Authors: Souradeep Chattopadhyay, Brendan Kennedy, Sai Munikoti, Soumik Sarkar, Karl Pazdernik
Abstract: In the critical task of making generative models trustworthy and robust, methods for Uncertainty Quantification (UQ) have begun to show encouraging potential. However, many of these methods rely on rigid heuristics that fail to generalize across tasks and modalities. Here, we propose a novel framework for UQ that is highly flexible and approaches or surpasses the performance of prior heuristic methods. We introduce Directional Concentration Uncertainty (DCU), a novel statistical procedure for quantifying the concentration of embeddings based on the von Mises-Fisher (vMF) distribution. Our method captures uncertainty by measuring the geometric dispersion of multiple generated outputs from a language model using continuous embeddings of the generated outputs without any task specific heuristics. In our experiments, we show that DCU matches or exceeds calibration levels of prior works like semantic entropy (Kuhn et al., 2023) and also generalizes well to more complex tasks in multi-modal domains. We present a framework for the wider potential of DCU and its implications for integration into UQ for multi-modal and agentic frameworks.
Authors: Ethan Seefried, Ran Eldegaway, Sanjay Das, Nathaniel Blanchard, Tirthankar Ghosal
Abstract: Decades of engineering drawings and technical records remain locked in legacy archives with inconsistent or missing metadata, making retrieval difficult and often manual. We present Blueprint, a layout-aware multimodal retrieval system designed for large-scale engineering repositories. Blueprint detects canonical drawing regions, applies region-restricted VLM-based OCR, normalizes identifiers (e.g., DWG, part, facility), and fuses lexical and dense retrieval with a lightweight region-level reranker. Deployed on ~770k unlabeled files, it automatically produces structured metadata suitable for cross-facility search. We evaluate Blueprint on a 5k-file benchmark with 350 expert-curated queries using pooled, graded (0/1/2) relevance judgments. Blueprint delivers a 10.1% absolute gain in Success@3 and an 18.9% relative improvement in nDCG@3 over the strongest vision-language baseline}, consistently outperforming across vision, text, and multimodal intents. Oracle ablations reveal substantial headroom under perfect region detection and OCR. We release all queries, runs, annotations, and code to facilitate reproducible evaluation on legacy engineering archives.
Authors: Michael Beebe, GodsGift Uzor, Manasa Chepuri, Divya Sree Vemula, Angel Ayala
Abstract: Small datasets like MNIST have historically been instrumental in advancing machine learning research by providing a controlled environment for rapid experimentation and model evaluation. However, their simplicity often limits their utility for distinguishing between advanced neural network architectures. To address these challenges, Greydanus et al. introduced the MNIST-1D dataset, a one-dimensional adaptation of MNIST designed to explore inductive biases in sequential data. This dataset maintains the advantages of small-scale datasets while introducing variability and complexity that make it ideal for studying advanced architectures. In this paper, we extend the exploration of MNIST-1D by evaluating the performance of Residual Networks (ResNet), Temporal Convolutional Networks (TCN), and Dilated Convolutional Neural Networks (DCNN). These models, known for their ability to capture sequential patterns and hierarchical features, were implemented and benchmarked alongside previously tested architectures such as logistic regression, MLPs, CNNs, and GRUs. Our experimental results demonstrate that advanced architectures like TCN and DCNN consistently outperform simpler models, achieving near-human performance on MNIST-1D. ResNet also shows significant improvements, highlighting the importance of leveraging inductive biases and hierarchical feature extraction in small structured datasets. Through this study, we validate the utility of MNIST-1D as a robust benchmark for evaluating machine learning architectures under computational constraints. Our findings emphasize the role of architectural innovations in improving model performance and offer insights into optimizing deep learning models for resource-limited environments.
Authors: Hannes Kath, Thiago S. Gouv\^ea, Daniel Sonntag
Abstract: Machine learning models excel with abundant annotated data, but annotation is often costly and time-intensive. Active learning (AL) aims to improve the performance-to-annotation ratio by using query methods (QMs) to iteratively select the most informative samples. While AL research focuses mainly on QM development, the evaluation of this iterative process lacks appropriate performance metrics. This work reviews eight years of AL evaluation literature and formally introduces the speed-up factor, a quantitative multi-iteration QM performance metric that indicates the fraction of samples needed to match random sampling performance. Using four datasets from diverse domains and seven QMs of various types, we empirically evaluate the speed-up factor and compare it with state-of-the-art AL performance metrics. The results confirm the assumptions underlying the speed-up factor, demonstrate its accuracy in capturing the described fraction, and reveal its superior stability across iterations.
Authors: Daniel Emerson, Nora Gaby-Biegel, Purva Joshi, Yoed Rabin, Rebecca D. Sandlin, Levent Burak Kara
Abstract: Designing cryoprotectant agent (CPA) cocktails for vitrification is challenging because formulations must be concentrated enough to suppress ice formation yet non-toxic enough to preserve cell viability. This tradeoff creates a large, multi-objective design space in which traditional discovery is slow, often relying on expert intuition or exhaustive experimentation. We present a data-efficient framework that accelerates CPA cocktail design by combining high-throughput screening with an active-learning loop based on multi-objective Bayesian optimization. From an initial set of measured cocktails, we train probabilistic surrogate models to predict concentration and viability and quantify uncertainty across candidate formulations. We then iteratively select the next experiments by prioritizing cocktails expected to improve the Pareto front, maximizing expected Pareto improvement under uncertainty, and update the models as new assay results are collected. Wet-lab validation shows that our approach efficiently discovers cocktails that simultaneously achieve high CPA concentrations and high post-exposure viability. Relative to a naive strategy and a strong baseline, our method improves dominated hypervolume by 9.5\% and 4.5\%, respectively, while reducing the number of experiments needed to reach high-quality solutions. In complementary synthetic studies, it recovers a comparably strong set of Pareto-optimal solutions using only 30\% of the evaluations required by the prior state-of-the-art multi-objective approach, which amounts to saving approximately 10 weeks of experimental time. Because the framework assumes only a suitable assay and defined formulation space, it can be adapted to different CPA libraries, objective definitions, and cell lines to accelerate cryopreservation development.
Authors: Yuchen Fang, James Demmel, Javad Lavaei
Abstract: We develop a worst-case complexity theory for stochastically preconditioned stochastic gradient descent (SPSGD) and its accelerated variants under heavy-tailed noise, a setting that encompasses widely used adaptive methods such as Adam, RMSProp, and Shampoo. We assume the stochastic gradient noise has a finite $p$-th moment for some $p \in (1,2]$, and measure convergence after $T$ iterations. While clipping and normalization are parallel tools for stabilizing training of SGD under heavy-tailed noise, there is a fundamental separation in their worst-case properties in stochastically preconditioned settings. We demonstrate that normalization guarantees convergence to a first-order stationary point at rate $\mathcal{O}(T^{-\frac{p-1}{3p-2}})$ when problem parameters are known, and $\mathcal{O}(T^{-\frac{p-1}{2p}})$ when problem parameters are unknown, matching the optimal rates for normalized SGD, respectively. In contrast, we prove that clipping may fail to converge in the worst case due to the statistical dependence between the stochastic preconditioner and the gradient estimates. To enable the analysis, we develop a novel vector-valued Burkholder-type inequality that may be of independent interest. These results provide a theoretical explanation for the empirical preference for normalization over clipping in large-scale model training.
Authors: Haiwen Guan, Moein Darman, Dibyajyoti Chakraborty, Troy Arcomano, Ashesh Chattopadhyay, Romit Maulik
Abstract: The proliferation of data-driven models in weather and climate sciences has marked a significant paradigm shift, with advanced models demonstrating exceptional skill in medium-range forecasting. However, these models are often limited by long-term instabilities, climatological drift, and substantial computational costs during training and inference, restricting their broader application for climate studies. Addressing these limitations, Guan et al. (2024) introduced LUCIE, a lightweight, physically consistent climate emulator utilizing a Spherical Fourier Neural Operator (SFNO) architecture. This model is able to reproduce accurate long-term statistics including climatological mean and seasonal variability. However, LUCIE's native resolution (~300 km) is inadequate for detailed regional impact assessments. To overcome this limitation, we introduce a deep learning-based downscaling framework, leveraging probabilistic diffusion-based generative models with conditional and posterior sampling frameworks. These models downscale coarse LUCIE outputs to 25 km resolution. They are trained on approximately 14,000 ERA5 timesteps spanning 2000-2009 and evaluated on LUCIE predictions from 2010 to 2020. Model performance is assessed through diverse metrics, including latitude-averaged RMSE, power spectrum, probability density functions and First Empirical Orthogonal Function of the zonal wind. We observe that the proposed approach is able to preserve the coarse-grained dynamics from LUCIE while generating fine-scaled climatological statistics at ~28km resolution.
Authors: Karish Grover, Hanqing Zeng, Yinglong Xia, Christos Faloutsos, Geoffrey J. Gordon
Abstract: Does text have an intrinsic curvature? Language is increasingly modeled in curved geometries - hyperbolic spaces for hierarchy, mixed-curvature manifolds for compositional structure - yet a basic scientific question remains unresolved: what does curvature mean for text itself, in a way that is native to language rather than an artifact of the embedding space we choose? We argue that text does indeed have curvature, and show how to detect it, define it, and use it. To this end, we propose Texture, a text-native, word-level discrete curvature signal, and make three contributions. (a) Existence: We provide empirical and theoretical certificates that semantic inference in natural corpora is non-flat, i.e. language has inherent curvature. (b) Definition: We define Texture by reconciling left- and right-context beliefs around a masked word through a Schrodinger bridge, yielding a curvature field that is positive where context focuses meaning and negative where it fans out into competing continuations. (c) Utility: Texture is actionable: it serves as a general-purpose measurement and control primitive enabling geometry without geometric training; we instantiate it on two representative tasks, improving long-context inference through curvature-guided compression and retrieval-augmented generation through curvature-guided routing. Together, our results establish a text-native curvature paradigm, making curvature measurable and practically useful.
Authors: Sadra Sabouri, Alireza Zolanvari, Sepand Haghighi
Abstract: Selecting an optimal classification model requires a robust and comprehensive understanding of the performance of the model. This paper provides a tutorial on the PyCM library, demonstrating its utility in conducting deep-dive evaluations of multi-class classifiers. By examining two different case scenarios, we illustrate how the choice of evaluation metrics can fundamentally shift the interpretation of a model's efficacy. Our findings emphasize that a multi-dimensional evaluation framework is essential for uncovering small but important differences in model performance. However, standard metrics may miss these subtle performance trade-offs.
Authors: Gabriel Franco, Lucas M. Tassis, Azalea Rohr, Mark Crovella
Abstract: Understanding the internal circuits that language models use to solve tasks remains a central challenge in mechanistic interpretability. Most prior work identifies circuits at the task level by averaging across many prompts, implicitly assuming a single stable mechanism per task. We show that this assumption can obscure a crucial source of structure: circuits are prompt-specific, even within a fixed task. Building on attention causal communication (ACC) (Franco & Crovella, 2025), we introduce ACC++, refinements that extract cleaner, lower-dimensional causal signals inside attention heads from a single forward pass. Like ACC, our approach does not require replacement models (e.g., SAEs) or activation patching; ACC++ further improves circuit precision by reducing attribution noise. Applying ACC++ to indirect object identification (IOI) in GPT-2, Pythia, and Gemma 2, we find there is no single circuit for IOI in any model: different prompt templates induce systematically different mechanisms. Despite this variation, prompts cluster into prompt families with similar circuits, and we propose a representative circuit for each family as a practical unit of analysis. Finally, we develop an automated interpretability pipeline that uses ACC++ signals to surface human-interpretable features and assemble mechanistic explanations for prompt families behavior. Together, our results recast circuits as a meaningful object of study by shifting the unit of analysis from tasks to prompts, enabling scalable circuit descriptions in the presence of prompt-specific mechanisms.
Authors: Ayse Tursucular, Ayush Mohanty, Nazal Mohamed, Nagi Gebraeel
Abstract: Networks of modern industrial systems are increasingly monitored by distributed sensors, where each system comprises multiple subsystems generating high dimensional time series data. These subsystems are often interdependent, making it important to understand how temporal patterns at one subsystem relate to others. This is challenging in decentralized settings where raw measurements cannot be shared and client observations are heterogeneous. In practical deployments each subsystem (client) operates a fixed proprietary model that cannot be modified or retrained, limiting existing approaches. Nonlinear dynamics further make cross client temporal interdependencies difficult to interpret because they are embedded in nonlinear state transition functions. We present a federated framework for learning temporal interdependencies across clients under these constraints. Each client maps high dimensional local observations to low dimensional latent states using a nonlinear state space model. A central server learns a graph structured neural state transition model over the communicated latent states using a Graph Attention Network. For interpretability we relate the Jacobian of the learned server side transition model to attention coefficients, providing the first interpretable characterization of cross client temporal interdependencies in decentralized nonlinear systems. We establish theoretical convergence guarantees to a centralized oracle and validate the framework through synthetic experiments demonstrating convergence, interpretability, scalability and privacy. Additional real world experiments show performance comparable to decentralized baselines.
Authors: Fei Wu, Jia Hu, Geyong Min, Shiqiang Wang
Abstract: Federated low-rank adaptation (FedLoRA) has facilitated communication-efficient and privacy-preserving fine-tuning of foundation models for downstream tasks. In practical federated learning scenarios, client heterogeneity in system resources and data distributions motivates heterogeneous LoRA ranks across clients. We identify a previously overlooked phenomenon in heterogeneous FedLoRA, termed rank collapse, where the energy of the global update concentrates on the minimum shared rank, resulting in suboptimal performance and high sensitivity to rank configurations. Through theoretical analysis, we reveal the root cause of rank collapse: a mismatch between rank-agnostic aggregation weights and rank-dependent client contributions, which systematically suppresses higher-rank updates at a geometric rate over rounds. Motivated by this insight, we propose raFLoRA, a rank-partitioned aggregation method that decomposes local updates into rank partitions and then aggregates each partition weighted by its effective client contributions. Extensive experiments across classification and reasoning tasks show that raFLoRA prevents rank collapse, improves model performance, and preserves communication efficiency compared to state-of-the-art FedLoRA baselines.
Authors: Peng Cheng, Jiucheng Zang, Qingnan Li, Liheng Ma, Yufei Cui, Yingxue Zhang, Boxing Chen, Ming Jian, Wen Tong
Abstract: Muon-style optimizers leverage Newton-Schulz (NS) iterations to orthogonalize updates, yielding update geometries that often outperform Adam-series methods. However, this orthogonalization discards magnitude information, rendering training sensitive to step-size hyperparameters and vulnerable to high-energy bursts. To mitigate this, we introduce TrasMuon (\textbf{T}rust \textbf{R}egion \textbf{A}daptive \textbf{S}caling \textbf{Muon}). TrasMuon preserves the near-isometric geometry of Muon while stabilizing magnitudes through (i) global RMS calibration and (ii) energy-based trust-region clipping. We demonstrate that while reintroducing adaptive scaling improves optimization efficiency, it typically exacerbates instability due to high-energy outliers. TrasMuon addresses this by defining a trust region based on relative energy ratios, confining updates to a stable zone. Empirical experiments on vision and language models demonstrate that TrasMuon converges faster than baselines. Furthermore, experiments without warmup stages confirm TrasMuon's superior stability and robustness.
Authors: Mohammad Pedramfar, Vaneet Aggarwal
Abstract: Optimizing monotone non-convex functions is a fundamental challenge across machine learning and combinatorial optimization. We introduce and study $\gamma$-weakly $\theta$-up-concavity, a novel first-order condition that characterizes a broad class of such functions. This condition provides a powerful unifying framework, strictly generalizing both DR-submodular functions and One-Sided Smooth (OSS) functions. Our central theoretical contribution demonstrates that $\gamma$-weakly $\theta$-up-concave functions are upper-linearizable: for any feasible point, we can construct a linear surrogate whose gains provably approximate the original non-linear objective. This approximation holds up to a constant factor, namely the approximation coefficient, dependent solely on $\gamma$, $\theta$, and the geometry of the feasible set. This linearizability yields immediate and unified approximation guarantees for a wide range of problems. Specifically, we obtain unified approximation guarantees for offline optimization as well as static and dynamic regret bounds in online settings via standard reductions to linear optimization. Moreover, our framework recovers the optimal approximation coefficient for DR-submodular maximization and significantly improves existing approximation coefficients for OSS optimization, particularly over matroid constraints.
Authors: Gabriel Franco, Carson Loughridge, Mark Crovella
Abstract: Identifying feature representations in language models is a central task in mechanistic interpretability. Several recent studies have made an implicit assumption that feature representations can be inferred in some cases from singular vectors of attention matrices. However, sound justification for this assumption is lacking. In this paper we address that question, asking: why and when do singular vectors align with features? First, we demonstrate that singular vectors robustly align with features in a model where features can be directly observed. We then show theoretically that such alignment is expected under a range of conditions. We close by asking how, operationally, alignment may be recognized in real models where feature representations are not directly observable. We identify sparse attention decomposition as a testable prediction of alignment, and show evidence that it emerges in a manner consistent with predictions in real models. Together these results suggest that alignment of singular vectors with features can be a sound and theoretically justified basis for feature identification in language models.
Authors: Abdallah Aaraba, Soumaya Cherkaoui, Ola Ahmad, Shengrui Wang
Abstract: Quantum reservoir computing offers a promising route for time series learning by modelling sequential data via rich quantum dynamics while the only training required happens at the level of a lightweight classical readout. However, studies featuring efficient and implementable quantum reservoir architectures along with model learning guarantees remain scarce in the literature. To close this gap, we introduce QuaRK, an end-to-end framework that couples a hardware-realistic quantum reservoir featurizer with a kernel-based readout scheme. Given a sequence of sample points, the reservoir injects the points one after the other to yield a compact feature vector from efficiently measured k-local observables using classical shadow tomography, after which a classical kernel-based readout learns the target mapping with explicit regularization and fast optimization. The resulting pipeline exposes clear computational knobs -- circuit width and depth as well as the measurement budget -- while preserving the flexibility of kernel methods to model nonlinear temporal functionals and being scalable to high-dimensional data. We further provide learning-theoretic generalization guarantees for dependent temporal data, linking design and resource choices to finite-sample performance, thereby offering principled guidance for building reliable temporal learners. Empirical experiments validate QuaRK and illustrate the predicted interpolation and generalization behaviours on synthetic beta-mixing time series tasks.
Authors: Nobutaka Ono
Abstract: In this paper, we propose a fast algorithm for element selection, a multiplication-free form of dimension reduction that produces a dimension-reduced vector by simply selecting a subset of elements from the input. Dimension reduction is a fundamental technique for reducing unnecessary model parameters, mitigating overfitting, and accelerating training and inference. A standard approach is principal component analysis (PCA), but PCA relies on matrix multiplications; on resource-constrained systems, the multiplication count itself can become a bottleneck. Element selection eliminates this cost because the reduction consists only of selecting elements, and thus the key challenge is to determine which elements should be retained. We evaluate a candidate subset through the minimum mean-squared error of linear regression that predicts a target vector from the selected elements, where the target may be, for example, a one-hot label vector in classification. When an explicit target is unavailable, the input itself can be used as the target, yielding a reconstruction-based criterion. The resulting optimization is combinatorial, and exhaustive search is impractical. To address this, we derive an efficient formula for the objective change caused by swapping a selected and an unselected element, using the matrix inversion lemma, and we perform a swap-based local search that repeatedly applies objective-decreasing swaps until no further improvement is possible. Experiments on MNIST handwritten-digit images demonstrate the effectiveness of the proposed method.
Authors: Roussel Desmond Nzoyem
Abstract: As breakthroughs in deep learning transform key industries, models are increasingly required to extrapolate on datapoints found outside the range of the training set, a challenge we coin as out-of-support (OoS) generalisation. However, neural networks frequently exhibit catastrophic failure on OoS samples, yielding unrealistic but overconfident predictions. We address this challenge by reformulating the OoS generalisation problem as a sequence modelling task in the weight space, wherein the training set is partitioned into concentric shells corresponding to discrete sequential steps. Our WeightCaster framework yields plausible, interpretable, and uncertainty-aware predictions without necessitating explicit inductive biases, all the while maintaining high computational efficiency. Emprical validation on a synthetic cosine dataset and real-world air quality sensor readings demonstrates performance competitive or superior to the state-of-the-art. By enhancing reliability beyond in-distribution scenarios, these results hold significant implications for the wider adoption of artificial intelligence in safety-critical applications.
Authors: Chongyang Li, Tianqian Zhang, Shouyin Liu
Abstract: Semantic Communication (SemCom) has emerged as a promising paradigm for 6G networks, aiming to extract and transmit task-relevant information rather than minimizing bit errors. However, applying SemCom to realistic downlink Multi-User Multi-Input Multi-Output (MU-MIMO) Orthogonal Frequency Division Multiplexing (OFDM) systems remains challenging due to severe Multi-User Interference (MUI) and frequency-selective fading. Existing Deep Joint Source-Channel Coding (DJSCC) schemes, primarily designed for point-to-point links, suffer from performance saturation in multi-user scenarios. To address these issues, we propose a scenario-adaptive MU-MIMO SemCom framework featuring an asymmetric architecture tailored for downlink transmission. At the transmitter, we introduce a scenario-aware semantic encoder that dynamically adjusts feature extraction based on Channel State Information (CSI) and Signal-to-Noise Ratio (SNR), followed by a neural precoding network designed to mitigate MUI in the semantic domain. At the receiver, a lightweight decoder equipped with a novel pilot-guided attention mechanism is employed to implicitly perform channel equalization and feature calibration using reference pilot symbols. Extensive simulation results over 3GPP channel models demonstrate that the proposed framework significantly outperforms DJSCC and traditional Separate Source-Channel Coding (SSCC) schemes in terms of Peak Signal-to-Noise Ratio (PSNR) and classification accuracy, particularly in low-SNR regimes, while maintaining low latency and computational cost on edge devices.
Authors: Hayato Suzuki, Shunnosuke Ikeda, Yuichi Takano
Abstract: Clustering serves as a vital tool for uncovering latent data structures, and achieving both high accuracy and interpretability is essential. To this end, existing methods typically construct binary decision trees by solving mixed-integer nonlinear optimization problems, often leading to significant computational costs and suboptimal solutions. Furthermore, binary decision trees frequently result in excessively deep structures, which makes them difficult to interpret. To mitigate these issues, we propose an interpretable clustering method based on optimal multiway-split decision trees, formulated as a 0-1 integer linear optimization problem. This reformulation renders the optimization problem more tractable compared to existing models. A key feature of our method is the integration of a one-dimensional K-means algorithm for the discretization of continuous variables, allowing for flexible and data-driven branching. Extensive numerical experiments on publicly available real-world datasets demonstrate that our method outperforms baseline methods in terms of clustering accuracy and interpretability. Our method yields multiway-split decision trees with concise decision rules while maintaining competitive performance across various evaluation metrics.
Authors: Mingqiao Zhang, Qiyao Peng, Yumeng Wang, Chunyuan Liu, Hongtao Liu
Abstract: The expanding integration of Large Language Models (LLMs) into recommender systems poses critical challenges to evaluation reliability. This paper identifies and investigates a previously overlooked issue: benchmark data leakage in LLM-based recommendation. This phenomenon occurs when LLMs are exposed to and potentially memorize benchmark datasets during pre-training or fine-tuning, leading to artificially inflated performance metrics that fail to reflect true model performance. To validate this phenomenon, we simulate diverse data leakage scenarios by conducting continued pre-training of foundation models on strategically blended corpora, which include user-item interactions from both in-domain and out-of-domain sources. Our experiments reveal a dual-effect of data leakage: when the leaked data is domain-relevant, it induces substantial but spurious performance gains, misleadingly exaggerating the model's capability. In contrast, domain-irrelevant leakage typically degrades recommendation accuracy, highlighting the complex and contingent nature of this contamination. Our findings reveal that data leakage acts as a critical, previously unaccounted-for factor in LLM-based recommendation, which could impact the true model performance. We release our code at https://github.com/yusba1/LLMRec-Data-Leakage.
Authors: Shuaibin Song, Kai Ming Ting, Kaifeng Zhang, Tianrun Liang
Abstract: Neighborhood Aggregation Strategy (NAS) is a widely used approach in graph embedding, underpinning both Graph Neural Networks (GNNs) and Weisfeiler-Lehman (WL) methods. However, NAS-based methods are identified to be prone to over-smoothing-the loss of node distinguishability with increased iterations-thereby limiting their effectiveness. This paper identifies two characteristics in a network, i.e., the distributions of nodes and node degrees that are critical for expressive representation but have been overlooked in existing methods. We show that these overlooked characteristics contribute significantly to over-smoothing of NAS-methods. To address this, we propose a novel weighted distribution-aware kernel that embeds nodes while taking their distributional characteristics into consideration. Our method has three distinguishing features: (1) it is the first method to explicitly incorporate both distributional characteristics; (2) it requires no optimization; and (3) it effectively mitigates the adverse effects of over-smoothing, allowing WL to preserve node distinguishability and expressiveness even after many iterations of embedding. Experiments demonstrate that our method achieves superior community detection performance via spectral clustering, outperforming existing graph embedding methods, including deep learning methods, on standard benchmarks.
Authors: Li Zhang, Nital Patel, Xiuqi Li, Jessica Lin
Abstract: Time series chain (TSC) is a recently introduced concept that captures the evolving patterns in large scale time series. Informally, a time series chain is a temporally ordered set of subsequences, in which consecutive subsequences in the chain are similar to one another, but the last and the first subsequences maybe be dissimilar. Time series chain has the great potential to reveal latent unusual evolving trend in the time series, or identify precursor of important events in a complex system. Unfortunately, existing definitions of time series chains only consider finding chains in a single time series. As a result, they are likely to miss unexpected evolving patterns in interrupted time series, or across two related time series. To address this limitation, in this work, we introduce a new definition called \textit{Joint Time Series Chain}, which is specially designed for the task of finding unexpected evolving trend across interrupted time series or two related time series. Our definition focuses on mitigating the robustness issues caused by the gap or interruption in the time series. We further propose an effective ranking criterion to identify the best chain. We demonstrate that our proposed approach outperforms existing TSC work in locating unusual evolving patterns through extensive empirical evaluations. We further demonstrate the utility of our work with a real-life manufacturing application from Intel. Our source code is publicly available at the supporting page https://github.com/lizhang-ts/JointTSC .
Authors: Stefan Behfar, Richard Mortier
Abstract: In real-world federated learning (FL) systems, client participation is intermittent, heterogeneous, and often correlated with data characteristics or resource constraints. Existing fairness approaches in FL primarily focus on equalizing loss or accuracy conditional on participation, implicitly assuming that clients have comparable opportunities to contribute over time. However, when participation itself is uneven, these objectives can lead to systematic under-representation of intermittently available clients, even if per-round performance appears fair. We propose cumulative utility parity, a fairness principle that evaluates whether clients receive comparable long-term benefit per participation opportunity, rather than per training round. To operationalize this notion, we introduce availability-normalized cumulative utility, which disentangles unavoidable physical constraints from avoidable algorithmic bias arising from scheduling and aggregation. Experiments on temporally skewed, non-IID federated benchmarks demonstrate that our approach substantially improves long-term representation parity, while maintaining near-perfect performance.
Authors: Valery Parfenov, Grigoriy Evseev, Andrey Veprikov, Nikolay Bushkov, Stanislav Moiseev, Aleksandr Beznosikov
Abstract: Fine-tuning large pretrained language models (LLMs) is a cornerstone of modern NLP, yet its growing memory demands (driven by backpropagation and large optimizer States) limit deployment in resource-constrained settings. Zero-order (ZO) methods bypass backpropagation by estimating directional derivatives from forward evaluations, offering substantial memory savings. However, classical ZO estimators suffer from high variance and an adverse dependence on the parameter dimensionality $d$, which has constrained their use to low-dimensional problems. In this work, we propose a policy-driven ZO framework that treats the sampling distribution over perturbation directions as a learnable policy and updates it to reduce the variance of directional estimates. We develop a practical algorithm implementing this idea and provide a theoretical analysis, showing that learned sampling distributions improve the quality of gradient information and relax the explicit dependence on $d$ in convergence bounds. Empirically, we validate the approach on challenging LLM fine-tuning benchmarks, demonstrating substantially improved performance compared to standard ZO baselines. Our results suggest that adaptive direction sampling is a promising route to make ZO fine-tuning viable at scale. The source code is available at https://github.com/brain-lab-research/zo_ldsd
Authors: Jiayi Huang, Amirmohammad Farzaneh, Osvaldo Simeone
Abstract: In safety-critical applications such as medical image segmentation, prediction systems must provide reliability guarantees that extend beyond conventional expected loss control. While risk-controlling prediction sets (RCPS) offer probabilistic guarantees on the expected risk, they fail to capture tail behavior and worst-case scenarios that are crucial in high-stakes settings. This paper introduces optimized certainty equivalent RCPS (OCE-RCPS), a novel framework that provides high-probability guarantees on general optimized certainty equivalent (OCE) risk measures, including conditional value-at-risk (CVaR) and entropic risk. OCE-RCPS leverages upper confidence bounds to identify prediction set parameters that satisfy user-specified risk tolerance levels with provable reliability. We establish theoretical guarantees showing that OCE-RCPS satisfies the desired probabilistic constraint for loss functions such as miscoverage and false negative rate. Experiments on image segmentation demonstrate that OCE-RCPS consistently meets target satisfaction rates across various risk measures and reliability configurations, while OCE-CRC fails to provide probabilistic guarantees.
Authors: Edward Chen, Natalie Dullerud, Pang Wei Koh, Thomas Niedermayr, Elizabeth Kidd, Sanmi Koyejo, Carlos Guestrin
Abstract: In complex clinical decision-making, clinicians must often track a variety of competing metrics defined by aim (ideal) and limit (strict) thresholds. Sifting through these high-dimensional tradeoffs to infer the optimal patient-specific strategy is cognitively demanding and historically prone to variability. In this paper, we address this challenge within the context of High-Dose-Rate (HDR) brachytherapy for cervical cancer, where planning requires strictly managing radiation hot spots while balancing tumor coverage against organ sparing. We present ALMo (Aim-Limit-defined Multi-Objective system), an interactive decision support system designed to infer and operationalize clinician intent. ALMo employs a novel optimization framework that minimizes manual input through automated parameter setup and enables flexible control over toxicity risks. Crucially, the system allows clinicians to navigate the Pareto surface of dosimetric tradeoffs by directly manipulating intuitive aim and limit values. In a retrospective evaluation of 25 clinical cases, ALMo generated treatment plans that consistently met or exceeded manual planning quality, with 65% of cases demonstrating dosimetric improvements. Furthermore, the system significantly enhanced efficiency, reducing average planning time to approximately 17 minutes, compared to the conventional 30-60 minutes. While validated in brachytherapy, ALMo demonstrates a generalized framework for streamlining interaction in multi-criteria clinical decision-making.
Authors: Binyu Zhao, Wei Zhang, Xingrui Yu, Zhaonian Zou, Ivor Tsang
Abstract: Class-incremental learning (CIL) with pre-trained models (PTMs) faces a critical trade-off between efficient adaptation and long-term stability. While analytic learning enables rapid, recursive closed-form updates, its efficacy is often compromised by accumulated errors and feature incompatibility. In this paper, we first conduct a systematic study to dissect the failure modes of PTM-based analytic CIL, identifying representation rigidity as the primary bottleneck. Motivated by these insights, we propose \textbf{VILA}, a novel dual-branch framework that advances analytic CIL via a two-level vision-language calibration strategy. Specifically, we coherently fuse plastic, task-adapted features with a frozen, universal semantic anchor at the feature level through geometric calibration, and leverage cross-modal priors at the decision level to rectify prediction bias. This confluence maintains analytic-learning's extreme efficiency while overcoming its inherent brittleness. Extensive experiments across eight benchmarks demonstrate that VILA consistently yields superior performance, particularly in fine-grained and long-sequence scenarios. Our framework harmonizes high-fidelity prediction with the simplicity of analytic learning. Our code is available at https://github.com/byzhaoAI/VILA
Authors: Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma
Abstract: Correlation Clustering (CC) is a fundamental unsupervised learning primitive whose strongest LP-based approximation guarantees require $\Theta(n^3)$ triangle inequality constraints and are prohibitive at scale. We initiate the study of \emph{sparsification--approximation trade-offs} for CC, asking how much edge information is needed to retain LP-based guarantees. We establish a structural dichotomy between pseudometric and general weighted instances. On the positive side, we prove that the VC dimension of the clustering disagreement class is exactly $n{-}1$, yielding additive $\varepsilon$-coresets of optimal size $\tilde{O}(n/\varepsilon^2)$; that at most $\binom{n}{2}$ triangle inequalities are active at any LP vertex, enabling an exact cutting-plane solver; and that a sparsified variant of LP-PIVOT, which imputes missing LP marginals via triangle inequalities, achieves a robust $\frac{10}{3}$-approximation (up to an additive term controlled by an empirically computable imputation-quality statistic $\overline{\Gamma}_w$) once $\tilde{\Theta}(n^{3/2})$ edges are observed, a threshold we prove is sharp. On the negative side, we show via Yao's minimax principle that without pseudometric structure, any algorithm observing $o(n)$ uniformly random edges incurs an unbounded approximation ratio, demonstrating that the pseudometric condition governs not only tractability but also the robustness of CC to incomplete information.
Authors: Aritra Das (Ashoka University), Yashas Shende (Ashoka University), Muskaan Chugh (Ashoka University), Reva Laxmi Chauhan (Ashoka University), Arghya Pathak (Ashoka University), Debayan Gupta (Ashoka University)
Abstract: Magnetic-anomaly navigation, leveraging small-scale variations in the Earth's magnetic field, is a promising alternative when GPS is unavailable or compromised. Airborne systems face a key challenge in extracting geomagnetic field data: the aircraft itself induces magnetic noise. Although the classical Tolles-Lawson model addresses this, it inadequately handles stochastically corrupted magnetic data required for navigation. To address stochastic noise, we propose a framework based on two physics-based constraints: divergence-free vector field and E(3)-equivariance. These ensure the learned magnetic field obeys Maxwell's equations and that outputs transform correctly with sensor position/orientation. The divergence-free constraint is implemented by training a neural network to output a vector potential $A$, with the magnetic field defined as its curl. For E(3)-equivariance, we use tensor products of geometric tensors representable via spherical harmonics with known rotational transformations. Enforcing physical consistency and restricting the admissible function space acts as an implicit regularizer that improves spatio-temporal performance. We present ablation studies evaluating each constraint alone and jointly across CNNs, MLPs, Liquid Time Constant models, and Contiformers. Continuous-time dynamics and long-term memory are critical for modelling magnetic time series; the Contiformer architecture, which provides both, outperforms state-of-the-art methods. To mitigate data scarcity, we generate synthetic datasets using the World Magnetic Model (WMM) with time-series conditional GANs, producing realistic, temporally consistent magnetic sequences across varied trajectories and environments. Experiments show that embedding these constraints significantly improves predictive accuracy and physical plausibility, outperforming classical and unconstrained deep learning approaches.
Authors: Sophie Ostmeier, Brian Axelrod, Maya Varma, Asad Aali, Yabin Zhang, Magdalini Paschali, Sanmi Koyejo, Curtis Langlotz, Akshay Chaudhari
Abstract: Large language models (LLMs) often generate plausible yet incorrect answers, posing risks in safety-critical settings such as medicine. Human evaluation is expensive, and LLM-as-judge approaches risk introducing hidden errors. Recent white-box methods detect contextual hallucinations using model internals, focusing on the localization of the attention mass, but two questions remain open: do these approaches extend to predicting answer correctness, and do they generalize out-of-domains? We introduce Head Entropy, a method that predicts answer correctness from attention entropy patterns, specifically measuring the spread of the attention mass. Using sparse logistic regression on per-head 2-Renyi entropies, Head Entropy matches or exceeds baselines in-distribution and generalizes substantially better on out-of-domains, it outperforms the closest baseline on average by +8.5% AUROC. We further show that attention patterns over the question/context alone, before answer generation, already carry predictive signal using Head Entropy with on average +17.7% AUROC over the closest baseline. We evaluate across 5 instruction-tuned LLMs and 3 QA datasets spanning general knowledge, multi-hop reasoning, and medicine.
Authors: Orin Levy, Yishay Mansour
Abstract: We present the first high-probability optimal regret bound for a policy optimization technique applied to the problem of stochastic contextual multi-armed bandit (CMAB) with general offline function approximation. Our algorithm is both efficient and achieves an optimal regret bound of $\widetilde{O}(\sqrt{ K|\mathcal{A}|\log|\mathcal{F}|})$, where $K$ is the number of rounds, $\mathcal{A}$ is the set of arms, and $\mathcal{F}$ is the function class used to approximate the losses. Our results bridge the gap between theory and practice, demonstrating that the widely used policy optimization methods for the contextual bandit problem can achieve a rigorously-proved optimal regret bound. We support our theoretical results with an empirical evaluation of our algorithm.
Authors: Orin Levy, Aviv Rosenberg, Alon Cohen, Yishay Mansour
Abstract: We introduce \texttt{OPO-CMDP}, the first policy optimization algorithm for stochastic Contextual Markov Decision Process (CMDPs) under general offline function approximation. Our approach achieves a high probability regret bound of $\widetilde{O}(H^4\sqrt{T|S||A|\log(|\mathcal{F}||\mathcal{P}|)}),$ where $S$ and $A$ denote the state and action spaces, $H$ the horizon length, $T$ the number of episodes, and $\mathcal{F}, \mathcal{P}$ the finite function classes used to approximate the losses and dynamics, respectively. This is the first regret bound with optimal dependence on $|S|$ and $|A|$, directly improving the current state-of-the-art (Qian, Hu, and Simchi-Levi, 2024). These results demonstrate that optimistic policy optimization provides a natural, computationally superior and theoretically near-optimal path for solving CMDPs.
Authors: Xin Yan, Zhenglin Wan, Feiyang Ye, Xingrui Yu, Hangyu Du, Yang You, Ivor Tsang
Abstract: Vision-Language-Action (VLA) models enable instruction-following embodied control, but their large compute and memory footprints hinder deployment on resource-constrained robots and edge platforms. While reducing weights to 1-bit precision through binarization can greatly improve efficiency, existing methods fail to narrow the distribution gap between binarized and full-precision weights, causing quantization errors to accumulate under long-horizon closed-loop execution and severely degrade actions. To fill this gap, we propose HBVLA, a VLA-tailored binarization framework. First, we use a policy-aware enhanced Hessian to identify weights that are truly critical for action generation. Then, we employ a sparse orthogonal transform for non-salient weights to induce a low-entropy intermediate state. Finally, we quantize both salient and non-salient weights in the Harr domain with group-wise 1-bit quantization. We have evaluated our approach on different VLAs: on LIBERO, quantized OpenVLA-OFT retains 92.2% of full-precision performance; on SimplerEnv, quantized CogAct retains 93.6%, significantly outperforming state-of-the-art binarization methods. We further validate our method on real-world evaluation suite and the results show that HBVLA incurs only marginal success-rate degradation compared to the full-precision model, demonstrating robust deployability under tight hardware constraints. Our work provides a practical foundation for ultra-low-bit quantization of VLAs, enabling more reliable deployment on hardware-limited robotic platforms.
Authors: Talha Ansar, Muhammad Mujtaba Abbas, Ramit Debnath, Vivek Dua, Waqar Muhammad Ashraf
Abstract: Industrial thermal power systems have coupled performance variables with hierarchical order of importance, making their simultaneous optimization computationally challenging or infeasible. This barrier limits the integrated and computationally scaleable operation optimization of industrial thermal power systems. To address this issue for large-scale engineering systems, we present a fully machine learning-powered bi-level optimization framework for data-driven optimization of industrial thermal power systems. The objective functions of upper and lower levels are approximated by artificial neural network (ANN) models and the lower-level problem is analytically embedded through Karush-Kuhn-Tucker (KKT) optimality conditions. The reformulated single level optimization framework integrating ANN models and KKT constraints (ANN-KKT) is validated on benchmark problems and on real-world power generation operation of 660 MW coal power plant and 395 MW gas turbine system. The results reveal a comparable solutions obtained from the proposed ANN-KKT framework to the bi-level solutions of the benchmark problems. Marginal computational time requirement (0.22 to 0.88 s) to compute optimal solutions yields 583 MW (coal) and 402 MW (gas turbine) of power output at optimal turbine heat rate of 7337 kJ/kWh and 7542 kJ/kWh, respectively. In addition, the method expands to delineate a feasible and robust operating envelope that accounts for uncertainty in operating variables while maximizing thermal efficiency in various scenarios. These results demonstrate that ANN-KKT offers a scalable and computationally efficient route for hierarchical, data-driven optimization of industrial thermal power systems, achieving energy-efficient operations of large-scale engineering systems and contributing to industry 5.0.
Authors: ZhiMing Li, JiaHe Feng
Abstract: We study matrix-free eigendecomposition under a matrix-vector product (MVP) oracle, where each step observes a covariance operator $C_k = C_{sig} + \sigma_k^2 I + E_k$. Standard stochastic approximation methods either use fixed steps that couple stability to $\|C_k\|_2$, or adapt steps in ways that slow down due to vanishing updates. We introduce a discrete double-bracket flow whose generator is invariant to isotropic shifts, yielding pathwise invariance to $\sigma_k^2 I$ at the discrete-time level. The resulting trajectory and a maximal stable step size $\eta_{max} \propto 1/\|C_e\|_2^2$ depend only on the trace-free covariance $C_e$. We establish global convergence via strict-saddle geometry for the diagonalization objective and an input-to-state stability analysis, with sample complexity scaling as $O(\|C_e\|_2^2 / (\Delta^2 \epsilon))$ under trace-free perturbations. An explicit characterization of degenerate blocks yields an accelerated $O(\log(1/\zeta))$ saddle-escape rate and a high-probability finite-time convergence guarantee.
Authors: Youwei Shu, Shaomian Zheng, Dingnan Jin, Wenjie Qu, Ziyao Guo, Qing Cui, Jun Zhou, Jiaheng Zhang
Abstract: Data quality is a crucial factor in large language models training. While prior work has shown that models trained on smaller, high-quality datasets can outperform those trained on much larger but noisy or low-quality corpora, systematic methods for industrial-scale data selection in instruction tuning remain underexplored. In this work, we study instruction-tuning data selection through the lens of semantic representation similarity and identify a key limitation of state-of-the-art LLM encoders: they produce highly redundant semantic embeddings. To mitigate this redundancy, we propose Compressed Representation Data Selection (CRDS), a novel framework with two variants. CRDS-R applies Rademacher random projection followed by concatenation of transformer hidden-layer representations, while CRDS-W employs whitening-based dimensionality reduction to improve representational quality. Experimental results demonstrate that both variants substantially enhance data quality and consistently outperform state-of-the-art representation-based selection methods. Notably, CRDS-W achieves strong performance using only 3.5% of the data, surpassing the full-data baseline by an average of 0.71% across four datasets. Our code is available at https://github.com/tdano1/CRDS.
Authors: Xiaoyun Yu, Li fan, Xiangfei Qiu, Nanqing Dong, Yonggui Huang, Honggang Qi, Geguang Pu, Wanli Ouyang, Xi Chen, Jilin Hu
Abstract: While Time Series Foundation Models (TSFMs) have demonstrated exceptional performance in generalized forecasting, their performance often degrades significantly when deployed in real-world vertical domains characterized by temporal distribution shifts and domain-specific periodic structures. Current solutions are primarily constrained by two paradigms: Domain-Adaptive Pretraining (DAPT), which improves short-term domain fitting but frequently disrupts previously learned global temporal patterns due to catastrophic forgetting; and Retrieval-Augmented Generation (RAG), which incorporates external knowledge but introduces substantial retrieval overhead. This creates a severe scalability bottleneck that fails to meet the high-efficiency requirements of real-time stream processing. To break this impasse, we propose Memory for Time Series (MEMTS), a lightweight and plug-and-play method for retrieval-free domain adaptation in time series forecasting. The key component of MEMTS is a Knowledge Persistence Module (KPM), which internalizes domain-specific temporal dynamics, such as recurring seasonal patterns and trends into a compact set of learnable latent prototypes. In doing so, it transforms fragmented historical observations into continuous, parameterized knowledge representations. This paradigm shift enables MEMTS to achieve accurate domain adaptation with constant-time inference and near-zero latency, while effectively mitigating catastrophic forgetting of general temporal patterns, all without requiring any architectural modifications to the frozen TSFM backbone. Extensive experiments on multiple datasets demonstrate the SOTA performance of MEMTS.
Authors: Marc Boubnovski Martell, Josefa Lia Stoisser, Lawrence Phillips, Aditya Misra, Robert Kitchen, Jesper Ferkinghoff-Borg, Jialin Yu, Philip Torr, Kaspar M\"arten
Abstract: Predicting transcriptional responses to unseen genetic perturbations is essential for understanding gene regulation and prioritizing large-scale perturbation experiments. Existing approaches either rely on static, potentially incomplete knowledge graphs, or prompt language models for functionally similar genes, retrieving associations shaped by symmetric co-occurrence in scientific text rather than directed regulatory logic. We introduce MechPert, a lightweight framework that encourages LLM agents to generate directed regulatory hypotheses rather than relying solely on functional similarity. Multiple agents independently propose candidate regulators with associated confidence scores; these are aggregated through a consensus mechanism that filters spurious associations, producing weighted neighborhoods for downstream prediction. We evaluate MechPert on Perturb-seq benchmarks across four human cell lines. For perturbation prediction in low-data regimes ($N=50$ observed perturbations), MechPert improves Pearson correlation by up to 10.5\% over similarity-based baselines. For experimental design, MechPert-selected anchor genes outperform standard network centrality heuristics by up to 46\% in well-characterized cell lines.
