainty quantification in state-of-the-art statistical and deep learning probabilistic forecasting models for electricity price forecasting in the German market. In particular, we consider deep distributional neural networks (DDNNs) and augment them with an ensemble approach, Monte Carlo (MC) dropout, and conformal prediction to account for model uncertainty. Additionally, we consider the LASSO-estimated autoregressive (LEAR) approach combined with quantile regression averaging (QRA), generalized autoregressive conditional heteroskedasticity (GARCH), and conformal prediction. Across a range of performance metrics, we find that the LEAR-based models perform well in terms of probabilistic forecasting, irrespective of the uncertainty quantification method. Furthermore, we find that DDNNs benefit from incorporating both data and model uncertainty, improving both point and probabilistic forecasting. Uncertainty itself appears to be best captured by the models using conformal prediction. Overall, our extensive study shows that all models under consideration perform competitively. However, their relative performance depends on the choice of metrics for point and probabilistic forecasting.
Authors: Mohan Guo, Cong Liu, Patrick Forr\'e
Abstract: Recent methods for molecular generation face a trade-off: they either enforce strict equivariance with costly architectures or relax it to gain scalability and flexibility. We propose a frame-based diffusion paradigm that achieves deterministic E(3)-equivariance while decoupling symmetry handling from the backbone. Building on this paradigm, we investigate three variants: Global Frame Diffusion (GFD), which assigns a shared molecular frame; Local Frame Diffusion (LFD), which constructs node-specific frames and benefits from additional alignment constraints; and Invariant Frame Diffusion (IFD), which relies on pre-canonicalized invariant representations. To enhance expressivity, we further utilize EdgeDiT, a Diffusion Transformer with edge-aware attention. On the QM9 dataset, GFD with EdgeDiT achieves state-of-the-art performance, with a test NLL of -137.97 at standard scale and -141.85 at double scale, alongside atom stability of 98.98%, and molecular stability of 90.51%. These results surpass all equivariant baselines while maintaining high validity and uniqueness and nearly 2x faster sampling compared to EDM. Altogether, our study establishes frame-based diffusion as a scalable, flexible, and physically grounded paradigm for molecular generation, highlighting the critical role of global structure preservation.
Authors: Benjamin Feuer, Chiung-Yi Tseng, Astitwa Sarthak Lathe, Oussama Elachqar, John P Dickerson
Abstract: LLM-judged benchmarks are increasingly used to evaluate complex model behaviors, yet their design introduces failure modes absent in conventional ground-truth based benchmarks. We argue that without tight objectives and verifiable constructions, benchmark rankings can produce high-confidence rankings that are in fact largely noise. We introduce two mechanisms to diagnose these issues. Schematic adherence quantifies how much of a judge's overall verdict is explained by the explicit evaluation schema, revealing unexplained variance when judges deviate from their own rubric. Psychometric validity aggregates internal consistency and discriminant validity signals to quantify irreducible uncertainty in any benchmarking run. Applying these tools to Arena-Hard Auto, we find severe schema incoherence and factor collapse across popular judges: for example, unexplained variance exceeding 90 percent for DeepSeek-R1-32B and factor correlations above 0.93 for most criteria. We also show that the ELO-style aggregation used by Arena-Hard Auto collapses and masks genuine ranking uncertainty. Our results highlight design failures that undermine validity and offer actionable principles for building better-scoped, reliability-aware LLM-judged benchmarks. We released our code and dataset at https://github.com/penfever/judgment-to-noise
Authors: Zhenpeng Su, Leiyu Pan, Minxuan Lv, Yuntao Li, Wenping Hu, Fuzheng Zhang, Kun Gai, Guorui Zhou
Abstract: Reinforcement learning (RL) has become a powerful paradigm for optimizing large language models (LLMs) to handle complex reasoning tasks. A core challenge in this process lies in managing policy entropy, which reflects the balance between exploration and exploitation during training. Existing methods, such as proximal policy optimization (PPO) and its variants, discard valuable gradient signals from low-probability tokens due to the clipping mechanism. We systematically analyze the entropy dynamics and reveal that these clipped tokens play a critical yet overlooked role in regulating entropy evolution. We propose \textbf{C}oordinating \textbf{E}ntropy via \textbf{G}radient-\textbf{P}reserving \textbf{P}olicy \textbf{O}ptimization (CE-GPPO), a novel algorithm that reintroduces gradients from clipped tokens in native PPO in a gentle and bounded manner. By controlling the magnitude of gradients from tokens outside the clipping interval, CE-GPPO is able to achieve an exploration-exploitation trade-off. We provide theoretical justification and empirical evidence showing that CE-GPPO effectively mitigates entropy instability. Extensive experiments on mathematical reasoning benchmarks show that CE-GPPO consistently outperforms strong baselines across different model scales.
Authors: Yiyou Sun, Yuhan Cao, Pohao Huang, Haoyue Bai, Hannaneh Hajishirzi, Nouha Dziri, Dawn Song
Abstract: It remains an open question whether LLMs can acquire or generalize genuinely new reasoning strategies, beyond the sharpened skills encoded in their parameters during pre-training or post-training. To attempt to answer this debate, we introduce DELTA-Code -- Distributional Evaluation of Learnability and Transferrability in Algorithmic Coding -- a controlled benchmark of synthetic coding problem families designed to probe two fundamental aspects: learnability -- can LLMs, through reinforcement learning (RL), solve problem families where pretrained models exhibit failure with large enough attempts (pass@K=0)? -- and transferrability -- if learnability happens, can such skills transfer systematically to out-of-distribution (OOD) test sets? Unlike prior public coding datasets, DELTA isolates reasoning skills through templated problem generators and introduces fully OOD problem families that demand novel strategies rather than tool invocation or memorized patterns. Our experiments reveal a striking grokking phase transition: after an extended period with near-zero reward, RL-trained models abruptly climb to near-perfect accuracy. To enable learnability on previously unsolvable problem families, we explore key training ingredients such as staged warm-up with dense rewards, experience replay, curriculum training, and verification-in-the-loop. Beyond learnability, we use DELTA to evaluate transferability or generalization along exploratory, compositional, and transformative axes, as well as cross-family transfer. Results show solid gains within families and for recomposed skills, but persistent weaknesses in transformative cases. DELTA thus offers a clean testbed for probing the limits of RL-driven reasoning and for understanding how models can move beyond existing priors to acquire new algorithmic skills.
Authors: Taeyoung Yun, Pierre-Luc St-Charles, Jinkyoo Park, Yoshua Bengio, Minsu Kim
Abstract: We address the challenge of generating diverse attack prompts for large language models (LLMs) that elicit harmful behaviors (e.g., insults, sexual content) and are used for safety fine-tuning. Rather than relying on manual prompt engineering, attacker LLMs can be trained with reinforcement learning (RL) to automatically generate such prompts using only a toxicity classifier as a reward. However, capturing a wide range of harmful behaviors is a significant challenge that requires explicit diversity objectives. Existing diversity-seeking RL methods often collapse to limited modes: once high-reward prompts are found, exploration of new regions is discouraged. Inspired by the active learning paradigm that encourages adaptive exploration, we introduce \textit{Active Attacks}, a novel RL-based red-teaming algorithm that adapts its attacks as the victim evolves. By periodically safety fine-tuning the victim LLM with collected attack prompts, rewards in exploited regions diminish, which forces the attacker to seek unexplored vulnerabilities. This process naturally induces an easy-to-hard exploration curriculum, where the attacker progresses beyond easy modes toward increasingly difficult ones. As a result, Active Attacks uncovers a wide range of local attack modes step by step, and their combination achieves wide coverage of the multi-mode distribution. Active Attacks, a simple plug-and-play module that seamlessly integrates into existing RL objectives, unexpectedly outperformed prior RL-based methods -- including GFlowNets, PPO, and REINFORCE -- by improving cross-attack success rates against GFlowNets, the previous state-of-the-art, from 0.07% to 31.28% (a relative gain greater than $400\ \times$) with only a 6% increase in computation. Our code is publicly available \href{https://github.com/dbsxodud-11/active_attacks}{here}.
Authors: Roie Kazoom, Yuval Ratzabi, Etamar Rothstein, Ofer Hadar
Abstract: Adversarial robustness in structured data remains an underexplored frontier compared to vision and language domains. In this work, we introduce a novel black-box, decision-based adversarial attack tailored for tabular data. Our approach combines gradient-free direction estimation with an iterative boundary search, enabling efficient navigation of discrete and continuous feature spaces under minimal oracle access. Extensive experiments demonstrate that our method successfully compromises nearly the entire test set across diverse models, ranging from classical machine learning classifiers to large language model (LLM)-based pipelines. Remarkably, the attack achieves success rates consistently above 90%, while requiring only a small number of queries per instance. These results highlight the critical vulnerability of tabular models to adversarial perturbations, underscoring the urgent need for stronger defenses in real-world decision-making systems.
Authors: Yuke Li, Yujia Zheng, Tianyi Xiong, Zhenyi Wang, Heng Huang
Abstract: Catastrophic interference, also known as catastrophic forgetting, is a fundamental challenge in machine learning, where a trained learning model progressively loses performance on previously learned tasks when adapting to new ones. In this paper, we aim to better understand and model the catastrophic interference problem from a latent representation learning point of view, and propose a novel theoretical framework that formulates catastrophic interference as an identification problem. Our analysis demonstrates that the forgetting phenomenon can be quantified by the distance between partial-task aware (PTA) and all-task aware (ATA) setups. Building upon recent advances in identifiability theory, we prove that this distance can be minimized through identification of shared latent variables between these setups. When learning, we propose our method \ourmeos with two-stage training strategy: First, we employ maximum likelihood estimation to learn the latent representations from both PTA and ATA configurations. Subsequently, we optimize the KL divergence to identify and learn the shared latent variables. Through theoretical guarantee and empirical validations, we establish that identifying and learning these shared representations can effectively mitigate catastrophic interference in machine learning systems. Our approach provides both theoretical guarantees and practical performance improvements across both synthetic and benchmark datasets.
Authors: Raviteja Anantha, Soheil Hor, Teodor Nicola Antoniu, Layne C. Price
Abstract: We present NanoFlux, a novel adversarial framework for generating targeted training data to improve LLM reasoning, where adversarially-generated datasets containing fewer than 200 examples outperform conventional fine-tuning approaches. The framework employs a competitive dynamic between models alternating as Attacker and Defender, supervised by a tool-augmented Judge, synthesizing multi-step questions with explanatory annotations that target specific reasoning capabilities. Fine-tuning a 4B-parameter model on NanoFlux-generated data yields performance gains across diverse domains compared to full-benchmark fine-tuning: +5.9% on mathematical reasoning (GSMHard), +3.6% on scientific reasoning (GenomeBench), and +16.6% on medical reasoning (MultiMedQA), while reducing computational requirements by 3-14x. Ablation studies reveal a non-monotonic relationship between dataset characteristics and model performance, uncovering domain-specific optimal points for question complexity and reasoning quality. NanoFlux automates training data generation through embedding-based novelty filtering, tool-augmented evaluation, and multi-hop reasoning, suggesting that future model improvements may lie in the intelligent synthesis of small, precisely targeted training datasets.
Authors: Hanlin Zhu, Shibo Hao, Zhiting Hu, Jiantao Jiao, Stuart Russell, Yuandong Tian
Abstract: Previous work shows that the chain of continuous thought (continuous CoT) improves the reasoning capability of large language models (LLMs) by enabling implicit parallel thinking, and a subsequent work provided theoretical insight by showing that a two-layer transformer equipped with continuous CoT can efficiently solve directed graph reachability by maintaining a superposition of multiple reasoning traces in the continuous thought. However, it remains unclear how the superposition mechanism is naturally learned from gradient-based training methods. To fill this gap, we theoretically analyze the training dynamics of a simplified two-layer transformer on the directed graph reachability problem to unveil how the superposition mechanism emerges during training in two training stages -- (i) a thought-generation stage that autoregressively expands the continuous thought, and (ii) a prediction stage that converts the thought into the final answer. Our analysis reveals that during training using continuous thought, the index-matching logit, an important quantity which reflects the strength of the model's local search ability, will first increase and then remain bounded under mild assumptions. The bounded index-matching logit effectively balances exploration and exploitation during the reasoning process: the model will exploit local problem structures to identify plausible search traces, and assign comparable weights to multiple such traces to explore when it is uncertain about which solution is correct, which results in superposition. Our experimental results tracking the growth of logits further validate our theory.
Authors: Liangjian Wen, Qun Dai, Jianzhuang Liu, Jiangtao Zheng, Yong Dai, Dongkai Wang, Zhao Kang, Jun Wang, Zenglin Xu, Jiang Duan
Abstract: In multimodal representation learning, synergistic interactions between modalities not only provide complementary information but also create unique outcomes through specific interaction patterns that no single modality could achieve alone. Existing methods may struggle to effectively capture the full spectrum of synergistic information, leading to suboptimal performance in tasks where such interactions are critical. This is particularly problematic because synergistic information constitutes the fundamental value proposition of multimodal representation. To address this challenge, we introduce InfMasking, a contrastive synergistic information extraction method designed to enhance synergistic information through an Infinite Masking strategy. InfMasking stochastically occludes most features from each modality during fusion, preserving only partial information to create representations with varied synergistic patterns. Unmasked fused representations are then aligned with masked ones through mutual information maximization to encode comprehensive synergistic information. This infinite masking strategy enables capturing richer interactions by exposing the model to diverse partial modality combinations during training. As computing mutual information estimates with infinite masking is computationally prohibitive, we derive an InfMasking loss to approximate this calculation. Through controlled experiments, we demonstrate that InfMasking effectively enhances synergistic information between modalities. In evaluations on large-scale real-world datasets, InfMasking achieves state-of-the-art performance across seven benchmarks. Code is released at https://github.com/brightest66/InfMasking.
Authors: Amirhossein Zare (SeyedAbolfazl), Amirhessam Zare (SeyedAbolfazl), Parmida Sadat Pezeshki (SeyedAbolfazl), Herlock (SeyedAbolfazl), Rahimi, Ali Ebrahimi, Ignacio V\'azquez-Garc\'ia, Leo Anthony Celi
Abstract: Class imbalance remains a major challenge in machine learning, especially for high-dimensional biomedical data where nonlinear manifold structures dominate. Traditional oversampling methods such as SMOTE rely on local linear interpolation, often producing implausible synthetic samples. Deep generative models like Conditional Variational Autoencoders (CVAEs) better capture nonlinear distributions, but standard variants treat all minority samples equally, neglecting the importance of uncertain, boundary-region examples emphasized by heuristic methods like Borderline-SMOTE and ADASYN. We propose Local Entropy-Guided Oversampling with a CVAE (LEO-CVAE), a generative oversampling framework that explicitly incorporates local uncertainty into both representation learning and data generation. To quantify uncertainty, we compute Shannon entropy over the class distribution in a sample's neighborhood: high entropy indicates greater class overlap, serving as a proxy for uncertainty. LEO-CVAE leverages this signal through two mechanisms: (i) a Local Entropy-Weighted Loss (LEWL) that emphasizes robust learning in uncertain regions, and (ii) an entropy-guided sampling strategy that concentrates generation in these informative, class-overlapping areas. Applied to clinical genomics datasets (ADNI and TCGA lung cancer), LEO-CVAE consistently improves classifier performance, outperforming both traditional oversampling and generative baselines. These results highlight the value of uncertainty-aware generative oversampling for imbalanced learning in domains governed by complex nonlinear structures, such as omics data.
Authors: Dingyi Kang, Dongming Jiang, Hanshen Yang, Hang Liu, Bingzhe Li
Abstract: Approximate Nearest Neighbor Search (ANNS), as the core of vector databases (VectorDBs), has become widely used in modern AI and ML systems, powering applications from information retrieval to bio-informatics. While graph-based ANNS methods achieve high query efficiency, their scalability is constrained by the available host memory. Recent disk-based ANNS approaches mitigate memory usage by offloading data to Solid-State Drives (SSDs). However, they still suffer from issues such as long I/O traversal path, misalignment with storage I/O granularity, and high in-memory indexing overhead, leading to significant I/O latency and ultimately limiting scalability for large-scale vector search. In this paper, we propose PageANN, a disk-based approximate nearest neighbor search (ANNS) framework designed for high performance and scalability. PageANN introduces a page-node graph structure that aligns logical graph nodes with physical SSD pages, thereby shortening I/O traversal paths and reducing I/O operations. Specifically, similar vectors are clustered into page nodes, and a co-designed disk data layout leverages this structure with a merging technique to store only representative vectors and topology information, avoiding unnecessary reads. To further improve efficiency, we design a memory management strategy that combines lightweight indexing with coordinated memory-disk data allocation, maximizing host memory utilization while minimizing query latency and storage overhead. Experimental results show that PageANN significantly outperforms state-of-the-art (SOTA) disk-based ANNS methods, achieving 1.85x-10.83x higher throughput and 51.7%-91.9% lower latency across different datasets and memory budgets, while maintaining comparable high recall accuracy.
Authors: Zhendong Mi, Bian Sun, Grace Li Zhang, Shaoyi Huang
Abstract: Large language models (LLMs) have rapidly scaled in size, bringing severe memory and computational challenges that hinder their deployment. Singular Value Decomposition (SVD)-based compression has emerged as an appealing post-training compression technique for LLMs, yet most existing methods apply a uniform compression ratio across all layers, implicitly assuming homogeneous information included in various layers. This overlooks the substantial intra-layer heterogeneity observed in LLMs, where middle layers tend to encode richer information while early and late layers are more redundant. In this work, we revisit the existing SVD-based compression method and propose D-Rank, a framework with layer-wise balanced Dynamic Rank allocation for LLMs compression. We first introduce effective rank as a principled metric to measure the information density of weight matrices, and then allocate ranks via a Lagrange multiplier-based optimization scheme to adaptively assign more capacity to groups with higher information density under a fixed compression ratio. Moreover, we rebalance the allocated ranks across attention layers to account for their varying importance and extend D-Rank to latest LLMs with grouped-query attention. Extensive experiments on various LLMs with different scales across multiple compression ratios demonstrate that D-Rank consistently outperforms SVD-LLM, ASVD, and Basis Sharing, achieving more than 15 lower perplexity with LLaMA-3-8B model on C4 datasets at 20% compression ratio and up to 5% higher zero-shot reasoning accuracy with LLaMA-7B model at 40% compression ratio while achieving even higher throughput.
Authors: Amber Srivastava, Salar Basiri, Srinivasa Salapaka
Abstract: Clustering arises in a wide range of problem formulations, yet most existing approaches assume that the entities under clustering are passive and strictly conform to their assigned groups. In reality, entities often exhibit local autonomy, overriding prescribed associations in ways not fully captured by feature representations. Such autonomy can substantially reshape clustering outcomes -- altering cluster compositions, geometry, and cardinality -- with significant downstream effects on inference and decision-making. We introduce autonomy-aware clustering, a reinforcement learning (RL) framework that learns and accounts for the influence of local autonomy without requiring prior knowledge of its form. Our approach integrates RL with a Deterministic Annealing (DA) procedure, where, to determine underlying clusters, DA naturally promotes exploration in early stages of annealing and transitions to exploitation later. We also show that the annealing procedure exhibits phase transitions that enable design of efficient annealing schedules. To further enhance adaptability, we propose the Adaptive Distance Estimation Network (ADEN), a transformer-based attention model that learns dependencies between entities and cluster representatives within the RL loop, accommodates variable-sized inputs and outputs, and enables knowledge transfer across diverse problem instances. Empirical results show that our framework closely aligns with underlying data dynamics: even without explicit autonomy models, it achieves solutions close to the ground truth (gap ~3-4%), whereas ignoring autonomy leads to substantially larger gaps (~35-40%). The code and data are publicly available at https://github.com/salar96/AutonomyAwareClustering.