Authors: Xiaoyu Tao, Mingyue Cheng, Chuang Jiang, Tian Gao, Huanjian Zhang, Yaguo Liu
Abstract: Time series forecasting has long been dominated by model-centric approaches that formulate prediction as a single-pass mapping from historical observations to future values. Despite recent progress, such formulations often struggle in complex and evolving settings, largely because most forecasting models lack the ability to autonomously acquire informative evidence, reason about potential future changes, or revise predictions through iterative decision processes. In this work, we propose Cast-R1, a learned time series forecasting framework that reformulates forecasting as a sequential decision-making problem. Cast-R1 introduces a memory-based state management mechanism that maintains decision-relevant information across interaction steps, enabling the accumulation of contextual evidence to support long-horizon reasoning. Building on this formulation, forecasting is carried out through a tool-augmented agentic workflow, in which the agent autonomously interacts with a modular toolkit to extract statistical features, invoke lightweight forecasting models for decision support, perform reasoning-based prediction, and iteratively refine forecasts through self-reflection. To train Cast-R1, we adopt a two-stage learning strategy that combines supervised fine-tuning with multi-turn reinforcement learning, together with a curriculum learning scheme that progressively increases task difficulty to improve policy learning. Extensive experiments on multiple real-world time series datasets demonstrate the effectiveness of Cast-R1. We hope this work provides a practical step towards further exploration of agentic paradigms for time series modeling. Our code is available at https://github.com/Xiaoyu-Tao/Cast-R1-TS.
Authors: Yutong Du, Zicheng Liu, Yi Huang, Bazargul Matkerim, Bo Qi, Yali Zong, Peixian Han
Abstract: Untrained neural networks (UNNs) offer high-fidelity electromagnetic inverse scattering reconstruction but are computationally limited by high-dimensional spatial-domain optimization. We propose a Real-Time Physics-Driven Fourier-Spectral (PDF) solver that achieves sub-second reconstruction through spectral-domain dimensionality reduction. By expanding induced currents using a truncated Fourier basis, the optimization is confined to a compact low-frequency parameter space supported by scattering measurements. The solver integrates a contraction integral equation (CIE) to mitigate high-contrast nonlinearity and a contrast-compensated operator (CCO) to correct spectral-induced attenuation. Furthermore, a bridge-suppressing loss is formulated to enhance boundary sharpness between adjacent scatterers. Numerical and experimental results demonstrate a 100-fold speedup over state-of-the-art UNNs with robust performance under noise and antenna uncertainties, enabling real-time microwave imaging applications.
Authors: Xiaoyu Tao, Yuchong Wu, Mingyue Cheng, Ze Guo, Tian Gao
Abstract: Time series anomaly detection is critical in many real-world applications, where effective solutions must localize anomalous regions and support reliable decision-making under complex settings. However, most existing methods frame anomaly detection as a purely discriminative prediction task with fixed feature inputs, rather than an evidence-driven diagnostic process. As a result, they often struggle when anomalies exhibit strong context dependence or diverse patterns. We argue that these limitations stem from the lack of adaptive feature preparation, reasoning-aware detection, and iterative refinement during inference. To address these challenges, we propose AnomaMind, an agentic time series anomaly detection framework that reformulates anomaly detection as a sequential decision-making process. AnomaMind operates through a structured workflow that progressively localizes anomalous intervals in a coarse-to-fine manner, augments detection through multi-turn tool interactions for adaptive feature preparation, and refines anomaly decisions via self-reflection. The workflow is supported by a set of reusable tool engines, enabling context-aware diagnostic analysis. A key design of AnomaMind is an explicitly designed hybrid inference mechanism for tool-augmented anomaly detection. In this mechanism, general-purpose models are responsible for autonomous tool interaction and self-reflective refinement, while core anomaly detection decisions are learned through reinforcement learning under verifiable workflow-level feedback, enabling task-specific optimization within a flexible reasoning framework. Extensive experiments across diverse settings demonstrate that AnomaMind consistently improves anomaly detection performance. The code is available at https://anonymous.4open.science/r/AnomaMind.
Authors: Guojian Zhan, Letian Tao, Pengcheng Wang, Yixiao Wang, Yiheng Li, Yuxin Chen, Masayoshi Tomizuka, Shengbo Eben Li
Abstract: Learning expressive and efficient policy functions is a promising direction in reinforcement learning (RL). While flow-based policies have recently proven effective in modeling complex action distributions with a fast deterministic sampling process, they still face a trade-off between expressiveness and computational burden, which is typically controlled by the number of flow steps. In this work, we propose mean velocity policy (MVP), a new generative policy function that models the mean velocity field to achieve the fastest one-step action generation. To ensure its high expressiveness, an instantaneous velocity constraint (IVC) is introduced on the mean velocity field during training. We theoretically prove that this design explicitly serves as a crucial boundary condition, thereby improving learning accuracy and enhancing policy expressiveness. Empirically, our MVP achieves state-of-the-art success rates across several challenging robotic manipulation tasks from Robomimic and OGBench. It also delivers substantial improvements in training and inference speed over existing flow-based policy baselines.
Authors: Jorge Carrasco-Pollo, Floor Eijkelboom, Jan-Willem van de Meent
Abstract: We introduce Pawsterior, a variational flow-matching framework for improved and extended simulation-based inference (SBI). Many SBI problems involve posteriors constrained by structured domains, such as bounded physical parameters or hybrid discrete-continuous variables, yet standard flow-matching methods typically operate in unconstrained spaces. This mismatch leads to inefficient learning and difficulty respecting physical constraints. Our contributions are twofold. First, generalizing the geometric inductive bias of CatFlow, we formalize endpoint-induced affine geometric confinement, a principle that incorporates domain geometry directly into the inference process via a two-sided variational model. This formulation improves numerical stability during sampling and leads to consistently better posterior fidelity, as demonstrated by improved classifier two-sample test performance across standard SBI benchmarks. Second, and more importantly, our variational parameterization enables SBI tasks involving discrete latent structure (e.g., switching systems) that are fundamentally incompatible with conventional flow-matching approaches. By addressing both geometric constraints and discrete latent structure, Pawsterior extends flow-matching to a broader class of structured SBI problems that were previously inaccessible.
Authors: Shalev Shaer, Yarin Bar, Drew Prinster, Yaniv Romano
Abstract: We propose a sequential test for detecting arbitrary distribution shifts that allows conformal test martingales (CTMs) to work under a fixed, reference-conditional setting. Existing CTM detectors construct test martingales by continually growing a reference set with each incoming sample, using it to assess how atypical the new sample is relative to past observations. While this design yields anytime-valid type-I error control, it suffers from test-time contamination: after a change, post-shift observations enter the reference set and dilute the evidence for distribution shift, increasing detection delay and reducing power. In contrast, our method avoids contamination by design by comparing each new sample to a fixed null reference dataset. Our main technical contribution is a robust martingale construction that remains valid conditional on the null reference data, achieved by explicitly accounting for the estimation error in the reference distribution induced by the finite reference set. This yields anytime-valid type-I error control together with guarantees of asymptotic power one and bounded expected detection delay. Empirically, our method detects shifts faster than standard CTMs, providing a powerful and reliable distribution-shift detector.
Authors: Weixuan Yuan, Zengrui Jin, Yichen Wang, Donglin Xie, Ziyi Ye, Chao Zhang, Xuesong Chen
Abstract: Tasks ranging from sleep staging to clinical diagnosis traditionally rely on standard polysomnography (PSG) devices, bedside monitors and wearable devices, which capture diverse nocturnal biosignals (e.g., EEG, EOG, ECG, SpO$_2$). However, heterogeneity across devices and frequent sensor dropout pose significant challenges for unified modelling of these multimodal signals. We present \texttt{sleep2vec}, a foundation model for diverse and incomplete nocturnal biosignals that learns a shared representation via cross-modal alignment. \texttt{sleep2vec} is contrastively pre-trained on 42,249 overnight recordings spanning nine modalities using a \textit{Demography, Age, Site \& History-aware InfoNCE} objective that incorporates physiological and acquisition metadata (\textit{e.g.}, age, gender, recording site) to dynamically weight negatives and mitigate cohort-specific shortcuts. On downstream sleep staging and clinical outcome assessment, \texttt{sleep2vec} consistently outperforms strong baselines and remains robust to any subset of available modalities and sensor dropout. We further characterize, to our knowledge for the first time, scaling laws for nocturnal biosignals with respect to modality diversity and model capacity. Together, these results show that unified cross-modal alignment, coupled with principled scaling, enables label-efficient, general-purpose modelling of real-world nocturnal biosignals.
Authors: Ouns El Harzli, Yoonsoo Nam, Ilja Kuzborskij, Bernardo Cuenca Grau, Ard A. Louis
Abstract: Algorithmic stability is a classical framework for analyzing the generalization error of learning algorithms. It predicts that an algorithm has small generalization error if it is insensitive to small perturbations in the training set such as the removal or replacement of a training point. While stability has been demonstrated for numerous well-known algorithms, this framework has had limited success in analyses of deep neural networks. In this paper we study the algorithmic stability of deep ReLU homogeneous neural networks that achieve zero training error using parameters with the smallest $L_2$ norm, also known as the minimum-norm interpolation, a phenomenon that can be observed in overparameterized models trained by gradient-based algorithms. We investigate sufficient conditions for such networks to be stable. We find that 1) such networks are stable when they contain a (possibly small) stable sub-network, followed by a layer with a low-rank weight matrix, and 2) such networks are not guaranteed to be stable even when they contain a stable sub-network, if the following layer is not low-rank. The low-rank assumption is inspired by recent empirical and theoretical results which demonstrate that training deep neural networks is biased towards low-rank weight matrices, for minimum-norm interpolation and weight-decay regularization.
Authors: Juntong Wang, Libin Chen, Xiyuan Wang, Shijia Kang, Haotong Yang, Da Zheng, Muhan Zhang
Abstract: Repository-level bug localization-the task of identifying where code must be modified to fix a bug-is a critical software engineering challenge. Standard Large Language Modles (LLMs) are often unsuitable for this task due to context window limitations that prevent them from processing entire code repositories. As a result, various retrieval methods are commonly used, including keyword matching, text similarity, and simple graph-based heuristics such as Breadth-First Search. Graph Neural Networks (GNNs) offer a promising alternative due to their ability to model complex, repository-wide dependencies; however, their application has been hindered by the lack of a dedicated benchmark. To address this gap, we introduce GREPO, the first GNN benchmark for repository-scale bug localization tasks. GREPO comprises 86 Python repositories and 47294 bug-fixing tasks, providing graph-based data structures ready for direct GNN processing. Our evaluation of various GNN architectures shows outstanding performance compared to established information retrieval baselines. This work highlights the potential of GNNs for bug localization and established GREPO as a foundation resource for future research, The code is available at https://github.com/qingpingmo/GREPO.
Authors: Zhimin Zhao
Abstract: Code generation has progressed more reliably than reinforcement learning, largely because code has an information structure that makes it learnable. Code provides dense, local, verifiable feedback at every token, whereas most reinforcement learning problems do not. This difference in feedback quality is not binary but graded. We propose a five-level hierarchy of learnability based on information structure and argue that the ceiling on ML progress depends less on model size than on whether a task is learnable at all. The hierarchy rests on a formal distinction among three properties of computational problems (expressibility, computability, and learnability). We establish their pairwise relationships, including where implications hold and where they fail, and present a unified template that makes the structural differences explicit. The analysis suggests why supervised learning on code scales predictably while reinforcement learning does not, and why the common assumption that scaling alone will solve remaining ML challenges warrants scrutiny.
Authors: Dat Le, Duc-Cuong Le, Anh-Son Nguyen, Tuan-Dung Bui, Thu-Trang Nguyen, Son Nguyen, Hieu Dinh Vo
Abstract: Automated Machine Learning (AutoML) has revolutionized the development of data-driven solutions; however, traditional frameworks often function as "black boxes", lacking the flexibility and transparency required for complex, real-world engineering tasks. Recent Large Language Model (LLM)-based agents have shifted toward code-driven approaches. However, they frequently suffer from hallucinated logic and logic entanglement, where monolithic code generation leads to unrecoverable runtime failures. In this paper, we present iML, a novel multi-agent framework designed to shift AutoML from black-box prompting to a code-guided, modular, and verifiable architectural paradigm. iML introduces three main ideas: (1) Code-Guided Planning, which synthesizes a strategic blueprint grounded in autonomous empirical profiling to eliminate hallucination; (2) Code-Modular Implementation, which decouples preprocessing and modeling into specialized components governed by strict interface contracts; and (3) Code-Verifiable Integration, which enforces physical feasibility through dynamic contract verification and iterative self-correction. We evaluate iML across MLE-BENCH and the newly introduced iML-BENCH, comprising a diverse range of real-world Kaggle competitions. The experimental results show iML's superiority over state-of-the-art agents, achieving a valid submission rate of 85% and a competitive medal rate of 45% on MLE-BENCH, with an average standardized performance score (APS) of 0.77. On iML-BENCH, iML significantly outperforms the other approaches by 38%-163% in APS. Furthermore, iML maintains a robust 70% success rate even under stripped task descriptions, effectively filling information gaps through empirical profiling. These results highlight iML's potential to bridge the gap between stochastic generation and reliable engineering, marking a meaningful step toward truly AutoML.
Authors: Adolfo Gonz\'alez, V\'ictor Parada
Abstract: Business environments characterized by structural demand intermittency, high variability, and multi-step planning horizons require robust and reproducible model selection mechanisms. Empirical evidence shows that no forecasting model is universally dominant and that relative rankings vary across error metrics, demand regimes, and forecast horizons, generating ambiguity in multi-SKU decision contexts. This study proposes AHSIV (Adaptive Hybrid Selector for Intermittency and Variability), a horizon-aware and regime-conditioned model selection framework designed to address horizon-induced ranking instability. The proposed approach integrates scaled and absolute error metrics adjusted through a Metric Degradation by Forecast Horizon (MDFH) procedure, structural demand classification, multi-objective Pareto dominance, and hierarchical bias refinement within a unified decision architecture. The empirical evaluation is conducted on the Walmart, M3, M4, and M5 datasets under multiple train-test partition schemes and twelve-step forecasting horizons. Results indicate that AHSIV achieves statistical equivalence with the strongest monometric baseline in terms of aggregated performance while increasing the frequency of horizon-specific best-model selection. The findings demonstrate that model selection in heterogeneous demand environments cannot be treated as a static ranking problem, and that horizon-consistent, structurally adaptive mechanisms provide a principled, operationally coherent solution for multi-SKU forecasting.
Authors: Sam Dauncey, Roger Wattenhofer
Abstract: Tokenization is a hardcoded compression step which remains in the training pipeline of Large Language Models (LLMs), despite a general trend towards architectures becoming increasingly end-to-end. Prior work has shown promising results at scale in bringing this compression step inside the LLMs' architecture with heuristics to draw token boundaries, and also attempts to learn these token boundaries with straight-through estimates, which treat the problem of drawing discrete token boundaries as a continuous one. We show that these token boundaries can instead be learned using score function estimates, which have tighter theoretical guarantees due to directly optimizing the problem of drawing discrete token boundaries to minimize loss. We observe that techniques from reinforcement learning, such as time discounting, are necessary to reduce the variance of this score function sufficiently to make it practicable. We demonstrate that the resultant method outperforms prior proposed straight-through estimates, both qualitatively and quantitatively at the $100$ million parameter scale.
Authors: Taiwei Shi, Sihao Chen, Bowen Jiang, Linxin Song, Longqi Yang, Jieyu Zhao
Abstract: Reinforcement learning has become the central approach for language models (LMs) to learn from environmental reward or feedback. In practice, the environmental feedback is usually sparse and delayed. Learning from such signals is challenging, as LMs must implicitly infer how observed failures should translate into behavioral changes for future iterations. We introduce Experiential Reinforcement Learning (ERL), a training paradigm that embeds an explicit experience-reflection-consolidation loop into the reinforcement learning process. Given a task, the model generates an initial attempt, receives environmental feedback, and produces a reflection that guides a refined second attempt, whose success is reinforced and internalized into the base policy. This process converts feedback into structured behavioral revision, improving exploration and stabilizing optimization while preserving gains at deployment without additional inference cost. Across sparse-reward control environments and agentic reasoning benchmarks, ERL consistently improves learning efficiency and final performance over strong reinforcement learning baselines, achieving gains of up to +81% in complex multi-step environments and up to +11% in tool-using reasoning tasks. These results suggest that integrating explicit self-reflection into policy training provides a practical mechanism for transforming feedback into durable behavioral improvement.
Authors: Yuhang Li, Reena Elangovan, Xin Dong, Priyadarshini Panda, Brucek Khailany
Abstract: Reinforcement learning with verifiable rewards (RLVR) has become a trending paradigm for training reasoning large language models (LLMs). However, due to the autoregressive decoding nature of LLMs, the rollout process becomes the efficiency bottleneck of RL training, consisting of up to 70\% of the total training time. In this work, we propose Quantized Reinforcement Learning (QuRL) that uses a quantized actor for accelerating the rollout. We address two challenges in QuRL. First, we propose Adaptive Clipping Range (ACR) that dynamically adjusts the clipping ratio based on the policy ratio between the full-precision actor and the quantized actor, which is essential for mitigating long-term training collapse. Second, we identify the weight update problem, where weight changes between RL steps are extremely small, making it difficult for the quantization operation to capture them effectively. We mitigate this problem through the invariant scaling technique that reduces quantization noise and increases weight update. We evaluate our method with INT8 and FP8 quantization experiments on DeepScaleR and DAPO, and achieve 20% to 80% faster rollout during training.
Authors: Ho-Hsuan Wang, Afnan Sultan, Andrea Volkamer, Dietrich Klakow
Abstract: Language models are widely used in chemistry for molecular property prediction and small-molecule generation, yet Natural Products (NPs) remain underexplored despite their importance in drug discovery. To address this gap, we develop NP-specific chemical language models (NPCLMs) by pre-training state-space models (Mamba and Mamba-2) and comparing them with transformer baselines (GPT). Using a dataset of about 1M NPs, we present the first systematic comparison of selective state-space models and transformers for NP-focused tasks, together with eight tokenization strategies including character-level, Atom-in-SMILES (AIS), byte-pair encoding (BPE), and NP-specific BPE. We evaluate molecule generation (validity, uniqueness, novelty) and property prediction (membrane permeability, taste, anti-cancer activity) using MCC and AUC-ROC. Mamba generates 1-2 percent more valid and unique molecules than Mamba-2 and GPT, with fewer long-range dependency errors, while GPT yields slightly more novel structures. For property prediction, Mamba variants outperform GPT by 0.02-0.04 MCC under random splits, while scaffold splits show comparable performance. Results demonstrate that domain-specific pre-training on about 1M NPs can match models trained on datasets over 100 times larger.
Authors: Zedong Wang, Yuyang Wang, Ijay Narang, Felix Wang, Yuzhou Wang, Siva Theja Maguluri
Abstract: Constant-stepsize stochastic approximation (SA) is widely used in learning for computational efficiency. For a fixed stepsize, the iterates typically admit a stationary distribution that is rarely tractable. Prior work shows that as the stepsize $\alpha \downarrow 0$, the centered-and-scaled steady state converges weakly to a Gaussian random vector. However, for fixed $\alpha$, this weak convergence offers no usable error bound for approximating the steady-state by its Gaussian limit. This paper provides explicit, non-asymptotic error bounds for fixed $\alpha$. We first prove general-purpose theorems that bound the Wasserstein distance between the centered-scaled steady state and an appropriate Gaussian distribution, under regularity conditions for drift and moment conditions for noise. To ensure broad applicability, we cover both i.i.d. and Markovian noise models. We then instantiate these theorems for three representative SA settings: (1) stochastic gradient descent (SGD) for smooth strongly convex objectives, (2) linear SA, and (3) contractive nonlinear SA. We obtain dimension- and stepsize-dependent, explicit bounds in Wasserstein distance of order $\alpha^{1/2}\log(1/\alpha)$ for small $\alpha$. Building on the Wasserstein approximation error, we further derive non-uniform Berry--Esseen-type tail bounds that compare the steady-state tail probability to Gaussian tails. We achieve an explicit error term that decays in both the deviation level and stepsize $\alpha$. We adapt the same analysis for SGD beyond strongly convexity and study general convex objectives. We identify a non-Gaussian (Gibbs) limiting law under the correct scaling, which is validated numerically, and provide a corresponding pre-limit Wasserstein error bound.
Authors: Liangyu Su, Jun Shu, Rui Liu, Deyu Meng, Zongben Xu
Abstract: Representing and predicting high-dimensional and spatiotemporally chaotic dynamical systems remains a fundamental challenge in dynamical systems and machine learning. Although data-driven models can achieve accurate short-term forecasts, they often lack stability, interpretability, and scalability in regimes dominated by broadband or continuous spectra. Koopman-based approaches provide a principled linear perspective on nonlinear dynamics, but existing methods rely on restrictive finite-dimensional assumptions or explicit spectral parameterizations that degrade in high-dimensional settings. Against these issues, we introduce KoopGen, a generator-based neural Koopman framework that models dynamics through a structured, state-dependent representation of Koopman generators. By exploiting the intrinsic Cartesian decomposition into skew-adjoint and self-adjoint components, KoopGen separates conservative transport from irreversible dissipation while enforcing exact operator-theoretic constraints during learning. Across systems ranging from nonlinear oscillators to high-dimensional chaotic and spatiotemporal dynamics, KoopGen improves prediction accuracy and stability, while clarifying which components of continuous-spectrum dynamics admit interpretable and learnable representations.
Authors: Chenyue Li, Wen Deng, Zhuotao Sun, Mengxi Jin, Hanzhe Cui, Han Li, Shentong Li, Man Kit Yu, Ming Long Lai, Yuhao Yang, Mengqian Lu, Binhang Yuan
Abstract: Subseasonal-to-seasonal (S2S) forecasts play an essential role in providing a decision-critical weeks-to-months planning window for climate resilience and sustainability, yet a growing bottleneck is the last-mile gap: translating scientific forecasts into trusted, actionable climate services, requiring reliable multimodal understanding and decision-facing reasoning under uncertainty. Meanwhile, multimodal large language models (MLLMs) and corresponding agentic paradigms have made rapid progress in supporting various workflows, but it remains unclear whether they can reliably generate decision-making deliverables from operational service products (e.g., actionable signal comprehension, decision-making handoff, and decision analysis & planning) under uncertainty. We introduce S2SServiceBench, a multimodal benchmark for last-mile S2S climate services curated from an operational climate-service system to evaluate this capability. S2SServiceBenchcovers 10 service products with about 150+ expert-selected cases in total, spanning six application domains - Agriculture, Disasters, Energy, Finance, Health, and Shipping. Each case is instantiated at three service levels, yielding around 500 tasks and 1,000+ evaluation items across climate resilience and sustainability applications. Using S2SServiceBench, we benchmark state-of-the-art MLLMs and agents, and analyze performance across products and service levels, revealing persistent challenges in S2S service plot understanding and reasoning - namely, actionable signal comprehension, operationalizing uncertainty into executable handoffs, and stable, evidence-grounded analysis and planning for dynamic hazards-while offering actionable guidance for building future climate-service agents.
Authors: Xinxing Zhou, Qingren Yao, Yiji Zhao, Chenghao Liu, Flora Salim, Xiaojie Yuan, Yanlong Wen, Ming Jin
Abstract: Most time series foundation models are pretrained by directly predicting future observations, which often yields weakly structured latent representations that capture surface noise rather than coherent and predictable temporal dynamics. In this work, we introduce EIDOS, a foundation model family that shifts pretraining from future value prediction to latent-space predictive learning. We train a causal Transformer to predict the evolution of latent representations, encouraging the emergence of structured and temporally coherent latent states. To ensure stable targets for latent-space learning, we design a lightweight aggregation branch to construct target representations. EIDOS is optimized via a joint objective that integrates latent-space alignment, observational grounding to anchor representations to the input signal, and direct forecasting supervision. On the GIFT-Eval benchmark, EIDOS mitigates structural fragmentation in the representation space and achieves state-of-the-art performance. These results demonstrate that constraining models to learn predictable latent dynamics is a principled step toward more robust and reliable time series foundation models.
Authors: Yue Wang, Areg Karapetyan, Djellel Difallah, Samer Madanat
Abstract: Spatio-temporal traffic forecasting is a core component of intelligent transportation systems, supporting various downstream tasks such as signal control and network-level traffic management. In real-world deployments, forecasting models must operate under structural and observational uncertainties, conditions that are rarely considered in model design. Recent approaches achieve strong short-term predictive performance by tightly coupling spatial and temporal modeling, often at the cost of increased complexity and limited modularity. In contrast, efficient time-series models capture long-range temporal dependencies without relying on explicit network structure. We propose UniST-Pred, a unified spatio-temporal forecasting framework that first decouples temporal modeling from spatial representation learning, then integrates both through adaptive representation-level fusion. To assess robustness of the proposed approach, we construct a dataset based on an agent-based, microscopic traffic simulator (MATSim) and evaluate UniST-Pred under severe network disconnection scenarios. Additionally, we benchmark UniST-Pred on standard traffic prediction datasets, demonstrating its competitive performance against existing well-established models despite a lightweight design. The results illustrate that UniST-Pred maintains strong predictive performance across both real-world and simulated datasets, while also yielding interpretable spatio-temporal representations under infrastructure disruptions. The source code and the generated dataset are available at https://anonymous.4open.science/r/UniST-Pred-EF27
Authors: Atsushi Shimizu, Shohei Taniguchi, Yutaka Matsuo
Abstract: Length generalization is the ability of language models to maintain performance on inputs longer than those seen during pretraining. In this work, we introduce a simple yet powerful position encoding (PE) strategy, Random Float Sampling (RFS), that generalizes well to lengths unseen during pretraining or fine-tuning. In particular, instead of selecting position indices from a predefined discrete set, RFS uses randomly sampled continuous values, thereby avoiding out-of-distribution (OOD) issues on unseen lengths by exposing the model to diverse indices during training. Since assigning indices to tokens is a common and fundamental procedure in widely used PEs, the advantage of RFS can easily be incorporated into, for instance, the absolute sinusoidal encoding, RoPE, and ALiBi. Experiments corroborate its effectiveness by showing that RFS results in superior performance in length generalization tasks as well as zero-shot commonsense reasoning benchmarks.
Authors: Kai Zhang, Xuanyu Cao, Khaled B. Letaief
Abstract: Decentralized federated learning (DFL) enables edge devices to collaboratively train models through local training and fully decentralized device-to-device (D2D) model exchanges. However, these energy-intensive operations often rapidly deplete limited device batteries, reducing their operational lifetime and degrading the learning performance. To address this limitation, we apply energy harvesting technique to DFL systems, allowing edge devices to extract ambient energy and operate sustainably. We first derive the convergence bound for wireless DFL with energy harvesting, showing that the convergence is influenced by partial device participation and transmission packet drops, both of which further depend on the available energy supply. To accelerate convergence, we formulate a joint device scheduling and power control problem and model it as a multi-agent Markov decision process (MDP). Traditional MDP algorithms (e.g., value or policy iteration) require a centralized coordinator with access to all device states and exhibit exponential complexity in the number of devices, making them impractical for large-scale decentralized networks. To overcome these challenges, we propose a fully decentralized policy iteration algorithm that leverages only local state information from two-hop neighboring devices, thereby substantially reducing both communication overhead and computational complexity. We further provide a theoretical analysis showing that the proposed decentralized algorithm achieves asymptotic optimality. Finally, comprehensive numerical experiments on real-world datasets are conducted to validate the theoretical results and corroborate the effectiveness of the proposed algorithm.
Authors: Yaqian Zhang, Bernhard Pfahringer, Eibe Frank, Albert Bifet
Abstract: Despite their success, large pretrained vision models remain vulnerable to catastrophic forgetting when adapted to new tasks in class-incremental settings. Parameter-efficient fine-tuning (PEFT) alleviates this by restricting trainable parameters, yet most approaches still rely on cross-entropy (CE) loss, a surrogate for the 0-1 loss, to learn from new data. We revisit this choice and revive the true objective (0-1 loss) through a reinforcement learning perspective. By formulating classification as a one-step Markov Decision Process, we derive an Expected Policy Gradient (EPG) method that directly minimizes misclassification error with a low-variance gradient estimation. Our analysis shows that CE can be interpreted as EPG with an additional sample-weighting mechanism: CE encourages exploration by emphasizing low-confidence samples, while EPG prioritizes high-confidence ones. Building on this insight, we propose adaptive entropy annealing (aEPG), a training strategy that transitions from exploratory (CE-like) to exploitative (EPG-like) learning. aEPG-based methods outperform CE-based methods across diverse benchmarks and with various PEFT modules. More broadly, we evaluate various entropy regularization methods and demonstrate that lower entropy of the output prediction distribution enhances adaptation in pretrained vision models.
Authors: Jae-Hwan Choi, Jiwoo Yoon, Dohyun Kwon, Jaewoong Choi
Abstract: We study Neural Optimal Transport in infinite-dimensional Hilbert spaces. In non-regular settings, Semi-dual Neural OT often generates spurious solutions that fail to accurately capture target distributions. We analytically characterize this spurious solution problem using the framework of regular measures, which generalize Lebesgue absolute continuity in finite dimensions. To resolve ill-posedness, we extend the semi-dual framework via a Gaussian smoothing strategy based on Brownian motion. Our primary theoretical contribution proves that under a regular source measure, the formulation is well-posed and recovers a unique Monge map. Furthermore, we establish a sharp characterization for the regularity of smoothed measures, proving that the success of smoothing depends strictly on the kernel of the covariance operator. Empirical results on synthetic functional data and time-series datasets demonstrate that our approach effectively suppresses spurious solutions and outperforms existing baselines.
Authors: Luigi Ciceri, Corrado Mio, Jianyi Lin, Gabriele Gianini
Abstract: Predicting flows that occur both through and around porous bodies is challenging due to coupled physics across fluid and porous regions and the need to generalize across diverse geometries and boundary conditions. We address this problem using two Physics Informed learning approaches: Physics Informed PointNets (PIPN) and Physics Informed Geometry Aware Neural Operator (P-IGANO). We enforce the incompressible Navier Stokes equations in the free-flow region and a Darcy Forchheimer extension in the porous region within a unified loss and condition the networks on geometry and material parameters. Datasets are generated with OpenFOAM on 2D ducts containing porous obstacles and on 3D windbreak scenarios with tree canopies and buildings. We first verify the pipeline via the method of manufactured solutions, then assess generalization to unseen shapes, and for PI-GANO, to variable boundary conditions and parameter settings. The results show consistently low velocity and pressure errors in both seen and unseen cases, with accurate reproduction of the wake structures. Performance degrades primarily near sharp interfaces and in regions with large gradients. Overall, the study provides a first systematic evaluation of PIPN/PI-GANO for simultaneous through-and-around porous flows and shows their potential to accelerate design studies without retraining per geometry.
Authors: Anton Korznikov, Andrey Galichin, Alexey Dontsov, Oleg Rogov, Ivan Oseledets, Elena Tutubalina
Abstract: Sparse Autoencoders (SAEs) have emerged as a promising tool for interpreting neural networks by decomposing their activations into sparse sets of human-interpretable features. Recent work has introduced multiple SAE variants and successfully scaled them to frontier models. Despite much excitement, a growing number of negative results in downstream tasks casts doubt on whether SAEs recover meaningful features. To directly investigate this, we perform two complementary evaluations. On a synthetic setup with known ground-truth features, we demonstrate that SAEs recover only $9\%$ of true features despite achieving $71\%$ explained variance, showing that they fail at their core task even when reconstruction is strong. To evaluate SAEs on real activations, we introduce three baselines that constrain SAE feature directions or their activation patterns to random values. Through extensive experiments across multiple SAE architectures, we show that our baselines match fully-trained SAEs in interpretability (0.87 vs 0.90), sparse probing (0.69 vs 0.72), and causal editing (0.73 vs 0.72). Together, these results suggest that SAEs in their current state do not reliably decompose models' internal mechanisms.
Authors: Xuanbo Su, Hao Luo, Yingfang Zhang, Lijun Zhang
Abstract: Activation steering provides parameter-efficient control over large language models (LLMs) at inference time, but many methods rely on off-distribution supervision and discrete masking, leading to brittle interventions. We propose ROAST (Rollout-based On-distribution Activation Steering Technique), which estimates steering directions from the model's own on-distribution rollouts via ROC and avoids hard sparsification via Continuous Soft Scaling (CSS) and Grouped Mean Normalization. Our empirical analysis reveals that while activation magnitude correlates moderately with directional consistency, the variance in magnitude is significant and often disproportionate to semantic quality. This suggests that high-magnitude activations risk dominating the global steering direction if not properly normalized. To address this, ROAST employs grouped normalization to balance contributions across samples, ensuring a more robust estimation of the consensus steering direction. Across models (0.6B to 32B), ROAST consistently improves performance on diverse tasks (e.g., +9.7% on GSM8K for Qwen3-0.6B and +12.1% on TruthfulQA for GLM4-32B), and analyses show that CSS better preserves activation energy.
Authors: Yuxuan Linghu, Zhiyuan Liu, Qi Deng
Abstract: Differentiating through the solution of a quadratic program (QP) is a central problem in differentiable optimization. Most existing approaches differentiate through the Karush--Kuhn--Tucker (KKT) system, but their computational cost and numerical robustness can degrade at scale. To address these limitations, we propose dXPP, a penalty-based differentiation framework that decouples QP solving from differentiation. In the solving step (forward pass), dXPP is solver-agnostic and can leverage any black-box QP solver. In the differentiation step (backward pass), we map the solution to a smooth approximate penalty problem and implicitly differentiate through it, requiring only the solution of a much smaller linear system in the primal variables. This approach bypasses the difficulties inherent in explicit KKT differentiation and significantly improves computational efficiency and robustness. We evaluate dXPP on various tasks, including randomly generated QPs, large-scale sparse projection problems, and a real-world multi-period portfolio optimization task. Empirical results demonstrate that dXPP is competitive with KKT-based differentiation methods and achieves substantial speedups on large-scale problems.
Authors: Rizhen Hu, Yuan Cao, Boao Kong, Mou Sun, Kun Yuan
Abstract: Sparse Mixture-of-Experts (MoE) models scale Transformers efficiently but suffer from expert overlap -- redundant representations across experts and routing ambiguity, resulting in severely underutilized model capacity. While architectural solutions like DeepSeekMoE promote specialization, they require substantial structural modifications and rely solely on intra-layer signals. In this paper, we propose two plug-and-play regularization losses that enhance MoE specialization and routing efficiency without modifying router or model architectures. First, an intra-layer specialization loss penalizes cosine similarity between experts' SwiGLU activations on identical tokens, encouraging experts to specialize in complementary knowledge. Second, a cross-layer coupling loss maximizes joint Top-$k$ routing probabilities across adjacent layers, establishing coherent expert pathways through network depth while reinforcing intra-layer expert specialization. Both losses are orthogonal to the standard load-balancing loss and compatible with both the shared-expert architecture in DeepSeekMoE and vanilla top-$k$ MoE architectures. We implement both losses as a drop-in Megatron-LM module. Extensive experiments across pre-training, fine-tuning, and zero-shot benchmarks demonstrate consistent task gains, higher expert specialization, and lower-entropy routing; together, these improvements translate into faster inference via more stable expert pathways.
Authors: Max Fomin
Abstract: Detecting prompt injection and jailbreak attacks is critical for deploying LLM-based agents safely. As agents increasingly process untrusted data from emails, documents, tool outputs, and external APIs, robust attack detection becomes essential. Yet current evaluation practices and production systems have fundamental limitations. We present a comprehensive analysis using a diverse benchmark of 18 datasets spanning harmful requests, jailbreaks, indirect prompt injections, and extraction attacks. We propose Leave-One-Dataset-Out (LODO) evaluation to measure true out-of-distribution generalization, revealing that the standard practice of train-test splits from the same dataset sources severely overestimates performance: aggregate metrics show an 8.4 percentage point AUC inflation, but per-dataset gaps range from 1% to 25% accuracy-exposing heterogeneous failure modes. To understand why classifiers fail to generalize, we analyze Sparse Auto-Encoder (SAE) feature coefficients across LODO folds, finding that 28% of top features are dataset-dependent shortcuts whose class signal depends on specific dataset compositions rather than semantic content. We systematically compare production guardrails (PromptGuard 2, LlamaGuard) and LLM-as-judge approaches on our benchmark, finding all three fail on indirect attacks targeting agents (7-37% detection) and that PromptGuard 2 and LlamaGuard cannot evaluate agentic tool injection due to architectural limitations. Finally, we show that LODO-stable SAE features provide more reliable explanations for classifier decisions by filtering dataset artifacts. We release our evaluation framework at https://github.com/maxf-zn/prompt-mining to establish LODO as the appropriate protocol for prompt attack detection research.
Authors: Yiran Guo, Zhongjian Qiao, Yingqi Xie, Jie Liu, Dan Ye, Ruiqing Zhang, Shuang Qiu, Lijie Xu
Abstract: Effective exploration is a key challenge in reinforcement learning for large language models: discovering high-quality trajectories within a limited sampling budget from the vast natural language sequence space. Existing methods face notable limitations: GRPO samples exclusively from the root, saturating high-probability trajectories while leaving deep, error-prone states under-explored. Tree-based methods blindly disperse budgets across trivial or unrecoverable states, causing sampling dilution that fails to uncover rare correct suffixes and destabilizes local baselines. To address this, we propose Deep Dense Exploration (DDE), a strategy that focuses exploration on $\textit{pivots}$-deep, recoverable states within unsuccessful trajectories. We instantiate DDE with DEEP-GRPO, which introduces three key innovations: (1) a lightweight data-driven utility function that automatically balances recoverability and depth bias to identify pivot states; (2) local dense resampling at each pivot to increase the probability of discovering correct subsequent trajectories; and (3) a dual-stream optimization objective that decouples global policy learning from local corrective updates. Experiments on mathematical reasoning benchmarks demonstrate that our method consistently outperforms GRPO, tree-based methods, and other strong baselines.
Authors: Nicolas Zumarraga, Thomas Kaar, Ning Wang, Maxwell A. Xu, Max Rosenblattl, Markus Kreft, Kevin O'Sullivan, Paul Schmiedmayer, Patrick Langer, Robert Jakob
Abstract: Time Series Language Models (TSLMs) are emerging as unified models for reasoning over continuous signals in natural language. However, long-context retrieval remains a major limitation: existing models are typically trained and evaluated on short sequences, while real-world time-series sensor streams can span millions of datapoints. This mismatch requires precise temporal localization under strict computational constraints, a regime that is not captured by current benchmarks. We introduce TS-Haystack, a long-context temporal retrieval benchmark comprising ten task types across four categories: direct retrieval, temporal reasoning, multi-step reasoning and contextual anomaly. The benchmark uses controlled needle insertion by embedding short activity bouts into longer longitudinal accelerometer recordings, enabling systematic evaluation across context lengths ranging from seconds to 2 hours per sample. We hypothesize that existing TSLM time series encoders overlook temporal granularity as context length increases, creating a task-dependent effect: compression aids classification but impairs retrieval of localized events. Across multiple model and encoding strategies, we observe a consistent divergence between classification and retrieval behavior. Learned latent compression preserves or improves classification accuracy at compression ratios up to 176$\times$, but retrieval performance degrades with context length, incurring in the loss of temporally localized information. These results highlight the importance of architectural designs that decouple sequence length from computational complexity while preserving temporal fidelity.
Authors: Jinbo Wang, Binghui Li, Zhanpeng Zhou, Mingze Wang, Yuxuan Sun, Jiaqi Zhang, Xunliang Cai, Lei Wu
Abstract: Batch size scheduling (BSS) plays a critical role in large-scale deep learning training, influencing both optimization dynamics and computational efficiency. Yet, its theoretical foundations remain poorly understood. In this work, we show that the functional scaling law (FSL) framework introduced in Li et al. (2025a) provides a principled lens for analyzing BSS. Specifically, we characterize the optimal BSS under a fixed data budget and show that its structure depends sharply on task difficulty. For easy tasks, optimal schedules keep increasing batch size throughout. In contrast, for hard tasks, the optimal schedule maintains small batch sizes for most of training and switches to large batches only in a late stage. To explain the emergence of late switching, we uncover a dynamical mechanism -- the fast catch-up effect -- which also manifests in large language model (LLM) pretraining. After switching from small to large batches, the loss rapidly aligns with the constant large-batch trajectory. Using FSL, we show that this effect stems from rapid forgetting of accumulated gradient noise, with the catch-up speed determined by task difficulty. Crucially, this effect implies that large batches can be safely deferred to late training without sacrificing performance, while substantially reducing data consumption. Finally, extensive LLM pretraining experiments -- covering both Dense and MoE architectures with up to 1.1B parameters and 1T tokens -- validate our theoretical predictions. Across all settings, late-switch schedules consistently outperform constant-batch and early-switch baselines.
Authors: Omin Kwon, Yeonjae Kim, Doyeon Kim, Minseo Kim, Yeonhong Park, Jae W. Lee
Abstract: Block diffusion LLMs are emerging as a promising next paradigm for language generation, but their use of KV caching makes memory access a dominant bottleneck in long-context settings. While dynamic sparse attention has been actively explored, existing methods designed for autoregressive LLMs rely on approximate importance estimation and perform poorly when adapted to block diffusion. This work identifies a key opportunity unique to block diffusion: attention at the first All-[MASK] denoising step reliably predicts important KV entries and budget requirements, enabling MAGE to perform a single exact attention pass per block and reuse it for training-free sparse denoising. Across long-context benchmarks including LongBench and Needle-in-a-Haystack, MAGE achieves near-lossless accuracy with a fraction of the KV budget while delivering up to 3-4x end-to-end speedup, consistently outperforming AR-oriented sparse attention baselines. A lightweight fine-tuning strategy further strengthens [MASK]-guided patterns with minimal cost, requiring only a few hours of training on a single NVIDIA H100 GPU for both 1.5B and 7B models.
Authors: Seyedsaman Emami, Daniel Hern\'andez-Lobato, Gonzalo Mart\'inez-Mu\~noz
Abstract: Multi-Task Learning (MTL) aims to boost predictive performance by sharing information across related tasks, yet conventional methods often suffer from negative transfer when unrelated or noisy tasks are forced to share representations. We propose Robust Multi-Task Boosting using Clustering and Local Ensembling (RMB-CLE), a principled MTL framework that integrates error-based task clustering with local ensembling. Unlike prior work that assumes fixed clusters or hand-crafted similarity metrics, RMB-CLE derives inter-task similarity directly from cross-task errors, which admit a risk decomposition into functional mismatch and irreducible noise, providing a theoretically grounded mechanism to prevent negative transfer. Tasks are grouped adaptively via agglomerative clustering, and within each cluster, a local ensemble enables robust knowledge sharing while preserving task-specific patterns. Experiments show that RMB-CLE recovers ground-truth clusters in synthetic data and consistently outperforms multi-task, single-task, and pooling-based ensemble methods across diverse real-world and synthetic benchmarks. These results demonstrate that RMB-CLE is not merely a combination of clustering and boosting but a general and scalable framework that establishes a new basis for robust multi-task learning.
Authors: Yaxuan Kong, Hoyoung Lee, Yoontae Hwang, Alejandro Lopez-Lira, Bradford Levy, Dhagash Mehta, Qingsong Wen, Chanyeol Choi, Yongjae Lee, Stefan Zohren
Abstract: Large Language Models (LLMs) are increasingly integrated into financial workflows, but evaluation practice has not kept up. Finance-specific biases can inflate performance, contaminate backtests, and make reported results useless for any deployment claim. We identify five recurring biases in financial LLM applications. They include look-ahead bias, survivorship bias, narrative bias, objective bias, and cost bias. These biases break financial tasks in distinct ways and they often compound to create an illusion of validity. We reviewed 164 papers from 2023 to 2025 and found that no single bias is discussed in more than 28 percent of studies. This position paper argues that bias in financial LLM systems requires explicit attention and that structural validity should be enforced before any result is used to support a deployment claim. We propose a Structural Validity Framework and an evaluation checklist with minimal requirements for bias diagnosis and future system design. The material is available at https://github.com/Eleanorkong/Awesome-Financial-LLM-Bias-Mitigation.
URLs: https://github.com/Eleanorkong/Awesome-Financial-LLM-Bias-Mitigation.
Authors: Pinqiao Wang, Sheng Li
Abstract: Tabular anomaly detection is often handled by single detectors or static ensembles, even though strong performance on tabular data typically comes from heterogeneous model families (e.g., tree ensembles, deep tabular networks, and tabular foundation models) that frequently disagree under distribution shift, missingness, and rare-anomaly regimes. We propose MAD, a Multi-Agent Debating framework that treats this disagreement as a first-class signal and resolves it through a mathematically grounded coordination layer. Each agent is a machine learning (ML)-based detector that produces a normalized anomaly score, confidence, and structured evidence, augmented by a large language model (LLM)-based critic. A coordinator converts these messages into bounded per-agent losses and updates agent influence via an exponentiated-gradient rule, yielding both a final debated anomaly score and an auditable debate trace. MAD is a unified agentic framework that can recover existing approaches, such as mixture-of-experts gating and learning-with-expert-advice aggregation, by restricting the message space and synthesis operator. We establish regret guarantees for the synthesized losses and show how conformal calibration can wrap the debated score to control false positives under exchangeability. Experiments on diverse tabular anomaly benchmarks show improved robustness over baselines and clearer traces of model disagreement
Authors: Manal Rahal, Bestoun S. Ahmed, Roger Renstr\"om, Robert Stener
Abstract: With the rapid increase in residential heat pump (HP) installations, optimizing hot water production in households is essential, yet it faces major technical and scalability challenges. Adapting production to actual household needs requires accurate forecasting of hot water demand to ensure comfort and, most importantly, to reduce energy waste. However, the conventional approach of training separate machine learning models for each household becomes computationally expensive at scale, particularly in cloud-connected HP deployments. This study introduces DELTAiF, a transfer learning (TL) based framework that provides scalable and accurate prediction of household hot water consumption. By predicting large hot water usage events, such as showers, DELTAiF enables adaptive yet scalable hot water production at the household level. DELTAiF leverages learned knowledge from a representative household and fine-tunes it across others, eliminating the need to train separate machine learning models for each HP installation. This approach reduces overall training time by approximately 67 percent while maintaining high predictive accuracy values between 0.874 and 0.991, and mean absolute percentage error values between 0.001 and 0.017. The results show that TL is particularly effective when the source household exhibits regular consumption patterns, enabling hot water demand forecasting at scale.