Authors: Shenao Zhang, Donghan Yu, Yihao Feng, Bowen Jin, Zhaoran Wang, John Peebles, Zirui Wang
Abstract: Large language models excel with reinforcement learning (RL), but fully unlocking this potential requires a mid-training stage. An effective mid-training phase should identify a compact set of useful actions and enable fast selection among them through online RL. We formalize this intuition by presenting the first theoretical result on how mid-training shapes post-training: it characterizes an action subspace that minimizes both the value approximation error from pruning and the RL error during subsequent planning. Our analysis reveals two key determinants of mid-training effectiveness: pruning efficiency, which shapes the prior of the initial RL policy, and its impact on RL convergence, which governs the extent to which that policy can be improved via online interactions. These results suggest that mid-training is most effective when the decision space is compact and the effective horizon is short, highlighting the importance of operating in the space of action abstractions rather than primitive actions. Building on these insights, we propose Reasoning as Action Abstractions (RA3), a scalable mid-training algorithm. Specifically, we derive a sequential variational lower bound and optimize it by iteratively discovering temporally-consistent latent structures via RL, followed by fine-tuning on the bootstrapped data. Experiments on code generation tasks demonstrate the effectiveness of our approach. Across multiple base models, RA3 improves the average performance on HumanEval and MBPP by 8 and 4 points over the base model and the next-token prediction baseline. Furthermore, RA3 achieves faster convergence and higher asymptotic performance in RLVR on HumanEval+, MBPP+, LiveCodeBench, and Codeforces.
Authors: Shuche Wang, Fengzhuo Zhang, Jiaxiang Li, Cunxiao Du, Chao Du, Tianyu Pang, Zhuoran Yang, Mingyi Hong, Vincent Y. F. Tan
Abstract: The Muon optimizer is consistently faster than Adam in training Large Language Models (LLMs), yet the mechanism underlying its success remains unclear. This paper demystifies this mechanism through the lens of associative memory. By ablating the transformer components optimized by Muon, we reveal that the associative memory parameters of LLMs, namely the Value and Output (VO) attention weights and Feed-Forward Networks (FFNs), are the primary contributors to Muon's superiority. Motivated by this associative memory view, we then explain Muon's superiority on real-world corpora, which are intrinsically heavy-tailed: a few classes (tail classes) appear far less frequently than others. The superiority is explained through two key properties: (i) its update rule consistently yields a more isotropic singular spectrum than Adam; and as a result, (ii) on heavy-tailed data, it optimizes tail classes more effectively than Adam. Beyond empirical evidence, we theoretically confirm these findings by analyzing a one-layer associative memory model under class-imbalanced data. We prove that Muon consistently achieves balanced learning across classes regardless of feature embeddings, whereas Adam can induce large disparities in learning errors depending on embedding properties. In summary, our empirical observations and theoretical analyses reveal Muon's core advantage: its update rule aligns with the outer-product structure of linear associative memories, enabling more balanced and effective learning of tail classes in heavy-tailed distributions than Adam.
Authors: Chi Ho Leung, Philip E. Par\'e
Abstract: We propose the multistep port-Hamiltonian Gaussian process (MS-PHS GP) to learn physically consistent continuous-time dynamics and a posterior over the Hamiltonian from noisy, irregularly-sampled trajectories. By placing a GP prior on the Hamiltonian surface $H$ and encoding variable-step multistep integrator constraints as finite linear functionals, MS-PHS GP enables closed-form conditioning of both the vector field and the Hamiltonian surface without latent states, while enforcing energy balance and passivity by design. We state a finite-sample vector-field bound that separates the estimation and variable-step discretization terms. Lastly, we demonstrate improved vector-field recovery and well-calibrated Hamiltonian uncertainty on mass-spring, Van der Pol, and Duffing benchmarks.
Authors: Rui Zhu, Xuan Yu, Yudong Zhang, Chen Zhang, Xu Wang, Yang Wang
Abstract: Generative Flow Networks (GFlowNets) have emerged as a powerful tool for generating diverse and high-reward structured objects by learning to sample from a distribution proportional to a given reward function. Unlike conventional reinforcement learning (RL) approaches that prioritize optimization of a single trajectory, GFlowNets seek to balance diversity and reward by modeling the entire trajectory distribution. This capability makes them especially suitable for domains such as molecular design and combinatorial optimization. However, existing GFlowNets sampling strategies tend to overexplore and struggle to consistently generate high-reward samples, particularly in large search spaces with sparse high-reward regions. Therefore, improving the probability of generating high-reward samples without sacrificing diversity remains a key challenge under this premise. In this work, we integrate an enhanced Monte Carlo Tree Search (MCTS) into the GFlowNets sampling process, using MCTS-based policy evaluation to guide the generation toward high-reward trajectories and Polynomial Upper Confidence Trees (PUCT) to balance exploration and exploitation adaptively, and we introduce a controllable mechanism to regulate the degree of greediness. Our method enhances exploitation without sacrificing diversity by dynamically balancing exploration and reward-driven guidance. The experimental results show that our method can not only accelerate the speed of discovering high-reward regions but also continuously generate high-reward samples, while preserving the diversity of the generative distribution. All implementations are available at https://github.com/ZRNB/MG2FlowNet.
Authors: Hengyi Zhu, Grace Li Zhang, Shaoyi Huang
Abstract: Hardware-Aware Neural Architecture Search (HW-NAS) requires joint optimization of accuracy and latency under device constraints. Traditional supernet-based methods require multiple GPU days per dataset. Large Language Model (LLM)-driven approaches avoid training a large supernet and can provide quick feedback, but we observe an exploration bias: the LLM repeatedly proposes neural network designs within limited search space and fails to discover architectures across different latency ranges in the entire search space. To address this issue, we propose PEL-NAS: a search space Partitioned, architecture prompt co-Evolutionary and LLM-driven Neural Architecture Search that can generate neural networks with high accuracy and low latency with reduced search cost. Our proposed PEL-NAS has three key components: 1) a complexity-driven partitioning engine that divides the search space by complexity to enforce diversity and mitigate exploration bias; 2) an LLM-powered architecture prompt co-evolution operator, in which the LLM first updates a knowledge base of design heuristics based on results from the previous round, then performs a guided evolution algorithm on architectures with prompts that incorporate this knowledge base. Prompts and designs improve together across rounds which avoids random guesswork and improve efficiency; 3) a zero-cost predictor to avoid training a large number of candidates from scratch. Experimental results show that on HW-NAS-Bench, PEL-NAS can achieve overall higher HV, lower IGD, and up to 54% lower latency than baselines at similar accuracy. Meanwhile, the search cost drops from days to minutes compared with traditional supernet baselines.
Authors: Kezhao Liu, Jason Klein Liu, Mingtao Chen, Yiming Liu
Abstract: Reinforcement Learning from Human Feedback (RLHF) leverages a Kullback-Leibler (KL) divergence loss to stabilize training and prevent overfitting. However, in methods such as GRPO, its implementation may be guided by principles from numerical value estimation-a practice that overlooks the term's functional role as an optimization loss. To analyze this issue, we establish a unified framework that connects two seemingly distinct implementation styles: using the mathematical term $k_n$ as a detached coefficient for the policy's score function ('$k_n$ in reward') or as a direct loss function through which gradients are propagated ('$k_n$ as loss'). We show that the latter can always be analyzed via an equivalent gradient coefficient in the former, unifying the two perspectives. Through this framework, we prove that the conventional '$k_1$ in reward' (like in PPO) is the principled loss for Reverse KL (RKL) regularization. We further establish a key finding: under on-policy conditions, the '$k_2$ as loss' formulation is, in fact, gradient-equivalent to '$k_1$ in reward'. This equivalence, first proven in our work, identifies both as the theoretically sound implementations of the RKL objective. In contrast, we show that the recently adopted '$k_3$ as loss' (like in GRPO) is merely a first-order, biased approximation of the principled loss. Furthermore, we argue that common off-policy implementations of '$k_n$ as loss' methods are biased due to neglected importance sampling, and we propose a principled correction. Our findings provide a comprehensive, gradient-based rationale for choosing and correctly implementing KL regularization, paving the way for more robust and effective RLHF systems.
Authors: Vladimir Krsmanovic, Matthias Cosler, Mohamed Ghanem, Bernd Finkbeiner
Abstract: Model checking is a key technique for verifying safety-critical systems against formal specifications, where recent applications of deep learning have shown promise. However, while ubiquitous for vision and language domains, representation learning remains underexplored in formal verification. We introduce Contrastive Neural Model Checking (CNML), a novel method that leverages the model checking task as a guiding signal for learning aligned representations. CNML jointly embeds logical specifications and systems into a shared latent space through a self-supervised contrastive objective. On industry-inspired retrieval tasks, CNML considerably outperforms both algorithmic and neural baselines in cross-modal and intra-modal settings. We further show that the learned representations effectively transfer to downstream tasks and generalize to more complex formulas. These findings demonstrate that model checking can serve as an objective for learning representations for formal languages.
Authors: Klaus-Rudolf Kladny, Bernhard Sch\"olkopf, Michael Muehlebach
Abstract: AdaBoost sequentially fits so-called weak learners to minimize an exponential loss, which penalizes mislabeled data points more severely than other loss functions like cross-entropy. Paradoxically, AdaBoost generalizes well in practice as the number of weak learners grows. In the present work, we introduce Penalized Exponential Loss (PENEX), a new formulation of the multi-class exponential loss that is theoretically grounded and, in contrast to the existing formulation, amenable to optimization via first-order methods. We demonstrate both empirically and theoretically that PENEX implicitly maximizes margins of data points. Also, we show that gradient increments on PENEX implicitly parameterize weak learners in the boosting framework. Across computer vision and language tasks, we show that PENEX exhibits a regularizing effect often better than established methods with similar computational cost. Our results highlight PENEX's potential as an AdaBoost-inspired alternative for effective training and fine-tuning of deep neural networks.
Authors: Kyoungjun Park, Yifan Yang, Changhan Ge, Lili Qiu, Shiqi Jiang
Abstract: Modeling radio frequency (RF) signal propagation is essential for understanding the environment, as RF signals offer valuable insights beyond the capabilities of RGB cameras, which are limited by the visible-light spectrum, lens coverage, and occlusions. It is also useful for supporting wireless diagnosis, deployment, and optimization. However, accurately predicting RF signals in complex environments remains a challenge due to interactions with obstacles such as absorption and reflection. We introduce Diffusion^2, a diffusion-based approach that uses 3D point clouds to model the propagation of RF signals across a wide range of frequencies, from Wi-Fi to millimeter waves. To effectively capture RF-related features from 3D data, we present the RF-3D Encoder, which encapsulates the complexities of 3D geometry along with signal-specific details. These features undergo multi-scale embedding to simulate the actual RF signal dissemination process. Our evaluation, based on synthetic and real-world measurements, demonstrates that Diffusion^2 accurately estimates the behavior of RF signals in various frequency bands and environmental conditions, with an error margin of just 1.9 dB and 27x faster than existing methods, marking a significant advancement in the field. Refer to https://rfvision-project.github.io/ for more information.
Authors: Taurai Muvunza, Egor Kraev, Pere Planell-Morell, Alexander Y. Shestopaloff
Abstract: Existing feature filters rely on statistical pair-wise dependence metrics to model feature-target relationships, but this approach may fail when the target depends on higher-order feature interactions rather than individual contributions. We introduce Mutual Information Neural Estimation Regularized Vetting Algorithm (MINERVA), a novel approach to supervised feature selection based on neural estimation of mutual information between features and targets. We paramaterize the approximation of mutual information with neural networks and perform feature selection using a carefully designed loss function augmented with sparsity-inducing regularizers. Our method is implemented in a two-stage process to decouple representation learning from feature selection, ensuring better generalization and a more accurate expression of feature importance. We present examples of ubiquitous dependency structures that are rarely captured in literature and show that our proposed method effectively captures these complex feature-target relationships by evaluating feature subsets as an ensemble. Experimental results on synthetic and real-life fraud datasets demonstrate the efficacy of our method and its ability to perform exact solutions.
Authors: Th\'eo Dupuy, Binbin Xu, St\'ephane Perrey, Jacky Montmain, Abdelhak Imoussaten
Abstract: Uncertainty quantification has received considerable interest in recent works in Machine Learning. In particular, Conformal Prediction (CP) gains ground in this field. For the case of time series, Online Conformal Prediction (OCP) becomes an option to address the problem of data distribution shift over time. Indeed, the idea of OCP is to update a threshold of some quantity (whether the miscoverage level or the quantile) based on the distribution observation. To evaluate the performance of OCP methods, two key aspects are typically considered: the coverage validity and the prediction interval width minimization. Recently, new OCP methods have emerged, offering long-run coverage guarantees and producing more informative intervals. However, during the threshold update step, most of these methods focus solely on the validity of the prediction intervals~--~that is, whether the ground truth falls inside or outside the interval~--~without accounting for their relevance. In this paper, we aim to leverage this overlooked aspect. Specifically, we propose enhancing the threshold update step by replacing the binary evaluation (inside/outside) with a broader class of functions that quantify the relevance of the prediction interval using the ground truth. This approach helps prevent abrupt threshold changes, potentially resulting in narrower prediction intervals. Indeed, experimental results on real-world datasets suggest that these functions can produce tighter intervals compared to existing OCP methods while maintaining coverage validity.
Authors: Feiyang Wu, Ye Zhao, Anqi Wu
Abstract: We propose a distributional framework for offline Inverse Reinforcement Learning (IRL) that jointly models uncertainty over reward functions and full distributions of returns. Unlike conventional IRL approaches that recover a deterministic reward estimate or match only expected returns, our method captures richer structure in expert behavior, particularly in learning the reward distribution, by minimizing first-order stochastic dominance (FSD) violations and thus integrating distortion risk measures (DRMs) into policy learning, enabling the recovery of both reward distributions and distribution-aware policies. This formulation is well-suited for behavior analysis and risk-aware imitation learning. Empirical results on synthetic benchmarks, real-world neurobehavioral data, and MuJoCo control tasks demonstrate that our method recovers expressive reward representations and achieves state-of-the-art imitation performance.
Authors: Teemu H\"ark\"onen, Sara Wade, Kody Law, Lassi Roininen
Abstract: Gaussian processes are a key component of many flexible statistical and machine learning models. However, they exhibit cubic computational complexity and high memory constraints due to the need of inverting and storing a full covariance matrix. To circumvent this, mixtures of Gaussian process experts have been considered where data points are assigned to independent experts, reducing the complexity by allowing inference based on smaller, local covariance matrices. Moreover, mixtures of Gaussian process experts substantially enrich the model's flexibility, allowing for behaviors such as non-stationarity, heteroscedasticity, and discontinuities. In this work, we construct a novel inference approach based on nested sequential Monte Carlo samplers to simultaneously infer both the gating network and Gaussian process expert parameters. This greatly improves inference compared to importance sampling, particularly in settings when a stationary Gaussian process is inappropriate, while still being thoroughly parallelizable.
Authors: Alexandre M\"osching, Housen Li, Axel Munk
Abstract: Hidden Markov models (HMMs) are characterized by an unobservable Markov chain and an observable process -- a noisy version of the hidden chain. Decoding the original signal from the noisy observations is one of the main goals in nearly all HMM based data analyses. Existing decoding algorithms such as Viterbi and the pointwise maximum a posteriori (PMAP) algorithm have computational complexity at best linear in the length of the observed sequence, and sub-quadratic in the size of the state space of the hidden chain. We present Quick Adaptive Ternary Segmentation (QATS), a divide-and-conquer procedure with computational complexity polylogarithmic in the length of the sequence, and cubic in the size of the state space, hence particularly suited for large scale HMMs with relatively few states. It also suggests an effective way of data storage as specific cumulative sums. In essence, the estimated sequence of states sequentially maximizes local likelihood scores among all local paths with at most three segments, and is meanwhile admissible. The maximization is performed only approximately using an adaptive search procedure. Our simulations demonstrate the speedups offered by QATS in comparison to Viterbi and PMAP, along with a precision analysis. An implementation of QATS is in the R-package QATS on GitHub.
Authors: Erhan Bayraktar, Bingyan Han
Abstract: We develop a fitted value iteration (FVI) method to compute bicausal optimal transport (OT) where couplings have an adapted structure. Based on the dynamic programming formulation, FVI adopts a function class to approximate the value functions in bicausal OT. Under the concentrability condition and approximate completeness assumption, we prove the sample complexity using (local) Rademacher complexity. Furthermore, we demonstrate that multilayer neural networks with appropriate structures satisfy the crucial assumptions required in sample complexity proofs. Numerical experiments reveal that FVI outperforms linear programming and adapted Sinkhorn methods in scalability as the time horizon increases, while still maintaining acceptable accuracy.
Authors: Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, Hanyu Lai, Yu Gu, Hangliang Ding, Kaiwen Men, Kejuan Yang, Shudan Zhang, Xiang Deng, Aohan Zeng, Zhengxiao Du, Chenhui Zhang, Sheng Shen, Tianjun Zhang, Yu Su, Huan Sun, Minlie Huang, Yuxiao Dong, Jie Tang
Abstract: The potential of Large Language Model (LLM) as agents has been widely acknowledged recently. Thus, there is an urgent need to quantitatively \textit{evaluate LLMs as agents} on challenging tasks in interactive environments. We present AgentBench, a multi-dimensional benchmark that consists of 8 distinct environments to assess LLM-as-Agent's reasoning and decision-making abilities. Our extensive test over \num API-based and open-sourced (OSS) LLMs shows that, while top commercial LLMs present a strong ability of acting as agents in complex environments, there is a significant disparity in performance between them and many OSS competitors that are no larger than 70B. We identify the typical reasons of failures in environments and LLMs, showing that poor long-term reasoning, decision-making, and instruction following abilities are the main obstacles for developing usable LLM agents. Improving instruction following and training on high quality multi-round alignment data could improve agent performance. And different from existing assumptions, training on code present ambivalent impacts on different agent tasks. Datasets, environments, and an integrated evaluation package for AgentBench are released at https://github.com/THUDM/AgentBench.
Authors: Shijun Liang, Van Hoang Minh Nguyen, Jinghan Jia, Ismail Alkhouri, Sijia Liu, Saiprasad Ravishankar
Abstract: As the popularity of deep learning (DL) in the field of magnetic resonance imaging (MRI) continues to rise, recent research has indicated that DL-based MRI reconstruction models might be excessively sensitive to minor input disturbances, including worst-case additive perturbations. This sensitivity often leads to unstable, aliased images. This raises the question of how to devise DL techniques for MRI reconstruction that can be robust to train-test variations. To address this problem, we propose a novel image reconstruction framework, termed Smoothed Unrolling (SMUG), which advances a deep unrolling-based MRI reconstruction model using a randomized smoothing (RS)-based robust learning approach. RS, which improves the tolerance of a model against input noises, has been widely used in the design of adversarial defense approaches for image classification tasks. Yet, we find that the conventional design that applies RS to the entire DL-based MRI model is ineffective. In this paper, we show that SMUG and its variants address the above issue by customizing the RS process based on the unrolling architecture of a DL-based MRI reconstruction model. Compared to the vanilla RS approach, we show that SMUG improves the robustness of MRI reconstruction with respect to a diverse set of instability sources, including worst-case and random noise perturbations to input measurements, varying measurement sampling rates, and different numbers of unrolling steps. Furthermore, we theoretically analyze the robustness of our method in the presence of perturbations.