Authors: Yilun Kuang, Yash Dagade, Deep Chakraborty, Erik Learned-Miller, Randall Balestriero, Tim G. J. Rudner, Yann LeCun
Abstract: Self-supervised learning aims to learn maximally informative representations, but explicit information maximization is hindered by the curse of dimensionality. Existing methods like VCReg address this by regularizing first and second-order feature statistics, which cannot fully achieve maximum entropy. We propose Radial-VCReg, which augments VCReg with a radial Gaussianization loss that aligns feature norms with the Chi distribution-a defining property of high-dimensional Gaussians. We prove that Radial-VCReg transforms a broader class of distributions towards normality compared to VCReg and show on synthetic and real-world datasets that it consistently improves performance by reducing higher-order dependencies and promoting more diverse and informative representations.
Authors: Boning Zhou, Ziyu Wang, Han Hong, Haoqi Hu
Abstract: Causal inference, a critical tool for informing business decisions, traditionally relies heavily on structured data. However, in many real-world scenarios, such data can be incomplete or unavailable. This paper presents a framework that leverages transformer-based language models to perform causal inference using unstructured text. We demonstrate the effectiveness of our framework by comparing causal estimates derived from unstructured text against those obtained from structured data across population, group, and individual levels. Our findings show consistent results between the two approaches, validating the potential of unstructured text in causal inference tasks. Our approach extends the applicability of causal inference methods to scenarios where only textual data is available, enabling data-driven business decision-making when structured tabular data is scarce.
Authors: Lamine Rihani
Abstract: Artificial intelligence/machine learning (AI/ML) systems and emerging quantum computing software present unprecedented testing challenges characterized by high-dimensional/continuous input spaces, probabilistic/non-deterministic output distributions, behavioral correctness defined exclusively over observable prediction behaviors and measurement outcomes, and critical quality dimensions, trustworthiness, fairness, calibration, robustness, error syndrome patterns, that manifest through complex multi-way interactions among semantically meaningful output properties rather than deterministic input-output mappings. This paper introduces reverse n-wise output testing, a mathematically principled paradigm inversion that constructs covering arrays directly over domain-specific output equivalence classes, ML confidence calibration buckets, decision boundary regions, fairness partitions, embedding clusters, ranking stability bands, quantum measurement outcome distributions (0-dominant, 1-dominant, superposition collapse), error syndrome patterns (bit-flip, phase-flip, correlated errors), then solves the computationally challenging black-box inverse mapping problem via gradient-free metaheuristic optimization to synthesize input feature configurations or quantum circuit parameters capable of eliciting targeted behavioral signatures from opaque models. The framework delivers synergistic benefits across both domains: explicit customer-centric prediction/measurement coverage guarantees, substantial improvements in fault detection rates for ML calibration/boundary failures and quantum error syndromes, enhanced test suite efficiency, and structured MLOps/quantum validation pipelines with automated partition discovery from uncertainty analysis and coverage drift monitoring.
Authors: Ruomeng Ding, Tianwei Gao, Thomas P. Zollo, Eitan Bachmat, Richard Zemel, Zhun Deng
Abstract: Eliciting information to reduce uncertainty about latent group-level properties from surveys and other collective assessments requires allocating limited questioning effort under real costs and missing data. Although large language models enable adaptive, multi-turn interactions in natural language, most existing elicitation methods optimize what to ask with a fixed respondent pool, and do not adapt respondent selection or leverage population structure when responses are partial or incomplete. To address this gap, we study adaptive group elicitation, a multi-round setting where an agent adaptively selects both questions and respondents under explicit query and participation budgets. We propose a theoretically grounded framework that combines (i) an LLM-based expected information gain objective for scoring candidate questions with (ii) heterogeneous graph neural network propagation that aggregates observed responses and participant attributes to impute missing responses and guide per-round respondent selection. This closed-loop procedure queries a small, informative subset of individuals while inferring population-level responses via structured similarity. Across three real-world opinion datasets, our method consistently improves population-level response prediction under constrained budgets, including a >12% relative gain on CES at a 10% respondent budget.
Authors: Kris Shengjun Dong, Sahil Modi, Dima Nikiforov, Sana Damani, Edward Lin, Siva Kumar Sastry Hari, Christos Kozyrakis
Abstract: Optimizing CUDA code across multiple generations of GPU architectures is challenging, as achieving peak performance requires an extensive exploration of an increasingly complex, hardware-specific optimization space. Traditional compilers are constrained by fixed heuristics, whereas finetuning Large Language Models (LLMs) can be expensive. However, agentic workflows for CUDA code optimization have limited ability to aggregate knowledge from prior exploration, leading to biased sampling and suboptimal solutions. We propose KernelBlaster, a Memory-Augmented In-context Reinforcement Learning (MAIC-RL) framework designed to improve CUDA optimization search capabilities of LLM-based GPU coding agents. KernelBlaster enables agents to learn from experience and make systematically informed decisions on future tasks by accumulating knowledge into a retrievable Persistent CUDA Knowledge Base. We propose a novel profile-guided, textual-gradient-based agentic flow for CUDA generation and optimization to achieve high performance across generations of GPU architectures. KernelBlaster guides LLM agents to systematically explore high-potential optimization strategies beyond naive rewrites. Compared to the PyTorch baseline, our method achieves geometric mean speedups of 1.43x, 2.50x, and 1.50x on KernelBench Levels 1, 2, and 3, respectively. We release KernelBlaster as an open-source agentic framework, accompanied by a test harness, verification components, and a reproducible evaluation pipeline.
Authors: Edwin Chen, Zulekha Bibi
Abstract: We introduce Machine Learning as a Tool (MLAT), a design pattern in which pre-trained statistical machine learning models are exposed as callable tools within large language model (LLM) agent workflows. This allows an orchestrating agent to invoke quantitative predictions when needed and reason about their outputs in context. Unlike conventional pipelines that treat ML inference as a static preprocessing step, MLAT positions the model as a first-class tool alongside web search, database queries, and APIs, enabling the LLM to decide when and how to use it based on conversational context. To validate MLAT, we present PitchCraft, a pilot production system that converts discovery call recordings into professional proposals with ML-predicted pricing. The system uses two agents: a Research Agent that gathers prospect intelligence via parallel tool calls, and a Draft Agent that invokes an XGBoost pricing model as a tool call and generates a complete proposal through structured outputs. The pricing model, trained on 70 examples combining real and human-verified synthetic data, achieves R^2 = 0.807 on held-out data with a mean absolute error of 3688 USD. The system reduces proposal generation time from multiple hours to under 10 minutes. We describe the MLAT framework, structured output architecture, training methodology under extreme data scarcity, and sensitivity analysis demonstrating meaningful learned relationships. MLAT generalizes to domains requiring quantitative estimation combined with contextual reasoning.
Authors: Songyuan Li, Jia Hu, Ahmed M. Abdelmoniem, Geyong Min, Haojun Huang, Jiwei Huang
Abstract: Recent Mixture-of-Experts (MoE)-based large language models (LLMs) such as Qwen-MoE and DeepSeek-MoE are transforming generative AI in natural language processing. However, these models require vast and diverse training data. Federated learning (FL) addresses this challenge by leveraging private data from heterogeneous edge devices for privacy-preserving MoE training. Nonetheless, traditional FL approaches require devices to host local MoE models, which is impractical for resource-constrained devices due to large model sizes. To address this, we propose DeepFusion, the first scalable federated MoE training framework that enables the fusion of heterogeneous on-device LLM knowledge via federated knowledge distillation, yielding a knowledge-abundant global MoE model. Specifically, DeepFusion features each device to independently configure and train an on-device LLM tailored to its own needs and hardware limitations. Furthermore, we propose a novel View-Aligned Attention (VAA) module that integrates multi-stage feature representations from the global MoE model to construct a predictive perspective aligned with on-device LLMs, thereby enabling effective cross-architecture knowledge distillation. By explicitly aligning predictive perspectives, VAA resolves the view-mismatch problem in traditional federated knowledge distillation, which arises from heterogeneity in model architectures and prediction behaviors between on-device LLMs and the global MoE model. Experiments with industry-level MoE models (Qwen-MoE and DeepSeek-MoE) and real-world datasets (medical and finance) demonstrate that DeepFusion achieves performance close to centralized MoE training. Compared with key federated MoE baselines, DeepFusion reduces communication costs by up to 71% and improves token perplexity by up to 5.28%.
Authors: Trishit Mondal, Ameya D. Jagtap
Abstract: Transformer architectures have revolutionized machine learning across a wide range of domains, from natural language processing to scientific computing. However, their growing deployment in high-stakes applications, such as computer vision, natural language processing, healthcare, autonomous systems, and critical areas of scientific computing including climate modeling, materials discovery, drug discovery, nuclear science, and robotics, necessitates a deeper and more rigorous understanding of their trustworthiness. In this work, we critically examine the foundational question: \textitHow trustworthy are transformer models?} We evaluate their reliability through a comprehensive review of interpretability, explainability, robustness against adversarial attacks, fairness, and privacy. We systematically examine the trustworthiness of transformer-based models in safety-critical applications spanning natural language processing, computer vision, and science and engineering domains, including robotics, medicine, earth sciences, materials science, fluid dynamics, nuclear science, and automated theorem proving; highlighting high-impact areas where these architectures are central and analyzing the risks associated with their deployment. By synthesizing insights across these diverse areas, we identify recurring structural vulnerabilities, domain-specific risks, and open research challenges that limit the reliable deployment of transformers.
Authors: Hani Beirami, M M Manjurul Islam
Abstract: We investigate how formal temporal logic specifications can enhance the safety and robustness of reinforcement learning (RL) control in aerospace applications. Using the open source AeroBench F-16 simulation benchmark, we train a Proximal Policy Optimization (PPO) agent to regulate engine throttle and track commanded airspeed. The control objective is encoded as a Signal Temporal Logic (STL) requirement to maintain airspeed within a prescribed band during the final seconds of each maneuver. To enforce this specification at run time, we introduce a conformal STL shield that filters the RL agent's actions using online conformal prediction. We compare three settings: (i) PPO baseline, (ii) PPO with a classical rule-based STL shield, and (iii) PPO with the proposed conformal shield, under both nominal conditions and a severe stress scenario involving aerodynamic model mismatch, actuator rate limits, measurement noise, and mid-episode setpoint jumps. Experiments show that the conformal shield preserves STL satisfaction while maintaining near baseline performance and providing stronger robustness guarantees than the classical shield. These results demonstrate that combining formal specification monitoring with data driven RL control can substantially improve the reliability of autonomous flight control in challenging environments.
Authors: Zhi Zhang, Zhen Han, Costas Mavromatis, Qi Zhu, Yunyi Zhang, Sheng Guan, Dingmin Wang, Xiong Zhou, Shuai Wang, Soji Adeshina, Vassilis Ioannidis, Huzefa Rangwala
Abstract: Reinforcement learning (RL) plays a central role in large language model (LLM) post-training. Among existing approaches, Group Relative Policy Optimization (GRPO) is widely used, especially for RL with verifiable rewards (RLVR) fine-tuning. In GRPO, each query prompts the LLM to generate a group of rollouts with a fixed group size $N$. When all rollouts in a group share the same outcome, either all correct or all incorrect, the group-normalized advantages become zero, yielding no gradient signal and wasting fine-tuning compute. We introduce Adaptive Efficient Rollout Optimization (AERO), an enhancement of GRPO. AERO uses an adaptive rollout strategy, applies selective rejection to strategically prune rollouts, and maintains a Bayesian posterior to prevent zero-advantage dead zones. Across three model configurations (Qwen2.5-Math-1.5B, Qwen2.5-7B, and Qwen2.5-7B-Instruct), AERO improves compute efficiency without sacrificing performance. Under the same total rollout budget, AERO reduces total training compute by about 48% while shortening wall-clock time per step by about 45% on average. Despite the substantial reduction in compute, AERO matches or improves Pass@8 and Avg@8 over GRPO, demonstrating a practical, scalable, and compute-efficient strategy for RL-based LLM alignment.
Authors: Mathias Jackermeier, Mattia Giuri, Jacques Cloete, Alessandro Abate
Abstract: We study instruction following in multi-task reinforcement learning, where an agent must zero-shot execute novel tasks not seen during training. In this setting, linear temporal logic (LTL) has recently been adopted as a powerful framework for specifying structured, temporally extended tasks. While existing approaches successfully train generalist policies, they often struggle to effectively capture the rich logical and temporal structure inherent in LTL specifications. In this work, we address these concerns with a novel approach to learn structured task representations that facilitate training and generalisation. Our method conditions the policy on sequences of Boolean formulae constructed from a finite automaton of the task. We propose a hierarchical neural architecture to encode the logical structure of these formulae, and introduce an attention mechanism that enables the policy to reason about future subgoals. Experiments in a variety of complex environments demonstrate the strong generalisation capabilities and superior performance of our approach.
Authors: Mehran Aghabozorgi, Alireza Moazeni, Yanshu Zhang, Ke Li
Abstract: Model-based reinforcement learning promises strong sample efficiency but often underperforms in practice due to compounding model error, unimodal world models that average over multi-modal dynamics, and overconfident predictions that bias learning. We introduce WIMLE, a model-based method that extends Implicit Maximum Likelihood Estimation (IMLE) to the model-based RL framework to learn stochastic, multi-modal world models without iterative sampling and to estimate predictive uncertainty via ensembles and latent sampling. During training, WIMLE weights each synthetic transition by its predicted confidence, preserving useful model rollouts while attenuating bias from uncertain predictions and enabling stable learning. Across $40$ continuous-control tasks spanning DeepMind Control, MyoSuite, and HumanoidBench, WIMLE achieves superior sample efficiency and competitive or better asymptotic performance than strong model-free and model-based baselines. Notably, on the challenging Humanoid-run task, WIMLE improves sample efficiency by over $50$\% relative to the strongest competitor, and on HumanoidBench it solves $8$ of $14$ tasks (versus $4$ for BRO and $5$ for SimbaV2). These results highlight the value of IMLE-based multi-modality and uncertainty-aware weighting for stable model-based RL.
Authors: Kensuke Ajimoto, Yuma Yamamoto, Yoshifumi Kusunoki, Tomoharu Nakashima
Abstract: This paper proposes a multi-class online fuzzy classifier for dynamic environments. A fuzzy classifier comprises a set of fuzzy if-then rules where human users determine the antecedent fuzzy sets beforehand. In contrast, the consequent real values are determined by learning from training data. In an online framework, not all training dataset patterns are available beforehand. Instead, only a few patterns are available at a time step, and the subsequent patterns become available at the following time steps. The conventional online fuzzy classifier considered only two-class problems. This paper investigates the extension to the conventional fuzzy classifiers for multi-class problems. We evaluate the performance of the multi-class online fuzzy classifiers through numerical experiments on synthetic dynamic data and also several benchmark datasets.
Authors: Abdelali Bouyahia, Fr\'ed\'eric LeBlanc, Mario Marchand
Abstract: Data augmentation is one of the most widely used techniques to improve generalization in modern machine learning, often justified by its ability to promote invariance to label-irrelevant transformations. However, its theoretical role remains only partially understood. In this work, we propose an information-theoretic framework that systematically accounts for the effect of augmentation on generalization and invariance learning. Our approach builds upon mutual information-based bounds, which relate the generalization gap to the amount of information a learning algorithm retains about its training data. We extend this framework by modeling the augmented distribution as a composition of the original data distribution with a distribution over transformations, which naturally induces an orbit-averaged loss function. Under mild sub-Gaussian assumptions on the loss function and the augmentation process, we derive a new generalization bound that decompose the expected generalization gap into three interpretable terms: (1) a distributional divergence between the original and augmented data, (2) a stability term measuring the algorithm dependence on training data, and (3) a sensitivity term capturing the effect of augmentation variability. To connect our bounds to the geometry of the augmentation group, we introduce the notion of group diameter, defined as the maximal perturbation that augmentations can induce in the input space. The group diameter provides a unified control parameter that bounds all three terms and highlights an intrinsic trade-off: small diameters preserve data fidelity but offer limited regularization, while large diameters enhance stability at the cost of increased bias and sensitivity. We validate our theoretical bounds with numerical experiments, demonstrating that it reliably tracks and predicts the behavior of the true generalization gap.
Authors: Prithwijit Chowdhury, Ahmad Mustafa, Mohit Prabhushankar, Ghassan AlRegib
Abstract: In geophysics, hydrocarbon prospect risking involves assessing the risks associated with hydrocarbon exploration by integrating data from various sources. Machine learning-based classifiers trained on tabular data have been recently used to make faster decisions on these prospects. The lack of transparency in the decision-making processes of such models has led to the emergence of explainable AI (XAI). LIME and SHAP are two such examples of these XAI methods which try to generate explanations of a particular decision by ranking the input features in terms of importance. However, explanations of the same scenario generated by these two different explanation strategies have shown to disagree or be different, particularly for complex data. This is because the definitions of "importance" and "relevance" differ for different explanation strategies. Thus, grounding these ranked features using theoretically backed causal ideas of necessity and sufficiency can prove to be a more reliable and robust way to improve the trustworthiness of the concerned explanation strategies.We propose a unified framework to generate counterfactuals as well as quantify necessity and sufficiency and use these to perform a robustness evaluation of the explanations provided by LIME and SHAP on high dimensional structured prospect risking data. This robustness test gives us deeper insights into the models capabilities to handle erronous data and which XAI module works best in pair with which model for our dataset for hydorcarbon indication.
Authors: Arnav Chavan, Nahush Lele, Udbhav Bamba, Sankalp Dayal, Aditi Raghunathan, Deepak Gupta
Abstract: Activation outliers in large-scale transformer models pose a fundamental challenge to model quantization, creating excessively large ranges that cause severe accuracy drops during quantization. We empirically observe that outlier severity intensifies with pre-training scale (e.g., progressing from CLIP to the more extensively trained SigLIP and SigLIP2). Through theoretical analysis as well as empirical correlation studies, we establish the direct link between these activation outliers and dominant singular values of the weights. Building on this insight, we propose Selective Spectral Decay ($S^2D$), a geometrically-principled conditioning method that surgically regularizes only the weight components corresponding to the largest singular values during fine-tuning. Through extensive experiments, we demonstrate that $S^2D$ significantly reduces activation outliers and produces well-conditioned representations that are inherently quantization-friendly. Models trained with $S^2D$ achieve up to 7% improved PTQ accuracy on ImageNet under W4A4 quantization and 4% gains when combined with QAT. These improvements also generalize across downstream tasks and vision-language models, enabling the scaling of increasingly large and rigorously trained models without sacrificing deployment efficiency.
Authors: Ian Su, Gaurav Purushothaman, Jey Narayan, Ruhika Goel, Kevin Zhu, Sunishchal Dev, Yash More, Maheep Chaudhary
Abstract: Reasoning-specialized models like OpenAI's 5.1 and DeepSeek-V3.2 allocate substantial inference compute to extended chain-of-thought (CoT) traces, yet reasoning tokens incur significant costs. How do different reasoning modalities of code, natural language, hybrid, or none do perform under token constraints? We introduce a framework that constrains models to reason exclusively through code, comments, both, or neither, then systematically ablates token budgets to 10\%, 30\%, 50\%, and 70\% of optimal. We evaluate four frontier models (GPT-5.1, Gemini 3 Flash, DeepSeek-V3.2, Grok 4.1) across mathematical benchmarks (AIME, GSM8K, HMMT). Our findings reveal: (1) \textbf{truncated reasoning can hurt} as DeepSeek-V3.2 achieves 53\% with no reasoning but only 17\% with truncated CoT at 50\% budget; (2) \textbf{code degrades gracefully} as Gemini's comments collapse to 0\% while code maintains 43-47\%; (3) \textbf{hybrid reasoning underperforms} single modalities; (4) \textbf{robustness is model-dependent} as Grok maintains 80-90\% at 30\% budget where OpenAI and DeepSeek collapse to 7-27\%. These results suggest incomplete reasoning chains actively mislead models, with implications for deploying reasoning-specialized systems under resource constraints.
Authors: Hasi Hays
Abstract: The Transformer architecture has become the foundation of modern deep learning, yet its core self-attention mechanism suffers from quadratic computational complexity and lacks grounding in biological neural computation. We propose Selective Synchronization Attention (SSA), a novel attention mechanism that replaces the standard dot-product self-attention with a closed-form operator derived from the steady-state solution of the Kuramoto model of coupled oscillators. In SSA, each token is represented as an oscillator characterized by a learnable natural frequency and phase; the synchronization strength between token pairs, determined by a frequency-dependent coupling and phase-locking condition, serves as the attention weight. This formulation provides three key advantages: (i) natural sparsity arising from the phase-locking threshold, whereby tokens with incompatible frequencies automatically receive zero attention weight without explicit masking; (ii) unified positional-semantic encoding through the natural frequency spectrum, eliminating the need for separate positional encodings; and (iii) a single-pass, closed-form computation that avoids iterative ODE integration, with all components (coupling, order parameter, synchronization) derived from the oscillatory framework. We instantiate SSA within the Oscillatory Synchronization Network (OSN), a drop-in replacement for the Transformer block. Analysis of the synchronization matrices reveals non-uniform, head-diverse coupling patterns even at initialization, demonstrating a stronger architectural inductive bias than the approximately uniform attention produced by randomly initialized Transformers.
Authors: Lei Chen, Yuan Meng, Xiaoyu Zhan, Zhi Wang, Wenwu Zhu
Abstract: Large Language Models (LLMs) offer strong capabilities but incur high inference costs due to dense computation and memory access. Training-free activation sparsity is a promising approach for efficient LLM inference, yet existing methods often rely solely on activation information and uniform sparsity ratios. This overlooks the critical interplay with weights and inter-block sensitivity variation, leading to suboptimal performance. We identify two key phenomena in modern LLMs: 1) less significant activations may align with highly important weights, and 2) sparsity sensitivity varies non-monotonically across model blocks. We propose Weight-aware Mixed-Granularity Training-free Activation Sparsity (WiSparse), which leverages both activation and weight information for adaptive sparsity allocation. Specifically, we introduce a weight-aware mechanism integrating activation magnitudes with precomputed weight norms to accurately identify salient channels. This is combined with a mixed-granularity allocation scheme: a global budget is distributed across blocks via evolutionary search to protect sensitive regions, then refined within blocks to minimize reconstruction error. We improve sparse kernels and demonstrate effectiveness on three representative models. Notably, at 50% sparsity, WiSparse preserves 97% of Llama3.1's dense performance, surpassing the strongest baseline by 2.23 percentage points while achieving a 21.4% acceleration in end-to-end inference speed. Our research advances the limits of training-free approaches for efficient LLM inference, pushing the boundaries of achievable speedup without training.
Authors: Huaming Du, Tao Hu, Yijie Huang, Yu Zhao, Guisong Liu, Tao Gu, Gang Kou, Carl Yang
Abstract: Revealing the underlying causal mechanisms in the real world is crucial for scientific and technological progress. Despite notable advances in recent decades, the lack of high-quality data and the reliance of traditional causal discovery algorithms (TCDA) on the assumption of no latent confounders, as well as their tendency to overlook the precise semantics of latent variables, have long been major obstacles to the broader application of causal discovery. To address this issue, we propose a novel causal modeling framework, TLVD, which integrates the metadata-based reasoning capabilities of large language models (LLMs) with the data-driven modeling capabilities of TCDA for inferring latent variables and their semantics. Specifically, we first employ a data-driven approach to construct a causal graph that incorporates latent variables. Then, we employ multi-LLM collaboration for latent variable inference, modeling this process as a game with incomplete information and seeking its Bayesian Nash Equilibrium (BNE) to infer the possible specific latent variables. Finally, to validate the inferred latent variables across multiple real-world web-based data sources, we leverage LLMs for evidence exploration to ensure traceability. We comprehensively evaluate TLVD on three de-identified real patient datasets provided by a hospital and two benchmark datasets. Extensive experimental results confirm the effectiveness and reliability of TLVD, with average improvements of 32.67% in Acc, 62.21% in CAcc, and 26.72% in ECit across the five datasets.
Authors: Hong Li, Zhen Zhou, Honggang Zhang, Yuping Luo, Xinyue Wang, Han Gong, Zhiyuan Liu
Abstract: Data-parallel (DP) training with synchronous all-reduce is a dominant paradigm for full-parameter fine-tuning of large language models (LLMs). While parameter synchronization guarantees numerical equivalence of model weights after each iteration, it does not necessarily imply alignment of worker-level optimization dynamics before gradient aggregation. This paper identifies and studies this latent mismatch, termed \emph{silent inconsistency}, where cross-worker divergence in losses and gradients can remain invisible under conventional aggregated monitoring signals. We propose a lightweight, model-agnostic diagnostic framework that quantifies worker-level consistency using training signals readily available in standard pipelines. Specifically, we introduce three complementary metrics: loss dispersion, gradient-norm dispersion, and gradient-direction consistency measured by inter-worker cosine similarity. The proposed metrics incur negligible overhead and require no modification to model architecture, synchronization mechanisms, or optimization algorithms. We validate the framework by fully fine-tuning the 1B-parameter \texttt{openPangu-Embedded-1B-V1.1} model on the \texttt{tatsu-lab/alpaca} dataset using an 8-NPU DP setup, under controlled perturbations of cross-rank stochasticity. Experimental results show that progressively desynchronized data shuffling and random seeds lead to substantial increases in loss/gradient dispersion and reduced directional alignment, despite smooth globally averaged loss curves. These findings demonstrate that the proposed indicators provide actionable visibility into hidden instability modes in large-scale DP fine-tuning, enabling more reliable diagnosis and configuration assessment.
Authors: Chang Liu, Yiran Zhao, Lawrence Liu, Yaoqi Ye, Csaba Szepesv\'ari, Lin F. Yang
Abstract: Reinforcement learning (RL) has enhanced the capabilities of large language models (LLMs) through reward-driven training. Nevertheless, this process can introduce excessively long responses, inflating inference latency and computational overhead. Prior length-control approaches typically rely on fixed heuristic reward shaping, which can misalign with the task objective and require brittle tuning. In this work, we propose LACONIC, a reinforcement learning method that enforces a target token budget during training. Specifically, we update policy models using an augmented objective that combines the task reward with a length-based cost. To balance brevity and task performance, the cost scale is adaptively adjusted throughout training. This yields robust length control while preserving task reward. We provide a theoretical guarantee that support the method. Across mathematical reasoning models and datasets, LACONIC preserves or improves pass@1 while reducing output length by over 50%. It maintains out-of-domain performance on general knowledge and multilingual benchmarks with 44% fewer tokens. Moreover, LACONIC integrates into standard RL-tuning with no inference changes and minimal deployment overhead.
Authors: Aadirupa Saha, Amith Bhat, Haipeng Luo
Abstract: We study $K$-armed Multiarmed Bandit (MAB) problem with $M$ heterogeneous data sources, each exhibiting unknown and distinct noise variances $\{\sigma_j^2\}_{j=1}^M$. The learner's objective is standard MAB regret minimization, with the additional complexity of adaptively selecting which data source to query from at each round. We propose Source-Optimistic Adaptive Regret minimization (SOAR), a novel algorithm that quickly prunes high-variance sources using sharp variance-concentration bounds, followed by a `balanced min-max LCB-UCB approach' that seamlessly integrates the parallel tasks of identifying the best arm and the optimal (minimum-variance) data source. Our analysis shows SOAR achieves an instance-dependent regret bound of $\tilde{O}\left({\sigma^*}^2\sum_{i=2}^K \frac{\log T}{\Delta_i} + \sqrt{K \sum_{j=1}^M \sigma_j^2}\right)$, up to preprocessing costs depending only on problem parameters, where ${\sigma^*}^2 := \min_j \sigma_j^2$ is the minimum source variance and $\Delta_i$ denotes the suboptimality gap of the $i$-th arm. This result is both surprising as despite lacking prior knowledge of the minimum-variance source among $M$ alternatives, SOAR attains the optimal instance-dependent regret of standard single-source MAB with variance ${\sigma^*}^2$, while incurring only an small (and unavoidable) additive cost of $\tilde O(\sqrt{K \sum_{j=1}^M \sigma_j^2})$ towards the optimal (minimum variance) source identification. Our theoretical bounds represent a significant improvement over some proposed baselines, e.g. Uniform UCB or Explore-then-Commit UCB, which could potentially suffer regret scaling with $\sigma_{\max}^2$ in place of ${\sigma^*}^2$-a gap that can be arbitrarily large when $\sigma_{\max} \gg \sigma^*$. Experiments on multiple synthetic problem instances and the real-world MovieLens\;25M dataset, demonstrating the superior performance of SOAR over the baselines.
Authors: Fabian Gr\"oger, Shuo Wen, Maria Brbi\'c
Abstract: The Platonic Representation Hypothesis suggests that representations from neural networks are converging to a common statistical model of reality. We show that the existing metrics used to measure representational similarity are confounded by network scale: increasing model depth or width can systematically inflate representational similarity scores. To correct these effects, we introduce a permutation-based null-calibration framework that transforms any representational similarity metric into a calibrated score with statistical guarantees. We revisit the Platonic Representation Hypothesis with our calibration framework, which reveals a nuanced picture: the apparent convergence reported by global spectral measures largely disappears after calibration, while local neighborhood similarity, but not local distances, retains significant agreement across different modalities. Based on these findings, we propose the Aristotelian Representation Hypothesis: representations in neural networks are converging to shared local neighborhood relationships.
Authors: Buze Zhang, Jinkai Tao, Zilang Zeng, Neil He, Ali Maatouk, Menglin Yang, Rex Ying
Abstract: Large Language Models (LLMs) have achieved remarkable progress, with Parameter-Efficient Fine-Tuning (PEFT) emerging as a key technique for downstream task adaptation. However, existing PEFT methods mainly operate in Euclidean space, fundamentally limiting their capacity to capture complex geometric structures inherent in language data. While alternative geometric spaces, like hyperbolic geometries for hierarchical data and spherical manifolds for circular patterns, offer theoretical advantages, forcing representations into a single manifold type ultimately limits expressiveness, even when curvature parameters are learnable. To address this, we propose Mixture of Space (MoS), a unified framework that leverages multiple geometric spaces simultaneously to learn richer, curvature-aware representations. Building on this scheme, we develop MoSLoRA, which extends Low-Rank Adaptation (LoRA) with heterogeneous geometric experts, enabling models to dynamically select or combine appropriate geometric spaces based on input context. Furthermore, to address the computational overhead of frequent manifold switching, we develop a lightweight routing mechanism. Moreover, we provide empirical insights into how curvature optimization impacts training stability and model performance. Our experiments across diverse benchmarks demonstrate that MoSLoRA consistently outperforms strong baselines, achieving up to 5.6% improvement on MATH500 and 15.9% on MAWPS.
Authors: Alejandro Francisco Queiruga
Abstract: Scaling laws can be understood from ground-up numerical analysis, where traditional function approximation theory can explain shifts in model architecture choices. GLU variants now dominate frontier LLMs and similar outer-product architectures are prevalent in ranking models. The success of these architectures has mostly been left as an empirical discovery. In this paper, we apply the tools of numerical analysis to expose a key factor: these models have an $x^2$ which enables \emph{asymptotically} faster scaling than MLPs. GLUs have piecewise quadratic functional forms that are sufficient to exhibit quadratic order of approximation. Our key contribution is to demonstrate that the $L(P)$ scaling slope is $L(P)\propto P^{-3}$ for GLUs but only $L(P)=P^{-2}$ for MLPs on function reconstruction problems. We provide a parameter construction and empirical verification of these slopes for 1D function approximation. From the first principles we discover, we make one stride and propose the ``Gated Quadratic Unit'' which has an even steeper $L(P)$ slope than the GLU and MLP. This opens the possibility of architecture design from first principles numerical theory to unlock superior scaling in large models. Replication code is available at https://github.com/afqueiruga/divine_scaling.
Authors: Kutay Tire, Yufan Zhang, Ege Onur Taga, Samet Oymak
Abstract: We explore the use of transformers for solving quadratic programs and how this capability benefits decision-making problems that involve covariance matrices. We first show that the linear attention mechanism can provably solve unconstrained QPs by tokenizing the matrix variables (e.g.~$A$ of the objective $\frac{1}{2}x^\top Ax+b^\top x$) row-by-row and emulating gradient descent iterations. Furthermore, by incorporating MLPs, a transformer block can solve (i) $\ell_1$-penalized QPs by emulating iterative soft-thresholding and (ii) $\ell_1$-constrained QPs when equipped with an additional feedback loop. Our theory motivates us to introduce Time2Decide: a generic method that enhances a time series foundation model (TSFM) by explicitly feeding the covariance matrix between the variates. We empirically find that Time2Decide uniformly outperforms the base TSFM model for the classical portfolio optimization problem that admits an $\ell_1$-constrained QP formulation. Remarkably, Time2Decide also outperforms the classical "Predict-then-Optimize (PtO)" procedure, where we first forecast the returns and then explicitly solve a constrained QP, in suitable settings. Our results demonstrate that transformers benefit from explicit use of second-order statistics, and this can enable them to effectively solve complex decision-making problems, like portfolio construction, in one forward pass.
Authors: Chaosheng Dong, Peiyao Xiao, Yijia Wang, Kaiyi Ji
Abstract: This paper presents DeepMTL2R, an open-source deep learning framework for Multi-task Learning to Rank (MTL2R), where multiple relevance criteria must be optimized simultaneously. DeepMTL2R integrates heterogeneous relevance signals into a unified, context-aware model by leveraging the self-attention mechanism of transformer architectures, enabling effective learning across diverse and potentially conflicting objectives. The framework includes 21 state-of-the-art multi-task learning algorithms and supports multi-objective optimization to identify Pareto-optimal ranking models. By capturing complex dependencies and long-range interactions among items and labels, DeepMTL2R provides a scalable and expressive solution for modern ranking systems and facilitates controlled comparisons across MTL strategies. We demonstrate its effectiveness on a publicly available dataset, report competitive performance, and visualize the resulting trade-offs among objectives. DeepMTL2R is available at \href{https://github.com/amazon-science/DeepMTL2R}{https://github.com/amazon-science/DeepMTL2R}.
URLs: https://github.com/amazon-science/DeepMTL2R, https://github.com/amazon-science/DeepMTL2R
Authors: Francesco Emanuele Stradi, Kalana Kalupahana, Matteo Castiglioni, Alberto Marchesi, Nicola Gatti
Abstract: We study the constrained variant of the \emph{multi-armed bandit} (MAB) problem, in which the learner aims not only at minimizing the total loss incurred during the learning dynamic, but also at controlling the violation of multiple \emph{unknown} constraints, under both \emph{full} and \emph{bandit feedback}. We consider a non-stationary environment that subsumes both stochastic and adversarial models and where, at each round, both losses and constraints are drawn from distributions that may change arbitrarily over time. In such a setting, it is provably not possible to guarantee both sublinear regret and sublinear violation. Accordingly, prior work has mainly focused either on settings with stochastic constraints or on relaxing the benchmark with fully adversarial constraints (\emph{e.g.}, via competitive ratios with respect to the optimum). We provide the first algorithms that achieve optimal rates of regret and \emph{positive} constraint violation when the constraints are stochastic while the losses may vary arbitrarily, and that simultaneously yield guarantees that degrade smoothly with the degree of adversariality of the constraints. Specifically, under \emph{full feedback} we propose an algorithm attaining $\widetilde{\mathcal{O}}(\sqrt{T}+C)$ regret and $\widetilde{\mathcal{O}}(\sqrt{T}+C)$ {positive} violation, where $C$ quantifies the amount of non-stationarity in the constraints. We then show how to extend these guarantees when only bandit feedback is available for the losses. Finally, when \emph{bandit feedback} is available for the constraints, we design an algorithm achieving $\widetilde{\mathcal{O}}(\sqrt{T}+C)$ {positive} violation and $\widetilde{\mathcal{O}}(\sqrt{T}+C\sqrt{T})$ regret.
Authors: Qinqi Lin, Ningning Ding, Lingjie Duan, Jianwei Huang
Abstract: Despite legal mandates for the right to be forgotten, AI operators routinely fail to comply with data deletion requests. While machine unlearning (MU) provides a technical solution to remove personal data's influence from trained models, ensuring compliance remains challenging due to the fundamental gap between MU's technical feasibility and regulatory implementation. In this paper, we introduce the first economic framework for auditing MU compliance, by integrating certified unlearning theory with regulatory enforcement. We first characterize MU's inherent verification uncertainty using a hypothesis-testing interpretation of certified unlearning to derive the auditor's detection capability, and then propose a game-theoretic model to capture the strategic interactions between the auditor and the operator. A key technical challenge arises from MU-specific nonlinearities inherent in the model utility and the detection probability, which create complex strategic couplings that traditional auditing frameworks do not address and that also preclude closed-form solutions. We address this by transforming the complex bivariate nonlinear fixed-point problem into a tractable univariate auxiliary problem, enabling us to decouple the system and establish the equilibrium existence, uniqueness, and structural properties without relying on explicit solutions. Counterintuitively, our analysis reveals that the auditor can optimally reduce the inspection intensity as deletion requests increase, since the operator's weakened unlearning makes non-compliance easier to detect. This is consistent with recent auditing reductions in China despite growing deletion requests. Moreover, we prove that although undisclosed auditing offers informational advantages for the auditor, it paradoxically reduces the regulatory cost-effectiveness relative to disclosed auditing.
Authors: Shishir Sharma, Doina Precup, Theodore J. Perkins
Abstract: The primary focus of multi-agent reinforcement learning (MARL) has been to study interactions among a fixed number of agents embedded in an environment. However, in the real world, the number of agents is neither fixed nor known a priori. Moreover, an agent can decide to create other agents (for example, a cell may divide, or a company may spin off a division). In this paper, we propose a framework that allows agents to create other agents; we call this a fluid-agent environment. We present game-theoretic solution concepts for fluid-agent games and empirically evaluate the performance of several MARL algorithms within this framework. Our experiments include fluid variants of established benchmarks such as Predator-Prey and Level-Based Foraging, where agents can dynamically spawn, as well as a new environment we introduce that highlights how fluidity can unlock novel solution strategies beyond those observed in fixed-population settings. We demonstrate that this framework yields agent teams that adjust their size dynamically to match environmental demands.
Authors: Qian Liyan, Zhang Yao, Yuan Ye, Zhang Zhaoke, Fang Jin, Jiang Shimiao, Zhang Jin, Li Ke, Liu Beijiang, Xu Chenglin, Zhang Yifan, Jia Xiaoqian, Qin Xiaoshuai, Huang Xingtao
Abstract: We introduce a Monte Carlo (MC) dataset of single- and two-track drift chamber events to advance Machine Learning (ML)-based track reconstruction. To enable standardized and comparable evaluation, we define track reconstruction specific metrics and report results for traditional track reconstruction algorithms and a Graph Neural Networks (GNNs) method, facilitating rigorous, reproducible validation for future research.
Authors: Isam Vrce, Andreas Kassler, G\"ok\c{c}e Aydos
Abstract: Sparsity is a well-studied technique for compressing deep neural networks (DNNs) without compromising performance. In deep reinforcement learning (DRL), neural networks with up to 5% of their original weights can still be trained with minimal performance loss compared to their dense counterparts. However, most existing methods rely on unstructured fine-grained sparsity, which limits hardware acceleration opportunities due to irregular computation patterns. Structured coarse-grained sparsity enables hardware acceleration, yet typically degrades performance and increases pruning complexity. In this work, we present, to the best of our knowledge, the first study on N:M structured sparsity in RL, which balances compression, performance, and hardware efficiency. Our framework enforces row-wise N:M sparsity throughout training for all networks in off-policy RL (TD3), maintaining compatibility with accelerators that support N:M sparse matrix operations. Experiments on continuous-control benchmarks show that RNM-TD3, our N:M sparse agent, outperforms its dense counterpart at 50%-75% sparsity (e.g., 2:4 and 1:4), achieving up to a 14% increase in performance at 2:4 sparsity on the Ant environment. RNM-TD3 remains competitive even at 87.5% sparsity (1:8), while enabling potential training speedups.
Authors: Matteo Bollini, Gianmarco Genalti, Francesco Emanuele Stradi, Matteo Castiglioni, Alberto Marchesi
Abstract: Algorithmic \emph{replicability} has recently been introduced to address the need for reproducible experiments in machine learning. A \emph{replicable online learning} algorithm is one that takes the same sequence of decisions across different executions in the same environment, with high probability. We initiate the study of algorithmic replicability in \emph{constrained} MAB problems, where a learner interacts with an unknown stochastic environment for $T$ rounds, seeking not only to maximize reward but also to satisfy multiple constraints. Our main result is that replicability can be achieved in constrained MABs. Specifically, we design replicable algorithms whose regret and constraint violation match those of non-replicable ones in terms of $T$. As a key step toward these guarantees, we develop the first replicable UCB-like algorithm for \emph{unconstrained} MABs, showing that algorithms that employ the optimism in-the-face-of-uncertainty principle can be replicable, a result that we believe is of independent interest.
Authors: Minh Nguyen
Abstract: Many real-world control problems, ranging from finance to robotics, evolve in continuous time with non-uniform, event-driven decisions. Standard discrete-time reinforcement learning (RL), based on fixed-step Bellman updates, struggles in this setting: as time gaps shrink, the $Q$-function collapses to the value function $V$, eliminating action ranking. Existing continuous-time methods reintroduce action information via an advantage-rate function $q$. However, they enforce optimality through complicated martingale losses or orthogonality constraints, which are sensitive to the choice of test processes. These approaches entangle $V$ and $q$ into a large, complex optimization problem that is difficult to train reliably. To address these limitations, we propose a novel decoupled continuous-time actor-critic algorithm with alternating updates: $q$ is learned from diffusion generators on $V$, and $V$ is updated via a Hamiltonian-based value flow that remains informative under infinitesimal time steps, where standard max/softmax backups fail. Theoretically, we prove rigorous convergence via new probabilistic arguments, sidestepping the challenge that generator-based Hamiltonians lack Bellman-style contraction under the sup-norm. Empirically, our method outperforms prior continuous-time and leading discrete-time baselines across challenging continuous-control benchmarks and a real-world trading task, achieving 21% profit over a single quarter$-$nearly doubling the second-best method.
Authors: Tianyi Ma, Yiyang Li, Yiyue Qian, Zheyuan Zhang, Zehong Wang, Chuxu Zhang, Yanfang Ye
Abstract: The opioid epidemic continues to ravage communities worldwide, straining healthcare systems, disrupting families, and demanding urgent computational solutions. To combat this lethal opioid crisis, graph learning methods have emerged as a promising paradigm for modeling complex drug-related phenomena. However, a significant gap remains: there is no comprehensive benchmark for systematically evaluating these methods across real-world opioid crisis scenarios. To bridge this gap, we introduce OPBench, the first comprehensive opioid benchmark comprising five datasets across three critical application domains: opioid overdose detection from healthcare claims, illicit drug trafficking detection from digital platforms, and drug misuse prediction from dietary patterns. Specifically, OPBench incorporates diverse graph structures, including heterogeneous graphs and hypergraphs, to preserve the rich and complex relational information among drug-related data. To address data scarcity, we collaborate with domain experts and authoritative institutions to curate and annotate datasets while adhering to privacy and ethical guidelines. Furthermore, we establish a unified evaluation framework with standardized protocols, predefined data splits, and reproducible baselines to facilitate fair and systematic comparison among graph learning methods. Through extensive experiments, we analyze the strengths and limitations of existing graph learning methods, thereby providing actionable insights for future research in combating the opioid crisis. Our source code and datasets are available at https://github.com/Tianyi-Billy-Ma/OPBench.
Authors: Karim Galliamov, Syed M Ahsan Kazmi, Adil Khan, Ad\'in Ram\'irez Rivera
Abstract: Concept Bottleneck Models (CBMs) aim to deliver interpretable predictions by routing decisions through a human-understandable concept layer, yet they often suffer reduced accuracy and concept leakage that undermines faithfulness. We introduce an explicit Information Bottleneck regularizer on the concept layer that penalizes $I(X;C)$ while preserving task-relevant information in $I(C;Y)$, encouraging minimal-sufficient concept representations. We derive two practical variants (a variational objective and an entropy-based surrogate) and integrate them into standard CBM training without architectural changes or additional supervision. Evaluated across six CBM families and three benchmarks, the IB-regularized models consistently outperform their vanilla counterparts. Information-plane analyses further corroborate the intended behavior. These results indicate that enforcing a minimal-sufficient concept bottleneck improves both predictive performance and the reliability of concept-level interventions. The proposed regularizer offers a theoretic-grounded, architecture-agnostic path to more faithful and intervenable CBMs, resolving prior evaluation inconsistencies by aligning training protocols and demonstrating robust gains across model families and datasets.
Authors: Rohit Raj Rai, Abhishek Dhaka, Amit Awekar
Abstract: Compressed Deep Learning (DL) models are essential for deployment in resource-constrained environments. But their performance often lags behind their large-scale counterparts. To bridge this gap, we propose Alignment Adapter (AlAd): a lightweight, sliding-window-based adapter. It aligns the token-level embeddings of a compressed model with those of the original large model. AlAd preserves local contextual semantics, enables flexible alignment across differing dimensionalities or architectures, and is entirely agnostic to the underlying compression method. AlAd can be deployed in two ways: as a plug-and-play module over a frozen compressed model, or by jointly fine-tuning AlAd with the compressed model for further performance gains. Through experiments on BERT-family models across three token-level NLP tasks, we demonstrate that AlAd significantly boosts the performance of compressed models with only marginal overhead in size and latency.
Authors: Adri\'an Javaloy, Antonio Vergari
Abstract: Orthogonality constraints are ubiquitous in robust and probabilistic machine learning. Unfortunately, current optimizers are computationally expensive and do not scale to problems with hundreds or thousands of constraints. One notable exception is the Landing algorithm (Ablin et al., 2024) which, however comes at the expense of temporarily relaxing orthogonality. In this work, we revisit and improve on the ideas behind Landing, enabling the inclusion of modern adaptive optimizers while ensuring that orthogonal constraints are effectively met. Remarkably, these improvements come at little to no cost, and reduce the number of required hyperparemeters. Our algorithm POGO is fast and GPU-friendly, consisting of only 5 matrix products, and in practice maintains orthogonality at all times. On several challenging benchmarks, POGO greatly outperforms recent optimizers and shows it can optimize problems with thousands of orthogonal matrices in minutes while alternatives would take hours. As such, POGO sets a milestone to finally exploit orthogonality constraints in ML at scale. A PyTorch implementation of POGO is publicly available at https://github.com/adrianjav/pogo.