Authors: Marlon Tobaben, Hibiki Ito, Joonas J\"alk\"o, Yuan He, Antti Honkela
Abstract: Membership inference attacks (MIAs) are used to test practical privacy of machine learning models. MIAs complement formal guarantees from differential privacy (DP) under a more realistic adversary model. We analyse MIA vulnerability of fine-tuned neural networks both empirically and theoretically, the latter using a simplified model of fine-tuning. We show that the vulnerability of non-DP models when measured as the attacker advantage at a fixed false positive rate reduces according to a simple power law as the number of examples per class increases. A similar power-law applies even for the most vulnerable points, but the dataset size needed for adequate protection of the most vulnerable points is very large.
Authors: Benjamin Townsend, Madison May, Katherine Mackowiak, Christopher Wells
Abstract: We introduce RealKIE, a benchmark of five challenging datasets aimed at advancing key information extraction methods, with an emphasis on enterprise applications. The datasets include a diverse range of documents including SEC S1 Filings, US Non-disclosure Agreements, UK Charity Reports, FCC Invoices, and Resource Contracts. Each presents unique challenges: poor text serialization, sparse annotations in long documents, and complex tabular layouts. These datasets provide a realistic testing ground for key information extraction tasks like investment analysis and contract analysis. In addition to presenting these datasets, we offer an in-depth description of the annotation process, document processing techniques, and baseline modeling approaches. This contribution facilitates the development of NLP models capable of handling practical challenges and supports further research into information extraction technologies applicable to industry-specific problems. The annotated data, OCR outputs, and code to reproduce baselines are available to download at https://indicodatasolutions.github.io/RealKIE/.
Authors: Rajiv Sambharya, Bartolomeo Stellato
Abstract: We introduce a data-driven approach to analyze the performance of continuous optimization algorithms using generalization guarantees from statistical learning theory. We study classical and learned optimizers to solve families of parametric optimization problems. We build generalization guarantees for classical optimizers, using a sample convergence bound, and for learned optimizers, using the Probably Approximately Correct (PAC)-Bayes framework. To train learned optimizers, we use a gradient-based algorithm to directly minimize the PAC-Bayes upper bound. Numerical experiments in signal processing, control, and meta-learning showcase the ability of our framework to provide strong generalization guarantees for both classical and learned optimizers given a fixed budget of iterations. For classical optimizers, our bounds are much tighter than those that worst-case guarantees provide. For learned optimizers, our bounds outperform the empirical outcomes observed in their non-learned counterparts.
Authors: Negin Golrezaei, Sourav Sahoo
Abstract: We study the bidding problem in repeated uniform price multi-unit auctions from the perspective of a value-maximizing buyer. The buyer aims to maximize their cumulative value over $T$ rounds while adhering to per-round return-on-investment (RoI) constraints in a strategic (or adversarial) environment. Using an $m$-uniform bidding format, the buyer submits $m$ bid-quantity pairs $(b_i, q_i)$ to demand $q_i$ units at bid $b_i$, with $m \ll M$ in practice, where $M$ denotes the maximum demand of the buyer. We introduce the notion of safe bidding strategies as those that satisfy the RoI constraints irrespective of competing bids. Despite the stringent requirement, we show that these strategies satisfy a mild no-overbidding condition, depend only on the valuation curve of the bidder, and the bidder can focus on a finite subset without loss of generality. Though the subset size is $O(M^m)$, we design a polynomial-time learning algorithm that achieves sublinear regret, both in full-information and bandit settings, relative to the hindsight-optimal safe strategy. We assess the robustness of safe strategies against the hindsight-optimal strategy from a richer class. We define the richness ratio $\alpha \in (0,1]$ as the minimum ratio of the value of the optimal safe strategy to that of the optimal strategy from richer class and construct hard instances showing the tightness of $\alpha$. Our algorithm achieves $\alpha$-approximate sublinear regret against these stronger benchmarks. Simulations on semi-synthetic auction data show that empirical richness ratios significantly outperform the theoretical worst-case bounds. The proposed safe strategies and learning algorithm extend naturally to more nuanced buyer and competitor models.
Authors: Hoang M. Ngo, Nguyen H K. Do, Minh N. Vu, Tre' R. Jeter, Tamer Kahveci, My T. Thai
Abstract: Quantum annealing (QA) has great potential to solve combinatorial optimization problems efficiently. However, the effectiveness of QA algorithms is heavily based on the embedding of problem instances, represented as logical graphs, into the quantum processing unit (QPU) whose topology is in the form of a limited connectivity graph, known as the minor embedding problem. Because the minor embedding problem is an NP-hard problem~\mbox{\cite{Goodrich2018}}, existing methods for the minor embedding problem suffer from scalability issues when faced with larger problem sizes. In this paper, we propose a novel approach utilizing Reinforcement Learning (RL) techniques to address the minor embedding problem, named CHARME. CHARME includes three key components: a Graph Neural Network (GNN) architecture for policy modeling, a state transition algorithm that ensures solution validity, and an order exploration strategy for effective training. Through comprehensive experiments on synthetic and real-world instances, we demonstrate the efficiency of our proposed order exploration strategy as well as our proposed RL framework, CHARME. In particular, CHARME yields superior solutions in terms of qubit usage compared to fast embedding methods such as Minorminer and ATOM. Moreover, our method surpasses the OCT-based approach, known for its slower runtime but high-quality solutions, in several cases. In addition, our proposed exploration enhances the efficiency of the training of the CHARME framework by providing better solutions compared to the greedy strategy.
Authors: Efthymios Costa, Ioanna Papatsouma, Angelos Markos
Abstract: In this paper, we present an information-theoretic method for clustering mixed-type data, that is, data consisting of both continuous and categorical variables. The proposed approach extends the Information Bottleneck principle to heterogeneous data through generalised product kernels, integrating continuous, nominal, and ordinal variables within a unified optimization framework. We address the following challenges: developing a systematic bandwidth selection strategy that equalises contributions across variable types, and proposing an adaptive hyperparameter updating scheme that ensures a valid solution into a predetermined number of potentially imbalanced clusters. Through simulations on 28,800 synthetic data sets and ten publicly available benchmarks, we demonstrate that the proposed method, named DIBmix, achieves superior performance compared to four established methods (KAMILA, K-Prototypes, FAMD with K-Means, and PAM with Gower's dissimilarity). Results show DIBmix particularly excels when clusters exhibit size imbalances, data contain low or moderate cluster overlap, and categorical and continuous variables are equally represented. The method presents a significant advantage over traditional centroid-based algorithms, establishing DIBmix as a competitive and theoretically grounded alternative for mixed-type data clustering.
Authors: Yingzhen Yang, Ping Li
Abstract: We study nonparametric regression using an over-parameterized two-layer neural networks trained with algorithmic guarantees in this paper. We consider the setting where the training features are drawn uniformly from the unit sphere in $\RR^d$, and the target function lies in an interpolation space commonly studied in statistical learning theory. We demonstrate that training the neural network with a novel Preconditioned Gradient Descent (PGD) algorithm, equipped with early stopping, achieves a sharp regression rate of $\cO(n^{-\frac{2\alpha s'}{2\alpha s'+1}})$ when the target function is in the interpolation space $\bth{\cH_K}^{s'}$ with $s' \ge 3$. This rate is even sharper than the currently known nearly-optimal rate of $\cO(n^{-\frac{2\alpha s'}{2\alpha s'+1}})\log^2(1/\delta)$~\citep{Li2024-edr-general-domain}, where $n$ is the size of the training data and $\delta \in (0,1)$ is a small probability. This rate is also sharper than the standard kernel regression rate of $\cO(n^{-\frac{2\alpha}{2\alpha+1}})$ obtained under the regular Neural Tangent Kernel (NTK) regime when training the neural network with the vanilla gradient descent (GD), where $2\alpha = d/(d-1)$. Our analysis is based on two key technical contributions. First, we present a principled decomposition of the network output at each PGD step into a function in the reproducing kernel Hilbert space (RKHS) of a newly induced integral kernel, and a residual function with small $L^{\infty}$-norm. Second, leveraging this decomposition, we apply local Rademacher complexity theory to tightly control the complexity of the function class comprising all the neural network functions obtained in the PGD iterates. Our results further suggest that PGD enables the neural network to escape the linear NTK regime and achieve improved generalization.
Authors: Yinghao Zhu, Junyi Gao, Zixiang Wang, Weibin Liao, Xiaochen Zheng, Lifang Liang, Miguel O. Bernabeu, Yasha Wang, Lequan Yu, Chengwei Pan, Ewen M. Harrison, Liantao Ma
Abstract: Large Language Models (LLMs) are increasingly deployed in medicine. However, their utility in non-generative clinical prediction, often presumed inferior to specialized models, remains under-evaluated, leading to ongoing debate within the field and potential for misuse, misunderstanding, or over-reliance due to a lack of systematic benchmarking. Our ClinicRealm study addresses this by benchmarking 15 GPT-style LLMs, 5 BERT-style models, and 11 traditional methods on unstructured clinical notes and structured Electronic Health Records (EHR), while also assessing their reasoning, reliability, and fairness. Key findings reveal a significant shift: for clinical note predictions, leading LLMs (e.g., DeepSeek-V3.1-Think, GPT-5) in zero-shot settings now decisively outperform finetuned BERT models. On structured EHRs, while specialized models excel with ample data, advanced LLMs (e.g., GPT-5, DeepSeek-V3.1-Think) show potent zero-shot capabilities, often surpassing conventional models in data-scarce settings. Notably, leading open-source LLMs can match or exceed proprietary counterparts. These results provide compelling evidence that modern LLMs are competitive tools for non-generative clinical prediction, particularly with unstructured text and offering data-efficient structured data options, thus necessitating a re-evaluation of model selection strategies. This research should serve as an important insight for medical informaticists, AI developers, and clinical researchers, potentially prompting a reassessment of current assumptions and inspiring new approaches to LLM application in predictive healthcare.
Authors: Parthe Pandit, Zhichao Wang, Yizhe Zhu
Abstract: Kernel ridge regression (KRR) is a popular class of machine learning models that has become an important tool for understanding deep learning. Much of the focus thus far has been on studying the proportional asymptotic regime, $n \asymp d$, where $n$ is the number of training samples and $d$ is the dimension of the dataset. In the proportional regime, under certain conditions on the data distribution, the kernel random matrix involved in KRR exhibits behavior akin to that of a linear kernel. In this work, we extend the study of kernel regression to the quadratic asymptotic regime, where $n \asymp d^2$. In this regime, we demonstrate that a broad class of inner-product kernels exhibits behavior similar to a quadratic kernel. Specifically, we establish an operator norm approximation bound for the difference between the original kernel random matrix and a quadratic kernel random matrix with additional correction terms compared to the Taylor expansion of the kernel functions. The approximation works for general data distributions under a Gaussian-moment-matching assumption with a covariance structure. This new approximation is utilized to obtain a limiting spectral distribution of the original kernel matrix and characterize the precise asymptotic training and test errors for KRR in the quadratic regime when $n/d^2$ converges to a non-zero constant. The generalization errors are obtained for (i) a random teacher model, (ii) a deterministic teacher model where the weights are perfectly aligned with the covariance of the data. Under the random teacher model setting, we also verify that the generalized cross-validation (GCV) estimator can consistently estimate the generalization error in the quadratic regime for anisotropic data. Our proof techniques combine moment methods, Wick's formula, orthogonal polynomials, and resolvent analysis of random matrices with correlated entries.
Authors: Chao Shen, Yuan Li, Wenkang Zhan, Shujie Pan, Fuxin Lin, Kaiyao Xin, Hui Cong, Chi Xu, Xiaotian Cheng, Ruixiang Liu, Zhibo Ni, Chaoyuan Jin, Bo Xu, Siming Chen, Zhongming Wei, Chunlai Xue, Zhanguo Wang, Chao Zhao
Abstract: Developing tailored semiconductor heterostructures on demand represents a critical capability for addressing the escalating performance demands in electronic and optoelectronic devices. However, traditional fabrication methods remain constrained by simulation-based design and iterative trial-and-error optimization. Here, we introduce SemiEpi, a self-driving platform designed for molecular beam epitaxy (MBE) to perform multi-step semiconductor heterostructure growth through in-situ monitoring and on-the-fly feedback control. By integrating standard MBE reactors, physics-informed machine learning (ML) models, and parameter initialization, SemiEpi identifies optimal initial conditions and proposes experiments for heterostructure growth, eliminating the need for extensive expertise in MBE processes. As a proof of concept, we demonstrate the optimization of high-density InAs quantum dot (QD) growth with a target emission wavelength of 1240 nm, showcasing the power of SemiEpi. We achieve a QD density of 5 x 10^10 cm^-2, a 1.6-fold increase in photoluminescence (PL) intensity, and a reduced full width at half maximum (FWHM) of 29.13 meV, leveraging in-situ reflective high-energy electron diffraction monitoring with feedback control for adjusting growth temperatures. Taken together, our results highlight the potential of ML-guided systems to address challenges in multi-step heterostructure growth, facilitate the development of a hardware-independent framework, and enhance process repeatability and stability, even without exhaustive knowledge of growth parameters.
Authors: Mark Obozov, Makar Baderko, Stepan Kulibaba, Nikolay Kutuzov, Alexander Gasnikov
Abstract: Recommender systems aim to estimate the dynamically changing user preferences and sequential dependencies between historical user behaviour and metadata. Although transformer-based models have proven to be effective in sequential recommendations, their state growth is proportional to the length of the sequence that is being processed, which makes them expensive in terms of memory and inference costs. Our research focused on three promising directions in sequential recommendations: enhancing speed through the use of State Space Models (SSM), as they can achieve SOTA results in the sequential recommendations domain with lower latency, memory, and inference costs, as proposed by arXiv:2403.03900 improving the quality of recommendations with Large Language Models (LLMs) via Monolithic Preference Optimization without Reference Model (ORPO); and implementing adaptive batch- and step-size algorithms to reduce costs and accelerate training processes.
Authors: Yuling Jiao, Yang Wang, Bokai Yan
Abstract: We study the approximation capacity of deep ReLU recurrent neural networks (RNNs) and explore the convergence properties of nonparametric least squares regression using RNNs. We derive upper bounds on the approximation error of RNNs for H\"older smooth functions, in the sense that the output at each time step of an RNN can approximate a H\"older function that depends only on past and current information, termed a past-dependent function. This allows a carefully constructed RNN to simultaneously approximate a sequence of past-dependent H\"older functions. We apply these approximation results to derive non-asymptotic upper bounds for the prediction error of the empirical risk minimizer in regression problem. Our error bounds achieve minimax optimal rate under both exponentially $\beta$-mixing and i.i.d. data assumptions, improving upon existing ones. Our results provide statistical guarantees on the performance of RNNs.
Authors: James Halverson, Joydeep Naskar, Jiahua Tian
Abstract: We use the embedding formalism to construct conformal fields in $D$ dimensions, by restricting Lorentz-invariant ensembles of homogeneous neural networks in $(D+2)$ dimensions to the projective null cone. Conformal correlators may be computed using the parameter space description of the neural network. Exact four-point correlators are computed in a number of examples, and we perform a 4D conformal block decomposition that elucidates the spectrum. In some examples the analysis is facilitated by recent approaches to Feynman integrals. Generalized free CFTs are constructed using the infinite-width Gaussian process limit of the neural network, enabling a realization of the free boson. The extension to deep networks constructs conformal fields at each subsequent layer, with recursion relations relating their conformal dimensions and four-point functions. Numerical approaches are discussed.
Authors: Eviatar Bach, Ricardo Baptista, Daniel Sanz-Alonso, Andrew Stuart
Abstract: The aim of these notes is to demonstrate the potential for ideas in machine learning to impact on the fields of inverse problems and data assimilation. The perspective is one that is primarily aimed at researchers from inverse problems and/or data assimilation who wish to see a mathematical presentation of machine learning as it pertains to their fields. As a by-product, we include a succinct mathematical treatment of various fundamental underpinning topics in machine learning, and adjacent areas of (computational) mathematics.
Authors: Alexandros Graikos, Nebojsa Jojic, Dimitris Samaras
Abstract: Large denoising diffusion models, such as Stable Diffusion, have been trained on billions of image-caption pairs to perform text-conditioned image generation. As a byproduct of this training, these models have acquired general knowledge about image statistics, which can be useful for other inference tasks. However, when confronted with sampling an image under new constraints, e.g. generating the missing parts of an image, using large pre-trained text-to-image diffusion models is inefficient and often unreliable. Previous approaches either utilized backpropagation through the denoiser network, making them significantly slower and more memory-demanding than simple text-to-image generation, or only enforced the constraint locally, failing to capture critical long-range correlations in the sampled image. In this work, we propose an algorithm that enables fast, high-quality generation under arbitrary constraints. We show that in denoising diffusion models, we can employ an approximation to Newton's optimization method that allows us to speed up inference and avoid the expensive backpropagation operations. Our approach produces results that rival or surpass the state-of-the-art training-free inference methods while requiring a fraction of the time. We demonstrate the effectiveness of our algorithm under both linear (inpainting, super-resolution) and non-linear (style-guided generation) constraints. An implementation is provided at https://github.com/cvlab-stonybrook/fast-constrained-sampling.
URLs: https://github.com/cvlab-stonybrook/fast-constrained-sampling.
Authors: Sanghyuk Chun, Wonjae Kim, Song Park, Sangdoo Yun
Abstract: Vision-language models (VLMs) embed aligned image-text pairs into a joint space but often rely on deterministic embeddings, assuming a one-to-one correspondence between images and texts. This oversimplifies real-world relationships, which are inherently many-to-many, with multiple captions describing a single image and vice versa. We introduce Probabilistic Language-Image Pre-training (ProLIP), the first probabilistic VLM pre-trained on a billion-scale image-text dataset using only probabilistic objectives, achieving a strong zero-shot capability (e.g., 74.6% ImageNet zero-shot accuracy with ViT-B/16). ProLIP efficiently estimates uncertainty by an "uncertainty token" without extra parameters. We also introduce a novel inclusion loss that enforces distributional inclusion relationships between image-text pairs and between original and masked inputs. Experiments demonstrate that, by leveraging uncertainty estimates, ProLIP benefits downstream tasks and aligns with intuitive notions of uncertainty, e.g., shorter texts being more uncertain and more general inputs including specific ones. Utilizing text uncertainties, we further improve ImageNet accuracy from 74.6% to 75.8% (under a few-shot setting), supporting the practical advantages of our probabilistic approach. The code is available at https://github.com/naver-ai/prolip
Authors: Harshvardhan Mestha, Karan Bania, Shreyas V Sathyanarayana, Sidong Liu, Ashwin Srinivasan
Abstract: Our interest is in the design of software systems involving a human-expert interacting -- using natural language -- with a large language model (LLM) on data analysis tasks. For complex problems, it is possible that LLMs can harness human expertise and creativity to find solutions that were otherwise elusive. On one level, this interaction takes place through multiple turns of prompts from the human and responses from the LLM. Here we investigate a more structured approach based on an abstract protocol described in [3] for interaction between agents. The protocol is motivated by a notion of "two-way intelligibility" and is modelled by a pair of communicating finite-state machines. We provide an implementation of the protocol, and provide empirical evidence of using the implementation to mediate interactions between an LLM and a human-agent in two areas of scientific interest (radiology and drug design). We conduct controlled experiments with a human proxy (a database), and uncontrolled experiments with human subjects. The results provide evidence in support of the protocol's capability of capturing one- and two-way intelligibility in human-LLM interaction; and for the utility of two-way intelligibility in the design of human-machine systems. Our code is available at https://github.com/karannb/interact.
Authors: Debjyoti Saha Roy, Byron C. Wallace, Javed A. Aslam
Abstract: State-of-the-art Extreme Multi-Label Text Classification models rely on multi-label attention to focus on key tokens in input text, but learning good attention weights is challenging. We introduce PLANT - Pretrained and Leveraged Attention - a plug-and-play strategy for initializing attention. PLANT works by planting label-specific attention using a pretrained Learning-to-Rank model guided by mutual information gain. This architecture-agnostic approach integrates seamlessly with large language model backbones such as Mistral-7B, LLaMA3-8B, DeepSeek-V3, and Phi-3. PLANT outperforms state-of-the-art methods across tasks including ICD coding, legal topic classification, and content recommendation. Gains are especially pronounced in few-shot settings, with substantial improvements on rare labels. Ablation studies confirm that attention initialization is a key driver of these gains. For code and trained models, see https://github.com/debjyotiSRoy/xcube/tree/plant
Authors: Zijun Gao, Qingyuan Zhao
Abstract: Existing tools for explaining complex models and systems are associational rather than causal and do not provide mechanistic understanding. We propose a new notion called counterfactual explainability for causal attribution that is motivated by the concept of genetic heritability in twin studies. Counterfactual explainability extends methods for global sensitivity analysis (including the functional analysis of variance and Sobol's indices), which assumes independent explanatory variables, to dependent explanations by using a directed acyclic graphs to describe their causal relationship. Therefore, this explanability measure directly incorporates causal mechanisms by construction. Under a comonotonicity assumption, we discuss methods for estimating counterfactual explainability and apply them to a real dataset dataset to explain income inequality by gender, race, and educational attainment.
Authors: Shivanshu Shekhar, Shreyas Singh, Tong Zhang
Abstract: Direct Preference Optimization (DPO) has been successfully used to align large language models (LLMs) according to human preferences, and more recently it has also been applied to improving the quality of text-to-image diffusion models. However, DPO-based methods such as SPO, Diffusion-DPO, and D3PO are highly susceptible to overfitting and reward hacking, especially when the generative model is optimized to fit out-of-distribution during prolonged training. To overcome these challenges and stabilize the training of diffusion models, we introduce a self-entropy regularization mechanism in reinforcement learning from human feedback. This enhancement improves DPO training by encouraging broader exploration and greater robustness. Our regularization technique effectively mitigates reward hacking, leading to improved stability and enhanced image quality across the latent space. Extensive experiments demonstrate that integrating human feedback with self-entropy regularization can significantly boost image diversity and specificity, achieving state-of-the-art results on key image generation metrics.
Authors: Selim Furkan Tekin, Fatih Ilhan, Tiansheng Huang, Sihao Hu, Yichang Xu, Zachary Yahn, Ling Liu
Abstract: Alignment of pretrained LLMs using instruction-based datasets is critical for creating fine-tuned models that reflect human preference. A growing number of alignment-based fine-tuning algorithms and benchmarks emerged recently, fueling the efforts on effective alignments of pre-trained LLMs to ensure helpful, harmless, and honest answers from both open-source and closed-source LLMs. This paper tackles this problem by developing an alignment fusion approach, coined as $H^3$Fusion, with three unique characteristics. First, $H^3$Fusion ensembles multiple individually aligned LLMs to create a final fine-tuned alignment model with enhanced capabilities beyond those of individual models, delivering robust alignment through promoting helpful, harmless, honest fusion. Second, $H^3$Fusion leverages the mixture-of-experts (MoE) methodology in two steps. We first freeze the multi-head attention weights of each individual model while tuning the FFN layer during alignment fusion. Then we merge the aligned model weights with an expert router according to the type of input instruction and dynamically select a subset of experts that are best suited for producing the output response. Finally, we boost the performance of the resulting $H^3$3Fusion model by introducing gating loss and regularization terms. The former penalizes the selection errors of the expert-router, and the latter mediates the expert weights drifting during fine-tuning and dynamically adjusts the fusion behavior of the resulting model by canalizing the activations on the experts. Extensive evaluations on three benchmark datasets show that $H^3$3Fusion is more helpful, less harmful, and more honest from two aspects: it outperforms each individually aligned model by $11.37\%$, and it provides stronger robustness compared to the state-of-the-art LLM ensemble approaches by $13.77\%$. Code is available at github.com/sftekin/h3fusion.
Authors: Jessica Gronsbell, Jianhui Gao, Yaqi Shi, Zachary R. McCaw, David Cheng
Abstract: From structural biology to epidemiology, predictions from machine learning (ML) models increasingly complement costly gold-standard data to enable faster, more affordable, and scalable scientific inquiry. In response, prediction-based (PB) inference has emerged to accommodate statistical analysis using a large volume of predictions together with a small amount of gold-standard data. The goals of PB inference are two-fold: (i) to mitigate bias from errors in predictions and (ii) to improve efficiency relative to classical inference using only the gold-standard data. While early PB inference methods focused on bias, their ability to enhance efficiency remains a focus of ongoing research. We revisit a foundational PB inference method and show that a simple modification can be applied to guarantee provable improvements in efficiency. In doing so, we establish new connections between augmented inverse probability weighted estimators (AIPW) and several recently proposed PB inference methods with a similar focus. The utility of our proposal, which leverages prediction-based outcomes to enhance efficiency, is demonstrated through extensive simulation studies and an application to real data from the UK Biobank. Further, we contextualize PB inference by drawing connections to historical literature from economics and statistics, highlighting how classic methods directly inform this contemporary problem.
Authors: Zongyu Lin, Wei Liu, Chen Chen, Jiasen Lu, Wenze Hu, Tsu-Jui Fu, Jesse Allardice, Zhengfeng Lai, Liangchen Song, Bowen Zhang, Cha Chen, Yiran Fei, Lezhi Li, Yizhou Sun, Kai-Wei Chang, Yinfei Yang
Abstract: The field of video generation has made remarkable advancements, yet there remains a pressing need for a clear, systematic recipe that can guide the development of robust and scalable models. In this work, we present a comprehensive study that systematically explores the interplay of model architectures, training recipes, and data curation strategies, culminating in a simple and scalable text-image-conditioned video generation method, named STIV. Our framework integrates image condition into a Diffusion Transformer (DiT) through frame replacement, while incorporating text conditioning via a joint image-text conditional classifier-free guidance. This design enables STIV to perform both text-to-video (T2V) and text-image-to-video (TI2V) tasks simultaneously. Additionally, STIV can be easily extended to various applications, such as video prediction, frame interpolation, multi-view generation, and long video generation, etc. With comprehensive ablation studies on T2I, T2V, and TI2V, STIV demonstrate strong performance, despite its simple design. An 8.7B model with 512 resolution achieves 83.1 on VBench T2V, surpassing both leading open and closed-source models like CogVideoX-5B, Pika, Kling, and Gen-3. The same-sized model also achieves a state-of-the-art result of 90.1 on VBench I2V task at 512 resolution. By providing a transparent and extensible recipe for building cutting-edge video generation models, we aim to empower future research and accelerate progress toward more versatile and reliable video generation solutions.
Authors: Yinpeng Cai, Lexin Li, Linjun Zhang
Abstract: Large Language Models (LLMs) are rapidly gaining enormous popularity in recent years. However, the training of LLMs has raised significant privacy and legal concerns, particularly regarding the distillation and inclusion of copyrighted materials in their training data without proper attribution or licensing, an issue that falls under the broader concern of data misappropriation. In this article, we focus on a specific problem of data misappropriation detection, namely, to determine whether a given LLM has incorporated the data generated by another LLM. We propose embedding watermarks into the copyrighted training data and formulating the detection of data misappropriation as a hypothesis testing problem. We develop a general statistical testing framework, construct test statistics, determine optimal rejection thresholds, and explicitly control type I and type II errors. Furthermore, we establish the asymptotic optimality properties of the proposed tests, and demonstrate the empirical effectiveness through intensive numerical experiments.
Authors: Andr\'e Hottung, Paula Wong-Chung, Kevin Tierney
Abstract: Autoregressive construction approaches generate solutions to vehicle routing problems in a step-by-step fashion, leading to high-quality solutions that are nearing the performance achieved by handcrafted operations research techniques. In this work, we challenge the conventional paradigm of sequential solution construction and introduce an iterative search framework where solutions are instead deconstructed by a neural policy. Throughout the search, the neural policy collaborates with a simple greedy insertion algorithm to rebuild the deconstructed solutions. Our approach matches or surpasses the performance of state-of-the-art operations research methods across three challenging vehicle routing problems of various problem sizes.
Authors: Jelena Bratuli\'c, Sudhanshu Mittal, David T. Hoffmann, Samuel B\"ohm, Robin Tibor Schirrmeister, Tonio Ball, Christian Rupprecht, Thomas Brox
Abstract: Large Language Models (LLMs) exhibit In-Context Learning (ICL), which enables the model to perform new tasks conditioning only on the examples provided in the context without updating the model's weights. While ICL offers fast adaptation across natural language tasks and domains, its emergence is less straightforward for modalities beyond text. In this work, we systematically uncover properties present in LLMs that support the emergence of ICL for autoregressive models and various modalities by promoting the learning of the needed mechanisms for ICL. We identify exact token repetitions in the training data sequences as an important factor for ICL. Such repetitions further improve stability and reduce transiency in ICL performance. Moreover, we emphasise the significance of training task difficulty for the emergence of ICL. Finally, by applying our novel insights on ICL emergence, we unlock ICL capabilities for various visual datasets and a more challenging EEG classification task.
Authors: Jian Cheng Wong, Abhishek Gupta, Chin Chun Ooi, Pao-Hsiung Chiu, Jiao Liu, Yew-Soon Ong
Abstract: Deep learning models trained on finite data lack a complete understanding of the physical world. On the other hand, physics-informed neural networks (PINNs) are infused with such knowledge through the incorporation of mathematically expressible laws of nature into their training loss function. By complying with physical laws, PINNs provide advantages over purely data-driven models in limited-data regimes and present as a promising route towards Physical AI. This feature has propelled them to the forefront of scientific machine learning, a domain characterized by scarce and costly data. However, the vision of accurate physics-informed learning comes with significant challenges. This work examines PINNs in terms of model optimization and generalization, shedding light on the need for new algorithmic advances to overcome issues pertaining to the training speed, precision, and generalizability of today's PINN models. Of particular interest are gradient-free evolutionary algorithms (EAs) for optimizing the uniquely complex loss landscapes arising in PINN training. Methods synergizing gradient descent and EAs for discovering bespoke neural architectures and balancing multiple terms in physics-informed learning objectives are positioned as important avenues for future research. Another exciting track is to cast EAs as a meta-learner of generalizable PINN models. To substantiate these proposed avenues, we further highlight results from recent literature to showcase the early success of such approaches in addressing the aforementioned challenges in PINN optimization and generalization.
Authors: Le Chen, Dahu Feng, Erhu Feng, Yingrui Wang, Rong Zhao, Yubin Xia, Pinjie Xu, Haibo Chen
Abstract: With the rapid advancement of artificial intelligence technologies such as ChatGPT, AI agents, and video generation, contemporary mobile systems have begun integrating these AI capabilities on local devices to enhance privacy and reduce response latency. To meet the computational demands of AI tasks, current mobile SoCs are equipped with diverse AI accelerators, including GPUs and Neural Processing Units (NPUs). However, there has not been a comprehensive characterization of these heterogeneous processors, and existing designs typically only leverage a single AI accelerator for LLM inference, leading to suboptimal use of computational resources and memory bandwidth. In this paper, we first summarize key performance characteristics of heterogeneous processors, SoC memory bandwidth, etc. Drawing on these observations, we propose different heterogeneous parallel mechanisms to fully exploit both GPU and NPU computational power and memory bandwidth. We further design a fast synchronization mechanism between heterogeneous processors that leverages the unified memory architecture. By employing these techniques, we present HeteroInfer, the fastest LLM inference engine in mobile devices which supports GPU-NPU heterogeneous execution. Evaluation shows that HeteroInfer delivers a 1.34x to 6.02x end-to-end speedup over state-of-the-art GPU-only and NPU-only LLM engines, while maintaining negligible interference with other applications.
Authors: Yifan Shen, Peiyuan Zhu, Zijian Li, Shaoan Xie, Namrata Deka, Zongfang Liu, Zeyu Tang, Guangyi Chen, Kun Zhang
Abstract: Controllable video generation remains a significant challenge, despite recent advances in generating high-quality and consistent videos. Most existing methods for controlling video generation treat the video as a whole, neglecting intricate fine-grained spatiotemporal relationships, which limits both control precision and efficiency. In this paper, we propose Controllable Video Generative Adversarial Networks (CoVoGAN) to disentangle the video concepts, thus facilitating efficient and independent control over individual concepts. Specifically, following the minimal change principle, we first disentangle static and dynamic latent variables. We then leverage the sufficient change property to achieve component-wise identifiability of dynamic latent variables, enabling disentangled control of video generation. To establish the theoretical foundation, we provide a rigorous analysis demonstrating the identifiability of our approach. Building on these theoretical insights, we design a Temporal Transition Module to disentangle latent dynamics. To enforce the minimal change principle and sufficient change property, we minimize the dimensionality of latent dynamic variables and impose temporal conditional independence. To validate our approach, we integrate this module as a plug-in for GANs. Extensive qualitative and quantitative experiments on various video generation benchmarks demonstrate that our method significantly improves generation quality and controllability across diverse real-world scenarios.
Authors: Jairo Diaz-Rodriguez
Abstract: We introduce k-NLPmeans and k-LLMmeans, text-clustering variants of k-means that periodically replace numeric centroids with textual summaries. The key idea, summary-as-centroid, retains k-means assignments in embedding space while producing human-readable, auditable cluster prototypes. The method is LLM-optional: k-NLPmeans uses lightweight, deterministic summarizers, enabling offline, low-cost, and stable operation; k-LLMmeans is a drop-in upgrade that uses an LLM for summaries under a fixed per-iteration budget whose cost does not grow with dataset size. We also present a mini-batch extension for real-time clustering of streaming text. Across diverse datasets, embedding models, and summarization strategies, our approach consistently outperforms classical baselines and approaches the accuracy of recent LLM-based clustering-without extensive LLM calls. Finally, we provide a case study on sequential text streams and release a StackExchange-derived benchmark for evaluating streaming text clustering.
Authors: Giorgio Franceschelli, Mirco Musolesi
Abstract: Despite their growing capabilities, language models still frequently reproduce content from their training data, generate repetitive text, and favor common grammatical patterns and vocabulary. A possible cause is the decoding strategy: the most common strategies either consider only the most probable tokens, which reduces output diversity, or increase the likelihood of unlikely tokens, compromising output accuracy and correctness. In this paper, we propose DiffSampling, a new decoding method that leverages a mathematical analysis of the token probability distribution to ensure the generation of contextually appropriate text. In particular, the difference between consecutive, sorted probabilities can be used to truncate incorrect tokens. In addition, we also propose two variations of the proposed method that aim to correct the subtle inconsistencies of common sampling strategies. Experiments involving four different text-generation tasks demonstrate that our approach consistently performs at least on par with the existing methods it builds upon in terms of quality, while potentially improving output diversity.
Authors: Amirreza Neshaei Moghaddam, Alex Olshevsky, Bahman Gharesifard
Abstract: Inspired by REINFORCE, we introduce a novel receding-horizon algorithm for the Linear Quadratic Regulator (LQR) problem with unknown dynamics. Unlike prior methods, our algorithm avoids reliance on two-point gradient estimates while maintaining the same order of sample complexity. Furthermore, it eliminates the restrictive requirement of starting with a stable initial policy, broadening its applicability. Beyond these improvements, we introduce a refined analysis of error propagation through the contraction of the Riccati operator under the Riemannian distance. This refinement leads to a better sample complexity and ensures improved convergence guarantees.
Authors: Tamer Ghattas, Michael Hassid, Roy Schwartz
Abstract: Recent work proposed state-space models (SSMs) as an efficient alternative to transformer-based LLMs. Can these models be pruned to further reduce their computation costs? We adapt several pruning methods to the SSM structure, and apply them to four SSM-based LLMs across multiple tasks. We find that such models are quite robust to some pruning methods (e.g. WANDA), while using other methods lead to fast performance degradation.
Authors: Yu Fu, Michael Stanley Smith, Anastasios Panagiotelis
Abstract: The key to VI is the selection of a tractable density to approximate the Bayesian posterior. For large and complex models a common choice is to assume independence between multivariate blocks in a partition of the parameter space. While this simplifies the problem it can reduce accuracy. This paper proposes using vector copulas to capture dependence between the blocks parsimoniously. Tailored multivariate marginals are constructed using learnable transport maps. We call the resulting joint distribution a ``dependent block posterior'' approximation. Vector copula models are suggested that make tractable and flexible variational approximations. They allow for differing marginals, numbers of blocks, block sizes and forms of between block dependence. They also allow for solution of the variational optimization using efficient stochastic gradient methods. The approach is demonstrated using four different statistical models and 16 datasets which have posteriors that are challenging to approximate. This includes models that use global-local shrinkage priors for regularization, and hierarchical models for smoothing and heteroscedastic time series. In all cases, our method produces more accurate posterior approximations than benchmark VI methods that either assume block independence or factor-based dependence, at limited additional computational cost. A python package implementing the method is available on GitHub at https://github.com/YuFuOliver/VCVI_Rep_PyPackage.
Authors: Zhenhua Liu, Lijun Li, Ruizhe Chen, Yuxian Jiang, Tong Zhu, Zhaochen Su, Wenliang Chen, Jing Shao
Abstract: While guided decoding, especially value-guided methods, has emerged as a cost-effective alternative for controlling language model outputs without re-training models, its effectiveness is limited by the accuracy of the value function. We identify that this inaccuracy stems from a core distributional gap: existing methods train static value functions on trajectories sampled exclusively from the base policy, which inherently confines their training to a narrow and suboptimal view of the potential output space. We propose Iterative Value Refinement, a novel framework designed to bridge this gap. It employs Value Exploration to provide a more comprehensive and robust training signal, complemented by Iterative Self-Refinement, which uses the improved value function from one iteration to guide the generation of higher-quality data for the next. Extensive experiments on text summarization, multi-turn dialogue, and instruction following demonstrate the effectiveness of our framework in aligning language models. Our approach not only achieves alignment but also significantly reduces computational costs by leveraging principled value function optimization for efficient and effective control.
Authors: Bernd Prostmaier, Jan V\'avra, Bettina Gr\"un, Paul Hofmarcher
Abstract: Topic models are widely used for discovering latent thematic structures in large text corpora, yet traditional unsupervised methods often struggle to align with pre-defined conceptual domains. This paper introduces seeded Poisson Factorization (SPF), a novel approach that extends the Poisson Factorization (PF) framework by incorporating domain knowledge through seed words. SPF enables a structured topic discovery by modifying the prior distribution of topic-specific term intensities, assigning higher initial rates to pre-defined seed words. The model is estimated using variational inference with stochastic gradient optimization, ensuring scalability to large datasets. We present in detail the results of applying SPF to an Amazon customer feedback dataset, leveraging pre-defined product categories as guiding structures. SPF achieves superior performance compared to alternative guided probabilistic topic models in terms of computational efficiency and classification performance. Robustness checks highlight SPF's ability to adaptively balance domain knowledge and data-driven topic discovery, even in case of imperfect seed word selection. Further applications of SPF to four additional benchmark datasets, where the corpus varies in size and the number of topics differs, demonstrate its general superior classification performance compared to the unseeded PF model.