Authors: Andrew Gracyk
Abstract: We present pseudo-differential enhanced physics-informed neural networks (PINNs), an extension of gradient enhancement but in Fourier space. Gradient enhancement of PINNs dictates that the PDE residual is taken to a higher differential order than prescribed by the PDE, added to the objective as an augmented term in order to improve training and overall learning fidelity. We propose the same procedure after application via Fourier transforms, since differentiating in Fourier space is multiplication with the Fourier wavenumber under suitable decay. Our methods are fast and efficient. Our methods oftentimes achieve superior PINN versus numerical error in fewer training iterations, potentially pair well with few samples in collocation, and can on occasion break plateaus in low collocation settings. Moreover, our methods are suitable for fractional derivatives. We establish that our methods improve spectral eigenvalue decay of the neural tangent kernel (NTK), and so our methods contribute towards the learning of high frequencies in early training, mitigating the effects of frequency bias up to the polynomial order and possibly greater with smooth activations. Our methods accommodate advanced techniques in PINNs, such as Fourier feature embeddings. A pitfall of discrete Fourier transforms via the Fast Fourier Transform (FFT) is mesh subjugation, and so we demonstrate compatibility of our methods for greater mesh flexibility and invariance on alternative Euclidean and non-Euclidean domains via Monte Carlo methods and otherwise.
Authors: Farzan Farnia, Mohammad Jalali, Azim Ospanov
Abstract: Deep generative models have achieved great success in producing high-quality samples, making them a central tool across machine learning applications. Beyond sample quality, an important yet less systematically studied question is whether trained generative models faithfully capture the diversity of the underlying data distribution. In this work, we address this question by directly comparing the diversity of samples generated by state-of-the-art models with that of test samples drawn from the target data distribution, using recently proposed reference-free entropy-based diversity scores, Vendi and RKE. Across multiple benchmark datasets, we find that test data consistently attains substantially higher Vendi and RKE diversity scores than the generated samples, suggesting a systematic downward diversity bias in modern generative models. To understand the origin of this bias, we analyze the finite-sample behavior of entropy-based diversity scores and show that their expected values increase with sample size, implying that diversity estimated from finite training sets could inherently underestimate the diversity of the true distribution. As a result, optimizing the generators to minimize divergence to empirical data distributions would induce a loss of diversity. Finally, we discuss potential diversity-aware regularization and guidance strategies based on Vendi and RKE as principled directions for mitigating this bias, and provide empirical evidence suggesting their potential to improve the results.
Authors: David Chanin, Adri\`a Garriga-Alonso
Abstract: Improving Sparse Autoencoders (SAEs) requires benchmarks that can precisely validate architectural innovations. However, current SAE benchmarks on LLMs are often too noisy to differentiate architectural improvements, and current synthetic data experiments are too small-scale and unrealistic to provide meaningful comparisons. We introduce SynthSAEBench, a toolkit for generating large-scale synthetic data with realistic feature characteristics including correlation, hierarchy, and superposition, and a standardized benchmark model, SynthSAEBench-16k, enabling direct comparison of SAE architectures. Our benchmark reproduces several previously observed LLM SAE phenomena, including the disconnect between reconstruction and latent quality metrics, poor SAE probing results, and a precision-recall trade-off mediated by L0. We further use our benchmark to identify a new failure mode: Matching Pursuit SAEs exploit superposition noise to improve reconstruction without learning ground-truth features, suggesting that more expressive encoders can easily overfit. SynthSAEBench complements LLM benchmarks by providing ground-truth features and controlled ablations, enabling researchers to precisely diagnose SAE failure modes and validate architectural improvements before scaling to LLMs.
Authors: Nihal V. Nayak, Paula Rodriguez-Diaz, Neha Hulkund, Sara Beery, David Alvarez-Melis
Abstract: Instruction fine-tuning of large language models (LLMs) often involves selecting a subset of instruction training data from a large candidate pool, using a small query set from the target task. Despite growing interest, the literature on targeted instruction selection remains fragmented and opaque: methods vary widely in selection budgets, often omit zero-shot baselines, and frequently entangle the contributions of key components. As a result, practitioners lack actionable guidance on selecting instructions for their target tasks. In this work, we aim to bring clarity to this landscape by disentangling and systematically analyzing the two core ingredients: data representation and selection algorithms. Our framework enables controlled comparisons across models, tasks, and budgets. We find that only gradient-based data representations choose subsets whose similarity to the query consistently predicts performance across datasets and models. While no single method dominates, gradient-based representations paired with a greedy round-robin selection algorithm tend to perform best on average at low budgets, but these benefits diminish at larger budgets. Finally, we unify several existing selection algorithms as forms of approximate distance minimization between the selected subset and the query set, and support this view with new generalization bounds. More broadly, our findings provide critical insights and a foundation for more principled data selection in LLM fine-tuning. The code is available at https://github.com/dcml-lab/targeted-instruction-selection.
URLs: https://github.com/dcml-lab/targeted-instruction-selection.
Authors: Killian Bakong (DI-ENS), Laurent Massouli\'e (Inria, ARGO, CMAP), Edouard Oyallon (MLIA), Kevin Scaman
Abstract: In this work we introduce methods to reduce the computational and memory costs of training deep neural networks. Our approach consists in replacing exact vector-jacobian products by randomized, unbiased approximations thereof during backpropagation. We provide a theoretical analysis of the trade-off between the number of epochs needed to achieve a target precision and the cost reduction for each epoch. We then identify specific unbiased estimates of vector-jacobian products for which we establish desirable optimality properties of minimal variance under sparsity constraints. Finally we provide in-depth experiments on multi-layer perceptrons, BagNets and Visual Transfomers architectures. These validate our theoretical results, and confirm the potential of our proposed unbiased randomized backpropagation approach for reducing the cost of deep learning.
Authors: Nozomu Fujisawa, Masaaki Kondo
Abstract: We systematically investigate the parameter-efficient fine-tuning design space under practical data and compute constraints, and propose D2-LoRA. D2-LoRA achieves 76.4 percent average accuracy across eight question answering and reading comprehension benchmarks using only 5k training samples per task and two epochs, while preserving algebraic mergeability at inference with near-exact numerical equivalence. The method combines signed low-rank residual updates with additive and subtractive components, together with a train-time column-wise projection that keeps each column close to its original norm. After training, the adapter is merged into a single weight matrix, adding zero inference latency. Compared with LoRA, D2-LoRA improves average accuracy by 2.2 percentage points; at matched parameter counts (LoRA rank 2r versus D2-LoRA rank r), the improvement is 1.6 points, indicating gains from architectural design rather than increased parameterization. Compared with DoRA, it matches or exceeds performance on most tasks. Beyond QA and reading comprehension, D2-LoRA improves generative tasks (plus 1.2 ROUGE-L and plus 1.1 percent win rate) and shows 36 percent lower training volatility. The merge preserves numerical fidelity (mean gap about 0.03 percentage points) and recovers about 1.91x evaluation throughput. Training overhead is 19 percent, comparable to DoRA, and decreases with longer input sequences. We provide a geometric analysis explaining how the projection stabilizes training, together with ablation studies isolating the contribution of each design component.
Authors: Rodrigo Carmo Terin
Abstract: Neural networks with positively homogeneous activations exhibit an exact continuous reparametrization symmetry: neuron-wise rescalings generate parameter-space orbits along which the input--output function is invariant. We interpret this symmetry as a gauge redundancy and introduce gauge-adapted coordinates that separate invariant and scale-imbalance directions. Inspired by gauge fixing in field theory, we introduce a soft orbit-selection (norm-balancing) functional acting only on redundant scale coordinates. We show analytically that it induces dissipative relaxation of imbalance modes to preserve the realized function. In controlled experiments, this orbit-selection penalty expands the stable learning-rate regime and suppresses scale drift without changing expressivity. These results establish a structural link between gauge-orbit geometry and optimization conditioning, providing a concrete connection between gauge-theoretic concepts and machine learning.
Authors: T. Matuli\'c, D. Ser\v{s}i\'c
Abstract: We propose a parameter-minimal neural architecture for solving differential equations by restricting the hypothesis class to Horner-factorized polynomials, yielding an implicit, differentiable trial solution with only a small set of learnable coefficients. Initial conditions are enforced exactly by construction by fixing the low-order polynomial degrees of freedom, so training focuses solely on matching the differential-equation residual at collocation points. To reduce approximation error without abandoning the low-parameter regime, we introduce a piecewise ("spline-like") extension that trains multiple small Horner models on subintervals while enforcing continuity (and first-derivative continuity) at segment boundaries. On illustrative ODE benchmarks and a heat-equation example, Horner networks with tens (or fewer) parameters accurately match the solution and its derivatives and outperform small MLP and sinusoidal-representation baselines under the same training settings, demonstrating a practical accuracy-parameter trade-off for resource-efficient scientific modeling.
Authors: Jonathan Lys, Vincent Gripon, Bastien Pasdeloup, Lukas Mauch, Fabien Cardinaux, Ghouthi Boukli Hacene
Abstract: Deep Learning architectures, and in particular Transformers, are conventionally viewed as a composition of layers. These layers are actually often obtained as the sum of two contributions: a residual path that copies the input and the output of a Transformer block. As a consequence, the inner representations (i.e. the input of these blocks) can be interpreted as iterative refinement of a propagated latent representation. Under this lens, many works suggest that the inner space is shared across layers, meaning that tokens can be decoded at early stages. Mechanistic interpretability even goes further by conjecturing that some layers act as refinement layers. Following this path, we propose inference-time inner looping, which prolongs refinement in pretrained off-the-shelf language models by repeatedly re-applying a selected block range. Across multiple benchmarks, inner looping yields modest but consistent accuracy improvements. Analyses of the resulting latent trajectories suggest more stable state evolution and continued semantic refinement. Overall, our results suggest that additional refinement can be obtained through simple test-time looping, extending computation in frozen pretrained models.
Authors: Stefano Woerner, Seong Joon Oh, Christian F. Baumgartner
Abstract: Current meta-learning methods are constrained to narrow task distributions with fixed feature and label spaces, limiting applicability. Moreover, the current meta-learning literature uses key terms like "universal" and "general-purpose" inconsistently and lacks precise definitions, hindering comparability. We introduce a theoretical framework for meta-learning which formally defines practical universality and introduces a distinction between algorithm-explicit and algorithm-implicit learning, providing a principled vocabulary for reasoning about universal meta-learning methods. Guided by this framework, we present TAIL, a transformer-based algorithm-implicit meta-learner that functions across tasks with varying domains, modalities, and label configurations. TAIL features three innovations over prior transformer-based meta-learners: random projections for cross-modal feature encoding, random injection label embeddings that extrapolate to larger label spaces, and efficient inline query processing. TAIL achieves state-of-the-art performance on standard few-shot benchmarks while generalizing to unseen domains. Unlike other meta-learning methods, it also generalizes to unseen modalities, solving text classification tasks despite training exclusively on images, handles tasks with up to 20$\times$ more classes than seen during training, and provides orders-of-magnitude computational savings over prior transformer-based approaches.
Authors: Sungwoo Kang
Abstract: The Winner Determination Problem (WDP) in combinatorial auctions is NP-hard, and no existing method reliably predicts which instances will defeat fast greedy heuristics. The ML-for-combinatorial-optimization community has focused on learning to \emph{replace} solvers, yet recent evidence shows that graph neural networks (GNNs) rarely outperform well-tuned classical methods on standard benchmarks. We pursue a different objective: learning to predict \emph{when} a given instance is hard for greedy allocation, enabling instance-dependent algorithm selection. We design a 20-dimensional structural feature vector and train a lightweight MLP hardness classifier that predicts the greedy optimality gap with mean absolute error 0.033, Pearson correlation 0.937, and binary classification accuracy 94.7\% across three random seeds. For instances identified as hard -- those exhibiting ``whale-fish'' trap structure where greedy provably fails -- we deploy a heterogeneous GNN specialist that achieves ${\approx}0\%$ optimality gap on all six adversarial configurations tested (vs.\ 3.75--59.24\% for greedy). A hybrid allocator combining the hardness classifier with GNN and greedy solvers achieves 0.51\% overall gap on mixed distributions. Our honest evaluation on CATS benchmarks confirms that GNNs do not outperform Gurobi (0.45--0.71 vs.\ 0.20 gap), motivating the algorithm selection framing. Learning \emph{when} to deploy expensive solvers is more tractable than learning to replace them.
Authors: Rotem Mulayoff, Sebastian U. Stich
Abstract: The dynamical stability of the iterates during training plays a key role in determining the minima obtained by optimization algorithms. For example, stable solutions of gradient descent (GD) correspond to flat minima, which have been associated with favorable features. While prior work often relies on linearization to determine stability, it remains unclear whether linearized dynamics faithfully capture the full nonlinear behavior. Recent work has shown that GD may stably oscillate near a linearly unstable minimum and still converge once the step size decays, indicating that linear analysis can be misleading. In this work, we explicitly study the effect of nonlinear terms. Specifically, we derive an exact criterion for stable oscillations of GD near minima in the multivariate setting. Our condition depends on high-order derivatives, generalizing existing results. Extending the analysis to stochastic gradient descent (SGD), we show that nonlinear dynamics can diverge in expectation even if a single batch is unstable. This implies that stability can be dictated by a single batch that oscillates unstably, rather than an average effect, as linear analysis suggests. Finally, we prove that if all batches are linearly stable, the nonlinear dynamics of SGD are stable in expectation.
Authors: Luuk Jacobs, Mohammad Ali Javidian
Abstract: Multi-Source Bayesian Optimization (MSBO) serves as a variant of the traditional Bayesian Optimization (BO) framework applicable to situations involving optimization of an objective black-box function over multiple information sources such as simulations, surrogate models, or real-world experiments. However, traditional MSBO assumes the input variables of the objective function to be independent and identically distributed, limiting its effectiveness in scenarios where causal information is available and interventions can be performed, such as clinical trials or policy-making. In the single-source domain, Causal Bayesian Optimization (CBO) extends standard BO with the principles of causality, enabling better modeling of variable dependencies. This leads to more accurate optimization, improved decision-making, and more efficient use of low-cost information sources. In this article, we propose a principled integration of the MSBO and CBO methodologies in the multi-source domain, leveraging the strengths of both to enhance optimization efficiency and reduce computational complexity in higher-dimensional problems. We present the theoretical foundations of both Causal and Multi-Source Bayesian Optimization, and demonstrate how their synergy informs our Multi-Source Causal Bayesian Optimization (MSCBO) algorithm. We compare the performance of MSCBO against its foundational counterparts for both synthetic and real-world datasets with varying levels of noise, highlighting the robustness and applicability of MSCBO. Based on our findings, we conclude that integrating MSBO with the causality principles of CBO facilitates dimensionality reduction and lowers operational costs, ultimately improving convergence speed, performance, and scalability.
Authors: Julien Siems, Riccardo Grazzi, Kirill Kalinin, Hitesh Ballani, Babak Rahmani
Abstract: Over the last years, state-tracking tasks, particularly permutation composition, have become a testbed to understand the limits of sequence models architectures like Transformers and RNNs (linear and non-linear). However, these are often sequence-to-sequence tasks: learning to map actions (permutations) to states, which is incompatible with the next-token prediction setting commonly used to train language models. We address this gap by converting permutation composition into code via REPL traces that interleave state-reveals through prints and variable transformations. We show that linear RNNs capable of state-tracking excel also in this setting, while Transformers still fail. Motivated by this representation, we investigate why tracking states in code is generally difficult: actions are not always fully observable. We frame this as tracking the state of a probabilistic finite-state automaton with deterministic state reveals and show that linear RNNs can be worse than non-linear RNNs at tracking states in this setup.
Authors: Elias Malomgr\'e, Pieter Simoens
Abstract: AI alignment is growing in importance, yet current approaches suffer from a critical structural flaw that entangles the safety objectives with the agent's policy. Methods such as Reinforcement Learning from Human Feedback and Direct Preference Optimization create opaque, single-use alignment artifacts, which we term Alignment Waste. We propose Interactionless Inverse Reinforcement Learning to decouple alignment artifact learning from policy optimization, producing an inspectable, editable, and model-agnostic reward model. Additionally, we introduce the Alignment Flywheel, a human-in-the-loop lifecycle that iteratively hardens the reward model through automated audits and refinement. This architecture transforms safety from a disposable expense into a durable, verifiable engineering asset.
Authors: Bardia Mohammadi, Nearchos Potamitis, Lars Klein, Akhil Arora, Laurent Bindschaedler
Abstract: LLM agents increasingly act on external systems, yet tool effects are immediate. Under failures, speculation, or contention, losing branches can leak unintended side effects with no safe rollback. We introduce Atomix, a runtime that provides progress-aware transactional semantics for agent tool calls. Atomix tags each call with an epoch, tracks per-resource frontiers, and commits only when progress predicates indicate safety; bufferable effects can be delayed, while externalized effects are tracked and compensated on abort. Across real workloads with fault injection, transactional retry improves task success, while frontier-gated commit strengthens isolation under speculation and contention.
Authors: Jonathan Gorard, Ammar Hakim, James Juno
Abstract: The traditional limitations of neural networks in reliably generalizing beyond the convex hulls of their training data present a significant problem for computational physics, in which one often wishes to solve PDEs in regimes far beyond anything which can be experimentally or analytically validated. In this paper, we show how it is possible to circumvent these limitations by constructing formally-verified neural network solvers for PDEs, with rigorous convergence, stability, and conservation properties, whose correctness can therefore be guaranteed even in extrapolatory regimes. By using the method of characteristics to predict the analytical properties of PDE solutions a priori (even in regions arbitrarily far from the training domain), we show how it is possible to construct rigorous extrapolatory bounds on the worst-case L^inf errors of shallow neural network approximations. Then, by decomposing PDE solutions into compositions of simpler functions, we show how it is possible to compose these shallow neural networks together to form deep architectures, based on ideas from compositional deep learning, in which the large L^inf errors in the approximations have been suppressed. The resulting framework, called BEACONS (Bounded-Error, Algebraically-COmposable Neural Solvers), comprises both an automatic code-generator for the neural solvers themselves, as well as a bespoke automated theorem-proving system for producing machine-checkable certificates of correctness. We apply the framework to a variety of linear and non-linear PDEs, including the linear advection and inviscid Burgers' equations, as well as the full compressible Euler equations, in both 1D and 2D, and illustrate how BEACONS architectures are able to extrapolate solutions far beyond the training data in a reliable and bounded way. Various advantages of the approach over the classical PINN approach are discussed.
Authors: Ryan DeWolfe, Pawe{\l} Pra{\l}at, Fran\c{c}ois Th\'eberge
Abstract: Clustering algorithms are an essential part of the unsupervised data science ecosystem, and extrinsic evaluation of clustering algorithms requires a method for comparing the detected clustering to a ground truth clustering. In a general setting, the detected and ground truth clusterings may have outliers (objects belonging to no cluster), overlapping clusters (objects may belong to more than one cluster), or both, but methods for comparing these clusterings are currently undeveloped. In this note, we define a pragmatic similarity measure for comparing clusterings with overlaps and outliers, show that it has several desirable properties, and experimentally confirm that it is not subject to several common biases afflicting other clustering comparison measures.
Authors: Ilia Mahrooghi, Aryo Lotfi, Emmanuel Abbe
Abstract: Reinforcement learning has emerged as a powerful paradigm for unlocking reasoning capabilities in large language models. However, relying on sparse rewards makes this process highly sample-inefficient, as models must navigate vast search spaces with minimal feedback. While classic curriculum learning aims to mitigate this by ordering data based on complexity, the right ordering for a specific model is often unclear. To address this, we propose Goldilocks, a novel teacher-driven data sampling strategy that aims to predict each question's difficulty for the student model. The teacher model selects questions of appropriate difficulty for the student model, i.e., questions that are neither too easy nor too hard (Goldilocks principle), while training the student with GRPO. By leveraging the student's performance on seen samples, the teacher continuously adapts to the student's evolving abilities. On OpenMathReasoning dataset, Goldilocks data sampling improves the performance of models trained with standard GRPO under the same compute budget.
Authors: Yu Huang, Zixin Wen, Yuejie Chi, Yuting Wei, Aarti Singh, Yingbin Liang, Yuxin Chen
Abstract: Reinforcement learning with verifiable rewards (RLVR) has been a main driver of recent breakthroughs in large reasoning models. Yet it remains a mystery how rewards based solely on final outcomes can help overcome the long-horizon barrier to extended reasoning. To understand this, we develop a theory of the training dynamics of RL for transformers on compositional reasoning tasks. Our theory characterizes how the effectiveness of RLVR is governed by the smoothness of the difficulty spectrum. When data contains abrupt discontinuities in difficulty, learning undergoes grokking-type phase transitions, producing prolonged plateaus before progress recurs. In contrast, a smooth difficulty spectrum leads to a relay effect: persistent gradient signals on easier problems elevate the model's capabilities to the point where harder ones become tractable, resulting in steady and continuous improvement. Our theory explains how RLVR can improve performance at the edge of competence, and suggests that appropriately designed data mixtures can yield scalable gains. As a technical contribution, our analysis develops and adapts tools from Fourier analysis on finite groups to our setting. We validate the predicted mechanisms empirically via synthetic experiments.
Authors: Mounvik K, N Harshit
Abstract: We introduce Web-Scale Multimodal Summarization, a lightweight framework for generating summaries by combining retrieved text and image data from web sources. Given a user-defined topic, the system performs parallel web, news, and image searches. Retrieved images are ranked using a fine-tuned CLIP model to measure semantic alignment with topic and text. Optional BLIP captioning enables image-only summaries for stronger multimodal coherence.The pipeline supports features such as adjustable fetch limits, semantic filtering, summary styling, and downloading structured outputs. We expose the system via a Gradio-based API with controllable parameters and preconfigured presets.Evaluation on 500 image-caption pairs with 20:1 contrastive negatives yields a ROC-AUC of 0.9270, an F1-score of 0.6504, and an accuracy of 96.99%, demonstrating strong multimodal alignment. This work provides a configurable, deployable tool for web-scale summarization that integrates language, retrieval, and vision models in a user-extensible pipeline.
Authors: Pedram Bakhtiarifard, Tong Chen, Jonathan Wensh{\o}j, Erik B Dam, Raghavendra Selvan
Abstract: Large-scale deep learning models are well-suited for compression. Methods like pruning, quantization, and knowledge distillation have been used to achieve massive reductions in the number of model parameters, with marginal performance drops across a variety of architectures and tasks. This raises the central question: \emph{Why are deep neural networks suited for compression?} In this work, we take up the perspective of algorithmic complexity to explain this behavior. We hypothesize that the parameters of trained models have more structure and, hence, exhibit lower algorithmic complexity compared to the weights at (random) initialization. Furthermore, that model compression methods harness this reduced algorithmic complexity to compress models. Although an unconstrained parameterization of model weights, $\mathbf{w} \in \mathbb{R}^n$, can represent arbitrary weight assignments, the solutions found during training exhibit repeatability and structure, making them algorithmically simpler than a generic program. To this end, we formalize the Kolmogorov complexity of $\mathbf{w}$ by $\mathcal{K}(\mathbf{w})$. We introduce a constrained parameterization $\widehat{\mathbf{w}}$, that partitions parameters into blocks of size $s$, and restricts each block to be selected from a set of $k$ reusable motifs, specified by a reuse pattern (or mosaic). The resulting method, $\textit{Mosaic-of-Motifs}$ (MoMos), yields algorithmically simpler model parameterization compared to unconstrained models. Empirical evidence from multiple experiments shows that the algorithmic complexity of neural networks, measured using approximations to Kolmogorov complexity, can be reduced during training. This results in models that perform comparably with unconstrained models while being algorithmically simpler.
Authors: Pramit Saha, Joshua Strong, Mohammad Alsharid, Divyanshu Mishra, J. Alison Noble
Abstract: Task-specialized models form the backbone of agentic healthcare systems, enabling the agents to answer clinical queries across tasks such as disease diagnosis, localization, and report generation. Yet, for a given task, a single "best" model rarely exists. In practice, each task is better served by multiple competing specialist models where different models excel on different data samples. As a result, for any given query, agents must reliably select the right specialist model from a heterogeneous pool of tool candidates. To this end, we introduce ToolSelect, which adaptively learns model selection for tools by minimizing a population risk over sampled specialist tool candidates using a consistent surrogate of the task-conditional selection loss. Concretely, we propose an Attentive Neural Process-based selector conditioned on the query and per-model behavioral summaries to choose among the specialist models. Motivated by the absence of any established testbed, we, for the first time, introduce an agentic Chest X-ray environment equipped with a diverse suite of task-specialized models (17 disease detection, 19 report generation, 6 visual grounding, and 13 VQA) and develop ToolSelectBench, a benchmark of 1448 queries. Our results demonstrate that ToolSelect consistently outperforms 10 SOTA methods across four different task families.
Authors: Farbod Siahkali, Ashwin Verma, Vijay Gupta
Abstract: Conformal prediction (CP) offers distribution-free marginal coverage guarantees under an exchangeability assumption, but these guarantees can fail if the data distribution shifts. We analyze the use of pseudo-calibration as a tool to counter this performance loss under a bounded label-conditional covariate shift model. Using tools from domain adaptation, we derive a lower bound on target coverage in terms of the source-domain loss of the classifier and a Wasserstein measure of the shift. Using this result, we provide a method to design pseudo-calibrated sets that inflate the conformal threshold by a slack parameter to keep target coverage above a prescribed level. Finally, we propose a source-tuned pseudo-calibration algorithm that interpolates between hard pseudo-labels and randomized labels as a function of classifier uncertainty. Numerical experiments show that our bounds qualitatively track pseudo-calibration behavior and that the source-tuned scheme mitigates coverage degradation under distribution shift while maintaining nontrivial prediction set sizes.
Authors: Olivier Jeunen, Shashank Gupta
Abstract: Off-policy evaluation (OPE) is essential for assessing ranking and recommendation systems without costly online interventions. Self-Normalised Inverse Propensity Scoring (SNIPS) is a standard tool for variance reduction in OPE, leveraging a multiplicative control variate. Recent advances in off-policy learning suggest that additive control variates (baseline corrections) may offer superior performance, yet theoretical guarantees for evaluation are lacking. This paper provides a definitive answer: we prove that $\beta^\star$-IPS, an estimator with an optimal additive baseline, asymptotically dominates SNIPS in Mean Squared Error. By analytically decomposing the variance gap, we show that SNIPS is asymptotically equivalent to using a specific -- but generally sub-optimal -- additive baseline. Our results theoretically justify shifting from self-normalisation to optimal baseline corrections for both ranking and recommendation.
Authors: Tianyi Ma, Yiyue Qian, Zehong Wang, Zheyuan Zhang, Chuxu Zhang, Yanfang Ye
Abstract: Hypergraph Neural Networks (HyGNNs) have demonstrated remarkable success in modeling higher-order relationships among entities. However, their performance often degrades on heterophilic hypergraphs, where nodes connected by the same hyperedge tend to have dissimilar semantic representations or belong to different classes. While several HyGNNs, including our prior work BHyGNN, have been proposed to address heterophily, their reliance on labeled data significantly limits their applicability in real-world scenarios where annotations are scarce or costly. To overcome this limitation, we introduce BHyGNN+, a self-supervised learning framework that extends BHyGNN for representation learning on heterophilic hypergraphs without requiring ground-truth labels. The core idea of BHyGNN+ is hypergraph duality, a structural transformation where the roles of nodes and hyperedges are interchanged. By contrasting augmented views of a hypergraph against its dual using cosine similarity, our framework captures essential structural patterns in a fully unsupervised manner. Notably, this duality-based formulation eliminates the need for negative samples, a common requirement in existing hypergraph contrastive learning methods that is often difficult to satisfy in practice. Extensive experiments on eleven benchmark datasets demonstrate that BHyGNN+ consistently outperforms state-of-the-art supervised and self-supervised baselines on both heterophilic and homophilic hypergraphs. Our results validate the effectiveness of leveraging hypergraph duality for self-supervised learning and establish a new paradigm for representation learning on challenging, unlabeled hypergraphs.
Authors: Martin Van Waerebeke, Marco Lorenzi, Kevin Scaman, El Mahdi El Mhamdi, Giovanni Neglia
Abstract: In machine unlearning, $(\varepsilon,\delta)-$unlearning is a popular framework that provides formal guarantees on the effectiveness of the removal of a subset of training data, the forget set, from a trained model. For strongly convex objectives, existing first-order methods achieve $(\varepsilon,\delta)-$unlearning, but they only use the forget set to calibrate injected noise, never as a direct optimization signal. In contrast, efficient empirical heuristics often exploit the forget samples (e.g., via gradient ascent) but come with no formal unlearning guarantees. We bridge this gap by presenting the Variance-Reduced Unlearning (VRU) algorithm. To the best of our knowledge, VRU is the first first-order algorithm that directly includes forget set gradients in its update rule, while provably satisfying ($(\varepsilon,\delta)-$unlearning. We establish the convergence of VRU and show that incorporating the forget set yields strictly improved rates, i.e. a better dependence on the achieved error compared to existing first-order $(\varepsilon,\delta)-$unlearning methods. Moreover, we prove that, in a low-error regime, VRU asymptotically outperforms any first-order method that ignores the forget set.Experiments corroborate our theory, showing consistent gains over both state-of-the-art certified unlearning methods and over empirical baselines that explicitly leverage the forget set.
Authors: Jivat Neet Kaur, Isaac Gibbs, Michael I. Jordan
Abstract: We consider the general problem of learning a predictor that satisfies multiple objectives of interest simultaneously, a broad framework that captures a range of specific learning goals including calibration, regret, and multiaccuracy. We work in an online setting where the data distribution can change arbitrarily over time. Existing approaches to this problem aim to minimize the set of objectives over the entire time horizon in a worst-case sense, and in practice they do not necessarily adapt to distribution shifts. Earlier work has aimed to alleviate this problem by incorporating additional objectives that target local guarantees over contiguous subintervals. Empirical evaluation of these proposals is, however, scarce. In this article, we consider an alternative procedure that achieves local adaptivity by replacing one part of the multi-objective learning method with an adaptive online algorithm. Empirical evaluations on datasets from energy forecasting and algorithmic fairness show that our proposed method improves upon existing approaches and achieves unbiased predictions over subgroups, while remaining robust under distribution shift.
Authors: Arik Reuter, Anish Dhir, Cristiana Diaconu, Jake Robertson, Ole Ossen, Frank Hutter, Adrian Weller, Mark van der Wilk, Bernhard Sch\"olkopf
Abstract: Estimating causal quantities traditionally relies on bespoke estimators tailored to specific assumptions. Recently proposed Causal Foundation Models (CFMs) promise a more unified approach by amortising causal discovery and inference in a single step. However, in their current state, they do not allow for the incorporation of any domain knowledge, which can lead to suboptimal predictions. We bridge this gap by introducing methods to condition CFMs on causal information, such as the causal graph or more readily available ancestral information. When access to complete causal graph information is too strict a requirement, our approach also effectively leverages partial causal information. We systematically evaluate conditioning strategies and find that injecting learnable biases into the attention mechanism is the most effective method to utilise full and partial causal information. Our experiments show that this conditioning allows a general-purpose CFM to match the performance of specialised models trained on specific causal structures. Overall, our approach addresses a central hurdle on the path towards all-in-one causal foundation models: the capability to answer causal queries in a data-driven manner while effectively leveraging any amount of domain expertise.
Authors: Alicja Maksymiuk, Alexandre Duplessis, Michael Bronstein, Alexander Tong, Fernanda Duarte, \.Ismail \.Ilkan Ceylan
Abstract: Macrocycles are ring-shaped molecules that offer a promising alternative to small-molecule drugs due to their enhanced selectivity and binding affinity against difficult targets. Despite their chemical value, they remain underexplored in generative modeling, likely owing to their scarcity in public datasets and the challenges of enforcing topological constraints in standard deep generative models. We introduce MacroGuide: Topological Guidance for Macrocycle Generation, a diffusion guidance mechanism that uses Persistent Homology to steer the sampling of pretrained molecular generative models toward the generation of macrocycles, in both unconditional and conditional (protein pocket) settings. At each denoising step, MacroGuide constructs a Vietoris-Rips complex from atomic positions and promotes ring formation by optimizing persistent homology features. Empirically, applying MacroGuide to pretrained diffusion models increases macrocycle generation rates from 1% to 99%, while matching or exceeding state-of-the-art performance on key quality metrics such as chemical validity, diversity, and PoseBusters checks.
Authors: Carolin Cissee, Raneen Younis, Zahra Ahmadi
Abstract: Multimodal learning seeks to integrate information from heterogeneous sources, where signals may be shared across modalities, specific to individual modalities, or emerge only through their interaction. While self-supervised multimodal contrastive learning has achieved remarkable progress, most existing methods predominantly capture redundant cross-modal signals, often neglecting modality-specific (unique) and interaction-driven (synergistic) information. Recent extensions broaden this perspective, yet they either fail to explicitly model synergistic interactions or learn different information components in an entangled manner, leading to incomplete representations and potential information leakage. We introduce \textbf{COrAL}, a principled framework that explicitly and simultaneously preserves redundant, unique, and synergistic information within multimodal representations. COrAL employs a dual-path architecture with orthogonality constraints to disentangle shared and modality-specific features, ensuring a clean separation of information components. To promote synergy modeling, we introduce asymmetric masking with complementary view-specific patterns, compelling the model to infer cross-modal dependencies rather than rely solely on redundant cues. Extensive experiments on synthetic benchmarks and diverse MultiBench datasets demonstrate that COrAL consistently matches or outperforms state-of-the-art methods while exhibiting low performance variance across runs. These results indicate that explicitly modeling the full spectrum of multimodal information yields more stable, reliable, and comprehensive embeddings.
Authors: Tim Mangliers, Bernhard M\"ossner, Benjamin Himpel
Abstract: Geometric deep learning (GDL) deals with supervised learning on data domains that go beyond Euclidean structure, such as data with graph or manifold structure. Due to the demand that arises from application-related data, there is a need to identify further topological and geometric structures with which these use cases can be made accessible to machine learning. There are various techniques, such as spectral convolution, that form the basic building blocks for some convolutional neural network-like architectures on non-Euclidean data. In this paper, the concept of spectral convolution on orbifolds is introduced. This provides a building block for making learning on orbifold structured data accessible using GDL. The theory discussed is illustrated using an example from music theory.
Authors: Xander Davies, Giorgi Giglemiani, Edmund Lau, Eric Winsor, Geoffrey Irving, Yarin Gal
Abstract: Frontier LLMs are safeguarded against attempts to extract harmful information via adversarial prompts known as "jailbreaks". Recently, defenders have developed classifier-based systems that have survived thousands of hours of human red teaming. We introduce Boundary Point Jailbreaking (BPJ), a new class of automated jailbreak attacks that evade the strongest industry-deployed safeguards. Unlike previous attacks that rely on white/grey-box assumptions (such as classifier scores or gradients) or libraries of existing jailbreaks, BPJ is fully black-box and uses only a single bit of information per query: whether or not the classifier flags the interaction. To achieve this, BPJ addresses the core difficulty in optimising attacks against robust real-world defences: evaluating whether a proposed modification to an attack is an improvement. Instead of directly trying to learn an attack for a target harmful string, BPJ converts the string into a curriculum of intermediate attack targets and then actively selects evaluation points that best detect small changes in attack strength ("boundary points"). We believe BPJ is the first fully automated attack algorithm that succeeds in developing universal jailbreaks against Constitutional Classifiers, as well as the first automated attack algorithm that succeeds against GPT-5's input classifier without relying on human attack seeds. BPJ is difficult to defend against in individual interactions but incurs many flags during optimisation, suggesting that effective defence requires supplementing single-interaction methods with batch-level monitoring.
Authors: Johannes Schmude, Sujit Roy, Liping Wang, Theodore van Kessel, Levente Klein, Marcus Freitag, Eloisa Bentivegna, Robert Manson-Sawko, Bjorn Lutjens, Manil Maskey, Campbell Watson, Rahul Ramachandran, Juan Bernabe-Moreno
Abstract: We show that AI foundation models that are pretrained on numerical solutions to a diverse corpus of partial differential equations can be adapted and fine-tuned to obtain skillful predictive weather emulators for the Martian atmosphere. We base our work on the Poseidon PDE foundation model for two-dimensional systems. We develop a method to extend Poseidon from two to three dimensions while keeping the pretraining information. Moreover, we investigate the performance of the model in the presence of sparse initial conditions. Our results make use of four Martian years (approx.~34 GB) of training data and a median compute budget of 13 GPU hours. We find that the combination of pretraining and model extension yields a performance increase of 34.4\% on a held-out year. This shows that PDEs-FMs can not only approximate solutions to (other) PDEs but also anchor models for real-world problems with complex interactions that lack a sufficient amount of training data or a suitable compute budget.
Authors: Daniil Dmitriev, Zhihan Huang, Yuting Wei
Abstract: Diffusion models over discrete spaces have recently shown striking empirical success, yet their theoretical foundations remain incomplete. In this paper, we study the sampling efficiency of score-based discrete diffusion models under a continuous-time Markov chain (CTMC) formulation, with a focus on $\tau$-leaping-based samplers. We establish sharp convergence guarantees for attaining $\varepsilon$ accuracy in Kullback-Leibler (KL) divergence for both uniform and masking noising processes. For uniform discrete diffusion, we show that the $\tau$-leaping algorithm achieves an iteration complexity of order $\tilde O(d/\varepsilon)$, with $d$ the ambient dimension of the target distribution, eliminating linear dependence on the vocabulary size $S$ and improving existing bounds by a factor of $d$; moreover, we establish a matching algorithmic lower bound showing that linear dependence on the ambient dimension is unavoidable in general. For masking discrete diffusion, we introduce a modified $\tau$-leaping sampler whose convergence rate is governed by an intrinsic information-theoretic quantity, termed the effective total correlation, which is bounded by $d \log S$ but can be sublinear or even constant for structured data. As a consequence, the sampler provably adapts to low-dimensional structure without prior knowledge or algorithmic modification, yielding sublinear convergence rates for various practical examples (such as hidden Markov models, image data, and random graphs). Our analysis requires no boundedness or smoothness assumptions on the score estimator beyond control of the score entropy loss.
Authors: Subham Sekhar Sahoo, Jean-Marie Lemercier, Zhihan Yang, Justin Deschenaux, Jingyu Liu, John Thickstun, Ante Jukic
Abstract: Diffusion language models are a promising alternative to autoregressive models due to their potential for faster generation. Among discrete diffusion approaches, Masked diffusion currently dominates, largely driven by strong perplexity on language modeling benchmarks. In this work, we present the first scaling law study of uniform-state and interpolating discrete diffusion methods. We also show that Masked diffusion models can be made approximately 12% more FLOPs-efficient when trained with a simple cross-entropy objective. We find that perplexity is informative within a diffusion family but can be misleading across families, where models with worse likelihood scaling may be preferable due to faster and more practical sampling, as reflected by the speed-quality Pareto frontier. These results challenge the view that Masked diffusion is categorically the future of diffusion language modeling and that perplexity alone suffices for cross-algorithm comparison. Scaling all methods to 1.7B parameters, we show that uniform-state diffusion remains competitive on likelihood-based benchmarks and outperforms autoregressive and Masked diffusion models on GSM8K, despite worse validation perplexity. We provide the code, model checkpoints, and video tutorials on the project page: http://s-sahoo.github.io/scaling-dllms
Authors: Cai Zhou, Zijie Chen, Zian Li, Jike Wang, Kaiyi Jiang, Pan Li, Rose Yu, Muhan Zhang, Stephen Bates, Tommi Jaakkola
Abstract: Many generative tasks in chemistry and science involve distributions invariant to group symmetries (e.g., permutation and rotation). A common strategy enforces invariance and equivariance through architectural constraints such as equivariant denoisers and invariant priors. In this paper, we challenge this tradition through the alternative canonicalization perspective: first map each sample to an orbit representative with a canonical pose or order, train an unconstrained (non-equivariant) diffusion or flow model on the canonical slice, and finally recover the invariant distribution by sampling a random symmetry transform at generation time. Building on a formal quotient-space perspective, our work provides a comprehensive theory of canonical diffusion by proving: (i) the correctness, universality and superior expressivity of canonical generative models over invariant targets; (ii) canonicalization accelerates training by removing diffusion score complexity induced by group mixtures and reducing conditional variance in flow matching. We then show that aligned priors and optimal transport act complementarily with canonicalization and further improves training efficiency. We instantiate the framework for molecular graph generation under $S_n \times SE(3)$ symmetries. By leveraging geometric spectra-based canonicalization and mild positional encodings, canonical diffusion significantly outperforms equivariant baselines in 3D molecule generation tasks, with similar or even less computation. Moreover, with a novel architecture Canon, CanonFlow achieves state-of-the-art performance on the challenging GEOM-DRUG dataset, and the advantage remains large in few-step generation.
Authors: Shangding Gu
Abstract: Large language models (LLMs) are increasingly deployed in privacy-critical and personalization-oriented scenarios, yet the role of context length in shaping privacy leakage and personalization effectiveness remains largely unexplored. We introduce a large-scale benchmark, PAPerBench, to systematically study how increasing context length influences both personalization quality and privacy protection in LLMs. The benchmark comprises approximately 29,000 instances with context lengths ranging from 1K to 256K tokens, yielding a total of 377K evaluation questions. It jointly evaluates personalization performance and privacy risks across diverse scenarios, enabling controlled analysis of long-context model behavior. Extensive evaluations across state-of-the-art LLMs reveal consistent performance degradation in both personalization and privacy as context length increases. We further provide a theoretical analysis of attention dilution under context scaling, explaining this behavior as an inherent limitation of soft attention in fixed-capacity Transformers. The empirical and theoretical findings together suggest a general scaling gap in current models -- long context, less focus. We release the benchmark to support reproducible evaluation and future research on scalable privacy and personalization. Code and data are available at https://github.com/SafeRL-Lab/PAPerBench
Authors: Dhruva Karkada, Daniel J. Korchinski, Andres Nava, Matthieu Wyart, Yasaman Bahri
Abstract: Although learned representations underlie neural networks' success, their fundamental properties remain poorly understood. A striking example is the emergence of simple geometric structures in LLM representations: for example, calendar months organize into a circle, years form a smooth one-dimensional manifold, and cities' latitudes and longitudes can be decoded by a linear probe. We show that the statistics of language exhibit a translation symmetry -- e.g., the co-occurrence probability of two months depends only on the time interval between them -- and we prove that the latter governs the aforementioned geometric structures in high-dimensional word embedding models. Moreover, we find that these structures persist even when the co-occurrence statistics are strongly perturbed (for example, by removing all sentences in which two months appear together) and at moderate embedding dimension. We show that this robustness naturally emerges if the co-occurrence statistics are collectively controlled by an underlying continuous latent variable. We empirically validate this theoretical framework in word embedding models, text embedding models, and large language models.
Authors: William Xion, Wolfgang Nejdl
Abstract: Dense retrieval is a promising approach for acquiring relevant context or world knowledge in open-domain natural language processing tasks and is now widely used in information retrieval applications. However, recent reports claim a broad preference for text generated by large language models (LLMs). This bias is called "source bias", and it has been hypothesized that lower perplexity contributes to this effect. In this study, we revisit this claim by conducting a controlled evaluation to trace the emergence of such preferences across training stages and data sources. Using parallel human- and LLM-generated counterparts of the SciFact and Natural Questions (NQ320K) datasets, we compare unsupervised checkpoints with models fine-tuned using in-domain human text, in-domain LLM-generated text, and MS MARCO. Our results show the following: 1) Unsupervised retrievers do not exhibit a uniform pro-LLM preference. The direction and magnitude depend on the dataset. 2) Across the settings tested, supervised fine-tuning on MS MARCO consistently shifts the rankings toward LLM-generated text. 3) In-domain fine-tuning produces dataset-specific and inconsistent shifts in preference. 4) Fine-tuning on LLM-generated corpora induces a pronounced pro-LLM bias. Finally, a retriever-centric perplexity probe involving the reattachment of a language modeling head to the fine-tuned dense retriever encoder indicates agreement with relevance near chance, thereby weakening the explanatory power of perplexity. Our study demonstrates that source bias is a training-induced phenomenon rather than an inherent property of dense retrievers.
Authors: Jie Zheng, Ruichen Zhang, Dusit Niyato, Haijun Zhang, Jiacheng Wang, Hongyang Du, Jiawen Kang, Zehui Xiong
Abstract: Enhancing future wireless networks presents a significant challenge for networking systems due to diverse user demands and the emergence of 6G technology. While reinforcement learning (RL) is a powerful framework, it often encounters difficulties with high-dimensional state spaces and complex environments, leading to substantial computational demands, distributed intelligence, and potentially inconsistent outcomes. Large language models (LLMs), with their extensive pretrained knowledge and advanced reasoning capabilities, offer promising tools to enhance RL in optimizing 6G wireless networks. We explore RL models augmented by LLMs, emphasizing their roles and the potential benefits of their synergy in wireless network optimization. We then examine LLM-enabled RL across various protocol layers: physical, data link, network, transport, and application layers. Additionally, we propose an LLM-assisted state representation and semantic extraction to enhance the multi-agent reinforcement learning (MARL) framework. This approach is applied to service migration and request routing, as well as topology graph generation in unmanned aerial vehicle (UAV)-satellite networks. Through case studies, we demonstrate that our framework effectively performs optimization of wireless network. Finally, we outline prospective research directions for LLM-enabled RL in wireless network optimization.
Authors: Joyjit Roy, Samaresh Kumar Singh
Abstract: Commercial insurance underwriting is a labor-intensive process that requires manual review of extensive documentation to assess risk and determine policy pricing. While AI offers substantial efficiency improvements, existing solutions lack comprehensive reasoning capabilities and internal mechanisms to ensure reliability within regulated, high-stakes environments. Full automation remains impractical and inadvisable in scenarios where human judgment and accountability are critical. This study presents a decision-negative, human-in-the-loop agentic system that incorporates an adversarial self-critique mechanism as a bounded safety architecture for regulated underwriting workflows. Within this system, a critic agent challenges the primary agent's conclusions prior to submitting recommendations to human reviewers. This internal system of checks and balances addresses a critical gap in AI safety for regulated workflows. Additionally, the research develops a formal taxonomy of failure modes to characterize potential errors by decision-negative agents. This taxonomy provides a structured framework for risk identification and risk management in high-stakes applications. Experimental evaluation using 500 expert-validated underwriting cases demonstrates that the adversarial critique mechanism reduces AI hallucination rates from 11.3% to 3.8% and increases decision accuracy from 92% to 96%. At the same time, the framework enforces strict human authority over all binding decisions by design. These findings indicate that adversarial self-critique supports safer AI deployment in regulated domains and offers a model for responsible integration where human oversight is indispensable.