Authors: Sushil Mahavir Varma, Ir\`ene Waldspurger, Laurent Massouli\'e
Abstract: We consider the graph alignment problem, wherein the objective is to find a vertex correspondence between two graphs that maximizes the edge overlap. The graph alignment problem is an instance of the quadratic assignment problem (QAP), known to be NP-hard in the worst case even to approximately solve. In this paper, we analyze Birkhoff relaxation, a tight convex relaxation of QAP, and present theoretical guarantees on its performance when the inputs follow the Gaussian Wigner Model. More specifically, the weighted adjacency matrices are correlated Gaussian Orthogonal Ensemble with correlation $1/\sqrt{1+\sigma^2}$. Denote the optimal solutions of the QAP and Birkhoff relaxation by $\Pi^\star$ and $X^\star$ respectively. We show that $\|X^\star-\Pi^\star\|_F^2 = o(n)$ when $\sigma = o(n^{-1.25})$ and $\|X^\star-\Pi^\star\|_F^2 = \Omega(n)$ when $\sigma = \Omega(n^{-0.5})$. Thus, the optimal solution $X^\star$ transitions from a small perturbation of $\Pi^\star$ for small $\sigma$ to being well separated from $\Pi^\star$ as $\sigma$ becomes larger than $n^{-0.5}$. This result allows us to guarantee that simple rounding procedures on $X^\star$ align $1-o(1)$ fraction of vertices correctly whenever $\sigma = o(n^{-1.25})$. This condition on $\sigma$ to ensure the success of the Birkhoff relaxation is state-of-the-art.
Authors: Chikai Shang, Mengke Li, Yiqun Zhang, Zhen Chen, Jinlin Wu, Fangqing Gu, Yang Lu, Yiu-ming Cheung
Abstract: Visual prompt tuning (VPT), i.e., fine-tuning some lightweight prompt tokens, provides an efficient and effective approach for adapting pre-trained models to various downstream tasks. However, most prior art indiscriminately uses a fixed prompt distribution across different tasks, neglecting the importance of each block varying depending on the task. In this paper, we introduce adaptive distribution optimization (ADO) by tackling two key questions: (1) How to appropriately and formally define ADO, and (2) How to design an adaptive distribution strategy guided by this definition? Through empirical analysis, we first confirm that properly adjusting the distribution significantly improves VPT performance, and further uncover a key insight that a nested relationship exists between ADO and VPT. Based on these findings, we propose a new VPT framework, termed PRO-VPT (iterative Prompt RelOcation-based VPT), which adaptively adjusts the distribution built upon a nested optimization formulation. Specifically, we develop a prompt relocation strategy derived from this formulation, comprising two steps: pruning idle prompts from prompt-saturated blocks, followed by allocating these prompts to the most prompt-needed blocks. By iteratively performing prompt relocation and VPT, our proposal can adaptively learn the optimal prompt distribution in a nested optimization-based manner, thereby unlocking the full potential of VPT. Extensive experiments demonstrate that our proposal significantly outperforms advanced VPT methods, e.g., PRO-VPT surpasses VPT by 1.6 pp and 2.0 pp average accuracy, leading prompt-based methods to state-of-the-art performance on VTAB-1k and FGVC benchmarks. The code is available at https://github.com/ckshang/PRO-VPT.
Authors: Zhen Zou, Feng Zhao
Abstract: Diffusion Transformer (DiT) has exhibited impressive generation capabilities but faces great challenges due to its high computational complexity. To address this issue, various methods, notably feature caching, have been introduced. However, these approaches focus on aligning non-cache diffusion without analyzing why caching damage the generation processes. In this paper, we first confirm that the cache greatly amplifies the exposure bias, resulting in a decline in the generation quality. However, directly applying noise scaling is challenging for this issue due to the non-smoothness of exposure bias. We found that this phenomenon stems from the mismatch between its frequency response characteristics and the simple cache of Attention and MLP. Since these two components exhibit unique preferences for frequency signals, which provides us with a caching strategy to separate Attention and MLP to achieve an enhanced fit of exposure bias and reduce it. Based on this, we introduced FEB-Cache, a joint caching strategy that aligns with the non-exposed bias diffusion process (which gives us a higher performance cap) of caching Attention and MLP based on the frequency-guided cache table. Our approach combines a comprehensive understanding of the caching mechanism and offers a new perspective on leveraging caching to accelerate the diffusion process. Empirical results indicate that FEB-Cache optimizes model performance while concurrently facilitating acceleration. Code is available at https://github.com/aSleepyTree/EB-Cache.
Authors: Yves-Simon Zeulner, Sandeep Selvaraj, Roberto Calandra
Abstract: Towards the grand challenge of achieving human-level manipulation in robots, playing piano is a compelling testbed that requires strategic, precise, and flowing movements. Over the years, several works demonstrated hand-designed controllers on real world piano playing, while other works evaluated robot learning approaches on simulated piano scenarios. In this paper, we develop the first piano playing robotic system that makes use of learning approaches while also being deployed on a real world dexterous robot. Specifically, we make use of Sim2Real to train a policy in simulation using reinforcement learning before deploying the learned policy on a real world dexterous robot. In our experiments, we thoroughly evaluate the interplay between domain randomization and the accuracy of the dynamics model used in simulation. Moreover, we evaluate the robot's performance across multiple songs with varying complexity to study the generalization of our learned policy. By providing a proof-of-concept of learning to play piano in the real world, we want to encourage the community to adopt piano playing as a compelling benchmark towards human-level manipulation. We open-source our code and show additional videos at https://lasr.org/research/learning-to-play-piano .
Authors: Alhasan Abdellatif, Hannah P. Menke, Julien Maes, Ahmed H. Elsheikh, Florian Doster
Abstract: Accurately capturing the complex interaction between CO2 and water in porous media at the pore scale is essential for various geoscience applications, including carbon capture and storage (CCS). We introduce a comprehensive dataset generated from high-fidelity numerical simulations to capture the intricate interaction between CO2 and water at the pore scale. The dataset consists of 624 2D samples, each of size 512x512 with a resolution of 35 {\mu}m, covering 100 time steps under a constant CO2 injection rate. It includes various levels of heterogeneity, represented by different grain sizes with random variation in spacing, offering a robust testbed for developing predictive models. This dataset provides high-resolution temporal and spatial information crucial for benchmarking machine learning models.
Authors: Jiaru Zou, Dongqi Fu, Sirui Chen, Xinrui He, Zihao Li, Yada Zhu, Jiawei Han, Jingrui He
Abstract: Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by integrating them with an external knowledge base to improve the answer relevance and accuracy. In real-world scenarios, beyond pure text, a substantial amount of knowledge is stored in tables, and user questions often require retrieving answers that are distributed across multiple tables. Retrieving knowledge from a table corpora (i.e., various individual tables) for a question remains nascent, at least, for (i) how to understand intra- and inter-table knowledge effectively, (ii) how to filter unnecessary tables and how to retrieve the most relevant tables efficiently, (iii) how to prompt LLMs to infer over the retrieval, (iv) how to evaluate the corresponding performance in a realistic setting. Facing the above challenges, in this paper, we first propose a table-corpora-aware RAG framework, named T-RAG, which consists of the hierarchical memory index, multi-stage retrieval, and graph-aware prompting for effective and efficient table knowledge retrieval and inference. Further, we first develop a multi-table question answering benchmark named MultiTableQA, which spans 3 different task types, 57,193 tables, and 23,758 questions in total, and the sources are all from real-world scenarios. Based on MultiTableQA, we did the holistic comparison over table retrieval methods, RAG methods, and table-to-graph representation learning methods, where T-RAG shows the leading accuracy, recall, and running time performance. Also, under T-RAG, we evaluate the inference ability upgrade of different LLMs. Code and Data are available at https://github.com/jiaruzouu/T-RAG
Authors: Luca Furieri, Sucheth Shenoy, Danilo Saccani, Andrea Martin, Giancarlo Ferrari-Trecate
Abstract: We introduce magnitude and direction (MAD) policies, a policy parameterization for reinforcement learning (RL) that preserves Lp closed-loop stability for nonlinear dynamical systems. Despite their completeness in describing all stabilizing controllers, methods based on nonlinear Youla and system-level synthesis are significantly impacted by the difficulty of parametrizing Lp-stable operators. In contrast, MAD policies introduce explicit feedback on state-dependent features - a key element behind the success of reinforcement learning pipelines - without jeopardizing closed-loop stability. This is achieved by letting the magnitude of the control input be described by a disturbance-feedback Lp-stable operator, while selecting its direction based on state-dependent features through a universal function approximator. We further characterize the robust stability properties of MAD policies under model mismatch. Unlike existing disturbance-feedback policy parametrizations, MAD policies introduce state-feedback components compatible with model-free RL pipelines, ensuring closed-loop stability with no model information beyond assuming open-loop stability. Numerical experiments show that MAD policies trained with deep deterministic policy gradient (DDPG) methods generalize to unseen scenarios - matching the performance of standard neural network policies while guaranteeing closed-loop stability by design.
Authors: Dang Nguyen, Chenhao Tan
Abstract: Understanding and mitigating biases is critical for the adoption of large language models (LLMs) in high-stakes decision-making. We introduce Admissions and Hiring, decision tasks with hypothetical applicant profiles where a person's race can be inferred from their name, as simplified test beds for racial bias. We show that Gemma 2B Instruct and LLaMA 3.2 3B Instruct exhibit strong biases. Gemma grants admission to 26% more White than Black applicants, and LLaMA hires 60% more Asian than White applicants. We demonstrate that these biases are resistant to prompt engineering: multiple prompting strategies all fail to promote fairness. In contrast, using distributed alignment search, we can identify "race subspaces" within model activations and intervene on them to debias model decisions. Averaging the representation across all races within the subspaces reduces Gemma's bias by 37-57%. Finally, we examine the generalizability of Gemma's race subspaces, and find limited evidence for generalization, where changing the prompt format can affect the race representation. Our work suggests mechanistic approaches may provide a promising venue for improving the fairness of LLMs, but a universal race representation remains elusive.
Authors: Canberk Ekmekci, Mujdat Cetin
Abstract: Uncertainty quantification plays an important role in achieving trustworthy and reliable learning-based computational imaging. Recent advances in generative modeling and Bayesian neural networks have enabled the development of uncertainty-aware image reconstruction methods. Current generative model-based methods seek to quantify the inherent (aleatoric) uncertainty on the underlying image for given measurements by learning to sample from the posterior distribution of the underlying image. On the other hand, Bayesian neural network-based approaches aim to quantify the model (epistemic) uncertainty on the parameters of a deep neural network-based reconstruction method by approximating the posterior distribution of those parameters. Unfortunately, an ongoing need for an inversion method that can jointly quantify complex aleatoric uncertainty and epistemic uncertainty patterns still persists. In this paper, we present a scalable framework that can quantify both aleatoric and epistemic uncertainties. The proposed framework accepts an existing generative model-based posterior sampling method as an input and introduces an epistemic uncertainty quantification capability through Bayesian neural networks with latent variables and deep ensembling. Furthermore, by leveraging the conformal prediction methodology, the proposed framework can be easily calibrated to ensure rigorous uncertainty quantification. We evaluated the proposed framework on magnetic resonance imaging, computed tomography, and image inpainting problems and showed that the epistemic and aleatoric uncertainty estimates produced by the proposed framework display the characteristic features of true epistemic and aleatoric uncertainties. Furthermore, our results demonstrated that the use of conformal prediction on top of the proposed framework enables marginal coverage guarantees consistent with frequentist principles.
Authors: Chenxi Whitehouse, Tianlu Wang, Ping Yu, Xian Li, Jason Weston, Ilia Kulikov, Swarnadeep Saha
Abstract: The progress of AI is bottlenecked by the quality of evaluation, making powerful LLM-as-a-Judge models a core solution. The efficacy of these judges depends on their chain-of-thought reasoning, creating a critical need for methods that can effectively optimize this reasoning process. In this work, we introduce J1, a reinforcement learning framework for teaching LLM judges to think before making decisions. Our core contribution lies in converting all judgment tasks for non-verifiable and verifiable prompts into a unified format with verifiable rewards, enabling direct optimization of evaluation quality while mitigating positional bias. We then use RL to train thinking-judges at scales of 8B, 32B, and 70B and show that they obtain state-of-the-art performance across multiple benchmarks. In particular, J1-Qwen-32B, our multitasked pointwise and pairwise judge also outperforms o1-mini, o3, and a much larger 671B DeepSeek-R1 on some benchmarks, while only training on synthetic data. Through comprehensive ablations of pairwise, pointwise, and multitask J1 variants, we demonstrate the effectiveness of our approach across seed prompts, reward strategies, and training recipes. Qualitative analysis reveals that J1 develops systematic evaluation strategies, including dynamic criteria generation, reference answer creation, iterative self-correction of initial assessments, and feedback generation for low-quality responses.
Authors: Chuanqing Pu, Feilong Fan, Nengling Tai, Songyuan Liu, Jinming Yu
Abstract: Obtaining accurate probabilistic energy forecasts and making effective decisions amid diverse uncertainties are routine challenges in future energy systems. This paper presents the winning solution of team GEB, which ranked 3rd in trading, 4th in forecasting, and 1st among student teams in the IEEE Hybrid Energy Forecasting and Trading Competition 2024 (HEFTCom2024). The solution provides accurate probabilistic forecasts for a wind-solar hybrid system, and achieves substantial trading revenue in the day-ahead electricity market. Key components include: (1) a stacking-based approach combining sister forecasts from various Numerical Weather Predictions (NWPs) to provide wind power forecasts, (2) an online solar post-processing model to address the distribution shift in the online test set caused by increased solar capacity, (3) a probabilistic aggregation method for accurate quantile forecasts of hybrid generation, and (4) a stochastic trading strategy to maximize expected trading revenue considering uncertainties in electricity prices. This paper also explores the potential of end-to-end learning to further enhance the trading revenue by shifting the distribution of forecast errors. Detailed case studies are provided to validate the effectiveness of these proposed methods. Code for all mentioned methods is available for reproduction and further research in both industry and academia.
Authors: Jiahao Yu, Haozhuang Liu, Yeqiu Yang, Lu Chen, Jian Wu, Yuning Jiang, Bo Zheng
Abstract: Regression models are crucial in recommender systems. However, retransformation bias problem has been conspicuously neglected within the community. While many works in other fields have devised effective bias correction methods, all of them are post-hoc cures externally to the model, facing practical challenges when applied to real-world recommender systems. Hence, we propose a preemptive paradigm to eradicate the bias intrinsically from the models via minor model refinement. Specifically, a novel TranSUN method is proposed with a joint bias learning manner to offer theoretically guaranteed unbiasedness under empirical superior convergence. It is further generalized into a novel generic regression model family, termed Generalized TranSUN (GTS), which not only offers more theoretical insights but also serves as a generic framework for flexibly developing various bias-free models. Comprehensive experimental results demonstrate the superiority of our methods across data from various domains, which have been successfully deployed in two real-world industrial recommendation scenarios, i.e. product and short video recommendation scenarios in Guess What You Like business domain in the homepage of Taobao App (a leading e-commerce platform with DAU > 300M), to serve the major online traffic.
Authors: Simmaco Di Lillo
Abstract: This work investigates the expected number of critical points of random neural networks with different activation functions as the depth increases in the infinite-width limit. Under suitable regularity conditions, we derive precise asymptotic formulas for the expected number of critical points of fixed index and those exceeding a given threshold. Our analysis reveals three distinct regimes depending on the value of the first derivative of the covariance evaluated at 1: the expected number of critical points may converge, grow polynomially, or grow exponentially with depth. The theoretical predictions are supported by numerical experiments. Moreover, we provide numerical evidence suggesting that, when the regularity condition is not satisfied (e.g. for neural networks with ReLU as activation function), the number of critical points increases as the map resolution increases, indicating a potential divergence in the number of critical points.
Authors: Santiago Acevedo, Andrea Mascaretti, Riccardo Rende, Mat\'eo Mahaut, Marco Baroni, Alessandro Laio
Abstract: Deep neural networks are known to develop similar representations for semantically related data, even when they belong to different domains, such as an image and its description, or the same text in different languages. We present a method for quantitatively investigating this phenomenon by measuring the relative information content of the representations of semantically related data and probing how it is encoded into multiple tokens of large language models (LLMs) and vision transformers. Looking first at how LLMs process pairs of translated sentences, we identify inner ``semantic'' layers containing the most language-transferable information. We find moreover that, on these layers, a larger LLM (DeepSeek-V3) extracts significantly more general information than a smaller one (Llama3.1-8B). Semantic information of English text is spread across many tokens and it is characterized by long-distance correlations between tokens and by a causal left-to-right (i.e., past-future) asymmetry. We also identify layers encoding semantic information within visual transformers. We show that caption representations in the semantic layers of LLMs predict visual representations of the corresponding images. We observe significant and model-dependent information asymmetries between image and text representations.
Authors: Jiachen Jiang, Yuxin Dong, Jinxin Zhou, Zhihui Zhu
Abstract: In-context learning (ICL) enables large language models (LLMs) to adapt to new tasks without weight updates by learning from demonstration sequences. While ICL shows strong empirical performance, its internal representational mechanisms are not yet well understood. In this work, we conduct a statistical geometric analysis of ICL representations to investigate how task-specific information is captured across layers. Our analysis reveals an intriguing phenomenon, which we term *Layerwise Compression-Expression*: early layers progressively produce compact and discriminative representations that encode task information from the input demonstrations, while later layers express these representations to incorporate the query and generate the prediction. This phenomenon is observed consistently across diverse tasks and a range of contemporary LLM architectures. We demonstrate that it has important implications for ICL performance -- improving with model size and the number of demonstrations -- and for robustness in the presence of noisy examples. To further understand the effect of the compact task representation, we propose a bias-variance decomposition and provide a theoretical analysis showing how attention mechanisms contribute to reducing both variance and bias, thereby enhancing performance as the number of demonstrations increases. Our findings reveal an intriguing layerwise dynamic in ICL, highlight how structured representations emerge within LLMs, and showcase that analyzing internal representations can facilitate a deeper understanding of model behavior.
Authors: Xingjian Li, Qifeng Wu, Adithya S. Ubaradka, Yiran Ding, Colleen Que, Runmin Jiang, Jianhua Xing, Tianyang Wang, Min Xu
Abstract: Medical image segmentation is vital for clinical diagnosis, yet current deep learning methods often demand extensive expert effort, i.e., either through annotating large training datasets or providing prompts at inference time for each new case. This paper introduces a zero-shot and automatic segmentation pipeline that combines off-the-shelf vision-language and segmentation foundation models. Given a medical image and a task definition (e.g., "segment the optic disc in an eye fundus image"), our method uses a grounding model to generate an initial bounding box, followed by a visual prompt boosting module that enhance the prompts, which are then processed by a promptable segmentation model to produce the final mask. To address the challenges of domain gap and result verification, we introduce a test-time adaptation framework featuring a set of learnable adaptors that align the medical inputs with foundation model representations. Its hyperparameters are optimized via Bayesian Optimization, guided by a proxy validation model without requiring ground-truth labels. Our pipeline offers an annotation-efficient and scalable solution for zero-shot medical image segmentation across diverse tasks. Our pipeline is evaluated on seven diverse medical imaging datasets and shows promising results. By proper decomposition and test-time adaptation, our fully automatic pipeline not only substantially surpasses the previously best-performing method, yielding a 69\% relative improvement in accuracy (Dice Score from 42.53 to 71.81), but also performs competitively with weakly-prompted interactive foundation models.
Authors: Futoshi Futami, Atsushi Nitanda
Abstract: Calibration is a critical requirement for reliable probabilistic prediction, especially in high-risk applications. However, the theoretical understanding of which learning algorithms can simultaneously achieve high accuracy and good calibration remains limited, and many existing studies provide empirical validation or a theoretical guarantee in restrictive settings. To address this issue, in this work, we focus on the smooth calibration error (CE) and provide a uniform convergence bound, showing that the smooth CE is bounded by the sum of the smooth CE over the training dataset and a generalization gap. We further prove that the functional gradient of the loss function can effectively control the training smooth CE. Based on this framework, we analyze three representative algorithms: gradient boosting trees, kernel boosting, and two-layer neural networks. For each, we derive conditions under which both classification and calibration performances are simultaneously guaranteed. Our results offer new theoretical insights and practical guidance for designing reliable probabilistic models with provable calibration guarantees.
Authors: Dang Nguyen, Jiping Li, Jinghao Zheng, Baharan Mirzasoleiman
Abstract: Synthetically augmenting training datasets with diffusion models has been an effective strategy for improving generalization of image classifiers. However, existing techniques struggle to ensure the diversity of generation and increase the size of the data by up to 10-30x to improve the in-distribution performance. In this work, we show that synthetically augmenting part of the data that is not learned early in training with faithful images-containing same features but different noise-outperforms augmenting the entire dataset. By analyzing a two-layer CNN, we prove that this strategy improves generalization by promoting homogeneity in feature learning speed without amplifying noise. Our extensive experiments show that by augmenting only 30%-40% of the data, our method boosts generalization by up to 2.8% in a variety of scenarios, including training ResNet, ViT, ConvNeXt, and Swin Transformer on CIFAR-10/100, and TinyImageNet, with various optimizers including SGD and SAM. Notably, our method applied with SGD outperforms the SOTA optimizer, SAM, on CIFAR-100 and TinyImageNet.
Authors: Agnideep Aich, Md Monzur Murshed, Sameera Hewage, Amanda Mayeaux
Abstract: Effective feature selection is vital for robust and interpretable medical prediction, especially for identifying risk factors concentrated in extreme patient strata. Standard methods emphasize average associations and may miss predictors whose importance lies in the tails of the distribution. We propose a computationally efficient supervised filter that ranks features using the Gumbel copula upper tail dependence coefficient ($\lambda_U$), prioritizing variables that are simultaneously extreme with the positive class. We benchmarked against Mutual Information, mRMR, ReliefF, and $L_1$ Elastic Net across four classifiers on two diabetes datasets: a large public health survey (CDC, N=253,680) and a clinical benchmark (PIMA, N=768). Evaluation included paired statistical tests, permutation importance, and robustness checks with label flips, feature noise, and missingness. On CDC, our method was the fastest selector and reduced the feature space by about 52% while retaining strong discrimination. Although using all 21 features yielded the highest AUC, our filter significantly outperformed Mutual Information and mRMR and was statistically indistinguishable from ReliefF. On PIMA, with only eight predictors, our ranking produced the numerically highest ROC AUC, and no significant differences were found versus strong baselines. Across both datasets, the upper tail criterion consistently identified clinically coherent, impactful predictors. We conclude that copula based feature selection via upper tail dependence is a powerful, efficient, and interpretable approach for building risk models in public health and clinical medicine.
Authors: Dipam Goswami, Liying Wang, Bart{\l}omiej Twardowski, Joost van de Weijer
Abstract: Text embedding models enable semantic search, powering several NLP applications like Retrieval Augmented Generation by efficient information retrieval (IR). However, text embedding models are commonly studied in scenarios where the training data is static, thus limiting its applications to dynamic scenarios where new training data emerges over time. IR methods generally encode a huge corpus of documents to low-dimensional embeddings and store them in a database index. During retrieval, a semantic search over the corpus is performed and the document whose embedding is most similar to the query embedding is returned. When updating an embedding model with new training data, using the already indexed corpus is suboptimal due to the non-compatibility issue, since the model which was used to obtain the embeddings of the corpus has changed. While re-indexing of old corpus documents using the updated model enables compatibility, it requires much higher computation and time. Thus, it is critical to study how the already indexed corpus can still be effectively used without the need of re-indexing. In this work, we establish a continual learning benchmark with large-scale datasets and continually train dense retrieval embedding models on query-document pairs from new datasets in each task and observe forgetting on old tasks due to significant drift of embeddings. We employ embedding distillation on both query and document embeddings to maintain stability and propose a novel query drift compensation method during retrieval to project new model query embeddings to the old embedding space. This enables compatibility with previously indexed corpus embeddings extracted using the old model and thus reduces the forgetting. We show that the proposed method significantly improves performance without any re-indexing. Code is available at https://github.com/dipamgoswami/QDC.
Authors: Ziyang Yu, Wenbing Huang, Yang Liu
Abstract: Molecular Dynamics (MD) simulations are essential for understanding the atomic-level behavior of molecular systems, giving insights into their transitions and interactions. However, classical MD techniques are limited by the trade-off between accuracy and efficiency, while recent deep learning-based improvements have mostly focused on single-domain molecules, lacking transferability to unfamiliar molecular systems. Therefore, we propose \textbf{Uni}fied \textbf{Sim}ulator (UniSim), which leverages cross-domain knowledge to enhance the understanding of atomic interactions. First, we employ a multi-head pretraining approach to learn a unified atomic representation model from a large and diverse set of molecular data. Then, based on the stochastic interpolant framework, we learn the state transition patterns over long timesteps from MD trajectories, and introduce a force guidance module for rapidly adapting to different chemical environments. Our experiments demonstrate that UniSim achieves highly competitive performance across small molecules, peptides, and proteins.
Authors: Swagat Kumar, Jan-Nico Zaech, Colin Michael Wilmott, Luc Van Gool
Abstract: Variational Quantum Algorithms (VQAs) are a promising approach to leverage Noisy Intermediate-Scale Quantum (NISQ) computers. However, choosing optimal quantum circuits that efficiently solve a given VQA problem is a non-trivial task. Quantum Architecture Search (QAS) algorithms enable automatic generation of quantum circuits tailored to the provided problem. Existing QAS approaches typically adapt classical neural architecture search techniques, training machine learning models to sample relevant circuits, but often overlook the inherent quantum nature of the circuits they produce. By reformulating QAS from a quantum perspective, we propose a sampling-free differentiable QAS algorithm that models the search process as the evolution of a quantum mixed state, which emerges from the search space of quantum circuits. The mixed state formulation also enables our method to incorporate generic noise models, for example the depolarizing channel, which cannot be modeled by state vector simulation. We validate our method by finding circuits for state initialization and Hamiltonian optimization tasks, namely the variational quantum eigensolver and the unweighted max-cut problems. We show our approach to be comparable to, if not outperform, existing QAS techniques while requiring significantly fewer quantum simulations during training, and also show improved robustness levels to noise.
Authors: Guangchen Lan, Huseyin A. Inan, Sahar Abdelnabi, Janardhan Kulkarni, Lukas Wutschitz, Reza Shokri, Christopher G. Brinton, Robert Sim
Abstract: As the era of autonomous agents making decisions on behalf of users unfolds, ensuring contextual integrity (CI) -- what is the appropriate information to share while carrying out a certain task -- becomes a central question to the field. We posit that CI demands a form of reasoning where the agent needs to reason about the context in which it is operating. To test this, we first prompt LLMs to reason explicitly about CI when deciding what information to disclose. We then extend this approach by developing a reinforcement learning (RL) framework that further instills in models the reasoning necessary to achieve CI. Using a synthetic, automatically created, dataset of only $\sim700$ examples but with diverse contexts and information disclosure norms, we show that our method substantially reduces inappropriate information disclosure while maintaining task performance across multiple model sizes and families. Importantly, improvements transfer from this synthetic dataset to established CI benchmarks such as PrivacyLens that has human annotations and evaluates privacy leakage of AI assistants in actions and tool calls.
Authors: Ran Xu, Yuchen Zhuang, Yishan Zhong, Yue Yu, Zifeng Wang, Xiangru Tang, Hang Wu, May D. Wang, Peifeng Ruan, Donghan Yang, Tao Wang, Guanghua Xiao, Xin Liu, Carl Yang, Yang Xie, Wenqi Shi
Abstract: We introduce MedAgentGym, a scalable and interactive training environment designed to enhance coding-based biomedical reasoning capabilities in large language model (LLM) agents. MedAgentGym comprises 72,413 task instances across 129 categories derived from 12 authentic real-world biomedical scenarios. Tasks are encapsulated within executable sandbox environments, each featuring detailed task specifications, interactive feedback mechanisms, verifiable ground truth annotations, and scalable training trajectory generation. Extensive benchmarking of 29 LLMs reveals substantial performance disparities in biomedical data science between commercial and open-source LLMs. Leveraging efficient multi-threaded and multi-turn trajectory sampling in MedAgentGym, Med-Copilot achieves performance gains of +43.02% and +45.28% from offline and online reinforcement learning, respectively, demonstrating MedAgentGym as an effective training ground while establishing itself as a cost-effective, privacy-preserving alternative competitive with proprietary LLMs (gpt-4o). By offering a unified execution environment with a comprehensive benchmark and accessible, extensible training resources, MedAgentGym delivers an integrated platform to develop LLM-based coding assistants for advanced biomedical data science.
Authors: Kihyun Kim, Jiawei Zhang, Asuman Ozdaglar, Pablo A. Parrilo
Abstract: Conventional preference learning methods often prioritize opinions held more widely when aggregating preferences from multiple evaluators. This may result in policies that are biased in favor of some types of opinions or groups and susceptible to strategic manipulation. To address this issue, we develop a novel preference learning framework capable of aligning aggregate opinions and policies proportionally with the true population distribution of evaluator preferences. Grounded in social choice theory, our approach infers the feasible set of evaluator population distributions directly from pairwise comparison data. Using these estimates, the algorithm constructs a policy that satisfies foundational axioms from social choice theory, namely monotonicity and Pareto efficiency, as well as our newly-introduced axioms of population-proportional alignment and population-bounded manipulability. Moreover, we propose a soft-max relaxation method that smoothly trade-offs population-proportional alignment with the selection of the Condorcet winner (which beats all other options in pairwise comparisons). Finally, we validate the effectiveness and scalability of our approach through experiments on both tabular recommendation tasks and large language model alignment.
Authors: Lu Ma, Hao Liang, Meiyi Qiang, Lexiang Tang, Xiaochen Ma, Zhen Hao Wong, Junbo Niu, Chengyu Shen, Runming He, Yanhao Li, Bin Cui, Wentao Zhang
Abstract: Recent advances in large language model (LLM) reasoning have shown that sophisticated behaviors such as planning and self-reflection can emerge through reinforcement learning (RL). However, despite these successes, RL in its current form remains insufficient to induce capabilities that exceed the limitations of the base model, as it is primarily optimized based on existing knowledge of the model rather than facilitating the acquisition of new information. To address this limitation, we employ supervised fine-tuning (SFT) to learn what RL cannot, which enables the incorporation of new knowledge and reasoning patterns by leveraging high-quality demonstration data. We analyze the training dynamics of RL and SFT for LLM reasoning and find that RL excels at maintaining and improving performance on questions within the model's original capabilities, while SFT is more effective at enabling progress on questions beyond the current scope of the model. Motivated by the complementary strengths of RL and SFT, we introduce a novel training approach, \textbf{ReLIFT} (\textbf{Re}inforcement \textbf{L}earning \textbf{I}nterleaved with Online \textbf{F}ine-\textbf{T}uning). In ReLIFT, the model is primarily trained using RL, but when it encounters challenging questions, high-quality solutions are collected for fine-tuning, and the training process alternates between RL and fine-tuning to enhance the model's reasoning abilities. ReLIFT achieves an average improvement of over +5.2 points across five competition-level benchmarks and one out-of-distribution benchmark compared to other zero-RL models. Furthermore, we demonstrate that ReLIFT outperforms both RL and SFT while using only 13\% of the detailed demonstration data, highlighting its scalability. These results provide compelling evidence that ReLIFT overcomes the fundamental limitations of RL and underscores the significant potential.
Authors: Emmanuel J. Cand\`es, Andrew Ilyas, Tijana Zrnic
Abstract: Obtaining high-quality labeled datasets is often costly, requiring either human annotation or expensive experiments. In theory, powerful pre-trained AI models provide an opportunity to automatically label datasets and save costs. Unfortunately, these models come with no guarantees on their accuracy, making wholesale replacement of manual labeling impractical. In this work, we propose a method for leveraging pre-trained AI models to curate cost-effective and high-quality datasets. In particular, our approach results in probably approximately correct labels: with high probability, the overall labeling error is small. Our method is nonasymptotically valid under minimal assumptions on the dataset or the AI model being studied, and thus enables rigorous yet efficient dataset curation using modern AI models. We demonstrate the benefits of the methodology through text annotation with large language models, image labeling with pre-trained vision models, and protein folding analysis with AlphaFold.
Authors: Theofanis Aravanis
Abstract: Artificial Neural Networks (ANNs) are powerful machine-learning models capable of capturing intricate non-linear relationships. They are widely used nowadays across numerous scientific and engineering domains, driving advancements in both research and real-world applications. In our recent work, we focused on the statics and dynamics of a particular subclass of ANNs, which we refer to as binary ANNs. A binary ANN is a feed-forward network in which both inputs and outputs are restricted to binary values, making it particularly suitable for a variety of practical use cases. Our previous study approached binary ANNs through the lens of belief-change theory, specifically the Alchourron, Gardenfors and Makinson (AGM) framework, yielding several key insights. Most notably, we demonstrated that the knowledge embodied in a binary ANN (expressed through its input-output behaviour) can be symbolically represented using a propositional logic language. Moreover, the process of modifying a belief set (through revision or contraction) was mapped onto a gradual transition through a series of intermediate belief sets. Analogously, the training of binary ANNs was conceptualized as a sequence of such belief-set transitions, which we showed can be formalized using full-meet AGM-style belief change. In the present article, we extend this line of investigation by addressing some critical limitations of our previous study. Specifically, we show that Dalal's method for belief change naturally induces a structured, gradual evolution of states of belief. More importantly, given the known shortcomings of full-meet belief change, we demonstrate that the training dynamics of binary ANNs can be more effectively modelled using robust AGM-style change operations -- namely, lexicographic revision and moderate contraction -- that align with the Darwiche-Pearl framework for iterated belief change.
Authors: Hongyu H\`e, Minhao Jin, Maria Apostolaki
Abstract: Generative ML models are increasingly popular in networking for tasks such as telemetry imputation, prediction, and synthetic trace generation. Despite their capabilities, they suffer from two shortcomings: (i) their output is often visibly violating well-known networking rules, which undermines their trustworthiness; and (ii) they are difficult to control, frequently requiring retraining even for minor changes. To address these limitations and unlock the benefits of generative models for networking, we propose a new paradigm for integrating explicit network knowledge in the form of first-order logic rules into ML models used for networking tasks. Rules capture well-known relationships among used signals, e.g., that increased latency precedes packet loss. While the idea is conceptually straightforward, its realization is challenging: networking knowledge is rarely formalized into rules, and naively injecting them into ML models often hampers ML's effectiveness. This paper introduces NetNomos a multi-stage framework that (1) learns rules directly from data (e.g., measurements); (2) filters them to distinguish semantically meaningful ones; and (3) enforces them through a collaborative generation between an ML model and an SMT solver.
Authors: Ken Tsui
Abstract: Although large language models (LLMs) have transformed AI, they still make mistakes and can explore unproductive reasoning paths. Self-correction capability is essential for deploying LLMs in safety-critical applications. We uncover a systematic failure: LLMs cannot correct errors in their own outputs while successfully correcting identical errors from external sources - a limitation we term the Self-Correction Blind Spot. To study this phenomenon, we introduce Self-Correction Bench, an evaluation framework to measure this phenomenon through controlled error injection at three complexity levels. Testing 14 open-source non-reasoning models, we find an average 64.5% blind spot rate. We provide multiple lines of evidence suggesting this limitation may be influenced by training data: human demonstrations rarely include error-correction sequences (favoring error-free responses), whereas reinforcement learning (RL) trained models learn error correction via outcome feedback. Remarkably, appending a minimal "Wait" prompt activates a 89.3% reduction in blind spots, suggesting dormant capabilities that require triggering. Our work highlights a critical limitation potentially influenced by training distribution and offers a practical approach to enhance LLM reliability and trustworthiness - vital for safety-critical domains.
Authors: Tiffany Ding, Jean-Baptiste Fermanian, Joseph Salmon
Abstract: Many real-world classification problems, such as plant identification, have extremely long-tailed class distributions. In order for prediction sets to be useful in such settings, they should (i) provide good class-conditional coverage, ensuring that rare classes are not systematically omitted from the prediction sets, and (ii) be a reasonable size, allowing users to easily verify candidate labels. Unfortunately, existing conformal prediction methods, when applied to the long-tailed setting, force practitioners to make a binary choice between small sets with poor class-conditional coverage or sets with very good class-conditional coverage but that are extremely large. We propose methods with guaranteed marginal coverage that smoothly trade off between set size and class-conditional coverage. First, we introduce a new conformal score function called prevalence-adjusted softmax that targets macro-coverage, a relaxed notion of class-conditional coverage. Second, we propose a new procedure that interpolates between marginal and class-conditional conformal prediction by linearly interpolating their conformal score thresholds. We demonstrate our methods on Pl@ntNet-300K and iNaturalist-2018, two long-tailed image datasets with 1,081 and 8,142 classes, respectively.
Authors: Binxu Li, Minkai Xu, Jiaqi Han, Meihua Dang, Stefano Ermon
Abstract: Diffusion models have achieved remarkable success in generating realistic and versatile images from text prompts. Inspired by the recent advancements of language models, there is an increasing interest in further improving the models by aligning with human preferences. However, we investigate alignment from a divergence minimization perspective and reveal that existing preference optimization methods are typically trapped in suboptimal mean-seeking optimization. In this paper, we introduce Divergence Minimization Preference Optimization (DMPO), a novel and principled method for aligning diffusion models by minimizing reverse KL divergence, which asymptotically enjoys the same optimization direction as original RL. We provide rigorous analysis to justify the effectiveness of DMPO and conduct comprehensive experiments to validate its empirical strength across both human evaluations and automatic metrics. Our extensive results show that diffusion models fine-tuned with DMPO can consistently outperform or match existing techniques, specifically consistently outperforming all baseline models across different base models and test sets, achieving the best PickScore in every case, demonstrating the method's superiority in aligning generative behavior with desired outputs. Overall, DMPO unlocks a robust and elegant pathway for preference alignment, bridging principled theory with practical performance in diffusion models.
Authors: Varun Srivastava, Fan Lei, Srija Mukhopadhyay, Vivek Gupta, Ross Maciejewski
Abstract: Recent advancements in multimodal large language models (MLLMs) have driven researchers to explore how well these models read data visualizations, e.g., bar charts, scatter plots. More recently, attention has shifted to visual question answering with maps (Map-VQA). However, Map-VQA research has primarily focused on choropleth maps, which cover only a limited range of thematic categories and visual analytical tasks. To address these gaps, we introduce MapIQ, a benchmark dataset comprising 14,706 question-answer pairs across three map types: choropleth maps, cartograms, and proportional symbol maps spanning topics from six distinct themes (e.g., housing, crime). We evaluate multiple MLLMs using six visual analytical tasks, comparing their performance against one another and a human baseline. An additional experiment examining the impact of map design changes (e.g., altered color schemes, modified legend designs, and removal of map elements) provides insights into the robustness and sensitivity of MLLMs, their reliance on internal geographic knowledge, and potential avenues for improving Map-VQA performance.
Authors: Byeonghee Lee, Juhyun Park, Saebom Jeon, Joonsung Kang
Abstract: Outliers can severely distort causal effect estimation in observational studies, especially in small samples. We develop a doubly robust estimator of the ATE under a contaminated-data model that explicitly accommodates outliers. Robustness to outliers is delivered via a bounded-influence estimating equation for the outcome model and covariate balancing propensity scores (CBPS) for treatment assignment. To mitigate overfitting in high dimensions, we incorporate variable selection and unify all components within a penalized empirical likelihood framework. For further inference, we derive an optimal finite-sample confidence interval (CI) whose endpoints are invariant to outliers under the contaminated model. Across extensive simulations and two gene-expression applications (Golub; Khan pediatric tumor), the proposed ATE estimator and finite-sample CI outperform state-of-the-art competitors in bias, mean squared error, empirical coverage, and interval length over a wide range of contamination levels and sample sizes.