Authors: Bowen Liu, Zhi Wu, Runquan Xie, Zhanhui Kang, Jia Li
Abstract: Scaling verifiable training signals remains a key bottleneck for Reinforcement Learning from Verifiable Rewards (RLVR). Logical reasoning is a natural substrate: constraints are formal and answers are programmatically checkable. However, prior synthesis pipelines either depend on expert-written code or operate within fixed templates/skeletons, which limits growth largely to instance-level perturbations. We propose SSLogic, an agentic meta-synthesis framework that scales at the task-family level by iteratively synthesizing and repairing executable Generator--Validator program pairs in a closed Generate--Validate--Repair loop, enabling continuous family evolution with controllable difficulty. To ensure reliability, we introduce a Multi-Gate Validation Protocol that combines multi-strategy consistency checks with Adversarial Blind Review, where independent agents must solve instances by writing and executing code to filter ambiguous or ill-posed tasks. Starting from 400 seed families, two evolution rounds expand to 953 families and 21,389 verifiable instances (from 5,718). Training on SSLogic-evolved data yields consistent gains over the seed baseline at matched training steps, improving SynLogic by +5.2, BBEH by +1.4, AIME25 by +3.0, and Brumo25 by +3.7.
Authors: Truong Xuan Khanh, Truong Quynh Hoa
Abstract: Despite recent advances in artificial intelligence, many systems exhibit stagnation in long-horizon adaptability despite continued performance optimization. This work argues that such limitations do not primarily arise from insufficient learning, data, or model capacity, but from a deeper structural property of how intelligence is optimized over time. We formulate intelligence as a trajectory-level phenomenon governed by multi-objective trade-offs, and introduce Trajectory-Dominant Pareto Optimization, a path-wise generalization of classical Pareto optimality in which dominance is defined over full trajectories. Within this framework, Pareto traps emerge as locally non-dominated regions of trajectory space that nevertheless restrict access to globally superior developmental paths under conservative local optimization. To characterize the rigidity of such constraints, we define the Trap Escape Difficulty Index (TEDI), a composite geometric measure capturing escape distance, structural constraints, and behavioral inertia. We show that dynamic intelligence ceilings arise as inevitable geometric consequences of trajectory-level dominance, independent of learning progress or architectural scale. We further introduce a formal taxonomy of Pareto traps and illustrate the resulting trajectory-level divergence using a minimal agent-environment model. Together, these results shift the locus of intelligence from terminal performance to optimization geometry, providing a principled framework for diagnosing and overcoming long-horizon developmental constraints in adaptive systems.
Authors: Khaleda Papry, Francesco Spinnato, Marco Fiore, Mirco Nanni, Israat Haque
Abstract: As 5G networks continue to evolve to deliver high speed, low latency, and reliable communications, ensuring uninterrupted service has become increasingly critical. While millimeter wave (mmWave) frequencies enable gigabit data rates, they are highly susceptible to environmental factors, often leading to radio link failures (RLF). Predictive models leveraging radio and weather data have been proposed to address this issue; however, many operate as black boxes, offering limited transparency for operational deployment. This work bridges that gap by introducing a framework that combines explainability based feature pruning with model refinement. Our framework can be integrated into state of the art predictors such as GNN Transformer and LSTM based architectures for RLF prediction, enabling the development of accurate and explainability guided models in 5G networks. It provides insights into the contribution of input features and the decision making logic of neural networks, leading to lighter and more scalable models. When applied to RLF prediction, our framework unveils that weather data contributes minimally to the forecast in extensive real world datasets, which informs the design of a leaner model with 50 percent fewer parameters and improved F1 scores with respect to the state of the art solution. Ultimately, this work empowers network providers to evaluate and refine their neural network based prediction models for better interpretability, scalability, and performance.
Authors: Le-Hung Hoang, Quang-Trung Luu, Dinh Thai Hoang, Diep N. Nguyen, Van-Dinh Nguyen
Abstract: Stacked intelligent metasurfaces (SIMs) have recently emerged as a powerful wave-domain technology that enables multi-stage manipulation of electromagnetic signals through multilayer programmable architectures. While SIMs offer unprecedented degrees of freedom for enhancing physical-layer security, their extremely large number of meta-atoms leads to a high-dimensional and strongly coupled optimization space, making conventional design approaches inefficient and difficult to scale. Moreover, existing deep reinforcement learning (DRL) techniques suffer from slow convergence and performance degradation in dynamic wireless environments with imperfect knowledge of passive eavesdroppers. To overcome these challenges, we propose a hybrid quantum proximal policy optimization (Q-PPO) framework for SIM-assisted secure communications, which jointly optimizes transmit power allocation and SIM phase shifts to maximize the average secrecy rate under power and quality-of-service constraints. Specifically, a parameterized quantum circuit is embedded into the actor network, forming a hybrid classical-quantum policy architecture that enhances policy representation capability and exploration efficiency in high-dimensional continuous action spaces. Extensive simulations demonstrate that the proposed Q-PPO scheme consistently outperforms DRL baselines, achieving approximately 15% higher secrecy rates and 30% faster convergence under imperfect eavesdropper channel state information. These results establish Q-PPO as a powerful optimization paradigm for SIM-enabled secure wireless networks.
Authors: Hyosoon Jang, Hyunjin Seo, Yunhui Jang, Seonghyun Park, Sungsoo Ahn
Abstract: Foundation models in molecular learning have advanced along two parallel tracks: protein models, which typically utilize evolutionary information to learn amino acid-level representations for folding, and small-molecule models, which focus on learning atom-level representations for property prediction tasks such as ADMET. Notably, cutting-edge protein-centric models such as Boltz now operate at atom-level granularity for protein-ligand co-folding, yet their atom-level expressiveness for small-molecule tasks remains unexplored. A key open question is whether these protein co-folding models capture transferable chemical physics or rely on protein evolutionary signals, which would limit their utility for small-molecule tasks. In this work, we investigate the quality of Boltz atom-level representations across diverse small-molecule benchmarks. Our results show that Boltz is competitive with specialized baselines on ADMET property prediction tasks and effective for molecular generation and optimization. These findings suggest that the representational capacity of cutting-edge protein-centric models has been underexplored and position Boltz as a strong baseline for atom-level representation learning for small molecules.
Authors: Aisha Aijaz, Raghava Mutharaju, Manohar Kumar
Abstract: Moral cognition is a crucial yet underexplored aspect of decision-making in AI models. Regardless of the application domain, it should be a consideration that allows for ethically aligned decision-making. This paper presents a multifaceted contribution to this research space. Firstly, a comparative analysis of techniques to instill ethical competence into AI models has been presented to gauge them on multiple performance metrics. Second, a novel mathematical discretization of morality and a demonstration of its real-life application have been conveyed and tested against other techniques on two datasets. This value is modeled as the risk of loss incurred by the least moral cases, or an Expected Moral Shortfall (EMS), which we direct the AI model to minimize in order to maximize its performance while retaining ethical competence. Lastly, the paper discusses the tradeoff between preliminary AI decision-making metrics such as model performance, complexity, and scale of ethical competence to recognize the true extent of practical social impact.
Authors: Md Muntasir Jahid Ayan, Md. Shahriar Rashid, Tazzina Afroze Hassan, Hossain Md. Mubashshir Jamil, Mahbubul Islam, Lisan Al Amin, Rupak Kumar Das, Farzana Akter, Faisal Quader
Abstract: The increasing complexity and frequency of cyber-threats demand intrusion detection systems (IDS) that are not only accurate but also interpretable. This paper presented a novel IDS framework that integrated Explainable Artificial Intelligence (XAI) to enhance transparency in deep learning models. The framework was evaluated experimentally using the benchmark dataset NSL-KDD, demonstrating superior performance compared to traditional IDS and black-box deep learning models. The proposed approach combined Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks for capturing temporal dependencies in traffic sequences. Our deep learning results showed that both CNN and LSTM reached 0.99 for accuracy, whereas LSTM outperformed CNN at macro average precision, recall, and F-1 score. For weighted average precision, recall, and F-1 score, both models scored almost similarly. To ensure interpretability, the XAI model SHapley Additive exPlanations (SHAP) was incorporated, enabling security analysts to understand and validate model decisions. Some notable influential features were srv_serror_rate, dst_host_srv_serror_rate, and serror_rate for both models, as pointed out by SHAP. We also conducted a trust-focused expert survey based on IPIP6 and Big Five personality traits via an interactive UI to evaluate the system's reliability and usability. This work highlighted the potential of combining performance and transparency in cybersecurity solutions and recommends future enhancements through adaptive learning for real-time threat detection.
Authors: Muyan Weng, Defu Cao, Wei Yang, Yashaswi Sharma, Yan Liu
Abstract: It is unclear whether strong forecasting performance reflects genuine temporal understanding or the ability to reason under contextual and event-driven conditions. We introduce TemporalBench, a multi-domain benchmark designed to evaluate temporal reasoning behavior under progressively richer informational settings. TemporalBench adopts a four-tier task taxonomy that examines historical structure interpretation, context-free forecasting, contextual temporal reasoning, and event-conditioned prediction across four real-world domains: retail, healthcare, energy, and physical systems. By controlling access to future targets and contextual information, the benchmark enables a diagnostic analysis of whether models can correctly interpret temporal patterns, align them with external context, and adapt predictions when conditions change. Extensive baseline experiments show that strong numerical forecasting accuracy does not reliably translate into robust contextual or event-aware temporal reasoning; instead, existing agent frameworks exhibit fragmented strengths and systematic failure modes that remain largely hidden under forecasting-only benchmarks. The TemporalBench dataset is publicly available at https://huggingface.co/datasets/Melady/TemporalBench, and we additionally provide a public leaderboard at https://huggingface.co/spaces/Melady/TemporalBench_Leaderboard.
URLs: https://huggingface.co/datasets/Melady/TemporalBench,, https://huggingface.co/spaces/Melady/TemporalBench_Leaderboard.
Authors: Nathanya Satriani, Djordje Slijep\v{c}evi\'c, Markus Schedl, Matthias Zeppelzauer
Abstract: Explanatory interactive learning (XIL) enables users to guide model training in machine learning (ML) by providing feedback on the model's explanations, thereby helping it to focus on features that are relevant to the prediction from the user's perspective. In this study, we explore the capability of this learning paradigm to mitigate bias and spurious correlations in visual classifiers, specifically in scenarios prone to data bias, such as gender classification. We investigate two methodologically different state-of-the-art XIL strategies, i.e., CAIPI and Right for the Right Reasons (RRR), as well as a novel hybrid approach that combines both strategies. The results are evaluated quantitatively by comparing segmentation masks with explanations generated using Gradient-weighted Class Activation Mapping (GradCAM) and Bounded Logit Attention (BLA). Experimental results demonstrate the effectiveness of these methods in (i) guiding ML models to focus on relevant image features, particularly when CAIPI is used, and (ii) reducing model bias (i.e., balancing the misclassification rates between male and female predictions). Our analysis further supports the potential of XIL methods to improve fairness in gender classifiers. Overall, the increased transparency and fairness obtained by XIL leads to slight performance decreases with an exception being CAIPI, which shows potential to even improve classification accuracy.
Authors: Mohammad Saiful Islam, Andriy Miranskyy
Abstract: Anomaly detection is important for keeping cloud systems reliable and stable. Deep learning has improved time-series anomaly detection, but most models are evaluated on one dataset at a time. This raises questions about whether these models can handle different types of telemetry, especially in large-scale and high-dimensional environments. In this study, we evaluate four deep learning models, GRU, TCN, Transformer, and TSMixer. We also include Isolation Forest as a classical baseline. The models are tested across four telemetry datasets: the Numenta Anomaly Benchmark, Microsoft Cloud Monitoring dataset, Exathlon dataset, and IBM Console dataset. These datasets differ in structure, dimensionality, and labelling strategy. They include univariate time series, synthetic multivariate workloads, and real-world production telemetry with over 100,000 features. We use a unified training and evaluation pipeline across all datasets. The evaluation includes NAB-style metrics to capture early detection behaviour for datasets where anomalies persist over contiguous time intervals. This enables window-based scoring in settings where anomalies occur over contiguous time intervals, even when labels are recorded at the point level. The unified setup enables consistent analysis of model behaviour under shared scoring and calibration assumptions. Our results demonstrate that anomaly detection performance in cloud systems is governed not only by model architecture, but critically by calibration stability and feature-space geometry. By releasing our preprocessing pipelines, benchmark configuration, and evaluation artifacts, we aim to support reproducible and deployment-aware evaluation of anomaly detection systems for cloud environments.
Authors: Edwyn Brient (CMM), Santiago Velasco-Forero (CMM), Rami Kassab
Abstract: High-resolution range profile (HRRP ) data are in vogue in radar automatic target recognition (RATR). With the interest in classifying models using HRRP, filling gaps in datasets using generative models has recently received promising contributions. Evaluating generated data is a challenging topic, even for explicit data like face images. However, the evaluation methods used in the state-ofthe-art of HRRP generation rely on classification models. Such models, called ''black-box'', do not allow either explainability on generated data or multi-level evaluation. This work focuses on decomposing HRRP data into three components: the mask, the features, and the noise. Using this decomposition, we propose two metrics based on the physical interpretation of those data. We take profit from an expensive dataset to evaluate our metrics on a challenging task and demonstrate the discriminative ability of those.
Authors: Edwyn Brient (CMM), Santiago Velasco-Forero (CMM), Rami Kassab
Abstract: High-resolution range profiles (HRRPs) enable fast onboard processing for radar automatic target recognition, but their strong sensitivity to acquisition conditions limits robustness across operational scenarios. Conditional HRRP generation can mitigate this issue, yet prior studies are constrained by small, highly specific datasets. We study HRRP synthesis on a largescale maritime database representative of coastal surveillance variability. Our analysis indicates that the fundamental scenario drivers are geometric: ship dimensions and the desired aspect angle. Conditioning on these variables, we train generative models and show that the synthesized signatures reproduce the expected line-of-sight geometric trend observed in real data. These results highlight the central role of acquisition geometry for robust HRRP generation.
Authors: Nicolas Bourriez, Alexandre Verine, Auguste Genovesio
Abstract: Conditional diffusion inversion provides a powerful framework for unpaired image-to-image translation. However, we demonstrate through an extensive analysis that standard deterministic inversion (e.g. DDIM) fails when the source domain is spectrally sparse compared to the target domain (e.g., super-resolution, sketch-to-image). In these contexts, the recovered latent from the input does not follow the expected isotropic Gaussian distribution. Instead it exhibits a signal with lower frequencies, locking target sampling to oversmoothed and texture-poor generations. We term this phenomenon spectral collapse. We observe that stochastic alternatives attempting to restore the noise variance tend to break the semantic link to the input, leading to structural drift. To resolve this structure-texture trade-off, we propose Orthogonal Variance Guidance (OVG), an inference-time method that corrects the ODE dynamics to enforce the theoretical Gaussian noise magnitude within the null-space of the structural gradient. Extensive experiments on microscopy super-resolution (BBBC021) and sketch-to-image (Edges2Shoes) demonstrate that OVG effectively restores photorealistic textures while preserving structural fidelity.
Authors: Zhehan Zhang, Meihua Qian, Li Luo, Siyu Huang, Chaoyi Zhou, Ripon Saha, Xinxin Song
Abstract: Assessing artistic creativity is foundational to creativity research and arts education, yet manual scoring (e.g., Torrance Tests of Creative Thinking) is labor-intensive at scale. Prior machine-learning approaches show promise for visual creativity scoring, but many rely mainly on image features and provide limited or no explanatory feedback. We propose a framework for automated creativity assessment of human paintings by fine-tuning the vision-language model Qwen2-VL-7B with multi-task learning. Our dataset contains 1000 human-created paintings scored on a 1-100 scale and paired with a short human-written description (content or artist explanation). Two expert raters evaluated each work using a five-dimension rubric (originality, color, texture, composition, content) and provided written critiques; we use an 80/20 train-test split. We add a lightweight regression head on the visual encoder output so the model can predict a numerical score and generate rubric-aligned feedback in a single forward pass. By embedding the structured rubric and the artwork description in the system prompt, we constrain the generated text to match the quantitative prediction. Experiments show strong accuracy, achieving Pearson r > 0.97 and MAE about 3.95 on the 100-point scale. Qualitative evaluation indicates the generated feedback is semantically close to expert critiques (average SBERT cosine similarity = 0.798). The proposed approach bridges computer vision and art assessment and offers a scalable tool for creativity research and classroom feedback.
Authors: Daesik Jang, Morgan Lindsay Heisler, Linzi Xing, Yifei Li, Edward Wang, Ying Xiong, Yong Zhang, Zhenan Fan
Abstract: Automatically generating and iteratively editing academic slide decks requires more than document summarization. It demands faithful content selection, coherent slide organization, layout-aware rendering, and robust multi-turn instruction following. However, existing benchmarks and evaluation protocols do not adequately measure these challenges. To address this gap, we introduce the Deck Edits and Compliance Kit Benchmark (DECKBench), an evaluation framework for multi-agent slide generation and editing. DECKBench is built on a curated dataset of paper to slide pairs augmented with realistic, simulated editing instructions. Our evaluation protocol systematically assesses slide-level and deck-level fidelity, coherence, layout quality, and multi-turn instruction following. We further implement a modular multi-agent baseline system that decomposes the slide generation and editing task into paper parsing and summarization, slide planning, HTML creation, and iterative editing. Experimental results demonstrate that the proposed benchmark highlights strengths, exposes failure modes, and provides actionable insights for improving multi-agent slide generation and editing systems. Overall, this work establishes a standardized foundation for reproducible and comparable evaluation of academic presentation generation and editing. Code and data are publicly available at https://github.com/morgan-heisler/DeckBench .
Authors: Tri Nguyen, Huy Hoang Bao Le, Lohith Srikanth Pentapalli, Laurah Turner, Kelly Cohen
Abstract: Detecting jailbreak attempts in clinical training large language models (LLMs) requires accurate modeling of linguistic deviations that signal unsafe or off-task user behavior. Prior work on the 2-Sigma clinical simulation platform showed that manually annotated linguistic features could support jailbreak detection. However, reliance on manual annotation limited both scalability and expressiveness. In this study, we extend this framework by using experts' annotations of four core linguistic features (Professionalism, Medical Relevance, Ethical Behavior, and Contextual Distraction) and training multiple general-domain and medical-domain BERT-based LLM models to predict these features directly from text. The most reliable feature regressor for each dimension was selected and used as the feature extractor in a second layer of classifiers. We evaluate a suite of predictive models, including tree-based, linear, probabilistic, and ensemble methods, to determine jailbreak likelihood from the extracted features. Across cross-validation and held-out evaluations, the system achieves strong overall performance, indicating that LLM-derived linguistic features provide an effective basis for automated jailbreak detection. Error analysis further highlights key limitations in current annotations and feature representations, pointing toward future improvements such as richer annotation schemes, finer-grained feature extraction, and methods that capture the evolving risk of jailbreak behavior over the course of a dialogue. This work demonstrates a scalable and interpretable approach for detecting jailbreak behavior in safety-critical clinical dialogue systems.
Authors: Datorien L. Anderson
Abstract: We present the PolyShapes-Ideal (PSI) dataset, a suite of diagnostic benchmarks designed to isolate topological invariance -- the ability to maintain structural identity across affine transformations -- from the textural correlations that dominate standard vision benchmarks. Through three diagnostic probes (polygon classification under noise, zero-shot font transfer from MNIST, and geometric collapse mapping under progressive deformation), we demonstrate that the Eidos architecture achieves >99% accuracy on PSI and 81.67% zero-shot transfer across 30 unseen typefaces without pre-training. These results validate the "Form-First" hypothesis: generalization in structurally constrained architectures is a property of geometric integrity, not statistical scale.
Authors: C\'edric Allier, Larissa Heinrich, Magdalena Schneider, Stephan Saalfeld
Abstract: Graph neural networks trained to predict observable dynamics can be used to decompose the temporal activity of complex heterogeneous systems into simple, interpretable representations. Here we apply this framework to simulated neural assemblies with thousands of neurons and demonstrate that it can jointly reveal the connectivity matrix, the neuron types, the signaling functions, and in some cases hidden external stimuli. In contrast to existing machine learning approaches such as recurrent neural networks and transformers, which emphasize predictive accuracy but offer limited interpretability, our method provides both reliable forecasts of neural activity and interpretable decomposition of the mechanisms governing large neural assemblies.
Authors: Hao Liu, Suhaib A. Fahmy
Abstract: Deploying Vision Transformers on edge devices is challenging due to their high computational complexity, while full offloading to cloud resources presents significant latency overheads. We propose a novel collaborative inference framework, which orchestrates a lightweight generalist ViT on an edge device and multiple medium-sized expert ViTs on a near-edge accelerator. A novel routing mechanism uses the edge model's Top-$\mathit{k}$ predictions to dynamically select the most relevant expert for samples with low confidence. We further design a progressive specialist training strategy to enhance expert accuracy on dataset subsets. Extensive experiments on the CIFAR-100 dataset using a real-world edge and near-edge testbed demonstrate the superiority of our framework. Specifically, the proposed training strategy improves expert specialization accuracy by 4.12% on target subsets and enhances overall accuracy by 2.76% over static experts. Moreover, our method reduces latency by up to 45% compared to edge execution, and energy consumption by up to 46% compared to just near-edge offload.
Authors: \'Ad\'am Jung, Domokos M. Kelen, Andr\'as A. Bencz\'ur
Abstract: A key challenge in probabilistic regression is ensuring that predictive distributions accurately reflect true empirical uncertainty. Minimizing overall prediction error often encourages models to prioritize informativeness over calibration, producing narrow but overconfident predictions. However, in safety-critical settings, trustworthy uncertainty estimates are often more valuable than narrow intervals. Realizing the problem, several recent works have focused on post-hoc corrections; however, existing methods either rely on weak notions of calibration (such as PIT uniformity) or impose restrictive parametric assumptions on the nature of the error. To address these limitations, we propose a novel nonparametric re-calibration algorithm based on conditional kernel mean embeddings, capable of correcting calibration error without restrictive modeling assumptions. For efficient inference with real-valued targets, we introduce a novel characteristic kernel over distributions that can be evaluated in $\mathcal{O}(n \log n)$ time for empirical distributions of size $n$. We demonstrate that our method consistently outperforms prior re-calibration approaches across a diverse set of regression benchmarks and model classes.
Authors: Simon Rosen, Siddarth Singh, Ebenezer Gelo, Helen Sarah Robertson, Ibrahim Suder, Victoria Williams, Benjamin Rosman, Geraud Nangue Tasse, Steven James
Abstract: Evaluating moral alignment in agents navigating conflicting, hierarchically structured human norms is a critical challenge at the intersection of AI safety, moral philosophy, and cognitive science. We introduce Morality Chains, a novel formalism for representing moral norms as ordered deontic constraints, and MoralityGym, a benchmark of 98 ethical-dilemma problems presented as trolley-dilemma-style Gymnasium environments. By decoupling task-solving from moral evaluation and introducing a novel Morality Metric, MoralityGym allows the integration of insights from psychology and philosophy into the evaluation of norm-sensitive reasoning. Baseline results with Safe RL methods reveal key limitations, underscoring the need for more principled approaches to ethical decision-making. This work provides a foundation for developing AI systems that behave more reliably, transparently, and ethically in complex real-world contexts.
Authors: Sohail Ali Farooqui, Zuhair Ahmed Khan Taha, Mohammed Mudassir Uddin, Shahnawaz Alam
Abstract: Unmanned aerial vehicles serve as primary sensing platforms for surveillance, traffic monitoring, and disaster response, making aerial object detection a central problem in applied computer vision. Current detectors struggle with UAV-specific challenges: targets spanning only a few pixels, cluttered backgrounds, heavy occlusion, and strict onboard computational budgets. This study introduces LAF-YOLOv10, built on YOLOv10n, integrating four complementary techniques to improve small-object detection in drone imagery. A Partial Convolution C2f (PC-C2f) module restricts spatial convolution to one quarter of backbone channels, reducing redundant computation while preserving discriminative capacity. An Attention-Guided Feature Pyramid Network (AG-FPN) inserts Squeeze-and-Excitation channel gates before multi-scale fusion and replaces nearest-neighbor upsampling with DySample for content-aware interpolation. An auxiliary P2 detection head at 160$\times$160 resolution extends localization to objects below 8$\times$8 pixels, while the P5 head is removed to redistribute parameters. Wise-IoU v3 replaces CIoU for bounding box regression, attenuating gradients from noisy annotations in crowded aerial scenes. The four modules address non-overlapping bottlenecks: PC-C2f compresses backbone computation, AG-FPN refines cross-scale fusion, the P2 head recovers spatial resolution, and Wise-IoU stabilizes regression under label noise. No individual component is novel; the contribution is the joint integration within a single YOLOv10 framework. Across three training runs (seeds 42, 123, 256), LAF-YOLOv10 achieves 35.1$\pm$0.3\% mAP@0.5 on VisDrone-DET2019 with 2.3\,M parameters, exceeding YOLOv10n by 3.3 points. Cross-dataset evaluation on UAVDT yields 35.8$\pm$0.4\% mAP@0.5. Benchmarks on NVIDIA Jetson Orin Nano confirm 24.3 FPS at FP16, demonstrating viability for embedded UAV deployment.
Authors: Xu Li, Simon Yu, Minzhou Pan, Yiyou Sun, Bo Li, Dawn Song, Xue Lin, Weiyan Shi
Abstract: LLM-based agents are becoming increasingly capable, yet their safety lags behind. This creates a gap between what agents can do and should do. This gap widens as agents engage in multi-turn interactions and employ diverse tools, introducing new risks overlooked by existing benchmarks. To systematically scale safety testing into multi-turn, tool-realistic settings, we propose a principled taxonomy that transforms single-turn harmful tasks into multi-turn attack sequences. Using this taxonomy, we construct MT-AgentRisk (Multi-Turn Agent Risk Benchmark), the first benchmark to evaluate multi-turn tool-using agent safety. Our experiments reveal substantial safety degradation: the Attack Success Rate (ASR) increases by 16% on average across open and closed models in multi-turn settings. To close this gap, we propose ToolShield, a training-free, tool-agnostic, self-exploration defense: when encountering a new tool, the agent autonomously generates test cases, executes them to observe downstream effects, and distills safety experiences for deployment. Experiments show that ToolShield effectively reduces ASR by 30% on average in multi-turn interactions. Our code is available at https://github.com/CHATS-lab/ToolShield.
Authors: Pooya Ashtari, Pourya Behmandpoor, Nikos Deligiannis, Aleksandra Pizurica
Abstract: Implicit neural representations (INRs) have emerged as powerful tools for encoding signals, yet dominant MLP-based designs often suffer from slow convergence, overfitting to noise, and poor extrapolation. We introduce FUTON (Fourier Tensor Network), which models signals as generalized Fourier series whose coefficients are parameterized by a low-rank tensor decomposition. FUTON implicitly expresses signals as weighted combinations of orthonormal, separable basis functions, combining complementary inductive biases: Fourier bases capture smoothness and periodicity, while the low-rank parameterization enforces low-dimensional spectral structure. We provide theoretical guarantees through a universal approximation theorem and derive an inference algorithm with complexity linear in the spectral resolution and the input dimension. On image and volume representation, FUTON consistently outperforms state-of-the-art MLP-based INRs while training 2--5$\times$ faster. On inverse problems such as image denoising and super-resolution, FUTON generalizes better and converges faster.
Authors: Shreyas Vinaya Sathyanarayana, Shah Rahil Kirankumar, Sharanabasava D. Hiremath, Bharath Ramsundar
Abstract: Large Language Models (LLMs) have shown remarkable potential in scientific domains like retrosynthesis; yet, they often lack the fine-grained control necessary to navigate complex problem spaces without error. A critical challenge is directing an LLM to avoid specific, chemically sensitive sites on a molecule - a task where unconstrained generation can lead to invalid or undesirable synthetic pathways. In this work, we introduce Protect$^*$, a neuro-symbolic framework that grounds the generative capabilities of Large Language Models (LLMs) in rigorous chemical logic. Our approach combines automated rule-based reasoning - using a comprehensive database of 55+ SMARTS patterns and 40+ characterized protecting groups - with the generative intuition of neural models. The system operates via a hybrid architecture: an ``automatic mode'' where symbolic logic deterministically identifies and guards reactive sites, and a ``human-in-the-loop mode'' that integrates expert strategic constraints. Through ``active state tracking,'' we inject hard symbolic constraints into the neural inference process via a dedicated protection state linked to canonical atom maps. We demonstrate this neuro-symbolic approach through case studies on complex natural products, including the discovery of a novel synthetic pathway for Erythromycin B, showing that grounding neural generation in symbolic logic enables reliable, expert-level autonomy.
Authors: Benjamin L. Badger
Abstract: The ability of machine learning models to store input information in hidden layer vector embeddings, analogous to the concept of `memory', is widely employed but not well characterized. We find that language model embeddings typically contain relatively little input information regardless of data and compute scale during training. In contrast, embeddings from autoencoders trained for input regeneration are capable of nearly perfect memory formation. The substitution of memory embeddings for token sequences leads to substantial computational efficiencies, motivating the introduction of a parallelizable encoder-decoder memory model architecture. Upon causal training these models contain information-poor embeddings incapable of arbitrary information access, but by combining causal and information retention objective functions they learn to form and decode information-rich memories. Training can be further streamlined by freezing a high fidelity encoder followed by a curriculum training approach where decoders first learn to process memories and then learn to additionally predict next tokens. We introduce the perspective that next token prediction training alone is poorly suited for accurate memory formation as the objective itself is non-invertible, motivating the use of combined objective functions for models where the entire input is not exposed.
Authors: Noriaki Hirose, Catherine Glossop, Dhruv Shah, Sergey Levine
Abstract: Robotic foundation models achieve strong generalization by leveraging internet-scale vision-language representations, but their massive computational cost creates a fundamental bottleneck: high inference latency. In dynamic environments, this latency breaks the control loop, rendering powerful models unsafe for real-time deployment. We propose AsyncVLA, an asynchronous control framework that decouples semantic reasoning from reactive execution. Inspired by hierarchical control, AsyncVLA runs a large foundation model on a remote workstation to provide high-level guidance, while a lightweight, onboard Edge Adapter continuously refines actions at high frequency. To bridge the domain gap between these asynchronous streams, we introduce an end-to-end finetuning protocol and a trajectory re-weighting strategy that prioritizes dynamic interactions. We evaluate our approach on real-world vision-based navigation tasks with communication delays up to 6 seconds. AsyncVLA achieves a 40% higher success rate than state-of-the-art baselines, effectively bridging the gap between the semantic intelligence of large models and the reactivity required for edge robotics.
Authors: Pavel Dvurechensky, Andrea Ebner, Johannes Carl Schnebel, Shimrit Shtern, Mathias Staudigl
Abstract: We are concerned with optimization in a broad sense through the lens of solving variational inequalities (VIs) -- a class of problems that are so general that they cover as particular cases minimization of functions, saddle-point (minimax) problems, Nash equilibrium problems, and many others. The key challenges in our problem formulation are the two-level hierarchical structure and finite-sum representation of the smooth operators in each level. For this setting, we are the first to prove convergence rates and complexity statements for variance-reduced stochastic algorithms approaching the solution of hierarchical VIs in Euclidean and Bregman setups.
Authors: Grant Norman, Conor Rowan, Kurt Maute, Alireza Doostan
Abstract: In this work, we investigate the use of data-driven equation discovery for dynamical systems to model and forecast continuous-time dynamics of unconstrained optimization problems. To avoid expensive evaluations of the objective function and its gradient, we leverage trajectory data on the optimization variables to learn the continuous-time dynamics associated with gradient descent, Newton's method, and ADAM optimization. The discovered gradient flows are then solved as a surrogate for the original optimization problem. To this end, we introduce the Learned Gradient Flow (LGF) optimizer, which is equipped to build surrogate models of variable polynomial order in full- or reduced-dimensional spaces at user-defined intervals in the optimization process. We demonstrate the efficacy of this approach on several standard problems from engineering mechanics and scientific machine learning, including two inverse problems, structural topology optimization, and two forward solves with different discretizations. Our results suggest that the learned gradient flows can significantly expedite convergence by capturing critical features of the optimization trajectory while avoiding expensive evaluations of the objective and its gradient.
Authors: Jintao Zhang, Kai Jiang, Chendong Xiang, Weiqi Feng, Yuezhou Hu, Haocheng Xi, Jianfei Chen, Jun Zhu
Abstract: Many training-free sparse attention methods are effective for accelerating diffusion models. Recently, several works suggest that making sparse attention trainable can further increase sparsity while preserving generation quality. We study three key questions: (1) when do the two common masking rules, i.e., Top-k and Top-p, fail, and how can we avoid these failures? (2) why can trainable sparse attention reach higher sparsity than training-free methods? (3) what are the limitations of fine-tuning sparse attention using the diffusion loss, and how can we address them? Based on this analysis, we propose SpargeAttention2, a trainable sparse attention method that achieves high sparsity without degrading generation quality. SpargeAttention2 includes (i) a hybrid masking rule that combines Top-k and Top-p for more robust masking at high sparsity, (ii) an efficient trainable sparse attention implementation, and (iii) a distillation-inspired fine-tuning objective to better preserve generation quality during fine-tuning using sparse attention. Experiments on video diffusion models show that SpargeAttention2 reaches 95% attention sparsity and a 16.2x attention speedup while maintaining generation quality, consistently outperforming prior sparse attention methods.
Authors: Sin-Han Yang, Cheng-Kuang Wu, Chieh-Yen Lin, Yun-Nung Chen, Hung-yi Lee, Shao-Hua Sun
Abstract: Large language models (LLMs) are widely deployed as general-purpose problem solvers, making accurate confidence estimation critical for reliable use. Prior work on LLM calibration largely focuses on response-level confidence, which estimates the correctness of a single generated output. However, this formulation is misaligned with many practical settings where the central question is how likely a model is to solve a query overall. We show that this mismatch results from the stochastic nature of modern LLM decoding, under which single-response correctness fails to reflect underlying model capability. To address this issue, we introduce capability calibration, which targets the model's expected accuracy on a query. We formally distinguish capability calibration from response calibration and show that the two differ both theoretically and empirically. We establish an empirical evaluation setup and study a range of confidence estimation methods. Our results demonstrate that capability-calibrated confidence improves pass@$k$ prediction and inference budget allocation, establishing a foundation with potential for diverse applications.
Authors: Yunfan Zhang, Kathleen McKeown, Smaranda Muresan
Abstract: Large Language Models (LLMs) with agentic web search capabilities show strong potential for tasks requiring real-time information access and complex fact retrieval, yet evaluating such systems remains challenging. We introduce \bench, a rigorous and regularly updated benchmark designed to assess the agentic web search abilities of LLMs. \bench automatically generates fresh question-answer pairs from recent news articles, ensuring that questions require information beyond an LLM's training data and enabling clear separation between internal knowledge and search capability. The benchmark features intentionally difficult questions requiring multi-hop search queries, page visits, and reasoning, making it well-suited for evaluating agentic search behavior. Our automated data curation and question generation pipeline enables frequent benchmark updates and supports construction of a large-scale training dataset for agentic web search models, addressing the scarcity of such data in the research community. To ensure reliable evaluation, we include a subset of human-verified samples in the test set. We evaluate a broad range of systems using \bench, including commercial and open-weight LLMs as well as LLM-based web search APIs. The leaderboard, datasets, and code are publicly available at livenewsbench.com.
Authors: Kun Gao, Katsumi Inoue, Yongzhi Cao, Hanpin Wang, Feng Yang
Abstract: Rule learning-based models are widely used in highly interpretable scenarios due to their transparent structures. Inductive logic programming (ILP), a form of machine learning, induces rules from facts while maintaining interpretability. Differentiable ILP models enhance this process by leveraging neural networks to improve robustness and scalability. However, most differentiable ILP methods rely on symbolic datasets, facing challenges when learning directly from raw data. Specifically, they struggle with explicit label leakage: The inability to map continuous inputs to symbolic variables without explicit supervision of input feature labels. In this work, we address this issue by integrating a self-supervised differentiable clustering model with a novel differentiable ILP model, enabling rule learning from raw data without explicit label leakage. The learned rules effectively describe raw data through its features. We demonstrate that our method intuitively and precisely learns generalized rules from time series and image data.
Authors: Chenwei Xu, Zhen Ye, Shang Wu, Weijian Li, Zihan Wang, Zhuofan Xia, Lie Lu, Pranav Maneriker, Fan Du, Manling Li, Han Liu
Abstract: We present \revise (\underline{Re}asoning with \underline{Vi}deo \underline{S}parsity), a multi-round agent for video question answering (VQA). Instead of uniformly sampling frames, \revise selects a small set of informative frames, maintains a summary-as-state across rounds, and stops early when confident. It supports proprietary vision-language models (VLMs) in a ``plug-and-play'' setting and enables reinforcement fine-tuning for open-source models. For fine-tuning, we introduce EAGER (Evidence-Adjusted Gain for Efficient Reasoning), an annotation-free reward with three terms: (1) Confidence gain: after new frames are added, we reward the increase in the log-odds gap between the correct option and the strongest alternative; (2) Summary sufficiency: at answer time we re-ask using only the last committed summary and reward success; (3) Correct-and-early stop: answering correctly within a small turn budget is rewarded. Across multiple VQA benchmarks, \revise improves accuracy while reducing frames, rounds, and prompt tokens, demonstrating practical sparse video reasoning.
Authors: Muhammad Baqer Mollah, Honggang Wang, Mohammad Ataul Karim, Hua Fang
Abstract: Millimeter wave (mmWave) communication, utilizing beamforming techniques to address the inherent path loss limitation, is considered as one of the key technologies to support ever increasing high throughput and low latency demands of connected vehicles. However, adopting standard defined beamforming approach in highly dynamic vehicular environments often incurs high beam training overheads and reduction in the available airtime for communications, which is mainly due to exchanging pilot signals and exhaustive beam measurements. To this end, we present a multi-modal sensing and fusion learning framework as a potential alternative solution to reduce such overheads. In this framework, we first extract the representative features from the sensing modalities by modality specific encoders, then, utilize multi-head cross-modal attention to learn dependencies and correlations between different modalities, and subsequently fuse the multimodal features to obtain predicted top-k beams so that the best line-of-sight links can be proactively established. To show the generalizability of the proposed framework, we perform a comprehensive experiment in four different vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) scenarios from real world multimodal and 60 GHz mmWave wireless sensing data. The experiment reveals that the proposed framework (i) achieves up to 96.72% accuracy on predicting top-15 beams correctly, (ii) incurs roughly 0.77 dB average power loss, and (iii) improves the overall latency and beam searching space overheads by 86.81% and 76.56% respectively for top-15 beams compared to standard defined approach.
Authors: Seungwoo Yoo, Juil Koo, Daehyeon Choi, Minhyuk Sung
Abstract: We propose DiffusionRollout, a novel selective rollout planning strategy for autoregressive diffusion models, aimed at mitigating error accumulation in long-horizon predictions of physical systems governed by partial differential equations (PDEs). Building on the recently validated probabilistic approach to PDE solving, we further explore its ability to quantify predictive uncertainty and demonstrate a strong correlation between prediction errors and standard deviations computed over multiple samples-supporting their use as a proxy for the model's predictive confidence. Based on this observation, we introduce a mechanism that adaptively selects step sizes during autoregressive rollouts, improving long-term prediction reliability by reducing the compounding effect of conditioning on inaccurate prior outputs. Extensive evaluation on long-trajectory PDE prediction benchmarks validates the effectiveness of the proposed uncertainty measure and adaptive planning strategy, as evidenced by lower prediction errors and longer predicted trajectories that retain a high correlation with their ground truths.
Authors: Anuj Kumar Yadav, Cemre Cadir, Yanina Shkel, Michael Gastpar
Abstract: We study parametric change-point detection, where the goal is to identify distributional changes in time series, under local differential privacy. In the non-private setting, we derive improved finite-sample accuracy guarantees for a change-point detection algorithm based on the generalized log-likelihood ratio test, via martingale methods. In the private setting, we propose two locally differentially private algorithms based on randomized response and binary mechanisms, and analyze their theoretical performance. We derive bounds on detection accuracy and validate our results through empirical evaluation. Our results characterize the statistical cost of local differential privacy in change-point detection and show how privacy degrades performance relative to a non-private benchmark. As part of this analysis, we establish a structural result for strong data processing inequalities (SDPI), proving that SDPI coefficients for R\'enyi divergences and their symmetric variants (Jeffreys-R\'enyi divergences) are achieved by binary input distributions. These results on SDPI coefficients are also of independent interest, with applications to statistical estimation, data compression, and Markov chain mixing.
Authors: Md. Kamrul Hossain, Walid Aljoby
Abstract: Intent-Based Networking (IBN) simplifies network management, but its reliability is challenged by "intent drift", where the network's state gradually deviates from its intended goal, often leading to silent failures. Conventional approaches struggle to detect the subtle, early stages of intent drift, raising alarms only when degradation is significant and failure is imminent, which limits their effectiveness for proactive assurance. To address this, we propose LEAD-Drift, a framework that detects intent drift in real time to enable proactive failure prevention. LEAD-Drift's core contribution is reformulating intent failure detection as a supervised learning problem by training a lightweight neural network on fixed-horizon labels to predict a future risk score. The model's raw output is then smoothed with an Exponential Moving Average (EMA) and passed through a statistically tuned threshold to generate robust, real-time alerts. Furthermore, we enhance the framework with two key features for operational intelligence: a multi-horizon modeling technique for dynamic time-to-failure estimation, and per-alert explainability using SHAP to identify root-cause KPIs. Our evaluation on a time-series dataset shows LEAD-Drift provides significantly earlier warnings, improving the average lead time by 7.3 minutes (+17.8\%) compared to a distance-based baseline. It also reduces alert noise by 80.2\% compared to a weighted-KPI heuristic, with only a minor trade-off in lead time. These results demonstrate that LEAD-Drift as a highly effective, interpretable, and operationally efficient solution for proactive network assurance in IBN.
Authors: Linjie Xu, Yanlin Zhang, Quan Gan, Minjie Wang, David Wipf
Abstract: Relational databases (RDBs) contain vast amounts of heterogeneous tabular information that can be exploited for predictive modeling purposes. But since the space of potential targets is vast across enterprise settings, how can we \textit{avoid retraining} a new model each time we wish to predict a new quantity of interest? Foundation models based on in-context learning (ICL) offer a convenient option, but so far are largely restricted to single-table operability. In generalizing to multiple interrelated tables, it is essential to compress variably-sized RDB neighborhoods into fixed-length ICL samples for consumption by the decoder. However, the details here are critical: unlike existing supervised learning RDB pipelines, we provide theoretical and empirical evidence that ICL-specific compression should be constrained \emph{within} high-dimensional RDB columns where all entities share units and roles, not \textit{across} columns where the relevance of heterogeneous data types cannot possibly be determined without label information. Conditioned on this restriction, we then demonstrate that encoder expressiveness is actually not compromised by excluding trainable parameters. Hence we arrive at a principled family of RDB encoders that can be seamlessly paired with already-existing single-table ICL foundation models, whereby no training or fine-tuning is required. From a practical standpoint, we develop scalable SQL primitives to implement the encoder stage, resulting in an easy-to-use open-source RDB foundation model\footnote{\label{foot: RDBLearn_learn} https://github.com/HKUSHXLab/rdblearn} capable of robust performance on unseen datasets out of the box.
Authors: Chan Hao Sien, Hezerul Abdul Karim, Nouar AlDahoul
Abstract: Soil-transmitted helminth (STH) infections continuously affect a large proportion of the global population, particularly in tropical and sub-tropical regions, where access to specialized diagnostic expertise is limited. Although manual microscopic diagnosis of parasitic eggs remains the diagnostic gold standard, the approach can be labour-intensive, time-consuming, and prone to human error. This paper aims to utilize a vision language model (VLM) such as Microsoft Florence that was fine-tuned to localize all parasitic eggs within microscopic images. The preliminary results show that our localization VLM performs comparatively better than the other object detection methods, such as EfficientDet, with an mIOU of 0.94. This finding demonstrates the potential of the proposed VLM to serve as a core component of an automated framework, offering a scalable engineering solution for intelligent parasitological diagnosis.
Authors: Yasaman Torabi, Parsa Razmara, Hamed Ajorlou, Bardia Baraeinejad
Abstract: Large Language Models (LLMs) have demonstrated strong semantic reasoning across multimodal domains. However, their integration with graph-based models of brain connectivity remains limited. In addition, most existing fMRI analysis methods rely on static Functional Connectivity (FC) representations, which obscure transient neural dynamics critical for neurodevelopmental disorders such as autism. Recent state-space approaches, including Mamba, model temporal structure efficiently, but are typically used as standalone feature extractors without explicit high-level reasoning. We propose NeuroMambaLLM, an end-to-end framework that integrates dynamic latent graph learning and selective state-space temporal modelling with LLMs. The proposed method learns the functional connectivity dynamically from raw Blood-Oxygen-Level-Dependent (BOLD) time series, replacing fixed correlation graphs with adaptive latent connectivity while suppressing motion-related artifacts and capturing long-range temporal dependencies. The resulting dynamic brain representations are projected into the embedding space of an LLM model, where the base language model remains frozen and lightweight low-rank adaptation (LoRA) modules are trained for parameter-efficient alignment. This design enables the LLM to perform both diagnostic classification and language-based reasoning, allowing it to analyze dynamic fMRI patterns and generate clinically meaningful textual reports.
Authors: Vashista Nobaub
Abstract: Large language models spend most of their inference cost on attention over long contexts, yet empirical behavior suggests that only a small subset of tokens meaningfully contributes to each query. We formalize this phenomenon by modeling attention as a projection onto the convex hull of key vectors and analyzing its entropic (softmax-like) relaxation. Our main theoretical contribution is a face-stability theorem showing that, under a strict complementarity margin (a support gap (\Delta) certified by KKT multipliers), entropic attention concentrates on a constant-size active face: the total mass assigned to inactive tokens decays exponentially as (\exp(-\Omega(\Delta/\varepsilon))), while the error on the active face scales linearly in the temperature/regularization parameter (\varepsilon). This yields a practical criterion for when sparse long-context decoding is safe and provides a principled knob to trade accuracy for compute. Building on these guarantees, we introduce Vashista Sparse Attention, a drop-in mechanism that maintains a small candidate set per query through a paging-style context selection strategy compatible with modern inference stacks. Across long-context evaluations, we observe stable constant-size effective support, strong wall-clock speedups, and minimal quality degradation in the regimes predicted by the support-gap diagnostics. Finally, we discuss deployment implications for privacy-sensitive and air-gapped settings, where interchangeable attention modules enable predictable latency and cost without external retrieval dependencies.
Authors: Suhas Suresh Bharadwaj, Reuben Thomas Thovelil
Abstract: Physics-Informed Neural Networks present a novel approach in SciML that integrates physical laws in the form of partial differential equations directly into the NN through soft constraints in the loss function. This work studies the application of PINNs to solve a one dimensional coupled electro-elastodynamic system modeling linear piezoelectricity in stress-charge form, governed by elastodynamic and electrodynamic equations. Our simulation employs a feedforward architecture, mapping space-time coordinates to mechanical displacement and electric potential. Our PINN model achieved global relative L2 errors of 2.34 and 4.87 percent for displacement and electric potential respectively. The results validate PINNs as effective mesh free solvers for coupled time-dependent PDE systems, though challenges remain regarding error accumulation and stiffness in coupled eigenvalue systems.
Authors: Zongcheng Han, Dongyan Cao, Haoran Sun, Yu Hong
Abstract: Recent advances in auto-regressive transformers have achieved remarkable success in generative modeling. However, text-to-3D generation remains challenging, primarily due to bottlenecks in learning discrete 3D representations. Specifically, existing approaches often suffer from information loss during encoding, causing representational distortion before the quantization process. This effect is further amplified by vector quantization, ultimately degrading the geometric coherence of text-conditioned 3D shapes. Moreover, the conventional two-stage training paradigm induces an objective mismatch between reconstruction and text-conditioned auto-regressive generation. To address these issues, we propose View-aware Auto-Regressive 3D (VAR-3D), which intergrates a view-aware 3D Vector Quantized-Variational AutoEncoder (VQ-VAE) to convert the complex geometric structure of 3D models into discrete tokens. Additionally, we introduce a rendering-supervised training strategy that couples discrete token prediction with visual reconstruction, encouraging the generative process to better preserve visual fidelity and structural consistency relative to the input text. Experiments demonstrate that VAR-3D significantly outperforms existing methods in both generation quality and text-3D alignment.
Authors: Fabrizio Falasca, Laure Zanna
Abstract: We introduce a flexible framework based on response theory and score matching to suppress spurious, noncausal dependencies in reduced-order neural emulators of turbulent systems, focusing on climate dynamics as a proof-of-concept. We showcase the approach using the stochastic Charney-DeVore model as a relevant prototype for low-frequency atmospheric variability. We show that the resulting causal constraints enhance neural emulators' ability to respond to both weak and strong external forcings, despite being trained exclusively on unforced data. The approach is broadly applicable to modeling complex turbulent dynamical systems in reduced spaces and can be readily integrated into general neural network architectures.
Authors: Naeem Shahabi Sani, Ferial Najiantabriz, Shayan Shafaei, Dean F. Hougen
Abstract: Learning in the presence of missing data can result in biased predictions and poor generalizability, among other difficulties, which data imputation methods only partially address. In neural networks, activation functions significantly affect performance yet typical options (e.g., ReLU, Swish) operate only on feature values and do not account for missingness indicators or confidence scores. We propose Three-Channel Evolved Activations (3C-EA), which we evolve using Genetic Programming to produce multivariate activation functions f(x, m, c) in the form of trees that take (i) the feature value x, (ii) a missingness indicator m, and (iii) an imputation confidence score c. To make these activations useful beyond the input layer, we introduce ChannelProp, an algorithm that deterministically propagates missingness and confidence values via linear layers based on weight magnitudes, retaining reliability signals throughout the network. We evaluate 3C-EA and ChannelProp on datasets with natural and injected (MCAR/MAR/MNAR) missingness at multiple rates under identical preprocessing and splits. Results indicate that integrating missingness and confidence inputs into the activation search improves classification performance under missingness.
Authors: Gabriel Romio, Mateus Begnini Melchiades, Bruno Castro da Silva, Gabriel de Oliveira Ramos
Abstract: Hierarchical Reinforcement Learning (HRL) frameworks like Option-Critic (OC) and Multi-updates Option Critic (MOC) have introduced significant advancements in learning reusable options. However, these methods underperform in multi-goal environments with sparse rewards, where actions must be linked to temporally distant outcomes. To address this limitation, we first propose MOC-HER, which integrates the Hindsight Experience Replay (HER) mechanism into the MOC framework. By relabeling goals from achieved outcomes, MOC-HER can solve sparse reward environments that are intractable for the original MOC. However, this approach is insufficient for object manipulation tasks, where the reward depends on the object reaching the goal rather than on the agent's direct interaction. This makes it extremely difficult for HRL agents to discover how to interact with these objects. To overcome this issue, we introduce Dual Objectives Hindsight Experience Replay (2HER), a novel extension that creates two sets of virtual goals. In addition to relabeling goals based on the object's final state (standard HER), 2HER also generates goals from the agent's effector positions, rewarding the agent for both interacting with the object and completing the task. Experimental results in robotic manipulation environments show that MOC-2HER achieves success rates of up to 90%, compared to less than 11% for both MOC and MOC-HER. These results highlight the effectiveness of our dual objective relabeling strategy in sparse reward, multi-goal tasks.
Authors: Sai Ravela, Jae Deok Kim, Kenneth Gee, Xingjian Yan, Samson Mercier, Lubna Albarghouty, Anamitra Saha
Abstract: We formulate Ensemble-Conditional Gaussian Processes (Ens-CGP), a finite-dimensional synthesis that centers ensemble-based inference on the conditional Gaussian law. Conditional Gaussian processes (CGP) arise directly from Gaussian processes under conditioning and, in linear-Gaussian settings, define the full posterior distribution for a Gaussian prior and linear observations. Classical Kalman filtering is a recursive algorithm that computes this same conditional law under dynamical assumptions; the conditional Gaussian law itself is therefore the underlying representational object, while the filter is one computational realization. In this sense, CGP provides the probabilistic foundation for Kalman-type methods as well as equivalent formulations as a strictly convex quadratic program (MAP estimation), RKHS-regularized regression, and classical regularization. Ens-CGP is the ensemble instantiation of this object, obtained by treating empirical ensemble moments as a (possibly low-rank) Gaussian prior and performing exact conditioning. By separating representation (GP -> CGP -> Ens-CGP) from computation (Kalman filters, EnKF variants, and iterative ensemble schemes), the framework links an earlier-established representational foundation for inference to ensemble-derived priors and clarifies the relationships among probabilistic, variational, and ensemble perspectives.
Authors: Harris Abdul Majid, Giannis Daras, Francesco Tudisco, Steven McDonagh
Abstract: In many scientific settings, acquiring complete observations of PDE coefficients and solutions can be expensive, hazardous, or impossible. Recent diffusion-based methods can reconstruct fields given partial observations, but require complete observations for training. We introduce Ambient Physics, a framework for learning the joint distribution of coefficient-solution pairs directly from partial observations, without requiring a single complete observation. The key idea is to randomly mask a subset of already-observed measurements and supervise on them, so the model cannot distinguish "truly unobserved" from "artificially unobserved", and must produce plausible predictions everywhere. Ambient Physics achieves state-of-the-art reconstruction performance. Compared with prior diffusion-based methods, it achieves a 62.51$\%$ reduction in average overall error while using 125$\times$ fewer function evaluations. We also identify a "one-point transition": masking a single already-observed point enables learning from partial observations across architectures and measurement patterns. Ambient Physics thus enables scientific progress in settings where complete observations are unavailable.
Authors: Daniel Chen, Zaria Zinn, Marcus Lowe
Abstract: We present a font classification system capable of identifying 394 font families from rendered text images. Our approach fine-tunes a DINOv2 Vision Transformer using Low-Rank Adaptation (LoRA), achieving approximately 86% top-1 accuracy while training fewer than 1% of the model's 87.2M parameters. We introduce a synthetic dataset generation pipeline that renders Google Fonts at scale with diverse augmentations including randomized colors, alignment, line wrapping, and Gaussian noise, producing training images that generalize to real-world typographic samples. The model incorporates built-in preprocessing to ensure consistency between training and inference, and is deployed as a HuggingFace Inference Endpoint. We release the model, dataset, and full training pipeline as open-source resources.
Authors: Zhanyu Tuo
Abstract: In this paper, we propose RPGD (RANSAC-P3P Gradient Descent), a human-pose-driven extrinsic calibration framework that robustly aligns MoCap-based 3D skeletal data with monocular or multi-view RGB cameras using only natural human motion. RPGD formulates extrinsic calibration as a coarse-to-fine problem tailored to human poses, combining the global robustness of RANSAC-P3P with Gradient-Descent-based refinement. We evaluate RPGD on three large-scale public 3D HPE datasets as well as on a self-collected in-the-wild dataset. Experimental results demonstrate that RPGD consistently recovers extrinsic parameters with accuracy comparable to the provided ground truth, achieving sub-pixel MPJPE reprojection error even in challenging, noisy settings. These results indicate that RPGD provides a practical and automatic solution for reliable extrinsic calibration of large-scale 3D HPE dataset collection.
Authors: Shaan Ul Haque, Zedong Wang, Zixuan Zhang, Siva Theja Maguluri
Abstract: Stochastic approximation (SA) is a method for finding the root of an operator perturbed by noise. There is a rich literature establishing the asymptotic normality of rescaled SA iterates under fairly mild conditions. However, these asymptotic results do not quantify the accuracy of the Gaussian approximation in finite time. In this paper, we establish explicit non-asymptotic bounds on the Wasserstein distance between the distribution of the rescaled iterate at time k and the asymptotic Gaussian limit for various choices of step-sizes including constant and polynomially decaying. As an immediate consequence, we obtain tail bounds on the error of SA iterates at any time. We obtain the sharp rates by first studying the convergence rate of the discrete Ornstein-Uhlenbeck (O-U) process driven by general noise, whose stationary distribution is identical to the limiting Gaussian distribution of the rescaled SA iterates. We believe that this is of independent interest, given its connection to sampling literature. The analysis involves adapting Stein's method for Gaussian approximation to handle the matrix weighted sum of i.i.d. random variables. The desired finite-time bounds for SA are obtained by characterizing the error dynamics between the rescaled SA iterate and the discrete time O-U process and combining it with the convergence rate of the latter process.
Authors: Aju Ani Justus, Ruchit Agrawal, Sudarsana Reddy Kadiri, Shrikanth Narayanan
Abstract: We present voice2mode, a method for classification of four singing phonation modes (breathy, neutral (modal), flow, and pressed) using embeddings extracted from large self-supervised speech models. Prior work on singing phonation has relied on handcrafted signal features or task-specific neural nets; this work evaluates the transferability of speech foundation models to singing phonation classification. voice2mode extracts layer-wise representations from HuBERT and two wav2vec2 variants, applies global temporal pooling, and classifies the pooled embeddings with lightweight classifiers (SVM, XGBoost). Experiments on a publicly available soprano dataset (763 sustained vowel recordings, four labels) show that foundation-model features substantially outperform conventional spectral baselines (spectrogram, mel-spectrogram, MFCC). HuBERT embeddings obtained from early layers yield the best result (~95.7% accuracy with SVM), an absolute improvement of ~12-15% over the best traditional baseline. We also show layer-wise behaviour: lower layers, which retain acoustic/phonetic detail, are more effective than top layers specialized for Automatic Speech Recognition (ASR).
Authors: Ruggiero Santeramo, Igor Zubarev, Florian Jug
Abstract: Breast cancer screening programmes increasingly seek to move from one-size-fits-all interval to risk-adapted and personalized strategies. Deep learning (DL) has enabled image-based risk models with stronger 1- to 5-year prediction than traditional clinical models, but leading systems (e.g., Mirai) typically use convolutional backbones, very high-resolution inputs (>1M pixels) and simple multi-view fusion, with limited explicit modelling of contralateral asymmetry. We hypothesised that combining complementary inductive biases (convolutional and transformer-based) with explicit contralateral asymmetry modelling would allow us to match state-of-the-art 3-year risk prediction performance even when operating on substantially lower-resolution mammograms, indicating that using less detailed images in a more structured way can recover state-of-the-art accuracy. We present MamaDino, a mammography-aware multi-view attentional DINO model. MamaDino fuses frozen self-supervised DINOv3 ViT-S features with a trainable CNN encoder at 512x512 resolution, and aggregates bilateral breast information via a BilateralMixer to output a 3-year breast cancer risk score. We train on 53,883 women from OPTIMAM (UK) and evaluate on matched 3-year case-control cohorts: an in-distribution test set from four screening sites and an external out-of-distribution cohort from an unseen site. At breast-level, MamaDino matches Mirai on both internal and external tests while using ~13x fewer input pixels. Adding the BilateralMixer improves discrimination to AUC 0.736 (vs 0.713) in-distribution and 0.677 (vs 0.666) out-of-distribution, with consistent performance across age, ethnicity, scanner, tumour type and grade. These findings demonstrate that explicit contralateral modelling and complementary inductive biases enable predictions that match Mirai, despite operating on substantially lower-resolution mammograms.
Authors: Yangxinyu Xie, Tao Wang, Soham Mallick, Yan Sun, Georgy Noarov, Mengxin Yu, Tanwi Mallick, Weijie J. Su, Edgar Dobriban
Abstract: While LLMs have seen substantial improvement in reasoning capabilities, they also sometimes overthink, generating unnecessary reasoning steps, particularly under uncertainty, given ill-posed or ambiguous queries. We introduce statistically principled early stopping methods that monitor uncertainty signals during generation to mitigate this issue. Our first approach is parametric: it models inter-arrival times of uncertainty keywords as a renewal process and applies sequential testing for stopping. Our second approach is nonparametric and provides finite-sample guarantees on the probability of halting too early on well-posed queries. We conduct empirical evaluations on reasoning tasks across several domains and models. Our results indicate that uncertainty-aware early stopping can improve both efficiency and reliability in LLM reasoning, and we observe especially significant gains for math reasoning.
Authors: Zexuan Sun, Garvesh Raskutti
Abstract: In the era of large language models (LLMs), fine-tuning pretrained models has become ubiquitous. Yet the theoretical underpinning remains an open question. A central question is why only a few epochs of fine-tuning are typically sufficient to achieve strong performance on many different tasks. In this work, we approach this question by developing a statistical framework, combining rigorous early stopping theory with the attention-based Neural Tangent Kernel (NTK) for LLMs, offering new theoretical insights on fine-tuning practices. Specifically, we formally extend classical NTK theory [Jacot et al., 2018] to non-random (i.e., pretrained) initializations and provide a convergence guarantee for attention-based fine-tuning. One key insight provided by the theory is that the convergence rate with respect to sample size is closely linked to the eigenvalue decay rate of the empirical kernel matrix induced by the NTK. We also demonstrate how the framework can be used to explain task vectors for multiple tasks in LLMs. Finally, experiments with modern language models on real-world datasets provide empirical evidence supporting our theoretical insights.
Authors: Guojie Liu, Yiqi Wang, Yanfeng Yang, Wenqi Fan, Songlei Jian, Jianfeng Zhang, Jie Yu
Abstract: Providing extensive context via prompting is vital for leveraging the capabilities of Large Language Models (LLMs). However, lengthy contexts significantly increase inference latency, as the computational cost of self-attention grows quadratically with sequence length. To mitigate this issue, context compression-particularly soft prompt compressio-has emerged as a widely studied solution, which converts long contexts into shorter memory embeddings via a trained compressor. Existing methods typically compress the entire context indiscriminately into a set of memory tokens, requiring the compressor to capture global dependencies and necessitating extensive pre-training data to learn effective patterns. Inspired by the chunking mechanism in human working memory and empirical observations of the spatial specialization of memory embeddings relative to original tokens, we propose Parallelized Iterative Compression (PIC). By simply modifying the Transformer's attention mask, PIC explicitly restricts the receptive field of memory tokens to sequential local chunks, thereby lowering the difficulty of compressor training. Experiments across multiple downstream tasks demonstrate that PIC consistently outperforms competitive baselines, with superiority being particularly pronounced in high compression scenarios (e.g., achieving relative improvements of 29.8\% in F1 score and 40.7\% in EM score on QA tasks at the $64\times$ compression ratio). Furthermore, PIC significantly expedites the training process. Specifically, when training the 16$\times$ compressor, it surpasses the peak performance of the competitive baseline while effectively reducing the training time by approximately 40\%.
Authors: Srikumar Nayak
Abstract: Named Entity Recognition (NER) has emerged as a critical component in automating financial transaction processing, particularly in extracting structured information from unstructured payment data. This paper presents a comprehensive analysis of state-of-the-art NER algorithms specifically designed for payment data extraction, including Conditional Random Fields (CRF), Bidirectional Long Short-Term Memory with CRF (BiLSTM-CRF), and transformer-based models such as BERT and FinBERT. We conduct extensive experiments on a dataset of 50,000 annotated payment transactions across multiple payment formats including SWIFT MT103, ISO 20022, and domestic payment systems. Our experimental results demonstrate that fine-tuned BERT models achieve an F1-score of 94.2% for entity extraction, outperforming traditional CRF-based approaches by 12.8 percentage points. Furthermore, we introduce PaymentBERT, a novel hybrid architecture combining domain-specific financial embeddings with contextual representations, achieving state-of-the-art performance with 95.7% F1-score while maintaining real-time processing capabilities. We provide detailed analysis of cross-format generalization, ablation studies, and deployment considerations. This research provides practical insights for financial institutions implementing automated sanctions screening, anti-money laundering (AML) compliance, and payment processing systems.
Authors: Even He, Zaizai Yan
Abstract: We study operator-norm covariance estimation from heavy-tailed samples that may include a small fraction of arbitrary outliers. A simple and widely used safeguard is \emph{Euclidean norm clipping}, but its accuracy depends critically on an unknown clipping level. We propose a cross-fitted clipped covariance estimator equipped with \emph{fully computable} Bernstein-type deviation certificates, enabling principled data-driven tuning via a selector (\emph{MinUpper}) that balances certified stochastic error and a robust hold-out proxy for clipping bias. The resulting procedure adapts to intrinsic complexity measures such as effective rank under mild tail regularity and retains meaningful guarantees under only finite fourth moments. Experiments on contaminated spiked-covariance benchmarks illustrate stable performance and competitive accuracy across regimes.
Authors: Mingqi Wu, Archer Y. Yang, Qiang Sun
Abstract: Iterative self-training (self-distillation) repeatedly refits a model on pseudo-labels generated by its own predictions. We study this procedure in overparameterized linear regression: an initial estimator is trained on noisy labels, and each subsequent iterate is trained on fresh covariates with noiseless pseudo-labels from the previous model. In the high-dimensional regime, we derive deterministic-equivalent recursions for the prediction risk and effective noise across iterations, and prove that the empirical quantities concentrate sharply around these limits. The recursion separates two competing forces: a systematic component that grows with iteration due to progressive signal forgetting, and a stochastic component that decays due to denoising via repeated data-dependent projections. Their interaction yields a $U$-shaped test-risk curve and an optimal early-stopping time. In spiked covariance models, iteration further acts as an iteration-dependent spectral filter that preserves strong eigendirections while suppressing weaker ones, inducing an implicit form of soft feature selection distinct from ridge regression. Finally, we propose an iterated generalized cross-validation criterion and prove its uniform consistency for estimating the risk along the self-training trajectory, enabling fully data-driven selection of the stopping time and regularization. Experiments on synthetic covariances validate the theory and illustrate the predicted denoising-forgetting trade-off.
Authors: Xuehao Cui, Ruibo Chen, Yihan Wu, Heng Huang
Abstract: Large language models now produce text indistinguishable from human writing, which increases the need for reliable provenance tracing. Multi-bit watermarking can embed identifiers into generated text, but existing methods struggle to keep both text quality and watermark strength while carrying long messages. We propose MC$^2$Mark, a distortion-free multi-bit watermarking framework designed for reliable embedding and decoding of long messages. Our key technical idea is Multi-Channel Colored Reweighting, which encodes bits through structured token reweighting while keeping the token distribution unbiased, together with Multi-Layer Sequential Reweighting to strengthen the watermark signal and an evidence-accumulation detector for message recovery. Experiments show that MC$^2$Mark improves detectability and robustness over prior multi-bit watermarking methods while preserving generation quality, achieving near-perfect accuracy for short messages and exceeding the second-best method by nearly 30% for long messages.
Authors: Mingfei Lu, Mengjia Wu, Feng Liu, Jiawei Xu, Weikai Li, Haoyang Wang, Zhengdong Hu, Ying Ding, Yizhou Sun, Jie Lu, Yi Zhang
Abstract: Memory is critical for enabling large language model (LLM) based agents to maintain coherent behavior over long-horizon interactions. However, existing agent memory systems suffer from two key gaps: they rely on a one-size-fits-all memory structure and do not model memory structure selection as a context-adaptive decision, limiting their ability to handle heterogeneous interaction patterns and resulting in suboptimal performance. We propose a unified framework, FluxMem, that enables adaptive memory organization for LLM agents. Our framework equips agents with multiple complementary memory structures. It explicitly learns to select among these structures based on interaction-level features, using offline supervision derived from downstream response quality and memory utilization. To support robust long-horizon memory evolution, we further introduce a three-level memory hierarchy and a Beta Mixture Model-based probabilistic gate for distribution-aware memory fusion, replacing brittle similarity thresholds. Experiments on two long-horizon benchmarks, PERSONAMEM and LoCoMo, demonstrate that our method achieves average improvements of 9.18% and 6.14%.
Authors: Sajjad Kachuee, Mohammad Sharifkhani
Abstract: Mixture-of-Experts (MoE) embedding models combine expert outputs using weighted linear summation, implicitly assuming a linear subspace structure in the embedding space. This assumption is shown to be inconsistent with the geometry of expert representations. Geometric analysis of a modern MoE embedding model reveals that expert outputs lie on a shared hyperspherical manifold characterized by tightly concentrated norms and substantial angular separation. Under this geometry, linear aggregation induces inward collapse toward the manifold interior, distorting vector magnitude and direction and reducing embedding comparability. To address this inconsistency, Spherical Barycentric Aggregation (SBA) is introduced as a geometry-preserving aggregation operator that separates radial and angular components to maintain hyperspherical structure while remaining fully compatible with existing routing mechanisms. Experiments on selected tasks from the Massive Text Embedding Benchmark (MTEB), including semantic similarity, clustering, and duplicate question detection, demonstrate consistent performance improvements with identical training cost and full stability. Additional geometric analyses confirm that SBA prevents aggregation-induced collapse and preserves hyperspherical consistency, highlighting the importance of geometry-aware aggregation in MoE embedding architectures.
Authors: Minghan Wang, Ye Bai, Thuy-Trang Vu, Ehsan Shareghi, Gholamreza Haffari
Abstract: Inference-time scaling (ITS) in latent reasoning models typically introduces stochasticity through heuristic perturbations, such as dropout or fixed Gaussian noise. While these methods increase trajectory diversity, their exploration behavior is not explicitly modeled and can be inefficient under finite sampling budgets. We observe that stronger perturbations do not necessarily translate into more effective candidate trajectories, as unguided noise may disrupt internal decision structure rather than steer it. To provide a more structured alternative, we model latent thought exploration as conditional sampling from learnable densities and instantiate this idea as a Gaussian Thought Sampler (GTS). GTS predicts context-dependent perturbation distributions over continuous reasoning states and is trained with GRPO-style policy optimization while keeping the backbone frozen. Experiments on GSM8K with two latent reasoning architectures show that GTS achieves more reliable inference-time scaling than heuristic baselines. These findings indicate that improving latent ITS requires structured and optimizable exploration mechanisms rather than simply amplifying stochasticity.
Authors: Yuan Cao, Dezhi Ran, Mengzhou Wu, Yuzhe Guo, Xin Chen, Ang Li, Gang Cao, Gong Zhi, Hao Yu, Linyi Li, Wei Yang, Tao Xie
Abstract: Post-training GUI agents in interactive environments is critical for developing generalization and long-horizon planning capabilities. However, training on real-world applications is hindered by high latency, poor reproducibility, and unverifiable rewards relying on noisy visual proxies. To address the limitations, we present GUI-GENESIS, the first framework to automatically synthesize efficient GUI training environments with verifiable rewards. GUI-GENESIS reconstructs real-world applications into lightweight web environments using multimodal code models and equips them with code-native rewards, executable assertions that provide deterministic reward signals and eliminate visual estimation noise. Extensive experiments show that GUI-GENESIS reduces environment latency by 10 times and costs by over $28,000 per epoch compared to training on real applications. Notably, agents trained with GUI-GENESIS outperform the base model by 14.54% and even real-world RL baselines by 3.27% on held-out real-world tasks. Finally, we observe that models can synthesize environments they cannot yet solve, highlighting a pathway for self-improving agents.
Authors: Kazuo Yano, Jonghyeok Lee, Tae Ishitomi, Hironobu Kawaguchi, Akira Koyama, Masakuni Ota, Yuki Ota, Nobuo Sato, Keita Shimada, Sho Takematsu, Ayaka Tobinai, Satomi Tsuji, Kazunori Yanagi, Keiko Yano, Manabu Harada, Yuki Matsuda, Kazunori Matsumoto, Kenichi Matsumura, Hamae Matsuo, Yumi Miyazaki, Kotaro Murai, Tatsuya Ohshita, Marie Seki, Shun Tanoue, Tatsuki Terakado, Yuko Ichimaru, Mirei Saito, Akihiro Otsuka, Koji Ara
Abstract: Large language models (LLMs) have achieved remarkable success in generating fluent and contextually appropriate text; however, their capacity to produce genuinely creative outputs remains limited. This paper posits that this limitation arises from a structural property of contemporary LLMs: when provided with rich context, the space of future generations becomes strongly constrained, and the generation process is effectively governed by near-deterministic dynamics. Recent approaches such as test-time scaling and context adaptation improve performance but do not fundamentally alter this constraint. To address this issue, we propose Algebraic Quantum Intelligence (AQI) as a computational framework that enables systematic expansion of semantic space. AQI is formulated as a noncommutative algebraic structure inspired by quantum theory, allowing properties such as order dependence, interference, and uncertainty to be implemented in a controlled and designable manner. Semantic states are represented as vectors in a Hilbert space, and their evolution is governed by C-values computed from noncommutative operators, thereby ensuring the coexistence and expansion of multiple future semantic possibilities. In this study, we implement AQI by extending a transformer-based LLM with more than 600 specialized operators. We evaluate the resulting system on creative reasoning benchmarks spanning ten domains under an LLM-as-a-judge protocol. The results show that AQI consistently outperforms strong baseline models, yielding statistically significant improvements and reduced cross-domain variance. These findings demonstrate that noncommutative algebraic dynamics can serve as a practical and reproducible foundation for machine creativity. Notably, this architecture has already been deployed in real-world enterprise environments.
Authors: Yi Li, Hongze Shen, Lexiang Tang, Xin Li, Xinpeng Ding, Yinsong Liu, Deqiang Jiang, Xing Sun, Xiaomeng Li
Abstract: Multimodal Large Language Models (MLLMs) have demonstrated exceptional capabilities in high-level visual understanding. However, extending these models to fine-grained dense prediction tasks, such as semantic segmentation and depth estimation, typically necessitates the incorporation of complex, task-specific decoders and other customizations. This architectural fragmentation increases model complexity and deviates from the generalist design of MLLMs, ultimately limiting their practicality. In this work, we challenge this paradigm by accommodating standard MLLMs to perform dense predictions without requiring additional task-specific decoders. The proposed model is called DenseMLLM, grounded in the standard architecture with a novel vision token supervision strategy for multiple labels and tasks. Despite its minimalist design, our model achieves highly competitive performance across a wide range of dense prediction and vision-language benchmarks, demonstrating that a standard, general-purpose MLLM can effectively support dense perception without architectural specialization.
Authors: Ahmed Ghorbel, Badr Moufad, Navid Bagheri Shouraki, Alain Oliviero Durmus, Thomas Hirtz, Eric Moulines, Jimmy Olsson, Yazid Janati
Abstract: Text-driven image and video editing can be naturally cast as inpainting problems, where masked regions are reconstructed to remain consistent with both the observed content and the editing prompt. Recent advances in test-time guidance for diffusion and flow models provide a principled framework for this task; however, existing methods rely on costly vector--Jacobian product (VJP) computations to approximate the intractable guidance term, limiting their practical applicability. Building upon the recent work of Moufad et al. (2025), we provide theoretical insights into their VJP-free approximation and substantially extend their empirical evaluation to large-scale image and video editing benchmarks. Our results demonstrate that test-time guidance alone can achieve performance comparable to, and in some cases surpass, training-based methods.
Authors: Keinichi Fujita, Yusuke Ijima
Abstract: Paralinguistic and non-linguistic aspects of speech strongly influence listener impressions. While most research focuses on absolute impression scoring, this study investigates relative voice impression estimation (RIE), a framework for predicting the perceptual difference between two utterances from the same speaker. The estimation target is a low-dimensional vector derived from subjective evaluations, quantifying the perceptual shift of the second utterance relative to the first along an antonymic axis (e.g., ``Dark--Bright''). To isolate expressive and prosodic variation, we used recordings of a professional speaker reading a text in various styles. We compare three modeling approaches: classical acoustic features commonly used for speech emotion recognition, self-supervised speech representations, and multimodal large language models (MLLMs). Our results demonstrate that models using self-supervised representations outperform methods with classical acoustic features, particularly in capturing complex and dynamic impressions (e.g., ``Cold--Warm'') where classical features fail. In contrast, current MLLMs prove unreliable for this fine-grained pairwise task. This study provides the first systematic investigation of RIE and demonstrates the strength of self-supervised speech models in capturing subtle perceptual variations.
Authors: Yue Chen, Muqing Jiang, Kaifeng Zheng, Jiaqi Liang, Chenrui Tie, Haoran Lu, Ruihai Wu, Hao Dong
Abstract: Articulated object manipulation is essential for various real-world robotic tasks, yet generalizing across diverse objects remains a major challenge. A key to generalization lies in understanding functional parts (e.g., door handles and knobs), which indicate where and how to manipulate across diverse object categories and shapes. Previous works attempted to achieve generalization by introducing foundation features, while these features are mostly 2D-based and do not specifically consider functional parts. When lifting these 2D features to geometry-profound 3D space, challenges arise, such as long runtimes, multi-view inconsistencies, and low spatial resolution with insufficient geometric information. To address these issues, we propose Part-Aware 3D Feature Field (PA3FF), a novel dense 3D feature with part awareness for generalizable articulated object manipulation. PA3FF is trained by 3D part proposals from a large-scale labeled dataset, via a contrastive learning formulation. Given point clouds as input, PA3FF predicts a continuous 3D feature field in a feedforward manner, where the distance between point features reflects the proximity of functional parts: points with similar features are more likely to belong to the same part. Building on this feature, we introduce the Part-Aware Diffusion Policy (PADP), an imitation learning framework aimed at enhancing sample efficiency and generalization for robotic manipulation. We evaluate PADP on several simulated and real-world tasks, demonstrating that PA3FF consistently outperforms a range of 2D and 3D representations in manipulation scenarios, including CLIP, DINOv2, and Grounded-SAM. Beyond imitation learning, PA3FF enables diverse downstream methods, including correspondence learning and segmentation tasks, making it a versatile foundation for robotic manipulation. Project page: https://pa3ff.github.io
URLs: https://pa3ff.github.io
Authors: Abubakarr Jaye, Nigel Boachie Kumankumah, Chidera Biringa, Anjel Shaileshbhai Patel, Sulaiman Vesal, Dayquan Julienne, Charlotte Siska, Manuel Ra\'ul Mel\'endez Luj\'an, Anthony Twum-Barimah, Mauricio Velazco, Tianwei Chen
Abstract: Long-horizon reasoning is a key challenge for autonomous agents, yet existing benchmarks evaluate agents on single tasks in isolation. Real organizational work requires managing many concurrent long-horizon tasks with interleaving, dependencies, and reprioritization. We introduce Multi-Horizon Task Environments (MHTEs): a distinct problem class requiring coherent execution across dozens of interleaved tasks (45+, 500-1500+ steps) within persistent execution contexts spanning hours. We identify four failure modes that cause baseline CUAs to degrade from 16.7% to 8.7% completion as load scales 25% to 100%, a pattern consistent across three independent implementations. These failure modes are context saturation (O(N) vs O(1) growth), memory interference, dependency complexity (DAGs vs. chains), and reprioritization overhead. We present CorpGen, an architecture-agnostic framework addressing these failures via hierarchical planning for multi-horizon goal alignment, sub-agent isolation preventing cross-task contamination, tiered memory (working, structured, semantic), and adaptive summarization. CorpGen simulates corporate environments through digital employees with persistent identities and realistic schedules. Across three CUA backends (UFO2, OpenAI CUA, hierarchical) on OSWorld Office, CorpGen achieves up to 3.5x improvement over baselines (15.2% vs 4.3%) with stable performance under increasing load, confirming that gains stem from architectural mechanisms rather than specific CUA implementations. Ablation studies show experiential learning provides the largest gains.
Authors: Vishnu Sai, Dheeraj Sai, Srinath B, Girish Varma, Priyesh Shukla
Abstract: Vision-Language Models (VLMs) face a critical memory bottleneck when processing long-form video content due to the linear growth of the Key-Value (KV) cache with sequence length. Existing solutions predominantly employ reactive eviction strategies that compute full attention matrices before discarding tokens, resulting in substantial computational waste. We propose Sali-Cache, a novel a priori optimization framework that implements dual-signal adaptive caching through proactive memory management. By integrating a temporal filter based on optical flow analysis for detecting inter-frame redundancy and a spatial filter leveraging saliency detection for identifying visually significant regions, Sali-Cache intelligently manages memory allocation before entering computationally expensive attention operations. Experimental evaluation on the LLaVA 1.6 architecture demonstrates that our method achieves a 2.20x compression ratio in effective memory usage while maintaining 100% accuracy across BLEU, ROUGE-L, and Exact Match metrics. Furthermore, under identical memory budget constraints, Sali-Cache preserves context-rich features over extended temporal durations without degrading model performance, enabling efficient processing of long-form video content on consumer-grade hardware.
Authors: Nafiseh Sadat Sajadi, Behnam Bahrak, Mahdi Jafari Siavoshani
Abstract: Predicting links in sparse, continuously evolving networks is a central challenge in network science. Conventional heuristic methods and deep learning models, including Graph Neural Networks (GNNs), are typically designed for static graphs and thus struggle to capture temporal dependencies. Snapshot-based techniques partially address this issue but often encounter data sparsity and class imbalance, particularly in networks with transient interactions such as telecommunication call detail records (CDRs). Temporal Graph Networks (TGNs) model dynamic graphs by updating node embeddings over time; however, their predictive accuracy under sparse conditions remains limited. In this study, we improve the TGN framework by extracting enclosing subgraphs around candidate links, enabling the model to jointly learn structural and temporal information. Experiments on a sparse CDR dataset show that our approach increases average precision by 2.6% over standard TGNs, demonstrating the advantages of integrating local topology for robust link prediction in dynamic networks.
Authors: Ala Emrani, Amir Najafi, Abolfazl Motahari
Abstract: Federated Learning provides a privacy-preserving paradigm for distributed learning, but suffers from statistical heterogeneity across clients. Personalized Federated Learning (PFL) mitigates this issue by considering client-specific models. A widely adopted approach in PFL decomposes neural networks into a shared feature extractor and client-specific heads. While effective, this design induces a fundamental tradeoff: deep or expressive shared components hinder personalization, whereas large local heads exacerbate overfitting under limited per-client data. Most existing methods rely on rigid, shallow heads, and therefore fail to navigate this tradeoff in a principled manner. In this work, we propose a boosting-inspired framework that enables a smooth control of this tradeoff. Instead of training a single personalized model, we construct an ensemble of $T$ models for each client. Across boosting iterations, the depth of the personalized component are progressively increased, while its effective complexity is systematically controlled via low-rank factorization or width shrinkage. This design simultaneously limits overfitting and substantially reduces per-client bias by allowing increasingly expressive personalization. We provide theoretical analysis that establishes generalization bounds with favorable dependence on the average local sample size and the total number of clients. Specifically, we prove that the complexity of the shared layers is effectively suppressed, while the dependence on the boosting horizon $T$ is controlled through parameter reduction. Notably, we provide a novel nonlinear generalization guarantee for decoupled PFL models. Extensive experiments on benchmark and real-world datasets (e.g., EMNIST, CIFAR-10/100, and Sent140) demonstrate that the proposed framework consistently outperforms state-of-the-art PFL methods under heterogeneous data distributions.
Authors: Saba Asaad, Ali Bereyhi
Abstract: Pinching antennas systems (PASSs) have recently been proposed as a novel flexible-antenna technology. These systems are implemented by attaching low-cost pinching elements to dielectric waveguides. As the direct link is bypassed through waveguides, PASSs can effectively compensate large-scale effects of the wireless channel. This work explores the potential gains of employing PASSs for over-the-air federated learning (OTA-FL). For a PASS-assisted server, we develop a low-complexity algorithmic approach, which jointly tunes the PASS parameters and schedules the mobile devices for minimal energy consumption in OTA-FL. We study the efficiency of the proposed design and compare it against the conventional OTA-FL setting with MIMO server. Numerical experiments demonstrate that using a single-waveguide PASS at the server within a moderately sized area, the required energy for model aggregation is drastically reduced as compared to the case with fully-digital MIMO server. This introduces PASS as a potential technology for energy-efficient distributed learning in next generations of wireless systems.
Authors: Osher Elhadad, Felipe Meneguzzi, Reuth Mirsky
Abstract: Understanding an agent's goals from its behavior is fundamental to aligning AI systems with human intentions. Existing goal recognition methods typically rely on an optimal goal-oriented policy representation, which may differ from the actor's true behavior and hinder the accurate recognition of their goal. To address this gap, this paper introduces Goal Recognition Alignment through Imitation Learning (GRAIL), which leverages imitation learning and inverse reinforcement learning to learn one goal-directed policy for each candidate goal directly from (potentially suboptimal) demonstration trajectories. By scoring an observed partial trajectory with each learned goal-directed policy in a single forward pass, GRAIL retains the one-shot inference capability of classical goal recognition while leveraging learned policies that can capture suboptimal and systematically biased behavior. Across the evaluated domains, GRAIL increases the F1-score by more than 0.5 under systematically biased optimal behavior, achieves gains of approximately 0.1-0.3 under suboptimal behavior, and yields improvements of up to 0.4 under noisy optimal trajectories, while remaining competitive in fully optimal settings. This work contributes toward scalable and robust models for interpreting agent goals in uncertain environments.
Authors: Lingxiang Hu, Yiding Sun, Tianle Xia, Wenwei Li, Ming Xu, Liqun Liu, Peng Shu, Huan Yu, Jie Jiang
Abstract: While Large Language Model (LLM) agents have achieved remarkable progress in complex reasoning tasks, evaluating their performance in real-world environments has become a critical problem. Current benchmarks, however, are largely restricted to idealized simulations, failing to address the practical demands of specialized domains like advertising and marketing analytics. In these fields, tasks are inherently more complex, often requiring multi-round interaction with professional marketing tools. To address this gap, we propose AD-Bench, a benchmark designed based on real-world business requirements of advertising and marketing platforms. AD-Bench is constructed from real user marketing analysis requests, with domain experts providing verifiable reference answers and corresponding reference tool-call trajectories. The benchmark categorizes requests into three difficulty levels (L1-L3) to evaluate agents' capabilities under multi-round, multi-tool collaboration. Experiments show that on AD-Bench, Gemini-3-Pro achieves Pass@1 = 68.0% and Pass@3 = 83.0%, but performance drops significantly on L3 to Pass@1 = 49.4% and Pass@3 = 62.1%, with a trajectory coverage of 70.1%, indicating that even state-of-the-art models still exhibit substantial capability gaps in complex advertising and marketing analysis scenarios. AD-Bench provides a realistic benchmark for evaluating and improving advertising marketing agents, the leaderboard and code can be found at https://github.com/Emanual20/adbench-leaderboard.
Authors: Zachary Bamberger, Till R. Saenger, Gilad Morad, Ofra Amir, Brandon M. Stewart, Amir Feder
Abstract: Inference-Time-Compute (ITC) methods like Best-of-N and Tree-of-Thoughts are meant to produce output candidates that are both high-quality and diverse, but their use of high-temperature sampling often fails to achieve meaningful output diversity. Moreover, existing ITC methods offer limited control over how to perform reasoning, which in turn limits their explainability. We present STATe-of-Thoughts (STATe), an interpretable ITC method that searches over high-level reasoning patterns. STATe replaces stochastic sampling with discrete and interpretable textual interventions: a controller selects actions encoding high-level reasoning choices, a generator produces reasoning steps conditioned on those choices, and an evaluator scores candidates to guide search. This structured approach yields three main advantages. First, action-guided textual interventions produce greater response diversity than temperature-based sampling. Second, in a case study on argument generation, STATe's explicit action sequences capture interpretable features that are highly predictive of output quality. Third, estimating the association between performance and action choices allows us to identify promising yet unexplored regions of the action space and steer generation directly toward them. Together, these results establish STATe as a practical framework for generating high-quality, diverse, and interpretable text. Our framework is available at https://github.com/zbambergerNLP/state-of-thoughts.
Authors: Nick Polson, Vadim Sokolov
Abstract: We study optimization for losses that admit a variance-mean scale-mixture representation. Under this representation, each EM iteration is a weighted least squares update in which latent variables determine observation and parameter weights; these play roles analogous to Adam's second-moment scaling and AdamW's weight decay, but are derived from the model. The resulting Scale Mixture EM (SM-EM) algorithm removes user-specified learning-rate and momentum schedules. On synthetic ill-conditioned logistic regression benchmarks with $p \in \{20, \ldots, 500\}$, SM-EM with Nesterov acceleration attains up to $13\times$ lower final loss than Adam tuned by learning-rate grid search. For a 40-point regularization path, sharing sufficient statistics across penalty values yields a $10\times$ runtime reduction relative to the same tuned-Adam protocol. For the base (non-accelerated) algorithm, EM monotonicity guarantees nonincreasing objective values; adding Nesterov extrapolation trades this guarantee for faster empirical convergence.
Authors: Md. Kamrul Hossain, Walid Aljoby
Abstract: In multi-intent intent-based networks, a single fault can trigger co-drift where multiple intents exhibit symptomatic KPI degradation, creating ambiguity about the true root-cause intent. We present MILD, a proactive framework that reformulates intent assurance from reactive drift detection to fixed-horizon failure prediction with intent-level disambiguation. MILD uses a teacher-augmented Mixture-of-Experts where a gated disambiguation module identifies the root-cause intent while per-intent heads output calibrated risk scores. On a benchmark with non-linear failures and co-drifts, MILD provides 3.8\%--92.5\% longer remediation lead time and improves intent-level root-cause disambiguation accuracy by 9.4\%--45.8\% over baselines. MILD also provides per-alert KPI explanations, enabling actionable diagnosis.
Authors: Zhenzhen Zhuang, Yuqing Fu, Jing Zhu, Zhangping Zhou, Jialiang Lin
Abstract: Automated scholarly paper review (ASPR) has entered the coexistence phase with traditional peer review, where artificial intelligence (AI) systems are increasingly incorporated into real-world manuscript evaluation. In parallel, research on automated and AI-assisted peer review has proliferated. Despite this momentum, empirical progress remains constrained by several critical limitations in existing datasets. While reviewers routinely evaluate figures, tables, and complex layouts to assess scientific claims, most existing datasets remain overwhelmingly text-centric. This bias is reinforced by a narrow focus on data from computer science venues. Furthermore, these datasets lack precise alignment between reviewer comments and specific manuscript versions, obscuring the iterative relationship between peer review and manuscript evolution. In response, we introduce FMMD, a multimodal and multidisciplinary open peer review dataset curated from F1000Research. The dataset bridges the current gap by integrating manuscript-level visual and structural data with version-specific reviewer reports and editorial decisions. By providing explicit alignment between reviewer comments and the exact article iteration under review, FMMD enables fine-grained analysis of the peer review lifecycle across diverse scientific domains. FMMD supports tasks such as multimodal issue detection and multimodal review comment generation. It provides a comprehensive empirical resource for the development of peer review research.
Authors: Chunlin Tian, Kahou Tam, Yebo Wu, Shuaihang Zhong, Li Li, Nicholas D. Lane, Chengzhong Xu
Abstract: Deploying large language models (LLMs) in real-time systems remains challenging due to their substantial computational demands and privacy concerns. We propose Floe, a hybrid federated learning framework designed for latency-sensitive, resource-constrained environments. Floe combines a cloud-based black-box LLM with lightweight small language models (SLMs) on edge devices to enable low-latency, privacy-preserving inference. Personal data and fine-tuning remain on-device, while the cloud LLM contributes general knowledge without exposing proprietary weights. A heterogeneity-aware LoRA adaptation strategy enables efficient edge deployment across diverse hardware, and a logit-level fusion mechanism enables real-time coordination between edge and cloud models. Extensive experiments demonstrate that Floe enhances user privacy and personalization. Moreover, it significantly improves model performance and reduces inference latency on edge devices under real-time constraints compared with baseline approaches.