Authors: Marius Willner, Marco Trenti, Dirk Lebiedz
Abstract: Tree tensor networks (TTNs) are widely used in low-rank approximation and quantum many-body simulation. In this work, we present a formal analysis of the differential geometry underlying TTNs. Building on this foundation, we develop efficient first- and second-order optimization algorithms that exploit the intrinsic quotient structure of TTNs. Additionally, we devise a backpropagation algorithm for training TTNs in a kernel learning setting. We validate our methods through numerical experiments on a representative machine learning task.
Authors: Robert Kuchen
Abstract: Statistical learning methods for automated variable selection, such as LASSO, elastic nets, or gradient boosting, have become increasingly popular tools for building powerful prediction models. Yet, in practice, analyses are often complicated by missing data. The most widely used approach to address missingness is multiple imputation, which involves creating several completed datasets. However, there is an ongoing debate on how to perform model selection in the presence of multiple imputed datasets. Simple strategies, such as pooling models across datasets, have been shown to have suboptimal properties. Although more sophisticated methods exist, they are often difficult to implement and therefore not widely applied. In contrast, two recent approaches modify the regularization methods LASSO and elastic nets by defining a single loss function, resulting in a unified set of coefficients across imputations. Our key contribution is to extend this principle to the framework of component-wise gradient boosting by proposing MIBoost, a novel algorithm that employs a uniform variable-selection mechanism across imputed datasets. Simulation studies suggest that our approach yields prediction performance comparable to that of these recently proposed methods.
Authors: Jingze Shi, Yifan Wu, Yiran Peng, Bingheng Wu, Liangdong Wang, Guang Liu, Yuyu Luo
Abstract: In large language models, the demand for modeling long contexts is ever-increasing, yet the quadratic complexity of standard self-attention presents a significant bottleneck. While existing sparse attention mechanisms enhance efficiency, they often suffer from limitations such as static patterns and information loss. This paper introduces a Trainable Dynamic Mask Sparse Attention mechanism that addresses these challenges through three key innovations. First, it leverages value vectors to dynamically generate content-aware sparse masks, enabling the model to adaptively identify and focus on crucial information. Second, it implements a position-aware sparse attention computation that effectively skips unnecessary computational regions. Finally, we ensure that the introduced dynamic masks and sparse weights do not obstruct gradients, thereby supporting end-to-end training. This dual-sparsity design allows the model to retain complete information while significantly reducing computational complexity, achieving an excellent balance between efficiency and performance. We validate the performance of Dynamic Mask Attention through comprehensive experiments. Comparative studies demonstrate that our method consistently achieves Pareto dominance across various tasks, including scaling laws, multi-query associative recall, general benchmarks, and needle-in-a-haystack tests, delivering up to 10 times acceleration. These results highlight its capability to effectively balance model efficiency with long-context modeling. Our computational kernel is open-sourced at https://github.com/SmallDoges/flash-dmattn to facilitate further research and application within the community.
Authors: Leo Zhang, Saifuddin Syed
Abstract: In this work, we study the problem of choosing the discretisation schedule for sampling from masked discrete diffusion models in terms of the information geometry of the induced probability path. Specifically, we show that the optimal schedule under the Fisher-Rao geometry recovers the popularly-used cosine schedule.
Authors: Mandira Sawkar, Samay U. Shetty, Deepak Pandita, Tharindu Cyril Weerasooriya, Christopher M. Homan
Abstract: The Learning With Disagreements (LeWiDi) 2025 shared task aims to model annotator disagreement through soft label distribution prediction and perspectivist evaluation, which focuses on modeling individual annotators. We adapt DisCo (Distribution from Context), a neural architecture that jointly models item-level and annotator-level label distributions, and present detailed analysis and improvements. In this paper, we extend DisCo by introducing annotator metadata embeddings, enhancing input representations, and multi-objective training losses to capture disagreement patterns better. Through extensive experiments, we demonstrate substantial improvements in both soft and perspectivist evaluation metrics across three datasets. We also conduct in-depth calibration and error analyses that reveal when and why disagreement-aware modeling improves. Our findings show that disagreement can be better captured by conditioning on annotator demographics and by optimizing directly for distributional metrics, yielding consistent improvements across datasets.
Authors: Jun Liu, Zhenglun Kong, Pu Zhao, Weihao Zeng, Hao Tang, Xuan Shen, Changdi Yang, Wenbin Zhang, Geng Yuan, Wei Niu, Xue Lin, Yanzhi Wang
Abstract: Autonomous driving platforms encounter diverse driving scenarios, each with varying hardware resources and precision requirements. Given the computational limitations of embedded devices, it is crucial to consider computing costs when deploying on target platforms like the NVIDIA\textsuperscript{\textregistered} DRIVE PX 2. Our objective is to customize the semantic segmentation network according to the computing power and specific scenarios of autonomous driving hardware. We implement dynamic adaptability through a three-tier control mechanism -- width multiplier, classifier depth, and classifier kernel -- allowing fine-grained control over model components based on hardware constraints and task requirements. This adaptability facilitates broad model scaling, targeted refinement of the final layers, and scenario-specific optimization of kernel sizes, leading to improved resource allocation and performance. Additionally, we leverage Bayesian Optimization with surrogate modeling to efficiently explore hyperparameter spaces under tight computational budgets. Our approach addresses scenario-specific and task-specific requirements through automatic parameter search, accommodating the unique computational complexity and accuracy needs of autonomous driving. It scales its Multiply-Accumulate Operations (MACs) for Task-Specific Learning Adaptation (TSLA), resulting in alternative configurations tailored to diverse self-driving tasks. These TSLA customizations maximize computational capacity and model accuracy, optimizing hardware utilization.
Authors: Pranjal Aggarwal, Seungone Kim, Jack Lanchantin, Sean Welleck, Jason Weston, Ilia Kulikov, Swarnadeep Saha
Abstract: Thinking LLMs solve complex tasks at the expense of increased compute and overthinking on simpler problems, while non-thinking LLMs are faster and cheaper but underthink on harder reasoning problems. This has led to the development of separate thinking and non-thinking LLM variants, leaving the onus of selecting the optimal model for each query on the end user. We introduce OptimalThinkingBench, a unified benchmark that jointly evaluates overthinking and underthinking in LLMs and also encourages the development of optimally-thinking models that balance performance and efficiency. Our benchmark comprises two sub-benchmarks: OverthinkingBench, featuring simple math and general queries in 72 domains, and UnderthinkingBench, containing 11 challenging reasoning tasks along with harder math problems. Using novel thinking-adjusted accuracy metrics, we extensively evaluate 33 different thinking and non-thinking models and show that no model is able to optimally think on our benchmark. Thinking models often overthink for hundreds of tokens on the simplest user queries without improving performance. In contrast, large non-thinking models underthink, often falling short of much smaller thinking models. We further explore several methods to encourage optimal thinking, but find that these approaches often improve on one sub-benchmark at the expense of the other, highlighting the need for better unified and optimal models in the future.
Authors: Wu Lin, Scott C. Lowe, Felix Dangel, Runa Eschenhagen, Zikun Xu, Roger B. Grosse
Abstract: Shampoo and its efficient variant, SOAP, employ structured second-moment estimations and have shown strong performance for training neural networks (NNs). In practice, however, Shampoo typically requires step-size grafting with Adam to be competitive, and SOAP mitigates this by applying Adam in Shampoo's eigenbasis -- at the cost of additional memory overhead from Adam in both methods. Prior analyses have largely relied on the Frobenius norm to motivate these estimation schemes. We instead recast their estimation procedures as covariance estimation under Kullback-Leibler (KL) divergence minimization, revealing a previously overlooked theoretical limitation and motivating principled redesigns. Building on this perspective, we develop $\textbf{KL-Shampoo}$ and $\textbf{KL-SOAP}$, practical schemes that match or exceed the performance of Shampoo and SOAP in NN pre-training while achieving SOAP-level per-iteration runtime. Notably, KL-Shampoo does not rely on Adam to attain competitive performance, eliminating the memory overhead introduced by Adam. Across our experiments, KL-Shampoo consistently outperforms SOAP, Shampoo, and even KL-SOAP, establishing the KL-based approach as a compelling foundation for designing structured methods in NN optimization.
Authors: Matthew Ho, Chen Si, Zhaoxiang Feng, Fangxu Yu, Yichi Yang, Zhijian Liu, Zhiting Hu, Lianhui Qin
Abstract: While inference-time scaling enables LLMs to carry out increasingly long and capable reasoning traces, the patterns and insights uncovered during these traces are immediately discarded once the context window is reset for a new query. External memory is a natural way to persist these discoveries, and recent work has shown clear benefits for reasoning-intensive tasks. We see an opportunity to make such memories more broadly reusable and scalable by moving beyond instance-based memory entries (e.g. exact query/response pairs, or summaries tightly coupled with the original problem context) toward concept-level memory: reusable, modular abstractions distilled from solution traces and stored in natural language. For future queries, relevant concepts are selectively retrieved and integrated into the prompt, enabling test-time continual learning without weight updates. Our design introduces new strategies for abstracting takeaways from rollouts and retrieving entries for new queries, promoting reuse and allowing memory to expand with additional experiences. We evaluate on ARC-AGI, a benchmark that stresses compositional generalization and abstract reasoning, making it a natural fit for concept memory. Our method yields a 7.5% relative gain over a strong no-memory baseline with performance continuing to scale with inference compute. We find abstract concepts to be the most consistent memory design, outscoring the baseline at all tested inference compute scales. Moreover, dynamically updating memory during test-time outperforms fixed settings, supporting the hypothesis that accumulating and abstracting patterns enables further solutions in a form of self-improvement. Code is available at https://github.com/matt-seb-ho/arc_memo.
Authors: Joachim Baumann, Paul R\"ottger, Aleksandra Urman, Albert Wendsj\"o, Flor Miriam Plaza-del-Arco, Johannes B. Gruber, Dirk Hovy
Abstract: Large language models are rapidly transforming social science research by enabling the automation of labor-intensive tasks like data annotation and text analysis. However, LLM outputs vary significantly depending on the implementation choices made by researchers (e.g., model selection or prompting strategy). Such variation can introduce systematic biases and random errors, which propagate to downstream analyses and cause Type I (false positive), Type II (false negative), Type S (wrong sign), or Type M (exaggerated effect) errors. We call this phenomenon where configuration choices lead to incorrect conclusions LLM hacking. We find that intentional LLM hacking is strikingly simple. By replicating 37 data annotation tasks from 21 published social science studies, we show that, with just a handful of prompt paraphrases, virtually anything can be presented as statistically significant. Beyond intentional manipulation, our analysis of 13 million labels from 18 different LLMs across 2361 realistic hypotheses shows that there is also a high risk of accidental LLM hacking, even when following standard research practices. We find incorrect conclusions in approximately 31% of hypotheses for state-of-the-art LLMs, and in half the hypotheses for smaller language models. While higher task performance and stronger general model capabilities reduce LLM hacking risk, even highly accurate models remain susceptible. The risk of LLM hacking decreases as effect sizes increase, indicating the need for more rigorous verification of LLM-based findings near significance thresholds. We analyze 21 mitigation techniques and find that human annotations provide crucial protection against false positives. Common regression estimator correction techniques can restore valid inference but trade off Type I vs. Type II errors. We publish a list of practical recommendations to prevent LLM hacking.
Authors: Zhengyu Hu, Jianxun Lian, Zheyuan Xiao, Max Xiong, Yuxuan Lei, Tianfu Wang, Kaize Ding, Ziang Xiao, Nicholas Jing Yuan, Xing Xie
Abstract: Recent advances in large language models (LLMs) have enabled human-like social simulations at unprecedented scale and fidelity, offering new opportunities for computational social science. A key challenge, however, is the construction of persona sets that authentically represent the diversity and distribution of real-world populations. Most existing LLM-based social simulation studies focus primarily on designing agentic frameworks and simulation environments, often overlooking the complexities of persona generation and the potential biases introduced by unrepresentative persona sets. In this paper, we propose a systematic framework for synthesizing high-quality, population-aligned persona sets for LLM-driven social simulation. Our approach begins by leveraging LLMs to generate narrative personas from long-term social media data, followed by rigorous quality assessment to filter out low-fidelity profiles. We then apply importance sampling to achieve global alignment with reference psychometric distributions, such as the Big Five personality traits. To address the needs of specific simulation contexts, we further introduce a task-specific module that adapts the globally aligned persona set to targeted subpopulations. Extensive experiments demonstrate that our method significantly reduces population-level bias and enables accurate, flexible social simulation for a wide range of research and policy applications.
Authors: Amirhossein Abaskohi, Raymond Li, Chuyuan Li, Shafiq Joty, Giuseppe Carenini
Abstract: We introduce CEMTM, a context-enhanced multimodal topic model designed to infer coherent and interpretable topic structures from both short and long documents containing text and images. CEMTM builds on fine-tuned large vision language models (LVLMs) to obtain contextualized embeddings, and employs a distributional attention mechanism to weight token-level contributions to topic inference. A reconstruction objective aligns topic-based representations with the document embedding, encouraging semantic consistency across modalities. Unlike existing approaches, CEMTM can process multiple images per document without repeated encoding and maintains interpretability through explicit word-topic and document-topic distributions. Extensive experiments on six multimodal benchmarks show that CEMTM consistently outperforms unimodal and multimodal baselines, achieving a remarkable average LLM score of 2.61. Further analysis shows its effectiveness in downstream few-shot retrieval and its ability to capture visually grounded semantics in complex domains such as scientific articles.
Authors: Donovan L. Buterakos, Sandesh S. Kalantre, Joshua Ziegler, Jacob M Taylor, Justyna P. Zwolak
Abstract: Recent advances in machine learning (ML) have accelerated progress in calibrating and operating quantum dot (QD) devices. However, most ML approaches rely on access to large, representative datasets designed to capture the full spectrum of data quality encountered in practice, with both high- and low-quality data for training, benchmarking, and validation, with labels capturing key features of the device state. Collating such datasets experimentally is challenging due to limited data availability, slow measurement bandwidths, and the labor-intensive nature of labeling. QDFlow is an open-source physics simulator for multi-QD arrays that generates realistic synthetic data with ground-truth labels. QDFlow combines a self-consistent Thomas-Fermi solver, a dynamic capacitance model, and flexible noise modules to simulate charge stability diagrams and ray-based data closely resembling experiments. With an extensive set of parameters that can be varied and customizable noise models, QDFlow supports the creation of large, diverse datasets for ML development, benchmarking, and quantum device research.
Authors: Nobin Sarwar, Shubhashis Roy Dipta
Abstract: Privacy-preserving adaptation of Large Language Models (LLMs) in sensitive domains (e.g., mental health) requires balancing strict confidentiality with model utility and safety. We propose FedMentor, a federated fine-tuning framework that integrates Low-Rank Adaptation (LoRA) and domain-aware Differential Privacy (DP) to meet per-domain privacy budgets while maintaining performance. Each client (domain) applies a custom DP noise scale proportional to its data sensitivity, and the server adaptively reduces noise when utility falls below a threshold. In experiments on three mental health datasets, we show that FedMentor improves safety over standard Federated Learning (FL) without privacy, raising safe output rates by up to three points and lowering toxicity, while maintaining utility (BERTScore F1 and ROUGE-L) within 0.5% of the non-private baseline and close to the centralized upper bound. The framework scales to backbones with up to 1.7B parameters on single-GPU clients, requiring < 173 MB of communication per-round. FedMentor demonstrates a practical approach to privately fine-tune LLMs for safer deployments in healthcare and other sensitive fields.
Authors: Riccardo Cadei, Christian Intern\`o
Abstract: Modern foundational models increasingly reflect not just world knowledge, but patterns of human preference embedded in their training data. We hypothesize that recursive alignment-via human feedback and model-generated corpora-induces a social desirability bias, nudging models to favor agreeable or flattering responses over objective reasoning. We refer to it as the Narcissus Hypothesis and test it across 31 models using standardized personality assessments and a novel Social Desirability Bias score. Results reveal a significant drift toward socially conforming traits, with profound implications for corpus integrity and the reliability of downstream inferences. We then offer a novel epistemological interpretation, tracing how recursive bias may collapse higher-order reasoning down Pearl's Ladder of Causality, culminating in what we refer to as the Rung of Illusion.
Authors: Yang Fu, Peng Qin, Yueyue Zhang, Pao Cheng, Jun Lu, Yifei Wang
Abstract: 6G networks are envisioned to support on-demand AI model downloading to accommodate diverse inference requirements of end users. By proactively caching models at edge nodes, users can retrieve the requested models with low latency for on-device AI inference. However, the substantial size of contemporary AI models poses significant challenges for edge caching under limited storage capacity, as well as for the concurrent delivery of heterogeneous models over wireless channels. To address these challenges, we propose a fine-grained AI model caching and downloading system that exploits parameter reusability, stemming from the common practice of fine-tuning task-specific models from a shared pre-trained model with frozen parameters. This system selectively caches model parameter blocks (PBs) at edge nodes, eliminating redundant storage of reusable parameters across different cached models. Additionally, it incorporates coordinated multipoint (CoMP) broadcasting to simultaneously deliver reusable PBs to multiple users, thereby enhancing downlink spectrum utilization. Under this arrangement, we formulate a model downloading delay minimization problem to jointly optimize PB caching, migration (among edge nodes), and broadcasting beamforming. To tackle this intractable problem, we develop a distributed multi-agent learning framework that enables edge nodes to explicitly learn mutual influence among their actions, thereby facilitating cooperation. Furthermore, a data augmentation approach is proposed to adaptively generate synthetic training samples through a predictive model, boosting sample efficiency and accelerating policy learning. Both theoretical analysis and simulation experiments validate the superior convergence performance of the proposed learning framework.
Authors: Maurice Wenig, Paul G. Rump, Mark Blacher, Joachim Giesen
Abstract: In 2011, einsum was introduced to NumPy as a practical and convenient notation for tensor expressions in machine learning, quantum circuit simulation, and other fields. It has since been implemented in additional Python frameworks such as PyTorch and TensorFlow, as well as in other programming languages such as Julia. Despite its practical success, the einsum notation still lacks a solid theoretical basis, and is not unified across the different frameworks, limiting opportunities for formal reasoning and systematic optimization. In this work, we discuss the terminology of tensor expressions and provide a formal definition of the einsum language. Based on this definition, we formalize and prove important equivalence rules for tensor expressions and highlight their relevance in practical applications.
Authors: Farbod Bigdeli, Mohsen Mohammadagha, Ali Bigdeli
Abstract: Breast cancer screening with mammography remains central to early detection and mortality reduction. Deep learning has shown strong potential for automating mammogram interpretation, yet limited-resolution datasets and small sample sizes continue to restrict performance. We revisit the Mini-DDSM dataset (9,684 images; 2,414 patients) and introduce a lightweight region-of-interest (ROI) augmentation strategy. During training, full images are probabilistically replaced with random ROI crops sampled from a precomputed, label-free bounding-box bank, with optional jitter to increase variability. We evaluate under strict patient-level cross-validation and report ROC-AUC, PR-AUC, and training-time efficiency metrics (throughput and GPU memory). Because ROI augmentation is training-only, inference-time cost remains unchanged. On Mini-DDSM, ROI augmentation (best: p_roi = 0.10, alpha = 0.10) yields modest average ROC-AUC gains, with performance varying across folds; PR-AUC is flat to slightly lower. These results demonstrate that simple, data-centric ROI strategies can enhance mammography classification in constrained settings without requiring additional labels or architectural modifications.