Authors: Samuele Marro, Jialin Yu, Emanuele La Malfa, Oishi Deb, Jiawei Li, Yibo Yang, Ebey Abraham, Sunando Sengupta, Eric Sommerlade, Michael Wooldridge, Philip Torr
Abstract: As frontier Large Language Models (LLMs) increasingly saturate new benchmarks shortly after they are published, benchmarking itself is at a juncture: if frontier models keep improving, it will become increasingly hard for humans to generate discriminative tasks, provide accurate ground-truth answers, or evaluate complex solutions. If benchmarking becomes infeasible, our ability to measure any progress in AI is at stake. We refer to this scenario as the post-comprehension regime. In this work, we propose Critique-Resilient Benchmarking, an adversarial framework designed to compare models even when full human understanding is infeasible. Our technique relies on the notion of critique-resilient correctness: an answer is deemed correct if no adversary has convincingly proved otherwise. Unlike standard benchmarking, humans serve as bounded verifiers and focus on localized claims, which preserves evaluation integrity beyond full comprehension of the task. Using an itemized bipartite Bradley-Terry model, we jointly rank LLMs by their ability to solve challenging tasks and to generate difficult yet solvable questions. We showcase the effectiveness of our method in the mathematical domain across eight frontier LLMs, showing that the resulting scores are stable and correlate with external capability measures. Our framework reformulates benchmarking as an adversarial generation-evaluation game in which humans serve as final adjudicators.
Authors: Saar Cohen
Abstract: Coalition formation concerns strategic collaborations of selfish agents that form coalitions based on their preferences. It is often assumed that coalitions are disjoint and preferences are fully known, which may not hold in practice. In this paper, we thus present a new model of coalition formation with possibly overlapping coalitions under partial information, where selfish agents may be part of multiple coalitions simultaneously and their full preferences are initially unknown. Instead, information about past interactions and associated utility feedback is stored in a fixed offline dataset, and we aim to efficiently infer the agents' preferences from this dataset. We analyze the impact of diverse dataset information constraints by studying two types of utility feedback that can be stored in the dataset: agent- and coalition-level utility feedback. For both feedback models, we identify assumptions under which the dataset covers sufficient information for an offline learning algorithm to infer preferences and use them to recover a partition that is (approximately) Nash stable, in which no agent can improve her utility by unilaterally deviating. Our additional goal is devising algorithms with low sample complexity, requiring only a small dataset to obtain a desired approximation to Nash stability. Under agent-level feedback, we provide a sample-efficient algorithm proven to obtain an approximately Nash stable partition under a sufficient and necessary assumption on the information covered by the dataset. However, under coalition-level feedback, we show that only under a stricter assumption is sufficient for sample-efficient learning. Still, in multiple cases, our algorithms' sample complexity bounds have optimality guarantees up to logarithmic factors. Finally, extensive experiments show that our algorithm converges to a low approximation level to Nash stability across diverse settings.
Authors: Fan Chen, Sinho Chewi, Constantinos Daskalakis, Alexander Rakhlin
Abstract: We show that high-accuracy guarantees for log-concave sampling -- that is, iteration and query complexities which scale as $\mathrm{poly}\log(1/\delta)$, where $\delta$ is the desired target accuracy -- are achievable using stochastic gradients with subexponential tails. Notably, this exhibits a separation with the problem of convex optimization, where stochasticity (even additive Gaussian noise) in the gradient oracle incurs $\mathrm{poly}(1/\delta)$ queries. We also give an information-theoretic argument that light-tailed stochastic gradients are necessary for high accuracy: for example, in the bounded variance case, we show that the minimax-optimal query complexity scales as $\Theta(1/\delta)$. Our framework also provides similar high accuracy guarantees under stochastic zeroth order (value) queries.
Authors: Dillon Davis, Huiji Gao, Thomas Legrand, Juan Manuel Caicedo Carvajal, Malay Haldar, Kedar Bellare, Moutupsi Paul, Soumyadip Banerjee, Liwei He, Stephanie Moyerman, Sanjeev Katariya
Abstract: Airbnb search must balance a worldwide, highly varied supply of homes with guests whose location, amenity, style, and price expectations differ widely. Meeting those expectations hinges on an efficient retrieval stage that surfaces only the listings a guest might realistically book, before resource intensive ranking models are applied to determine the best results. Unlike many recommendation engines, our system faces a distinctive challenge, location retrieval, that sits upstream of ranking and determines which geographic areas are queried in order to filter inventory to a candidate set. The preexisting approach employs a deep bayesian bandit based system to predict a rectangular retrieval bounds area that can be used for filtering. The purpose of this paper is to demonstrate the methodology, challenges, and impact of rearchitecting search to retrieve from the subset of most bookable high precision rectangular map cells defined by dividing the world into 25M uniform cells.
Authors: Morgan Byrd, Donghoon Baek, Kartik Garg, Hyunyoung Jung, Daesol Cho, Maks Sorokin, Robert Wright, Sehoon Ha
Abstract: This paper presents Adaptive Whole-body Loco-Manipulation, AdaptManip, a fully autonomous framework for humanoid robots to perform integrated navigation, object lifting, and delivery. Unlike prior imitation learning-based approaches that rely on human demonstrations and are often brittle to disturbances, AdaptManip aims to train a robust loco-manipulation policy via reinforcement learning without human demonstrations or teleoperation data. The proposed framework consists of three coupled components: (1) a recurrent object state estimator that tracks the manipulated object in real time under limited field-of-view and occlusions; (2) a whole-body base policy for robust locomotion with residual manipulation control for stable object lifting and delivery; and (3) a LiDAR-based robot global position estimator that provides drift-robust localization. All components are trained in simulation using reinforcement learning and deployed on real hardware in a zero-shot manner. Experimental results show that AdaptManip significantly outperforms baseline methods, including imitation learning-based approaches, in adaptability and overall success rate, while accurate object state estimation improves manipulation performance even under occlusion. We further demonstrate fully autonomous real-world navigation, object lifting, and delivery on a humanoid robot.
Authors: Shuofei Qiao, Yunxiang Wei, Xuehai Wang, Bin Wu, Boyang Xue, Ningyu Zhang, Hossein A. Rahmani, Yanshan Wang, Qiang Zhang, Keyan Ding, Jeff Z. Pan, Huajun Chen, Emine Yilmaz
Abstract: The rapid evolution of Large Language Models has catalyzed a surge in scientific idea production, yet this leap has not been accompanied by a matching advance in idea evaluation. The fundamental nature of scientific evaluation needs knowledgeable grounding, collective deliberation, and multi-criteria decision-making. However, existing idea evaluation methods often suffer from narrow knowledge horizons, flattened evaluation dimensions, and the inherent bias in LLM-as-a-Judge. To address these, we regard idea evaluation as a knowledge-grounded, multi-perspective reasoning problem and introduce InnoEval, a deep innovation evaluation framework designed to emulate human-level idea assessment. We apply a heterogeneous deep knowledge search engine that retrieves and grounds dynamic evidence from diverse online sources. We further achieve review consensus with an innovation review board containing reviewers with distinct academic backgrounds, enabling a multi-dimensional decoupled evaluation across multiple metrics. We construct comprehensive datasets derived from authoritative peer-reviewed submissions to benchmark InnoEval. Experiments demonstrate that InnoEval can consistently outperform baselines in point-wise, pair-wise, and group-wise evaluation tasks, exhibiting judgment patterns and consensus highly aligned with human experts.
Authors: Tingting Tang, Yongqin Wang, Murali Annavaram
Abstract: Secure Multi-party Computation (MPC) enables untrusted parties to jointly compute a function without revealing their inputs. Its application to machine learning (ML) has gained significant attention, particularly for secure inference services deployed across multiple cloud virtual machines (VMs), where each VM acts as an MPC party. Model providers secret-share model weights, and users secret-share inputs, ensuring that each server operates only on random shares. While MPC provides strong cryptographic guarantees, it incurs substantial computational and communication overhead. Deep neural networks rely heavily on convolutional and fully connected layers, which require costly matrix multiplications in MPC. To reduce this cost, we propose leveraging low-rank decomposition (LRD) for linear layers, replacing one large matrix multiplication with two smaller ones. Each matrix multiplication in MPC incurs a round of communication, meaning decomposing one matrix multiplication into two leads to an additional communication round. Second, the added matrix multiplication requires an additional truncation step to maintain numerical precision. Since truncation itself requires communication and computation, these overheads can offset the gains from decomposition. To address this, we introduce two complementary optimizations: truncation skipping and efficient linear layer concatenation. Truncation skipping removes the extra truncation induced by LRD, while linear layer concatenation pipelines operations to hide the additional communication round. Together, these techniques mitigate the main overheads of LRD in MPC and improve overall efficiency. Our approach is broadly applicable across MPC protocols. Experiments show up to 25% speedup in n-PC and 33% in 3-PC protocols over full-rank baselines, along with up to 52% GPU energy savings and 88% reduction in offline-phase latency.
Authors: William L. Tong, Ege Cakar, Cengiz Pehlevan
Abstract: Recent years have witnessed meteoric progress in reasoning models: neural networks that generate intermediate reasoning traces (RTs) before producing a final output. Despite the rapid advancement, our understanding of how RTs support reasoning, and the limits of this paradigm, remain incomplete. To promote greater clarity, we introduce PITA: a novel large-scale dataset of over 23 million statements in propositional logic and their corresponding proofs. As a benchmark for robust reasoning, we focus on length generalization: if a model is trained to determine truth or falsity on statements with proofs up to fixed length, how well does it generalize to statements requiring longer proofs? We propose notions of (1) task depth and (2) task breadth, which measure respectively (1) the number of steps required to solve an example from a task and (2) the number of unique examples across a task. We vary these quantities across subsets of PITA, and find that RT models generalize well on broad and shallow subsets, while deteriorating on narrow and deep subsets relative to non-RT baselines. To determine whether our results are idiosyncratic to PITA or indicative of general phenomena, we compare our results to a simple synthetic task based on syllogisms. Our resulting theory suggests fundamental scalings that limit how well RT models perform on deep tasks, and highlights their generalization strengths on broad tasks. Our findings overall identify fundamental benefits and limitations inherent in using reasoning traces.
Authors: Harri Vanhems, Yue Zhao, Peng Shi, Archer Y. Yang
Abstract: Standard regression methods typically optimize a single pointwise objective, such as mean squared error, which conflates the learning of ordering with the learning of scale. This coupling renders models vulnerable to outliers and heavy-tailed noise. We propose CAIRO (Calibrate After Initial Rank Ordering), a framework that decouples regression into two distinct stages. In the first stage, we learn a scoring function by minimizing a scale-invariant ranking loss; in the second, we recover the target scale via isotonic regression. We theoretically characterize a class of "Optimal-in-Rank-Order" objectives -- including variants of RankNet and Gini covariance -- and prove that they recover the ordering of the true conditional mean under mild assumptions. We further show that subsequent monotone calibration guarantees recovery of the true regression function. Empirically, CAIRO combines the representation learning of neural networks with the robustness of rank-based statistics. It matches the performance of state-of-the-art tree ensembles on tabular benchmarks and significantly outperforms standard regression objectives in regimes with heavy-tailed or heteroskedastic noise.
Authors: Dongrui Liu, Yi Yu, Jie Zhang, Guanxu Chen, Qihao Lin, Hanxi Zhu, Lige Huang, Yijin Zhou, Peng Wang, Shuai Shao, Boxuan Zhang, Zicheng Liu, Jingwei Sun, Yu Li, Yuejin Xie, Jiaxuan Guo, Jia Xu, Chaochao Lu, Bowen Zhou, Xia Hu, Jing Shao
Abstract: To understand and identify the unprecedented risks posed by rapidly advancing artificial intelligence (AI) models, Frontier AI Risk Management Framework in Practice presents a comprehensive assessment of their frontier risks. As Large Language Models (LLMs) general capabilities rapidly evolve and the proliferation of agentic AI, this version of the risk analysis technical report presents an updated and granular assessment of five critical dimensions: cyber offense, persuasion and manipulation, strategic deception, uncontrolled AI R\&D, and self-replication. Specifically, we introduce more complex scenarios for cyber offense. For persuasion and manipulation, we evaluate the risk of LLM-to-LLM persuasion on newly released LLMs. For strategic deception and scheming, we add the new experiment with respect to emergent misalignment. For uncontrolled AI R\&D, we focus on the ``mis-evolution'' of agents as they autonomously expand their memory substrates and toolsets. Besides, we also monitor and evaluate the safety performance of OpenClaw during the interaction on the Moltbook. For self-replication, we introduce a new resource-constrained scenario. More importantly, we propose and validate a series of robust mitigation strategies to address these emerging threats, providing a preliminary technical and actionable pathway for the secure deployment of frontier AI. This work reflects our current understanding of AI frontier risks and urges collective action to mitigate these challenges.
Authors: Furkan Mumcu, Yasin Yilmaz
Abstract: Deploying large language model (LLM) agents in shared environments introduces a fundamental tension between individual alignment and collective stability: locally rational decisions can impose negative externalities that degrade system-level performance. We propose Socially-Weighted Alignment (SWA), a game-theoretic framework that modifies inference-time decision making by interpolating between an agent's private objective and an estimate of group welfare via a social weight $\lambda\in[0,1]$. In a shared-resource congestion game with $n$ agents and congestion severity $\beta$, we show that SWA induces a critical threshold $\lambda^*=(n-\beta)/(n-1)$ above which agents no longer have marginal incentive to increase demand under overload, yielding a phase transition from persistent congestion to stable operation near capacity. We further provide an inference-time algorithmic instantiation of SWA that does not require parameter updates or multi-agent reinforcement learning, and use a multi-agent simulation to empirically validate the predicted threshold behavior.
Authors: Somjit Roy, Prateek Jaiswal, Anirban Bhattacharya, Debdeep Pati, Bani K. Mallick
Abstract: We study Gaussian Process Thompson Sampling (GP-TS) for sequential decision-making over compact, continuous action spaces and provide a frequentist regret analysis based on fractional Gaussian process posteriors, without relying on domain discretization as in prior work. We show that the variance inflation commonly assumed in existing analyses of GP-TS can be interpreted as Thompson Sampling with respect to a fractional posterior with tempering parameter $\alpha \in (0,1)$. We derive a kernel-agnostic regret bound expressed in terms of the information gain parameter $\gamma_t$ and the posterior contraction rate $\epsilon_t$, and identify conditions on the Gaussian process prior under which $\epsilon_t$ can be controlled. As special cases of our general bound, we recover regret of order $\tilde{\mathcal{O}}(T^{\frac{1}{2}})$ for the squared exponential kernel, $\tilde{\mathcal{O}}(T^{\frac{2\nu+3d}{2(2\nu+d)}} )$ for the Mat\'ern-$\nu$ kernel, and a bound of order $\tilde{\mathcal{O}}(T^{\frac{2\nu+3d}{2(2\nu+d)}})$ for the rational quadratic kernel. Overall, our analysis provides a unified and discretization-free regret framework for GP-TS that applies broadly across kernel classes.
Authors: Baixiao Huang, Baiyu Huang, Yu Hou
Abstract: Quadruped robots are employed in various scenarios in building construction. However, autonomous stair climbing across different indoor staircases remains a major challenge for robot dogs to complete building construction tasks. In this project, we employed a two-stage end-to-end deep reinforcement learning (RL) approach to optimize a robot's performance on U-shaped stairs. The training robot-dog modality, Unitree Go2, was first trained to climb stairs on Isaac Lab's pyramid-stair terrain, and then to climb a U-shaped indoor staircase using the learned policies. This project explores end-to-end RL methods that enable robot dogs to autonomously climb stairs. The results showed (1) the successful goal reached for robot dogs climbing U-shaped stairs with a stall penalty, and (2) the transferability from the policy trained on U-shaped stairs to deployment on straight, L-shaped, and spiral stair terrains, and transferability from other stair models to deployment on U-shaped terrain.
Authors: Thanh Dang, Jiaming Liang
Abstract: We study two log-concave sampling problems: constrained sampling and composite sampling. First, we consider sampling from a target distribution with density proportional to $\exp(-f(x))$ supported on a convex set $K \subset \mathbb{R}^d$, where $f$ is convex. The main challenge is enforcing feasibility without degrading mixing. Using an epigraph transformation, we reduce this task to sampling from a nearly uniform distribution over a lifted convex set in $\mathbb{R}^{d+1}$. We then solve the lifted problem using a proximal sampler. Assuming only a separation oracle for $K$ and a subgradient oracle for $f$, we develop an implementation of the proximal sampler based on the cutting-plane method and rejection sampling. Unlike existing constrained samplers that rely on projection, reflection, barrier functions, or mirror maps, our approach enforces feasibility using only minimal oracle access, resulting in a practical and unbiased sampler without knowing the geometry of the constraint set. Second, we study composite sampling, where the target is proportional to $\exp(-f(x)-h(x))$ with closed and convex $f$ and $h$. This composite structure is standard in Bayesian inference with $f$ modeling data fidelity and $h$ encoding prior information. We reduce composite sampling via an epigraph lifting of $h$ to constrained sampling in $\mathbb{R}^{d+1}$, which allows direct application of the constrained sampling algorithm developed in the first part. This reduction results in a double epigraph lifting formulation in $\mathbb{R}^{d+2}$, on which we apply a proximal sampler. By keeping $f$ and $h$ separate, we further demonstrate how different combinations of oracle access (such as subgradient and proximal) can be leveraged to construct separation oracles for the lifted problem. For both sampling problems, we establish mixing time bounds measured in R\'enyi and $\chi^2$ divergences.
Authors: Aryan Das, Tanishq Rachamalla, Koushik Biswas, Swalpa Kumar Roy, Vinay Kumar Verma
Abstract: We introduce a novel uncertainty-aware multimodal segmentation framework that leverages both radiological images and associated clinical text for precise medical diagnosis. We propose a Modality Decoding Attention Block (MoDAB) with a lightweight State Space Mixer (SSMix) to enable efficient cross-modal fusion and long-range dependency modelling. To guide learning under ambiguity, we propose the Spectral-Entropic Uncertainty (SEU) Loss, which jointly captures spatial overlap, spectral consistency, and predictive uncertainty in a unified objective. In complex clinical circumstances with poor image quality, this formulation improves model reliability. Extensive experiments on various publicly available medical datasets, QATA-COVID19, MosMed++, and Kvasir-SEG, demonstrate that our method achieves superior segmentation performance while being significantly more computationally efficient than existing State-of-the-Art (SoTA) approaches. Our results highlight the importance of incorporating uncertainty modelling and structured modality alignment in vision-language medical segmentation tasks. Code: https://github.com/arya-domain/UA-VLS
Authors: Dennis Gross
Abstract: Safe and interpretable sequential decision-making is critical in healthcare, yet reinforcement learning (RL) policies for sepsis treatment optimization remain opaque and difficult to verify. Standard probabilistic model checkers operate on the full state space, which becomes infeasible for larger MDPs, and cannot explain why a learned policy makes particular decisions. COOL-MC wraps the model checker Storm but adds three key capabilities: it constructs only the reachable state space induced by a trained policy, yielding a smaller discrete-time Markov chain amenable to verification even when full-MDP analysis is intractable; it automatically labels states with clinically meaningful atomic propositions; and it integrates explainability methods with probabilistic computation tree logic (PCTL) queries to reveal which features drive decisions across treatment trajectories. We demonstrate COOL-MC's capabilities on the ICU-Sepsis MDP, a benchmark derived from approximately 17,000 sepsis patient records, which serves as a case study for applying COOL-MC to the formal analysis of sepsis treatment policies. Our analysis establishes hard bounds via full MDP verification, trains a safe RL policy that achieves optimal survival probability, and analyzes its behavior via PCTL verification and explainability on the induced DTMC. This reveals, for instance, that our trained policy relies predominantly on prior dosing history rather than the patient's evolving condition, a weakness that is invisible to standard evaluation but is exposed by COOL-MC's integration of formal verification and explainability. Our results illustrate how COOL-MC could serve as a tool for clinicians to investigate and debug sepsis treatment policies before deployment.
Authors: Sukumar Kishanthan, Kumar Thushalika, Buddhi Jayasekara, Asela Hevapathige
Abstract: Large language models (LLMs) demonstrate strong mathematical reasoning in English, but whether these capabilities reflect genuine multilingual reasoning or reliance on translation-based processing in low-resource languages like Sinhala and Tamil remains unclear. We examine this fundamental question by evaluating whether LLMs genuinely reason mathematically in these languages or depend on implicit translation to English-like representations. Using a taxonomy of six math problem types, from basic arithmetic to complex unit conflict and optimization problems, we evaluate four prominent large language models. To avoid translation artifacts that confound language ability with translation quality, we construct a parallel dataset where each problem is natively authored by fluent speakers with mathematical training in all three languages. Our analysis demonstrates that while basic arithmetic reasoning transfers robustly across languages, complex reasoning tasks show significant degradation in Tamil and Sinhala. The pattern of failures varies by model and problem type, suggesting that apparent multilingual competence may not reflect uniform reasoning capabilities across languages. These findings challenge the common assumption that models exhibiting strong multilingual performance can reason equally effectively across languages, and highlight the need for fine-grained, type-aware evaluation in multilingual settings.
Authors: Guy Freund, Tom Jurgenson, Matan Sudry, Erez Karpas
Abstract: Robotic knot-tying represents a fundamental challenge in robotics due to the complex interactions between deformable objects and strict topological constraints. We present TWISTED-RL, a framework that improves upon the previous state-of-the-art in demonstration-free knot-tying (TWISTED), which smartly decomposed a single knot-tying problem into manageable subproblems, each addressed by a specialized agent. Our approach replaces TWISTED's single-step inverse model that was learned via supervised learning with a multi-step Reinforcement Learning policy conditioned on abstract topological actions rather than goal states. This change allows more delicate topological state transitions while avoiding costly and ineffective data collection protocols, thus enabling better generalization across diverse knot configurations. Experimental results demonstrate that TWISTED-RL manages to solve previously unattainable knots of higher complexity, including commonly used knots such as the Figure-8 and the Overhand. Furthermore, the increase in success rates and drop in planning time establishes TWISTED-RL as the new state-of-the-art in robotic knot-tying without human demonstrations.
Authors: Gabriel Roccabruna, Olha Khomyn, Giuseppe Riccardi
Abstract: AI agents need to plan to achieve complex goals that involve orchestrating perception, sub-goal decomposition, and execution. These plans consist of ordered steps structured according to a Temporal Execution Order (TEO, a directed acyclic graph that ensures each step executes only after its preconditions are satisfied. Existing research on foundational models' understanding of temporal execution is limited to automatically derived annotations, approximations of the TEO as a linear chain, or text-only inputs. To address this gap, we introduce MATEO (MultimodAl Temporal Execution Order), a benchmark designed to assess and improve the temporal reasoning abilities of Large Vision Language Models (LVLMs) required for real-world planning. We acquire a high-quality professional multimodal recipe corpus, authored through a standardized editorial process that decomposes instructions into discrete steps, each paired with corresponding images. We collect TEO annotations as graphs by designing and using a scalable crowdsourcing pipeline. Using MATEO, we evaluate six state-of-the-art LVLMs across model scales, varying language context, multimodal input structure, and fine-tuning strategies.
Authors: Nathan Kirk
Abstract: Low-discrepancy designs play a central role in quasi-Monte Carlo methods and are increasingly influential in other domains such as machine learning, robotics and computer graphics, to name a few. In recent years, one such low-discrepancy construction method called subset selection has received a lot of attention. Given a large population, one optimally selects a small low-discrepancy subset with respect to a discrepancy-based objective. Versions of this problem are known to be NP-hard. In this text, we establish, for the first time, that the subset selection problem with respect to kernel discrepancies is also NP-hard. Motivated by this intractability, we propose a Bayesian Optimization procedure for the subset selection problem utilizing the recent notion of deep embedding kernels. We demonstrate the performance of the BO algorithm to minimize discrepancy measures and note that the framework is broadly applicable any design criteria.
Authors: Naveen Vakada, Kartik Hegde, Arvind Krishna Sridhar, Yinyi Guo, Erik Visser
Abstract: Long-duration audio is increasingly common in industrial and consumer settings, yet reviewing multi-hour recordings is impractical, motivating systems that answer natural-language queries with precise temporal grounding and minimal hallucination. Existing audio-language models show promise, but long-audio question answering remains difficult due to context-length limits. We introduce LongAudio-RAG (LA-RAG), a hybrid framework that grounds Large Language Model (LLM) outputs in retrieved, timestamped acoustic event detections rather than raw audio. Multi-hour streams are converted into structured event records stored in an SQL database, and at inference time the system resolves natural-language time references, classifies intent, retrieves only the relevant events, and generates answers using this constrained evidence. To evaluate performance, we construct a synthetic long-audio benchmark by concatenating recordings with preserved timestamps and generating template-based question-answer pairs for detection, counting, and summarization tasks. Finally, we demonstrate the practicality of our approach by deploying it in a hybrid edge-cloud environment, where the audio grounding model runs on-device on IoT-class hardware while the LLM is hosted on a GPU-backed server. This architecture enables low-latency event extraction at the edge and high-quality language reasoning in the cloud. Experiments show that structured, event-level retrieval significantly improves accuracy compared to vanilla Retrieval-Augmented Generation (RAG) or text-to-SQL approaches.
Authors: Aswathi Varma, Suprosanna Shit, Chinmay Prabhakar, Daniel Scholz, Hongwei Bran Li, Bjoern Menze, Daniel Rueckert, Benedikt Wiestler
Abstract: Vision Transformers (ViTs) have emerged as the state-of-the-art architecture in representation learning, leveraging self-attention mechanisms to excel in various tasks. ViTs split images into fixed-size patches, constraining them to a predefined size and necessitating pre-processing steps like resizing, padding, or cropping. This poses challenges in medical imaging, particularly with irregularly shaped structures like tumors. A fixed bounding box crop size produces input images with highly variable foreground-to-background ratios. Resizing medical images can degrade information and introduce artefacts, impacting diagnosis. Hence, tailoring variable-sized crops to regions of interest can enhance feature representation capabilities. Moreover, large images are computationally expensive, and smaller sizes risk information loss, presenting a computation-accuracy tradeoff. We propose VariViT, an improved ViT model crafted to handle variable image sizes while maintaining a consistent patch size. VariViT employs a novel positional embedding resizing scheme for a variable number of patches. We also implement a new batching strategy within VariViT to reduce computational complexity, resulting in faster training and inference times. In our evaluations on two 3D brain MRI datasets, VariViT surpasses vanilla ViTs and ResNet in glioma genotype prediction and brain tumor classification. It achieves F1-scores of 75.5% and 76.3%, respectively, learning more discriminative features. Our proposed batching strategy reduces computation time by up to 30% compared to conventional architectures. These findings underscore the efficacy of VariViT in image representation learning. Our code can be found here: https://github.com/Aswathi-Varma/varivit
Authors: Erkan Karabulut, Daniel Daza, Paul Groth, Martijn C. Schut, Victoria Degeler
Abstract: Association Rule Mining (ARM) is a fundamental task for knowledge discovery in tabular data and is widely used in high-stakes decision-making. Classical ARM methods rely on frequent itemset mining, leading to rule explosion and poor scalability, while recent neural approaches mitigate these issues but suffer from degraded performance in low-data regimes. Tabular foundation models (TFMs), pretrained on diverse tabular data with strong in-context generalization, provide a basis for addressing these limitations. We introduce a model-agnostic association rule learning framework that extracts association rules from any conditional probabilistic model over tabular data, enabling us to leverage TFMs. We then introduce TabProbe, an instantiation of our framework that utilizes TFMs as conditional probability estimators to learn association rules out-of-the-box without frequent itemset mining. We evaluate our approach on tabular datasets of varying sizes based on standard ARM rule quality metrics and downstream classification performance. The results show that TFMs consistently produce concise, high-quality association rules with strong predictive performance and remain robust in low-data settings without task-specific training. Source code is available at https://github.com/DiTEC-project/tabprobe.
Authors: Luke Antoncich, Yuben Moodley, Ugo Varetto, Jingbo Wang, Jonathan Wurtz, Jing Chen, Pascal Jahan Elahi, Casey R. Myers
Abstract: Biomarker-based prediction of clinical outcomes is challenging due to nonlinear relationships, correlated features, and the limited size of many medical datasets. Classical machine-learning methods can struggle under these conditions, motivating the search for alternatives. In this work, we investigate quantum reservoir computing (QRC), using both noiseless emulation and hardware execution on the neutral-atom Rydberg processor \textit{Aquila}. We evaluate performance with six classical machine-learning models and use SHAP to generate feature subsets. We find that models trained on emulated quantum features achieve mean test accuracies comparable to those trained on classical features, but have higher training accuracies and greater variability over data splits, consistent with overfitting. When comparing hardware execution of QRC to noiseless emulation, the models are more robust over different data splits and often exhibit statistically significant improvements in mean test accuracy. This combination of improved accuracy and increased stability is suggestive of a regularising effect induced by hardware execution. To investigate the origin of this behaviour, we examine the statistical differences between hardware and emulated quantum feature distributions. We find that hardware execution applies a structured, time-dependent transformation characterised by compression toward the mean and a progressive reduction in mutual information relative to emulation.
Authors: Matthaios Chatzopoulos, Phaedon-Stelios Koutsourelakis
Abstract: Inverse problems and inverse design in multiphase media, i.e., recovering or engineering microstructures to achieve target macroscopic responses, require operating on discrete-valued material fields, rendering the problem non-differentiable and incompatible with gradient-based methods. Existing approaches either relax to continuous approximations, compromising physical fidelity, or employ separate heavyweight models for forward and inverse tasks. We propose GenPANIS, a unified generative framework that preserves exact discrete microstructures while enabling gradient-based inference through continuous latent embeddings. The model learns a joint distribution over microstructures and PDE solutions, supporting bidirectional inference (forward prediction and inverse recovery) within a single architecture. The generative formulation enables training with unlabeled data, physics residuals, and minimal labeled pairs. A physics-aware decoder incorporating a differentiable coarse-grained PDE solver preserves governing equation structure, enabling extrapolation to varying boundary conditions and microstructural statistics. A learnable normalizing flow prior captures complex posterior structure for inverse problems. Demonstrated on Darcy flow and Helmholtz equations, GenPANIS maintains accuracy on challenging extrapolative scenarios - including unseen boundary conditions, volume fractions, and microstructural morphologies, with sparse, noisy observations. It outperforms state-of-the-art methods while using 10 - 100 times fewer parameters and providing principled uncertainty quantification.
Authors: Attila Lischka, Bal\'azs Kulcs\'ar
Abstract: Emergency situations that require the evacuation of urban areas can arise from man-made causes (e.g., terrorist attacks or industrial accidents) or natural disasters, the latter becoming more frequent due to climate change. As a result, effective and fast methods to develop evacuation plans are of great importance. In this work, we identify and propose the Bus Evacuation Orienteering Problem (BEOP), an NP-hard combinatorial optimization problem with the goal of evacuating as many people from an affected area by bus in a short, predefined amount of time. The purpose of bus-based evacuation is to reduce congestion and disorder that arises in purely car-focused evacuation scenarios. To solve the BEOP, we propose a deep reinforcement learning-based method utilizing graph learning, which, once trained, achieves fast inference speed and is able to create evacuation routes in fractions of seconds. We can bound the gap of our evacuation plans using an MILP formulation. To validate our method, we create evacuation scenarios for San Francisco using real-world road networks and travel times. We show that we achieve near-optimal solution quality and are further able to investigate how many evacuation vehicles are necessary to achieve certain bus-based evacuation quotas given a predefined evacuation time while keeping run time adequate.
Authors: Markus Gross, Hans-Martin Rieser
Abstract: Finding optimal measurement operators is crucial for the performance of quantum reservoir computers (QRCs), since they employ a fixed quantum feature map. We formulate the training of both stateless (quantum extreme learning machines, QELMs) and stateful (memory dependent) QRCs in the framework of kernel ridge regression. This approach renders an optimal measurement operator that minimizes prediction error for a given reservoir and training dataset. For large qubit numbers, this method is more efficient than the conventional training of QRCs. We discuss efficiency and practical implementation strategies, including Pauli basis decomposition and operator diagonalization, to adapt the optimal observable to hardware constraints. Numerical experiments on image classification and time series prediction tasks demonstrate the effectiveness of this approach, which can also be applied to other quantum ML models.
Authors: Lukas Struppek, Adam Gleave, Kellin Pelrine
Abstract: As the capabilities of large language models continue to advance, so does their potential for misuse. While closed-source models typically rely on external defenses, open-weight models must primarily depend on internal safeguards to mitigate harmful behavior. Prior red-teaming research has largely focused on input-based jailbreaking and parameter-level manipulations. However, open-weight models also natively support prefilling, which allows an attacker to predefine initial response tokens before generation begins. Despite its potential, this attack vector has received little systematic attention. We present the largest empirical study to date of prefill attacks, evaluating over 20 existing and novel strategies across multiple model families and state-of-the-art open-weight models. Our results show that prefill attacks are consistently effective against all major contemporary open-weight models, revealing a critical and previously underexplored vulnerability with significant implications for deployment. While certain large reasoning models exhibit some robustness against generic prefilling, they remain vulnerable to tailored, model-specific strategies. Our findings underscore the urgent need for model developers to prioritize defenses against prefill attacks in open-weight LLMs.
Authors: Lunjun Zhang, Ryan Chen, Bradly C. Stadie
Abstract: Building agentic systems that can autonomously self-improve from experience is a longstanding goal of AI. Large language models (LLMs) today primarily self-improve via two mechanisms: self-reflection for context updates, and reinforcement learning (RL) for weight updates. In this work, we propose Evolutionary System Prompt Learning (E-SPL), a method for jointly improving model contexts and model weights. In each RL iteration, E-SPL selects multiple system prompts and runs rollouts with each in parallel. It applies RL updates to model weights conditioned on each system prompt, and evolutionary updates to the system prompt population via LLM-driven mutation and crossover. Each system prompt has a TrueSkill rating for evolutionary selection, updated from relative performance within each RL iteration batch. E-SPL encourages a natural division between declarative knowledge encoded in prompts and procedural knowledge encoded in weights, resulting in improved performance across reasoning and agentic tasks. For instance, in an easy-to-hard (AIME $\rightarrow$ BeyondAIME) generalization setting, E-SPL improves RL success rate from 38.8% $\rightarrow$ 45.1% while also outperforming reflective prompt evolution (40.0%). Overall, our results show that coupling reinforcement learning with system prompt evolution yields consistent gains in sample efficiency and generalization. Code: https://github.com/LunjunZhang/E-SPL
Authors: Ait Haddou Marwan
Abstract: The performance of quantum classifiers is typically analyzed through global state distinguishability or the trainability of variational models. This study investigates how much class information remains accessible under locality-constrained measurements in the presence of noise. The authors formulate binary quantum classification as constrained quantum state discrimination and introduce a locality-restricted distinguishability measure quantifying the maximum bias achievable by observables acting on at most $k$ subsystems. For $n$-qubit systems subject to independent depolarizing noise, the locally accessible signal is governed by a Pauli-weight-dependent contraction mechanism. This motivates a computable predictor, the $k$-local Pauli-accessible amplitude $A_{k}(p)$, which lower bounds the optimal $k$-local classification advantage. Numerical experiments on four-qubit encodings demonstrate quantitative agreement between empirical accuracy and the prediction across noise levels. The research identifies an operational breakdown threshold where $k$-local classifiers become indistinguishable from random guessing despite persistent global distinguishability.
Authors: S\"onke Tenckhoff, Mario Koddenbrock, Erik Rodner
Abstract: We present LLMStructBench, a novel benchmark for evaluating Large Language Models (LLMs) on extracting structured data and generating valid JavaScript Object Notation (JSON) outputs from natural-language text. Our open dataset comprises diverse, manually verified parsing scenarios of varying complexity and enables systematic testing across 22 models and five prompting strategies. We further introduce complementary performance metrics that capture both token-level accuracy and document-level validity, facilitating rigorous comparison of model, size, and prompting effects on parsing reliability. In particular, we show that choosing the right prompting strategy is more important than standard attributes such as model size. This especially ensures structural validity for smaller or less reliable models but increase the number of semantic errors. Our benchmark suite is an step towards future research in the area of LLM applied to parsing or Extract, Transform and Load (ETL) applications.
Authors: Erik Burman, Mats G. Larson, Karl Larsson, Jonatan Vallin
Abstract: We develop an interpolation-based reduced-order modeling framework for parameter-dependent partial differential equations arising in control, inverse problems, and uncertainty quantification. The solution is discretized in the physical domain using finite element methods, while the dependence on a finite-dimensional parameter is approximated separately. We establish existence, uniqueness, and regularity of the parametric solution and derive rigorous error estimates that explicitly quantify the interplay between spatial discretization and parameter approximation. In low-dimensional parameter spaces, classical interpolation schemes yield algebraic convergence rates based on Sobolev regularity in the parameter variable. In higher-dimensional parameter spaces, we replace classical interpolation by extreme learning machine (ELM) surrogates and obtain error bounds under explicit approximation and stability assumptions. The proposed framework is applied to inverse problems in quantitative photoacoustic tomography, where we derive potential and parameter reconstruction error estimates and demonstrate substantial computational savings compared to standard approaches, without sacrificing accuracy.
Authors: Shih-Fang Chen, Jun-Cheng Chen, I-Hong Jhuo, Yen-Yu Lin
Abstract: The human visual system tracks objects by integrating current observations with previously observed information, adapting to target and scene changes, and reasoning about occlusion at fine granularity. In contrast, recent generic object trackers are often optimized for training targets, which limits robustness and generalization in unseen scenarios, and their occlusion reasoning remains coarse, lacking detailed modeling of occlusion patterns. To address these limitations in generalization and occlusion perception, we propose GOT-JEPA, a model-predictive pretraining framework that extends JEPA from predicting image features to predicting tracking models. Given identical historical information, a teacher predictor generates pseudo-tracking models from a clean current frame, and a student predictor learns to predict the same pseudo-tracking models from a corrupted version of the current frame. This design provides stable pseudo supervision and explicitly trains the predictor to produce reliable tracking models under occlusions, distractors, and other adverse observations, improving generalization to dynamic environments. Building on GOT-JEPA, we further propose OccuSolver to enhance occlusion perception for object tracking. OccuSolver adapts a point-centric point tracker for object-aware visibility estimation and detailed occlusion-pattern capture. Conditioned on object priors iteratively generated by the tracker, OccuSolver incrementally refines visibility states, strengthens occlusion handling, and produces higher-quality reference labels that progressively improve subsequent model predictions. Extensive evaluations on seven benchmarks show that our method effectively enhances tracker generalization and robustness.
Authors: Laur\`ene Vaugrante, Anietta Weckauff, Thilo Hagendorff
Abstract: Recent research has demonstrated that large language models (LLMs) fine-tuned on incorrect trivia question-answer pairs exhibit toxicity - a phenomenon later termed "emergent misalignment". Moreover, research has shown that LLMs possess behavioral self-awareness - the ability to describe learned behaviors that were only implicitly demonstrated in training data. Here, we investigate the intersection of these phenomena. We fine-tune GPT-4.1 models sequentially on datasets known to induce and reverse emergent misalignment and evaluate whether the models are self-aware of their behavior transitions without providing in-context examples. Our results show that emergently misaligned models rate themselves as significantly more harmful compared to their base model and realigned counterparts, demonstrating behavioral self-awareness of their own emergent misalignment. Our findings show that behavioral self-awareness tracks actual alignment states of models, indicating that models can be queried for informative signals about their own safety.
Authors: Fengyuan Cao, Xinyu Liang, Fredrik Cumlin, Victor Ungureanu, Chandan K. A. Reddy, Christian Schuldt, Saikat Chatterjee
Abstract: Designing a speech quality assessment (SQA) system for estimating mean-opinion-score (MOS) of multi-rate speech with varying sampling frequency (16-48 kHz) is a challenging task. The challenge arises due to the limited availability of a MOS-labeled training dataset comprising multi-rate speech samples. While self-supervised learning (SSL) models have been widely adopted in SQA to boost performance, a key limitation is that they are pretrained on 16 kHz speech and therefore discard high-frequency information present in higher sampling rates. To address this issue, we propose a spectrogram-augmented SSL method that incorporates high-frequency features (up to 48 kHz sampling rate) through a parallel-branch architecture. We further introduce a two-step training scheme: the model is first pre-trained on a large 48 kHz dataset and then fine-tuned on a smaller multi-rate dataset. Experimental results show that leveraging high-frequency information overlooked by SSL features is crucial for accurate multi-rate SQA, and that the proposed two-step training substantially improves generalization when multi-rate data is limited.
Authors: Ivan Diliso, Roberto Barile, Claudia d'Amato, Nicola Fanizzi
Abstract: Datasets for the experimental evaluation of knowledge graph refinement algorithms typically contain only ground facts, retaining very limited schema level knowledge even when such information is available in the source knowledge graphs. This limits the evaluation of methods that rely on rich ontological constraints, reasoning or neurosymbolic techniques and ultimately prevents assessing their performance in large-scale, real-world knowledge graphs. In this paper, we present \resource{} the first resource that provides a workflow for extracting datasets including both schema and ground facts, ready for machine learning and reasoning services, along with the resulting curated suite of datasets. The workflow also handles inconsistencies detected when keeping both schema and facts and also leverage reasoning for entailing implicit knowledge. The suite includes newly extracted datasets from KGs with expressive schemas while simultaneously enriching existing datasets with schema information. Each dataset is serialized in OWL making it ready for reasoning services. Moreover, we provide utilities for loading datasets in tensor representations typical of standard machine learning libraries.
Authors: Ana F. Rodrigues, Lucas Ferraz, Laura Balbi, Pedro Giesteira Cotovio, Catia Pesquita
Abstract: Effective representations of protein sequences are widely recognized as a cornerstone of machine learning-based protein design. Yet, protein bioengineering poses unique challenges for sequence representation, as experimental datasets typically feature few mutations, which are either sparsely distributed across the entire sequence or densely concentrated within localized regions. This limits the ability of sequence-level representations to extract functionally meaningful signals. In addition, comprehensive comparative studies remain scarce, despite their crucial role in clarifying which representations best encode relevant information and ultimately support superior predictive performance. In this study, we systematically evaluate multiple ProtBERT and ESM2 embedding variants as sequence representations, using the adeno-associated virus capsid as a case study and prototypical example of bioengineering, where functional optimization is targeted through highly localized sequence variation within an otherwise large protein. Our results reveal that, prior to fine-tuning, amino acid-level embeddings outperform sequence-level representations in supervised predictive tasks, whereas the latter tend to be more effective in unsupervised settings. However, optimal performance is only achieved when embeddings are fine-tuned with task-specific labels, with sequence-level representations providing the best performance. Moreover, our findings indicate that the extent of sequence variation required to produce notable shifts in sequence representations exceeds what is typically explored in bioengineering studies, showing the need for fine-tuning in datasets characterized by sparse or highly localized mutations.
Authors: Hang Zou, Yu Tian, Bohao Wang, Lina Bariah, Samson Lasaulce, Chongwen Huang, M\'erouane Debbah
Abstract: Large language models (LLMs) and multimodal models have become powerful general-purpose reasoning systems. However, radio-frequency (RF) signals, which underpin wireless systems, are still not natively supported by these models. Existing LLM-based approaches for telecom focus mainly on text and structured data, while conventional RF deep-learning models are built separately for specific signal-processing tasks, highlighting a clear gap between RF perception and high-level reasoning. To bridge this gap, we introduce RF-GPT, a radio-frequency language model (RFLM) that utilizes the visual encoders of multimodal LLMs to process and understand RF spectrograms. In this framework, complex in-phase/quadrature (IQ) waveforms are mapped to time-frequency spectrograms and then passed to pretrained visual encoders. The resulting representations are injected as RF tokens into a decoder-only LLM, which generates RF-grounded answers, explanations, and structured outputs. To train RF-GPT, we perform supervised instruction fine-tuning of a pretrained multimodal LLM using a fully synthetic RF corpus. Standards-compliant waveform generators produce wideband scenes for six wireless technologies, from which we derive time-frequency spectrograms, exact configuration metadata, and dense captions. A text-only LLM then converts these captions into RF-grounded instruction-answer pairs, yielding roughly 12,000 RF scenes and 0.625 million instruction examples without any manual labeling. Across benchmarks for wideband modulation classification, overlap analysis, wireless-technology recognition, WLAN user counting, and 5G NR information extraction, RF-GPT achieves strong multi-task performance, whereas general-purpose VLMs with no RF grounding largely fail.
Authors: Xiang Xiang Wang, Guo-Wei Wei
Abstract: We propose a multi-dimensional persistent sheaf Laplacian (MPSL) framework on simplicial complexes for image analysis. The proposed method is motivated by the strong sensitivity of commonly used dimensionality reduction techniques, such as principal component analysis (PCA), to the choice of reduced dimension. Rather than selecting a single reduced dimension or averaging results across dimensions, we exploit complementary advantages of multiple reduced dimensions. At a given dimension, image samples are regarded as simplicial complexes, and persistent sheaf Laplacians are utilized to extract a multiscale localized topological spectral representation for individual image samples. Statistical summaries of the resulting spectra are then aggregated across scales and dimensions to form multiscale multi-dimensional image representations. We evaluate the proposed framework on the COIL20 and ETH80 image datasets using standard classification protocols. Experimental results show that the proposed method provides more stable performance across a wide range of reduced dimensions and achieves consistent improvements to PCA-based baselines in moderate dimensional regimes.
Authors: Pierre-Alexandre Mattei, Bruno Loureiro
Abstract: Temperature scaling is a simple method that allows to control the uncertainty of probabilistic models. It is mostly used in two contexts: improving the calibration of classifiers and tuning the stochasticity of large language models (LLMs). In both cases, temperature scaling is the most popular method for the job. Despite its popularity, a rigorous theoretical analysis of the properties of temperature scaling has remained elusive. We investigate here some of these properties. For classification, we show that increasing the temperature increases the uncertainty in the model in a very general sense (and in particular increases its entropy). However, for LLMs, we challenge the common claim that increasing temperature increases diversity. Furthermore, we introduce two new characterisations of temperature scaling. The first one is geometric: the tempered model is shown to be the information projection of the original model onto the set of models with a given entropy. The second characterisation clarifies the role of temperature scaling as a submodel of more general linear scalers such as matrix scaling and Dirichlet calibration: we show that temperature scaling is the only linear scaler that does not change the hard predictions of the model.