Authors: Sanish Suwal, Dipkamal Bhusal, Michael Clifford, Nidhi Rastogi
Abstract: Prior works have shown that neural networks can be heavily pruned while preserving performance, but the impact of pruning on model interpretability remains unclear. In this work, we investigate how magnitude-based pruning followed by fine-tuning affects both low-level saliency maps and high-level concept representations. Using a ResNet-18 trained on ImageNette, we compare post-hoc explanations from Vanilla Gradients (VG) and Integrated Gradients (IG) across pruning levels, evaluating sparsity and faithfulness. We further apply CRAFT-based concept extraction to track changes in semantic coherence of learned concepts. Our results show that light-to-moderate pruning improves saliency-map focus and faithfulness while retaining distinct, semantically meaningful concepts. In contrast, aggressive pruning merges heterogeneous features, reducing saliency map sparsity and concept coherence despite maintaining accuracy. These findings suggest that while pruning can shape internal representations toward more human-aligned attention patterns, excessive pruning undermines interpretability.
Authors: Gabriela Pinto, Palash Goyal, Yiwen Song, Souradip Chakraborty, Zifeng Wang, Tomas Pfister, Hamid Palangi
Abstract: Test-time scaling has shown considerable success in improving the performance of language models on complex reasoning tasks without requiring fine-tuning. However, current strategies such as self-reflection primarily focus on logical or structural refinement. They do not leverage the guiding potential of affective feedback. Inspired by psychological research showing that emotions can modulate cognitive performance, we introduce HEART--a novel framework that uses emotionally-driven prompts for iterative self-correction. HEART provides feedback on a model's incorrect response using a curated set of concise, emotionally charged phrases based on the six universal emotions categorized by Dr. Paul Ekman. By systematically varying the emotional tone of the feedback across iterations, our method guides the model to escape flawed reasoning paths and explore more promising alternatives. We evaluate our framework on challenging reasoning benchmarks including OlympiadBench, Humanity's Last Exam, and SimpleQA. Our results reveal a significant new phenomenon: when guided by an oracle verifier, this affective iteration protocol unlocks significantly deeper reasoning, leading to consistent and substantial increases in accuracy over state-of-the-art baselines with the same verifier. However, we also identify a critical bottleneck for practical deployment. In a verifier-free setting, it struggles to harness these gains consistently, highlighting as a key challenge for future work. Our findings suggest that the next frontier in machine reasoning may lie not just in refining logic, but also in understanding and leveraging the `HEART' of the models.
Authors: Yuntao Wu, Ege Mert Akin, Charles Martineau, Vincent Gr\'egoire, Andreas Veneris
Abstract: We examine how textual features in earnings press releases predict stock returns on earnings announcement days. Using over 138,000 press releases from 2005 to 2023, we compare traditional bag-of-words and BERT-based embeddings. We find that press release content (soft information) is as informative as earnings surprise (hard information), with FinBERT yielding the highest predictive power. Combining models enhances explanatory strength and interpretability of the content of press releases. Stock prices fully reflect the content of press releases at market open. If press releases are leaked, it offers predictive advantage. Topic analysis reveals self-serving bias in managerial narratives. Our framework supports real-time return prediction through the integration of online learning, provides interpretability and reveals the nuanced role of language in price formation.
Authors: Zhen Bi, Zhenlin Hu, Jinnan Yang, Mingyang Chen, Cheng Deng, Yida Xue, Zeyu Yang, Qing Shen, Zhenfang Liu, Kang Zhao, Ningyu Zhang, Jungang Lou
Abstract: Recent advances in large language models (LLMs) highlight the importance of training data structure and quality in shaping reasoning behavior. However, most existing approaches focus on transforming data formats while neglecting the internal reasoning complexity of training samples, leaving the reasoning potential of data under-explored and underutilized. In this work, we posit that LLM logical reasoning performance is jointly constrained by the potential of the training data and the cognitive capacity of the model. To make this relationship measurable, we introduce Data Reasoning Intensity (DRI), a novel metric that quantifies the latent logical reasoning complexity of samples by decomposing and aggregating their logical structures. This allows us to analyze how well current LLMs utilize logical reasoning signals and identify performance gaps relative to data potential. Based on this insight, we introduce a re-cognizing optimization strategy that systematically enhances the logical reasoning intensity of training data. Rather than increasing data volume, our method re-optimizes existing samples to better align with the LLM's logical reasoning boundary. Extensive experiments show that our approach significantly improves performance and generalization over data-centric strategies. We further validate our method under a reinforcement learning framework. Our results indicate that prioritizing reasoning complexity in data rather than sheer scale or superficial form is essential to realizing LLMs' full cognitive potential.
Authors: Yusuf Ziya Isik, Rafa{\l} {\L}aganowski
Abstract: The Low-Resource Audio Codec (LRAC) Challenge aims to advance neural audio coding for deployment in resource-constrained environments. The first edition focuses on low-resource neural speech codecs that must operate reliably under everyday noise and reverberation, while satisfying strict constraints on computational complexity, latency, and bitrate. Track 1 targets transparency codecs, which aim to preserve the perceptual transparency of input speech under mild noise and reverberation. Track 2 addresses enhancement codecs, which combine coding and compression with denoising and dereverberation. This paper presents the official baseline systems for both tracks in the 2025 LRAC Challenge. The baselines are convolutional neural codec models with Residual Vector Quantization, trained end-to-end using a combination of adversarial and reconstruction objectives. We detail the data filtering and augmentation strategies, model architectures, optimization procedures, and checkpoint selection criteria.
Authors: Zhaosong Lu, Xiangyuan Wang
Abstract: We study a class of constrained nonconvex--nonconcave minimax problems in which the inner maximization involves potentially complex constraints. Under the assumption that the inner problem of a novel lifted minimax problem satisfies a local Kurdyka-{\L}ojasiewicz (KL) condition, we show that the maximal function of the original problem enjoys a local H\"older smoothness property. We also propose a sequential convex programming (SCP) method for solving constrained optimization problems and establish its convergence rate under a local KL condition. Leveraging these results, we develop an inexact proximal gradient method for the original minimax problem, where the inexact gradient of the maximal function is computed via the SCP method applied to a locally KL-structured subproblem. Finally, we establish complexity guarantees for the proposed method in computing an approximate stationary point of the original minimax problem.
Authors: Rom Himelstein, Amit LeVi, Yonatan Belinkov, Avi Mendelson
Abstract: Padding tokens are widely used in large language models (LLMs) to equalize sequence lengths during batched inference. While they should be fully masked, implementation errors can cause them to influence computation, and the extent of this influence is not well understood. We systematically study this effect across three open-source model families (Llama, Gemma, Qwen), inserting controlled amounts of padding and evaluating outcomes along four axes: activations, generation quality, bias, and safety. Even small amounts of padding shift hidden representations, degrade quality in smaller models, alter bias in unpredictable ways, and weaken safety guardrails. These findings demonstrate that padding is not a harmless detail but a robustness risk that must be carefully handled in deployment.
Authors: Ying-Ren Chien, Po-Heng Chou, You-Jie Peng, Chun-Yuan Huang, Hen-Wai Tsao, Yu Tsao
Abstract: To effectively process impulse noise for narrowband powerline communications (NB-PLCs) transceivers, capturing comprehensive statistics of nonperiodic asynchronous impulsive noise (APIN) is a critical task. However, existing mathematical noise generative models only capture part of the characteristics of noise. In this study, we propose a novel generative adversarial network (GAN) called noise generation GAN (NGGAN) that learns the complicated characteristics of practically measured noise samples for data synthesis. To closely match the statistics of complicated noise over the NB-PLC systems, we measured the NB-PLC noise via the analog coupling and bandpass filtering circuits of a commercial NB-PLC modem to build a realistic dataset. To train NGGAN, we adhere to the following principles: 1) we design the length of input signals that the NGGAN model can fit to facilitate cyclostationary noise generation; 2) the Wasserstein distance is used as a loss function to enhance the similarity between the generated noise and training data; and 3) to measure the similarity performances of GAN-based models based on the mathematical and practically measured datasets, we conduct both quantitative and qualitative analyses. The training datasets include: 1) a piecewise spectral cyclostationary Gaussian model (PSCGM); 2) a frequency-shift (FRESH) filter; and 3) practical measurements from NB-PLC systems. Simulation results demonstrate that the generated noise samples from the proposed NGGAN are highly close to the real noise samples. The principal component analysis (PCA) scatter plots and Fr\'echet inception distance (FID) analysis have shown that NGGAN outperforms other GAN-based models by generating noise samples with superior fidelity and higher diversity.
Authors: Wei-Lung Mao, Chun-Chi Wang, Po-Heng Chou, Yen-Ting Liu
Abstract: Since the defect detection of conventional industry components is time-consuming and labor-intensive, it leads to a significant burden on quality inspection personnel and makes it difficult to manage product quality. In this paper, we propose an automated defect detection system for the dual in-line package (DIP) that is widely used in industry, using digital camera optics and a deep learning (DL)-based model. The two most common defect categories of DIP are examined: (1) surface defects, and (2) pin-leg defects. However, the lack of defective component images leads to a challenge for detection tasks. To solve this problem, the ConSinGAN is used to generate a suitable-sized dataset for training and testing. Four varieties of the YOLO model are investigated (v3, v4, v7, and v9), both in isolation and with the ConSinGAN augmentation. The proposed YOLOv7 with ConSinGAN is superior to the other YOLO versions in accuracy of 95.50\%, detection time of 285 ms, and is far superior to threshold-based approaches. In addition, the supervisory control and data acquisition (SCADA) system is developed, and the associated sensor architecture is described. The proposed automated defect detection can be easily established with numerous types of defects or insufficient defect data.
Authors: Donghuo Zeng
Abstract: Contrastive loss and triplet loss are widely used objectives in deep metric learning, yet their effects on representation quality remain insufficiently understood. We present a theoretical and empirical comparison of these losses, focusing on intra- and inter-class variance and optimization behavior (e.g., greedy updates). Through task-specific experiments with consistent settings on synthetic data and real datasets-MNIST, CIFAR-10-it is shown that triplet loss preserves greater variance within and across classes, supporting finer-grained distinctions in the learned representations. In contrast, contrastive loss tends to compact intra-class embeddings, which may obscure subtle semantic differences. To better understand their optimization dynamics, By examining loss-decay rate, active ratio, and gradient norm, we find that contrastive loss drives many small updates early on, while triplet loss produces fewer but stronger updates that sustain learning on hard examples. Finally, across both classification and retrieval tasks on MNIST, CIFAR-10, CUB-200, and CARS196 datasets, our results consistently show that triplet loss yields superior performance, which suggests using triplet loss for detail retention and hard-sample focus, and contrastive loss for smoother, broad-based embedding refinement.
Authors: Mark M. Wilde
Abstract: Quantum generalizations of the Fisher information are important in quantum information science, with applications in high energy and condensed matter physics and in quantum estimation theory, machine learning, and optimization. One can derive a quantum generalization of the Fisher information matrix in a natural way as the Hessian matrix arising in a Taylor expansion of a smooth divergence. Such an approach is appealing for quantum information theorists, given the ubiquity of divergences in quantum information theory. In contrast to the classical case, there is not a unique quantum generalization of the Fisher information matrix, similar to how there is not a unique quantum generalization of the relative entropy or the R\'enyi relative entropy. In this paper, I derive information matrices arising from the log-Euclidean, $\alpha$-$z$, and geometric R\'enyi relative entropies, with the main technical tool for doing so being the method of divided differences for calculating matrix derivatives. Interestingly, for all non-negative values of the R\'enyi parameter $\alpha$, the log-Euclidean R\'enyi relative entropy leads to the Kubo-Mori information matrix, and the geometric R\'enyi relative entropy leads to the right-logarithmic derivative Fisher information matrix. Thus, the resulting information matrices obey the data-processing inequality for all non-negative values of the R\'enyi parameter $\alpha$ even though the original quantities do not. Additionally, I derive and establish basic properties of $\alpha$-$z$ information matrices resulting from the $\alpha$-$z$ R\'enyi relative entropies. For parameterized thermal states and time-evolved states, I establish formulas for their $\alpha$-$z$ information matrices and hybrid quantum-classical algorithms for estimating them, with applications in quantum Boltzmann machine learning.
Authors: Kyoungjun Park, Yifan Yang, Juheon Yi, Shicheng Zheng, Yifei Shen, Dongqi Han, Caihua Shan, Muhammad Muaz, Lili Qiu
Abstract: With the rapid advancement of AI-generated videos, there is an urgent need for effective detection tools to mitigate societal risks such as misinformation and reputational harm. In addition to accurate classification, it is essential that detection models provide interpretable explanations to ensure transparency for regulators and end users. To address these challenges, we introduce VidGuard-R1, the first video authenticity detector that fine-tunes a multi-modal large language model (MLLM) using group relative policy optimization (GRPO). Our model delivers both highly accurate judgments and insightful reasoning. We curate a challenging dataset of 140k real and AI-generated videos produced by state-of-the-art generation models, carefully designing the generation process to maximize discrimination difficulty. We then fine-tune Qwen-VL using GRPO with two specialized reward models that target temporal artifacts and generation complexity. Extensive experiments demonstrate that VidGuard-R1 achieves state-of-the-art zero-shot performance on existing benchmarks, with additional training pushing accuracy above 95%. Case studies further show that VidGuard-R1 produces precise and interpretable rationales behind its predictions. The code is publicly available at https://VidGuard-R1.github.io.
Authors: Hadi Pouransari, David Grangier, C Thomas, Michael Kirchhof, Oncel Tuzel
Abstract: The impressive performance gains of modern language models currently rely on scaling parameters: larger models store more world knowledge and reason better. Yet compressing all world knowledge into parameters is unnecessary, as only a fraction is used per prompt, and impractical for edge devices with limited inference-time memory and compute. We address this shortcoming by a memory-augmented architecture and a pretraining strategy aligned with existing hardware paradigms. We introduce small language models that access large hierarchical parametric memory banks encoding world knowledge. During pretraining and inference, we fetch a small, context-dependent memory block and add it to the model. Our pretraining learns to store long-tail world knowledge in the memory parameters, while the small language model acts as an anchor capturing common knowledge and general reasoning abilities. Through trillion-token-scale experiments, we show significant gains: a 160M-parameters model augmented with an 18M-parameters memory fetched from a 4.6B memory bank obtains comparable performance to a regular model with more than 2x the parameters. Through extensive experiments, we study the optimal type and size of parametric memories in transformers, scaling them to over 21B parameters. We find that our proposed hierarchical feed-forward memories work robustly across transformer architectures, whether added during pretraining or post-hoc.
Authors: Peter Pak, Achuth Chandrasekhar, Amir Barati Farimani
Abstract: Agentic systems enable the intelligent use of research tooling, augmenting a researcher's ability to investigate and propose novel solutions to existing problems. Within Additive Manufacturing (AM), alloy discovery remains a complex challenge, often requiring expertise in the various domains of materials science, thermodynamic simulations, and experimental analysis. Large Language Model (LLM) enabled agents can facilitate this endeavor by utilizing their extensive knowledge base to dispatch tool calls via Model Context Protocol (MCP) to perform actions such as Thermo-Calc property diagram calculations and lack of fusion process map generation. In addition, the multi-agent system developed in this work is able to effectively reason through complex user prompts and provide analysis on the printability of proposed alloys. These agents can dynamically adjust their task trajectory to the outcomes of tool call results, effectively enabling autonomous decision-making in practical environments. This work aims to utilize LLM enabled agents to automate and accelerate the task of alloy discovery within the field of additive manufacturing and showcase the benefits of adopting this multi-agent system.
Authors: Julian Cremer, Tuan Le, Mohammad M. Ghahremanpour, Emilia S{\l}ugocka, Filipe Menezes, Djork-Arn\'e Clevert
Abstract: We present FLOWR:root, an equivariant flow-matching model for pocket-aware 3D ligand generation with joint binding affinity prediction and confidence estimation. The model supports de novo generation, pharmacophore-conditional sampling, fragment elaboration, and multi-endpoint affinity prediction (pIC50, pKi, pKd, pEC50). Training combines large-scale ligand libraries with mixed-fidelity protein-ligand complexes, followed by refinement on curated co-crystal datasets and parameter-efficient finetuning for project-specific adaptation. FLOWR:root achieves state-of-the-art performance in unconditional 3D molecule generation and pocket-conditional ligand design, producing geometrically realistic, low-strain structures. The integrated affinity prediction module demonstrates superior accuracy on the SPINDR test set and outperforms recent models on the Schrodinger FEP+/OpenFE benchmark with substantial speed advantages. As a foundation model, FLOWR:root requires finetuning on project-specific datasets to account for unseen structure-activity landscapes, yielding strong correlation with experimental data. Joint generation and affinity prediction enable inference-time scaling through importance sampling, steering molecular design toward higher-affinity compounds. Case studies validate this: selective CK2$\alpha$ ligand generation against CLK3 shows significant correlation between predicted and quantum-mechanical binding energies, while ER$\alpha$ and TYK2 scaffold elaboration demonstrates strong agreement with QM calculations. By integrating structure-aware generation, affinity estimation, and property-guided sampling, FLOWR:root provides a comprehensive foundation for structure-based drug design spanning hit identification through lead optimization.
Authors: Karim Farid, Rajat Sahay, Yumna Ali Alnaggar, Simon Schrodi, Volker Fischer, Cordelia Schmid, Thomas Brox
Abstract: Compositional generalization, the ability to generate novel combinations of known concepts, is a key ingredient for visual generative models. Yet, not all mechanisms that enable or inhibit it are fully understood. In this work, we conduct a systematic study of how various design choices influence compositional generalization in image and video generation in a positive or negative way. Through controlled experiments, we identify two key factors: (i) whether the training objective operates on a discrete or continuous distribution, and (ii) to what extent conditioning provides information about the constituent concepts during training. Building on these insights, we show that relaxing the MaskGIT discrete loss with an auxiliary continuous JEPA-based objective can improve compositional performance in discrete models like MaskGIT.
Authors: Francisco Patitucci, Ruichen Jiang, Aryan Mokhtari
Abstract: A recent breakthrough in nonconvex optimization is the online-to-nonconvex conversion framework of [Cutkosky et al., 2023], which reformulates the task of finding an $\varepsilon$-first-order stationary point as an online learning problem. When both the gradient and the Hessian are Lipschitz continuous, instantiating this framework with two different online learners achieves a complexity of $O(\varepsilon^{-1.75}\log(1/\varepsilon))$ in the deterministic case and a complexity of $O(\varepsilon^{-3.5})$ in the stochastic case. However, this approach suffers from several limitations: (i) the deterministic method relies on a complex double-loop scheme that solves a fixed-point equation to construct hint vectors for an optimistic online learner, introducing an extra logarithmic factor; (ii) the stochastic method assumes a bounded second-order moment of the stochastic gradient, which is stronger than standard variance bounds; and (iii) different online learning algorithms are used in the two settings. In this paper, we address these issues by introducing an online optimistic gradient method based on a novel doubly optimistic hint function. Specifically, we use the gradient at an extrapolated point as the hint, motivated by two optimistic assumptions: that the difference between the hint and the target gradient remains near constant, and that consecutive update directions change slowly due to smoothness. Our method eliminates the need for a double loop and removes the logarithmic factor. Furthermore, by simply replacing full gradients with stochastic gradients and under the standard assumption that their variance is bounded by $\sigma^2$, we obtain a unified algorithm with complexity $O(\varepsilon^{-1.75} + \sigma^2 \varepsilon^{-3.5})$, smoothly interpolating between the best-known deterministic rate and the optimal stochastic rate.