Authors: Artem Chuprov, Egor E. Nuzhin, Alexey A. Tsukanov, Nikolay V. Brilliantov
Abstract: We report a novel hybrid method of simultaneous atomistic simulation of solids in critical regions (contacts surfaces, cracks areas, etc.), along with continuum modeling of other parts. The continuum is treated in terms of quasi-atoms of different size, comprising composite medium. The parameters of interaction potential between the quasi-atoms are optimized to match elastic properties of the composite medium to those of the atomic one. The optimization method coincides conceptually with the online Machine Learning (ML) methods, making it computationally very efficient. Such an approach allows a straightforward application of standard software packages for molecular dynamics (MD), supplemented by the ML-based optimizer. The new method is applied to model systems with a simple, pairwise Lennard-Jones potential, as well with multi-body Tersoff potential, describing covalent bonds. Using LAMMPS software we simulate collision of particles of different size. Comparing simulation results, obtained by the novel method, with full-atomic simulations, we demonstrate its accuracy, validity and overwhelming superiority in computational speed. Furthermore, we compare our method with other hybrid methods, specifically, with the closest one -- AtC (Atomic to Continuum) method. We demonstrate a significant superiority of our approach in computational speed and implementation convenience. Finally, we discuss a possible extension of the method for modeling other phenomena.
Authors: Ram\'on Nartallo-Kaluarachchi, Renaud Lambiotte, Alain Goriely
Abstract: Recurrent neural networks (RNNs) provide a theoretical framework for understanding computation in biological neural circuits, yet classical results, such as Hopfield's model of associative memory, rely on symmetric connectivity that restricts network dynamics to gradient-like flows. In contrast, biological networks support rich time-dependent behaviour facilitated by their asymmetry. Here we introduce a general framework, which we term drift-diffusion matching, for training continuous-time RNNs to represent arbitrary stochastic dynamical systems within a low-dimensional latent subspace. Allowing asymmetric connectivity, we show that RNNs can faithfully embed the drift and diffusion of a given stochastic differential equation, including nonlinear and nonequilibrium dynamics such as chaotic attractors. As an application, we construct RNN realisations of stochastic systems that transiently explore various attractors through both input-driven switching and autonomous transitions driven by nonequilibrium currents, which we interpret as models of associative and sequential (episodic) memory. To elucidate how these dynamics are encoded in the network, we introduce decompositions of the RNN based on its asymmetric connectivity and its time-irreversibility. Our results extend attractor neural network theory beyond equilibrium, showing that asymmetric neural populations can implement a broad class of dynamical computations within low-dimensional manifolds, unifying ideas from associative memory, nonequilibrium statistical mechanics, and neural computation.
Authors: Moloud Arian Maram, Georgios Bletsos, Thanh Tung Nguyen, Ahmed Hassan, Michael Palm, Thomas Rung
Abstract: Adjoint-based shape optimization of ship hulls is a powerful tool for addressing high-dimensional design problems in naval architecture, particularly in minimizing the ship resistance. However, its application to vessels that employ complex propulsion systems introduces significant challenges. They arise from the need for transient simulations extending over long periods of time with small time steps and from the reverse temporal propagation of the primal and adjoint solutions. These challenges place considerable demands on the required storage and computing power, which significantly hamper the use of adjoint methods in the industry. To address this issue, we propose a machine learning-assisted optimization framework that employs a Conditional Variational Autoencoder-based surrogate model of the propulsion system. The surrogate model replicates the time-averaged flow field induced by a Voith Schneider Propeller and replaces the geometrically and time-resolved propeller with a data-driven approximation. Primal flow verification examples demonstrate that the surrogate model achieves significant computational savings while maintaining the necessary accuracy of the resolved propeller. Optimization studies show that ignoring the propulsion system can yield designs that perform worse than the initial shape. In contrast, the proposed method produces shapes that achieve more than an 8\% reduction in resistance.
Authors: Brandon Yee, Wilson Collins, Maximilian Rutkowski
Abstract: We extend the Prometheus framework for unsupervised phase transition discovery from 2D classical systems to 3D classical and quantum many-body systems, addressing scalability in higher dimensions and generalization to quantum fluctuations. For the 3D Ising model ($L \leq 32$), the framework detects the critical temperature within 0.01\% of literature values ($T_c/J = 4.511 \pm 0.005$) and extracts critical exponents with $\geq 70\%$ accuracy ($\beta = 0.328 \pm 0.015$, $\gamma = 1.24 \pm 0.06$, $\nu = 0.632 \pm 0.025$), correctly identifying the 3D Ising universality class via $\chi^2$ comparison ($p = 0.72$) without analytical guidance. For quantum systems, we developed quantum-aware VAE (Q-VAE) architectures using complex-valued wavefunctions and fidelity-based loss. Applied to the transverse field Ising model, we achieve 2\% accuracy in quantum critical point detection ($h_c/J = 1.00 \pm 0.02$) and successfully discover ground state magnetization as the order parameter ($r = 0.97$). Notably, for the disordered transverse field Ising model, we detect exotic infinite-randomness criticality characterized by activated dynamical scaling $\ln \xi \sim |h - h_c|^{-\psi}$, extracting a tunneling exponent $\psi = 0.48 \pm 0.08$ consistent with theoretical predictions ($\psi = 0.5$). This demonstrates that unsupervised learning can identify qualitatively different types of critical behavior, not just locate critical points. Our systematic validation across classical thermal transitions ($T = 0$ to $T > 0$) and quantum phase transitions ($T = 0$, varying $h$) establishes that VAE-based discovery generalizes across fundamentally different physical domains, providing robust tools for exploring phase diagrams where analytical solutions are unavailable.
Authors: Richard Bergna, Stefan Depeweg, Sergio Calvo-Ordo\~nez, Jonathan Plenk, Alvaro Cartea, Jose Miguel Hern\'andez-Lobato
Abstract: Reliable uncertainty estimates are crucial for deploying pretrained models; yet, many strong methods for quantifying uncertainty require retraining, Monte Carlo sampling, or expensive second-order computations and may alter a frozen backbone's predictions. To address this, we introduce Gaussian Process Activations (GAPA), a post-hoc method that shifts Bayesian modeling from weights to activations. GAPA replaces standard nonlinearities with Gaussian-process activations whose posterior mean exactly matches the original activation, preserving the backbone's point predictions by construction while providing closed-form epistemic variances in activation space. To scale to modern architectures, we use a sparse variational inducing-point approximation over cached training activations, combined with local k-nearest-neighbor subset conditioning, enabling deterministic single-pass uncertainty propagation without sampling, backpropagation, or second-order information. Across regression, classification, image segmentation, and language modeling, GAPA matches or outperforms strong post-hoc baselines in calibration and out-of-distribution detection while remaining efficient at test time.
Authors: Sidharthenee Nayak, Victor Sam Moses Babu, Chandrashekhar Narayan Bhende, Pratyush Chakraborty, Mayukha Pal
Abstract: In recent times, there has been considerable interest in fault detection within electrical power systems, garnering attention from both academic researchers and industry professionals. Despite the development of numerous fault detection methods and their adaptations over the past decade, their practical application remains highly challenging. Given the probabilistic nature of fault occurrences and parameters, certain decision-making tasks could be approached from a probabilistic standpoint. Protective systems are tasked with the detection, classification, and localization of faulty voltage and current line magnitudes, culminating in the activation of circuit breakers to isolate the faulty line. An essential aspect of designing effective fault detection systems lies in obtaining reliable data for training and testing, which is often scarce. Leveraging deep learning techniques, particularly the powerful capabilities of pattern classifiers in learning, generalizing, and parallel processing, offers promising avenues for intelligent fault detection. To address this, our paper proposes an anomaly-based approach for fault detection in electrical power systems, employing deep autoencoders. Additionally, we utilize Convolutional Autoencoders (CAE) for dimensionality reduction, which, due to its fewer parameters, requires less training time compared to conventional autoencoders. The proposed method demonstrates superior performance and accuracy compared to alternative detection approaches by achieving an accuracy of 97.62% and 99.92% on simulated and publicly available datasets.
Authors: Junyi Li, Tim Foissner, Floran Martin, Antti Piippo, Marko Hinkkanen
Abstract: This paper presents a physics-informed neural network approach for dynamic modeling of saturable synchronous machines, including cases with spatial harmonics. We introduce an architecture that incorporates gradient networks directly into the fundamental machine equations, enabling accurate modeling of the nonlinear and coupled electromagnetic constitutive relationship. By learning the gradient of the magnetic field energy, the model inherently satisfies energy balance (reciprocity conditions). The proposed architecture can universally approximate any physically feasible magnetic behavior and offers several advantages over lookup tables and standard machine learning models: it requires less training data, ensures monotonicity and reliable extrapolation, and produces smooth outputs. These properties further enable robust model inversion and optimal trajectory generation, often needed in control applications. We validate the proposed approach using measured and finite-element method (FEM) datasets from a 5.6-kW permanent-magnet (PM) synchronous reluctance machine. Results demonstrate accurate and physically consistent models, even with limited training data.
Authors: Nicola\"i Gouraud, C\^ome Cattin, Thomas Pl\'e, Olivier Adjoua, Louis Lagard\`ere, Jean-Philip Piquemal
Abstract: Following our previous work (J. Phys. Chem. Lett., 2026, 17, 5, 1288-1295), we propose the DMTS-NC approach, a distilled multi-time-step (DMTS) strategy using non conservative (NC) forces to further accelerate atomistic molecular dynamics simulations using foundation neural network models. There, a dual-level reversible reference system propagator algorithm (RESPA) formalism couples a target accurate conservative potential to a simplified distilled representation optimized for the production of non-conservative forces. Despite being non-conservative, the distilled architecture is designed to enforce key physical priors, such as equivariance under rotation and cancellation of atomic force components. These choices facilitate the distillation process and therefore improve drastically the robustness of simulation, significantly limiting the "holes" in the simpler potential, thus achieving excellent agreement with the forces data. Overall, the DMTS-NC scheme is found to be more stable and efficient than its conservative counterpart with additional speedups reaching 15-30% over DMTS. Requiring no finetuning steps, it is easier to implement and can be pushed to the limit of the systems physical resonances to maintain accuracy while providing maximum efficiency. As for DMTS, DMTS-NC is applicable to any neural network potential.
Authors: Ayush Shrivastava, Kirtan Gangani, Laksh Jain, Mayank Goel, Nipun Batra
Abstract: Vision language models (VLMs) achieve strong performance on RGB imagery, but they do not generalize to thermal images. Thermal sensing plays a critical role in settings where visible light fails, including nighttime surveillance, search and rescue, autonomous driving, and medical screening. Unlike RGB imagery, thermal images encode physical temperature rather than color or texture, requiring perceptual and reasoning capabilities that existing RGB-centric benchmarks do not evaluate. We introduce ThermEval-B, a structured benchmark of approximately 55,000 thermal visual question answering pairs designed to assess the foundational primitives required for thermal vision language understanding. ThermEval-B integrates public datasets with our newly collected ThermEval-D, the first dataset to provide dense per-pixel temperature maps with semantic body-part annotations across diverse indoor and outdoor environments. Evaluating 25 open-source and closed-source VLMs, we find that models consistently fail at temperature-grounded reasoning, degrade under colormap transformations, and default to language priors or fixed responses, with only marginal gains from prompting or supervised fine-tuning. These results demonstrate that thermal understanding requires dedicated evaluation beyond RGB-centric assumptions, positioning ThermEval as a benchmark to drive progress in thermal vision language modeling.
Authors: Meet Gandhi, George P. Kontoudis
Abstract: Gaussian Processes (GPs) are a powerful tool for probabilistic modeling, but their performance is often constrained in complex, largescale real-world domains due to the limited expressivity of classical kernels. Quantum computing offers the potential to overcome this limitation by embedding data into exponentially large Hilbert spaces, capturing complex correlations that remain inaccessible to classical computing approaches. In this paper, we propose a Distributed Quantum Gaussian Process (DQGP) method in a multiagent setting to enhance modeling capabilities and scalability. To address the challenging non-Euclidean optimization problem, we develop a Distributed consensus Riemannian Alternating Direction Method of Multipliers (DR-ADMM) algorithm that aggregates local agent models into a global model. We evaluate the efficacy of our method through numerical experiments conducted on a quantum simulator in classical hardware. We use real-world, non-stationary elevation datasets of NASA's Shuttle Radar Topography Mission and synthetic datasets generated by Quantum Gaussian Processes. Beyond modeling advantages, our framework highlights potential computational speedups that quantum hardware may provide, particularly in Gaussian processes and distributed optimization.
Authors: Max Sobol Mark, Jacky Liang, Maria Attarian, Chuyuan Fu, Debidatta Dwibedi, Dhruv Shah, Aviral Kumar
Abstract: Many robot tasks require attending to the history of past observations. For example, finding an item in a room requires remembering which places have already been searched. However, the best-performing robot policies typically condition only on the current observation, limiting their applicability to such tasks. Naively conditioning on past observations often fails due to spurious correlations: policies latch onto incidental features of training histories that do not generalize to out-of-distribution trajectories upon deployment. We analyze why policies latch onto these spurious correlations and find that this problem stems from limited coverage over the space of possible histories during training, which grows exponentially with horizon. Existing regularization techniques provide inconsistent benefits across tasks, as they do not fundamentally address this coverage problem. Motivated by these findings, we propose Big Picture Policies (BPP), an approach that conditions on a minimal set of meaningful keyframes detected by a vision-language model. By projecting diverse rollouts onto a compact set of task-relevant events, BPP substantially reduces distribution shift between training and deployment, without sacrificing expressivity. We evaluate BPP on four challenging real-world manipulation tasks and three simulation tasks, all requiring history conditioning. BPP achieves 70% higher success rates than the best comparison on real-world evaluations.
Authors: Avinandan Bose, Shuyue Stella Li, Faeze Brahman, Pang Wei Koh, Simon Shaolei Du, Yulia Tsvetkov, Maryam Fazel, Lin Xiao, Asli Celikyilmaz
Abstract: Cold-start personalization requires inferring user preferences through interaction when no user-specific historical data is available. The core challenge is a routing problem: each task admits dozens of preference dimensions, yet individual users care about only a few, and which ones matter depends on who is asking. With a limited question budget, asking without structure will miss the dimensions that matter. Reinforcement learning is the natural formulation, but in multi-turn settings its terminal reward fails to exploit the factored, per-criterion structure of preference data, and in practice learned policies collapse to static question sequences that ignore user responses. We propose decomposing cold-start elicitation into offline structure learning and online Bayesian inference. Pep (Preference Elicitation with Priors) learns a structured world model of preference correlations offline from complete profiles, then performs training-free Bayesian inference online to select informative questions and predict complete preference profiles, including dimensions never asked about. The framework is modular across downstream solvers and requires only simple belief models. Across medical, mathematical, social, and commonsense reasoning, Pep achieves 80.8% alignment between generated responses and users' stated preferences versus 68.5% for RL, with 3-5x fewer interactions. When two users give different answers to the same question, Pep changes its follow-up 39-62% of the time versus 0-28% for RL. It does so with ~10K parameters versus 8B for RL, showing that the bottleneck in cold-start elicitation is the capability to exploit the factored structure of preference data.
Authors: Xiaosheng Zhao, Yuan-Sen Ting, Rosemary F. G. Wyse, Alexander S. Szalay, Yang Huang, L\'aszl\'o Dobos, Tam\'as Budav\'ari, Viska Wei
Abstract: Cross-survey generalization is a critical challenge in stellar spectral analysis, particularly in cases such as transferring from low- to moderate-resolution surveys. We investigate this problem using pre-trained models, focusing on simple neural networks such as multilayer perceptrons (MLPs), with a case study transferring from LAMOST low-resolution spectra (LRS) to DESI medium-resolution spectra (MRS). Specifically, we pre-train MLPs on either LRS or their embeddings and fine-tune them for application to DESI stellar spectra. We compare MLPs trained directly on spectra with those trained on embeddings derived from transformer-based models (self-supervised foundation models pre-trained for multiple downstream tasks). We also evaluate different fine-tuning strategies, including residual-head adapters, LoRA, and full fine-tuning. We find that MLPs pre-trained on LAMOST LRS achieve strong performance, even without fine-tuning, and that modest fine-tuning with DESI spectra further improves the results. For iron abundance, embeddings from a transformer-based model yield advantages in the metal-rich ([Fe/H] > -1.0) regime, but underperform in the metal-poor regime compared to MLPs trained directly on LRS. We also show that the optimal fine-tuning strategy depends on the specific stellar parameter under consideration. These results highlight that simple pre-trained MLPs can provide competitive cross-survey generalization, while the role of spectral foundation models for cross-survey stellar parameter estimation requires further exploration.
Authors: \c{S}. \.Ilker Birbil, Sinan Y{\i}ld{\i}r{\i}m, Samet \c{C}opur, M. Hakan Aky\"uz
Abstract: We propose a new regression algorithm that learns from a set of input-output pairs. Our algorithm is designed for populations where the relation between the input variables and the output variable exhibits a heterogeneous behavior across the predictor space. The algorithm starts with generating subsets that are concentrated around random points in the input space. This is followed by training a local predictor for each subset. Those predictors are then combined in a novel way to yield an overall predictor. We call this algorithm "LEarning with Subset Stacking" or LESS, due to its resemblance to the method of stacking regressors. We offer bagging and boosting variants of LESS and test against the state-of-the-art methods on several datasets. Our comparison shows that LESS is highly competitive.
Authors: Vincent Liu, Prabhat Nagarajan, Andrew Patterson, Martha White
Abstract: Offline reinforcement learning algorithms often require careful hyperparameter tuning. Before deployment, we need to select amongst a set of candidate policies. However, there is limited understanding about the fundamental limits of this offline policy selection (OPS) problem. In this work we provide clarity on when sample efficient OPS is possible, primarily by connecting OPS to off-policy policy evaluation (OPE) and Bellman error (BE) estimation. We first show a hardness result, that in the worst case, OPS is just as hard as OPE, by proving a reduction of OPE to OPS. As a result, no OPS method can be more sample efficient than OPE in the worst case. We then connect BE estimation to the OPS problem, showing how BE can be used as a tool for OPS. While BE-based methods generally require stronger requirements than OPE, when those conditions are met they can be more sample efficient. Building on this insight, we propose a BE method for OPS, called Identifiable BE Selection (IBES), that has a straightforward method for selecting its own hyperparameters. We conclude with an empirical study comparing OPE and IBES, and by showing the difficulty of OPS on an offline Atari benchmark dataset.
Authors: Edwin V. Bonilla, Pantelis Elinas, He Zhao, Maurizio Filippone, Vassili Kitsios, Terry O'Kane
Abstract: Estimating the structure of Bayesian networks as directed acyclic graphs (DAGs) from observational data is a fundamental challenge, particularly in causal discovery. Bayesian approaches excel by quantifying uncertainty and addressing identifiability, but key obstacles remain: (i) representing distributions over DAGs and (ii) estimating a posterior in the underlying combinatorial space. We introduce PIVID, a method that jointly infers a distribution over permutations and DAGs using variational inference and continuous relaxations of discrete distributions. Through experiments on synthetic and real-world datasets, we show that PIVID can outperform deterministic and Bayesian approaches, achieving superior accuracy-uncertainty trade-offs while scaling efficiently with the number of variables.
Authors: Yong Liu, Zirui Zhu, Chaoyu Gong, Minhao Cheng, Cho-Jui Hsieh, Yang You
Abstract: While fine-tuning large language models (LLMs) for specific tasks often yields impressive results, it comes at the cost of memory inefficiency due to back-propagation in gradient-based training. Memory-efficient Zeroth-order (MeZO) optimizers, recently proposed to address this issue, only require forward passes during training, making them more memory-friendly. However, compared with exact gradients, ZO-based gradients usually exhibit an estimation error, which can significantly hurt the optimization process, leading to slower convergence and suboptimal solutions. In addition, we find that the estimation error will hurt more when adding to large weights instead of small weights. Based on this observation, this paper introduces Sparse MeZO, a novel memory-efficient zeroth-order optimization approach that applies ZO only to a carefully chosen subset of parameters. We propose a simple yet effective parameter selection scheme that yields significant performance gains with Sparse-MeZO. Additionally, we develop a memory-optimized implementation for sparse masking, ensuring the algorithm requires only inference-level memory consumption, allowing Sparse-MeZO to fine-tune LLaMA-30b on a single A100 GPU. Experimental results illustrate that Sparse-MeZO consistently improves both performance and convergence speed over MeZO without any overhead. For example, it achieves a 9\% absolute accuracy improvement and 3.5x speedup over MeZO on the RTE task. Code is available at https://github.com/NUS-HPC-AI-Lab/SparseMeZO.
Authors: Subhojyoti Mukherjee, Anusha Lalitha, Kousha Kalantari, Aniket Deshmukh, Ge Liu, Yifei Ma, Branislav Kveton
Abstract: Learning of preference models from human feedback has been central to recent advances in artificial intelligence. Motivated by the cost of obtaining high-quality human annotations, we study efficient human preference elicitation for learning preference models. The key idea in our work is to generalize optimal designs, an approach to computing optimal information-gathering policies, to lists of items that represent potential questions with answers. The policy is a distribution over the lists and we elicit preferences from them proportionally to their probabilities. To show the generality of our ideas, we study both absolute and ranking feedback models on items in the list. We design efficient algorithms for both and analyze them. Finally, we demonstrate that our algorithms are practical by evaluating them on existing question-answering problems.
Authors: Akira Kitaoka
Abstract: A data-driven inverse optimization problem (DDIOP) seeks to estimate an objective function (i.e., weights) that is consistent with observed optimal-solution data, and is important in many applications, including those involving mixed integer linear programs (MILPs). In the DDIOP for MILPs, the prediction loss on features (PLF), defined as the discrepancy between observed and predicted feature values, becomes discontinuous with respect to the weights, which makes it difficult to apply gradient-based optimization. To address this issue, we focus on a Lipschitz continuous and convex suboptimality loss. By exploiting its convex and piecewise-linear structure and the interiority of the minimum set, we show that a broad class of gradient-based optimization methods, including projected subgradient descent (PSGD), reaches the minimum suboptimality loss value in a finite number of iterations, thereby exactly solving the DDIOP for MILPs. Furthermore, as a corollary, we show that PSGD attains the minimum PLF in finitely many iterations. We also derive an upper bound on the number of iterations required for PSGD to reach finite convergence, and confirm the finite-step behavior through numerical experiments.
Authors: Shenghui Li, Fanghua Ye, Meng Fang, Jiaxu Zhao, Yun-Hin Chan, Edith C. H. Ngai, Thiemo Voigt
Abstract: Over the past few years, the landscape of Artificial Intelligence (AI) has been reshaped by the emergence of Foundation Models (FMs). Pre-trained on massive datasets, these models exhibit exceptional performance across diverse downstream tasks through adaptation techniques like fine-tuning and prompt learning. More recently, the synergy of FMs and Federated Learning (FL) has emerged as a promising paradigm, often termed Federated Foundation Models (FedFM), allowing for collaborative model adaptation while preserving data privacy. This survey paper provides a systematic review of the current state of the art in FedFM, offering insights and guidance into the evolving landscape. Specifically, we present a comprehensive multi-tiered taxonomy based on three major dimensions, namely efficiency, adaptability, and trustworthiness. To facilitate practical implementation and experimental research, we undertake a thorough review of existing libraries and benchmarks. Furthermore, we discuss the diverse real-world applications of this paradigm across multiple domains. Finally, we outline promising research directions to foster future advancements in FedFM. Overall, this survey serves as a resource for researchers and practitioners, offering a thorough understanding of FedFM's role in revolutionizing privacy-preserving AI and pointing toward future innovations in this promising area. A periodically updated paper collection on FM-FL is available at https://github.com/lishenghui/awesome-fm-fl.
Authors: Divyansha Lachi, Mehdi Azabou, Vinam Arora, Eva Dyer
Abstract: Graph neural networks (GNNs) are often trained on individual datasets, requiring specialized models and significant hyperparameter tuning due to the unique structures and features of each dataset. This approach limits the scalability and generalizability of GNNs, as models must be tailored for each specific graph type. To address these challenges, we introduce GraphFM, a scalable multi-graph pretraining approach designed for learning across diverse graph datasets. GraphFM uses a Perceiver-based encoder with learned latent tokens to compress domain-specific features into a shared latent space, enabling generalization across graph domains. We propose new techniques for scaling up graph training on datasets of different sizes, allowing us to train GraphFM on 152 distinct graph datasets, containing a total of 7.4 million nodes and 189 million edges. This allows us to study the effect of scale on pretraining across domains such as molecules, citation networks, and product graphs, and show that training on diverse datasets improves performance over single-source pretraining. Additionally, pretraining with a mixture of synthetic and real graphs enhances adaptability and stability, leading to competitive performance with state-of-the-art models across various node classification tasks. This approach reduces the burden of dataset-specific training and provides a single generalist model capable of performing across multiple diverse graph structures and tasks. Code is available at https://github.com/nerdslab/GraphFM.
Authors: Wuxin Wang, Weicheng Ni, Ben Fei, Tao Han, Lilan Huang, Taikang Yuan, Xiaoyong Li, Lei Bai, Boheng Duan, Kaijun Ren
Abstract: Research on Artificial Intelligence (AI)-based Data Assimilation (DA) is expanding rapidly. However, the absence of an objective, comprehensive, and real-world benchmark hinders the fair comparison of diverse methods. Here, we introduce DABench, a benchmark designed for contributing to the development and evaluation of AI-based DA methods. By integrating real-world observations, DABench provides an objective and fair platform for validating long-term closed-loop DA cycles, supporting both deterministic and ensemble configurations. Furthermore, we assess the efficacy of AI-based DA in generating initial conditions for the advanced AI-based weather forecasting model to produce accurate medium-range global weather forecasting. Our dual-validation, utilizing both reanalysis data and independent radiosonde observations, demonstrates that AI-based DA achieves performance competitive with state-of-the-art AI-driven four-dimensional variational frameworks across both global weather DA and medium-range forecasting metrics. We invite the research community to utilize DABench to accelerate the advancement of AI-based DA for global weather forecasting.
Authors: Aitian Ma, Dongsheng Luo, Mo Sha
Abstract: Recently, there has been a growing interest in Long-term Time Series Forecasting (LTSF), which involves predicting long-term future values by analyzing a large amount of historical time-series data to identify patterns and trends. There exist significant challenges in LTSF due to its complex temporal dependencies and high computational demands. Although Transformer-based models offer high forecasting accuracy, they are often too compute-intensive to be deployed on devices with hardware constraints. On the other hand, the linear models aim to reduce the computational overhead by employing either decomposition methods in the time domain or compact representations in the frequency domain. In this paper, we propose MixLinear, an ultra-lightweight multivariate time series forecasting model specifically designed for resource-constrained devices. MixLinear effectively captures both temporal and frequency domain features by modeling intra-segment and inter-segment variations in the time domain and extracting frequency variations from a low-dimensional latent space in the frequency domain. By reducing the parameter scale of a downsampled $n$-length input/output one-layer linear model from $O(n^2)$ to $O(n)$, MixLinear achieves efficient computation without sacrificing accuracy. Extensive evaluations with four benchmark datasets show that MixLinear attains forecasting performance comparable to, or surpassing, state-of-the-art models with significantly fewer parameters ($0.1K$), which makes it well-suited for deployment on devices with limited computational capacity.
Authors: Branislav Kveton, Boris Oreshkin, Youngsuk Park, Aniket Deshmukh, Rui Song
Abstract: Posterior sampling in contextual bandits with a Gaussian prior can be implemented exactly or approximately using the Laplace approximation. The Gaussian prior is computationally efficient but it cannot describe complex distributions. In this work, we propose approximate posterior sampling algorithms for contextual bandits with a diffusion model prior. The key idea is to sample from a chain of approximate conditional posteriors, one for each stage of the reverse diffusion process, which are obtained by the Laplace approximation. Our approximations are motivated by posterior sampling with a Gaussian prior, and inherit its simplicity and efficiency. They are asymptotically consistent and perform well empirically on a variety of contextual bandit problems.
Authors: Zhaomin Wu, Jizhou Guo, Junyi Hou, Bingsheng He, Lixin Fan, Qiang Yang
Abstract: Prominent Large Language Model (LLM) services from providers like OpenAI and Google excel at general tasks but often underperform on domain-specific applications. Current customization services for these LLMs typically require users to upload data for fine-tuning, posing significant privacy risks. While differentially private (DP) data synthesis presents a potential alternative, its application commonly results in low effectiveness due to the introduction of excessive noise on data for DP. To overcome this, we introduce Llamdex, a novel framework that facilitates LLM customization as a service, where the client uploads pre-trained domain-specific models rather than data. This client-uploaded model, optionally protected by DP with much lower noise, is inserted into the base LLM via connection modules. Significantly, these connecting modules are trained without requiring sensitive domain data, enabling clients to customize LLM services while preserving data privacy. Experiments demonstrate that Llamdex improves domain-specific accuracy by up to 26% over state-of-the-art private data synthesis methods under identical privacy constraints and, by obviating the need for users to provide domain context within queries, maintains inference efficiency comparable to the original LLM service.
Authors: Zhihan Huang, Yuting Wei, Yuxin Chen
Abstract: The denoising diffusion probabilistic model (DDPM) has emerged as a mainstream generative model in generative AI. While sharp convergence guarantees have been established for the DDPM, the iteration complexity is, in general, proportional to the ambient data dimension, resulting in overly conservative theory that fails to explain its practical efficiency. This has motivated the recent work Li and Yan (2024a) to investigate how the DDPM can achieve sampling speed-ups through automatic exploitation of intrinsic low dimensionality of data. We strengthen this line of work by demonstrating, in some sense, optimal adaptivity to unknown low dimensionality. For a broad class of data distributions with intrinsic dimension $k$, we prove that the iteration complexity of the DDPM scales nearly linearly with $k$, which is optimal when using KL divergence to measure distributional discrepancy. Notably, our work is closely aligned with the independent concurrent work Potaptchik et al. (2024) -- posted two weeks prior to ours -- in establishing nearly linear-$k$ convergence guarantees for the DDPM.
Authors: Gene Yu, Ce Guo, Wayne Luk
Abstract: Time series causal discovery is essential for understanding dynamic systems, yet many existing methods remain sensitive to noise, non-stationarity, and sampling variability. We propose the Validated Consensus-Driven Framework (VCDF), a simple and method-agnostic layer that improves robustness by evaluating the stability of causal relations across blocked temporal subsets. VCDF requires no modification to base algorithms and can be applied to methods such as VAR-LiNGAM and PCMCI. Experiments on synthetic datasets show that VCDF improves VAR-LiNGAM by approximately 0.08-0.12 in both window and summary F1 scores across diverse data characteristics, with gains most pronounced for moderate-to-long sequences. The framework also benefits from longer sequences, yielding up to 0.18 absolute improvement on time series of length 1000 and above. Evaluations on simulated fMRI data and IT-monitoring scenarios further demonstrate enhanced stability and structural accuracy under realistic noise conditions. VCDF provides an effective reliability layer for time series causal discovery without altering underlying modeling assumptions.
Authors: Vima Gupta, Jae Hyung Ju, Kartik Sinha, Ada Gavrilovska, Anand Padmanabha Iyer
Abstract: Selective parameter activation provided by Mixture-of-Expert (MoE) models have made them a popular choice in modern foundational models. However, MoEs face a fundamental tension when employed for serving. Batching, critical for performance in serving, forces the activation of all experts, thereby negating MoEs' benefits and exacerbating memory bandwidth bottlenecks. Existing work on efficient MoE inference are unable to resolve this tension even with extensive workload-specific tuning. We present LYNX, a system that enables efficient MoE inference in a workload-agnostic fashion. Exploiting several key observations that we uncover in this work, LYNX provides a light-weight run-time dynamic expert remapping technique that depends only on information already available in the models. Our evaluation of LYNX on four state-of-the-art model families across nine benchmarks shows that it achieves up to 1.23x improvement in throughput while simultaneously improving accuracy by up to 4% in the majority of the tasks, and incurs only a negligible accuracy loss of less than 1% points in significantly hard tasks. Further, LYNX is complementary to existing techniques where it additionally boosts their performance by up to 1.38x.
Authors: Kaizhao Liang, Lizhang Chen, Bo Liu, Qiang Liu
Abstract: AdamW has been the default optimizer for transformer pretraining. For many years, our community searched for faster and more stable optimizers with only constrained positive outcomes. In this work, we propose a \textbf{one-line modification in Pytorch} to any momentum-based optimizer, which we rename cautious optimizer, e.g. C-AdamW and C-Lion. Our theoretical result shows that this modification preserves Adam's Hamiltonian function and it does not break the convergence guarantee under the Lyapunov analysis. In addition, a whole new family of optimizers is revealed by our theoretical insight. Among them, we pick the simplest one for empirical experiments, showing not only consistent speed-up on LLM pretraining, but also image classification, with minimum extra tuning on hyperparameters. Code is available at https://github.com/kyleliang919/C-Optim.
Authors: Nianze Tao, Minori Abe
Abstract: Generating novel molecules with higher properties than the training space, namely the out-of-distribution generation, is important for de novo drug design. However, it is not easy for distribution learning-based models, for example diffusion models, to solve this challenge as these methods are designed to fit the distribution of training data as close as possible. In this paper, we show that Bayesian flow network, especially ChemBFN model, is capable of intrinsically generating high quality out-of-distribution samples that meet several scenarios. A reinforcement learning strategy is added to the ChemBFN and a controllable ordinary differential equation solver-like generating process is employed that accelerate the sampling processes. Most importantly, we introduce a semi-autoregressive strategy during training and inference that enhances the model performance and surpass the state-of-the-art models. A theoretical analysis of out-of-distribution generation in ChemBFN with semi-autoregressive approach is included as well.
Authors: Ali Bereyhi, Ben Liang, Gary Boudreau, Ali Afana
Abstract: Error accumulation is effective for gradient sparsification in distributed settings: initially-unselected gradient entries are eventually selected as their accumulated error exceeds a certain level. The accumulation essentially behaves as a scaling of the learning rate for the selected entries. Although this property prevents the slow-down of lateral movements in distributed gradient descent, it can deteriorate convergence in some settings. This work proposes a novel sparsification scheme that controls the learning rate scaling of error accumulation. The development of this scheme follows two major steps: first, gradient sparsification is formulated as an inverse probability (inference) problem, and the Bayesian optimal sparsification mask is derived as a maximum-a-posteriori estimator. Using the prior distribution inherited from Top-k, we derive a new sparsification algorithm which can be interpreted as a regularized form of Top-k. We call this algorithm regularized Top-k (RegTop-k). It utilizes past aggregated gradients to evaluate posterior statistics of the next aggregation. It then prioritizes the local accumulated gradient entries based on these posterior statistics. We validate our derivation through various numerical experiments. In distributed linear regression, it is observed that while Top-k remains at a fixed distance from the global optimum, RegTop-k converges to the global optimum at significantly higher compression ratios. We further demonstrate the generalization of this observation by employing RegTop-k in distributed training of ResNet-18 on CIFAR-10, as well as fine-tuning of multiple computer vision models on the ImageNette dataset. Our numerical results confirm that as the compression ratio increases, RegTop-k sparsification noticeably outperforms Top-k.
Authors: Federico Errica, Henrik Christiansen, Viktor Zaverkin, Mathias Niepert, Francesco Alesiani
Abstract: For almost 70 years, researchers have typically selected the width of neural networks' layers either manually or through automated hyperparameter tuning methods such as grid search and, more recently, neural architecture search. This paper challenges the status quo by introducing an easy-to-use technique to learn an unbounded width of a neural network's layer during training. The method jointly optimizes the width and the parameters of each layer via standard backpropagation. We apply the technique to a broad range of data domains such as tables, images, text, sequences, and graphs, showing how the width adapts to the task's difficulty. A by product of our width learning approach is the easy truncation of the trained network at virtually zero cost, achieving a smooth trade-off between performance and compute resources. Alternatively, one can dynamically compress the network until performances do not degrade. In light of recent foundation models trained on large datasets, requiring billions of parameters and where hyper-parameter tuning is unfeasible due to huge training costs, our approach introduces a viable alternative for width learning.
Authors: Wenxuan Xie, Fanpu Cao
Abstract: In recent work on time-series prediction, Transformers and even large language models have garnered significant attention due to their strong capabilities in sequence modeling. However, in practical deployments, time-series prediction often requires operation in resource-constrained environments, such as edge devices, which are unable to handle the computational overhead of large models. To address such scenarios, some lightweight models have been proposed, but they exhibit poor performance on non-stationary sequences. In this paper, we propose $\textit{SWIFT}$, a lightweight model that is not only powerful, but also efficient in deployment and inference for Long-term Time Series Forecasting (LTSF). Our model is based on three key points: (i) Utilizing wavelet transform to perform lossless downsampling of time series. (ii) Achieving cross-band information fusion with a learnable filter. (iii) Using only one shared linear layer or one shallow MLP for sub-series' mapping. We conduct comprehensive experiments, and the results show that $\textit{SWIFT}$ achieves state-of-the-art (SOTA) performance on multiple datasets, offering a promising method for edge computing and deployment in this task. Moreover, it is noteworthy that the number of parameters in $\textit{SWIFT-Linear}$ is only 25\% of what it would be with a single-layer linear model for time-domain prediction. Our code is available at https://github.com/LancelotXWX/SWIFT.
Authors: Adela DePavia, Jose Cruzado, Jiayou Liang, Vasileios Charisopoulos, Rebecca Willett
Abstract: Adaptive optimization algorithms -- such as Adagrad, Adam, and their variants -- have found widespread use in machine learning, signal processing and many other settings. Several methods in this family are not rotationally equivariant, meaning that simple reparameterizations (i.e. change of basis) can drastically affect their convergence. However, their sensitivity to the choice of parameterization has not been systematically studied; it is not clear how to identify a "favorable" change of basis in which these methods perform best. In this paper we propose a reparameterization method and demonstrate both theoretically and empirically its potential to improve their convergence behavior. Our method is an orthonormal transformation based on the expected gradient outer product (EGOP) matrix, which can be approximated using either full-batch or stochastic gradient oracles. We show that for a broad class of functions, the sensitivity of adaptive algorithms to choice-of-basis is influenced by the decay of the EGOP matrix spectrum. We illustrate the potential impact of EGOP reparameterization by presenting empirical evidence and theoretical arguments that common machine learning tasks with "natural" data exhibit EGOP spectral decay.
Authors: Markus Krimmel, Jenna Wiens, Karsten Borgwardt, Dexiong Chen
Abstract: Existing graph generative models often face a critical trade-off between sample quality and generation speed. We introduce Autoregressive Noisy Filtration Modeling (ANFM), a flexible autoregressive framework that addresses both challenges. ANFM leverages filtration, a concept from topological data analysis, to transform graphs into short sequences of subgraphs. We identify exposure bias as a potential hurdle in autoregressive graph generation and propose noise augmentation and reinforcement learning as effective mitigation strategies, which allow ANFM to learn both edge addition and deletion operations. This unique capability enables ANFM to correct errors during generation by modeling non-monotonic graph sequences. Our results show that ANFM matches state-of-the-art diffusion models in quality while offering over 100 times faster inference, making it a promising approach for high-throughput graph generation. The source code is publicly available at https://github.com/BorgwardtLab/anfm .
Authors: Reena Elangovan, Charbel Sakr, Anand Raghunathan, Brucek Khailany
Abstract: Post-training quantization (PTQ) is a promising approach to reducing the storage and computational requirements of large language models (LLMs) without additional training cost. Recent PTQ studies have primarily focused on quantizing only weights to sub-8-bits while maintaining activations at 8-bits or higher. Accurate sub-8-bit quantization for both weights and activations without relying on quantization-aware training remains a significant challenge. We propose a novel quantization method called block clustered quantization (BCQ) wherein each operand tensor is decomposed into blocks (a block is a group of contiguous scalars), blocks are clustered based on their statistics, and a dedicated optimal quantization codebook is designed for each cluster. As a specific embodiment of this approach, we propose a PTQ algorithm called Locally-Optimal BCQ (LO-BCQ) that iterates between the steps of block clustering and codebook design to greedily minimize the quantization mean squared error. When weight and activation scalars are encoded to W4A4 format (with 0.5-bits of overhead for storing scaling factors and codebook selectors), we advance the current state-of-the-art by demonstrating <1% loss in inference accuracy across several LLMs and downstream tasks.
Authors: Sophia Sklaviadis, Thomas Moellenhoff, Andre Martins, Mario Figueiredo
Abstract: From a variational perspective, many statistical learning criteria involve seeking a distribution that balances empirical risk and regularization. In this paper, we broaden this perspective by introducing a new general class of variational methods based on Fenchel-Young (FY) losses, treated as divergences that generalize (and encompass) the familiar Kullback-Leibler divergence at the core of classical variational learning. Our proposed formulation -- FY variational learning -- includes as key ingredients new notions of FY free energy, FY evidence, FY evidence lower bound, and FY posterior. We derive alternating minimization and gradient backpropagation algorithms to compute (or lower bound) the FY evidence, which enables learning a wider class of models than previous variational formulations. This leads to generalized FY variants of classical algorithms, such as an FY expectation-maximization (FYEM) algorithm, and latent-variable models, such as an FY variational autoencoder (FYVAE). Our new methods are shown to be empirically competitive, often outperforming their classical counterparts, and most importantly, to have qualitatively novel features. For example, FYEM has an adaptively sparse E-step, while the FYVAE can support models with sparse observations and sparse posteriors.
Authors: Xun Deng, Han Zhong, Rui Ai, Fuli Feng, Zheng Wang, Xiangnan He
Abstract: Direct Preference Optimization (DPO) has emerged as a promising approach for aligning large language models with human preferences. While prior work mainly extends DPO from the aspect of the objective function, we instead improve DPO from the largely overlooked but critical aspect of data selection. Specifically, we address the issue of parameter shrinkage caused by noisy data by proposing a novel margin-maximization principle for dataset curation in DPO training. To further mitigate the noise in different reward models, we propose a Bayesian Aggregation approach that unifies multiple margin sources (external and implicit) into a single preference probability. Extensive experiments in diverse settings demonstrate the consistently high data efficiency of our approach. Remarkably, by using just 10\% of the Ultrafeedback dataset, our approach achieves 3\% to 8\% improvements across various Llama, Mistral, and Qwen models on the AlpacaEval2 benchmark. Furthermore, our approach seamlessly extends to iterative DPO, yielding a roughly 3\% improvement with 25\% online data, revealing the high redundancy in this presumed high-quality data construction manner. These results highlight the potential of data selection strategies for advancing preference optimization.
Authors: Sharan Mourya, Hannes Leipold, Bibhas Adhikari
Abstract: In this paper, we apply quantum machine learning (QML) to predict the stock prices of multiple assets using a contextual quantum neural network. Our approach captures recent trends to predict future stock price distributions, moving beyond traditional models that focus on entire historical data, enhancing adaptability and precision. Utilizing the principles of quantum superposition, we introduce a new training technique called the quantum batch gradient update (QBGU), which accelerates the standard stochastic gradient descent (SGD) in quantum applications and improves convergence. Consequently, we propose a quantum multi-task learning (QMTL) architecture, specifically, the share-and-specify ansatz, that integrates task-specific operators controlled by quantum labels, enabling the simultaneous and efficient training of multiple assets on the same quantum circuit as well as enabling efficient portfolio representation with logarithmic overhead in the number of qubits. This architecture represents the first of its kind in quantum finance, offering superior predictive power and computational efficiency for multi-asset stock price forecasting. Through extensive experimentation on S\&P 500 data for Apple, Google, Microsoft, and Amazon stocks, we demonstrate that our approach not only outperforms quantum single-task learning (QSTL) models but also effectively captures inter-asset correlations, leading to enhanced prediction accuracy. Our findings highlight the transformative potential of QML in financial applications, paving the way for more advanced, resource-efficient quantum algorithms in stock price prediction and other complex financial modeling tasks.
Authors: Seongho Son, William Bankes, Sangwoong Yoon, Shyam Sundhar Ramesh, Xiaohang Tang, Ilija Bogunovic
Abstract: We introduce Robust Multi-Objective Decoding (RMOD), a novel inference-time algorithm that robustly aligns Large Language Models (LLMs) to multiple human objectives (e.g., instruction-following, helpfulness, safety) by maximizing the worst-case rewards. RMOD formulates the robust decoding problem as a maximin two-player game between adversarially computed reward weights and the sampling policy, solvable through a Nash equilibrium. We demonstrate that this game reduces to a convex optimization problem to identify the worst-case reward weights, with the optimal sampling policy analytically derived. For practical applications, we propose an efficient algorithm of RMOD tailored for contemporary LLMs, introducing minimal computational overhead compared to standard non-robust Controlled Decoding methods. Experimental results across a range of popular alignment datasets with up to 10 objectives show the effectiveness of RMOD and its distilled version, consistently outperforming baselines in worst-case rewards and win rates.
Authors: Amit Attia, Tomer Koren
Abstract: The learning rate in stochastic gradient methods is a critical hyperparameter that is notoriously costly to tune via standard grid search, especially for training modern large-scale models with billions of parameters. We identify a theoretical advantage of learning rate annealing schemes that decay the learning rate to zero at a polynomial rate, such as the widely-used cosine schedule, by demonstrating their increased robustness to initial parameter misspecification due to a coarse grid search. We present an analysis in a stochastic convex optimization setup demonstrating that the convergence rate of stochastic gradient descent with annealed schedules depends sublinearly on the multiplicative misspecification factor $\rho$ (i.e., the grid resolution), achieving a rate of $O(\rho^{1/(2p+1)}/\sqrt{T})$ where $p$ is the degree of polynomial decay and $T$ is the number of steps. This is in contrast to the $O(\rho/\sqrt{T})$ rate obtained under the inverse-square-root and fixed stepsize schedules, which depend linearly on $\rho$. Experiments confirm the increased robustness compared to tuning with a fixed stepsize, that has significant implications for the computational overhead of hyperparameter search in practical training scenarios.
Authors: Zongyue Qin, Shichang Zhang, Mingxuan Ju, Tong Zhao, Neil Shah, Yizhou Sun
Abstract: Link prediction is a crucial graph-learning task with applications including citation prediction and product recommendation. Distilling Graph Neural Networks (GNNs) teachers into Multi-Layer Perceptrons (MLPs) students has emerged as an effective approach to achieve strong performance and reducing computational cost by removing graph dependency. However, existing distillation methods only use standard GNNs and overlook alternative teachers such as specialized model for link prediction (GNN4LP) and heuristic methods (e.g., common neighbors). This paper first explores the impact of different teachers in GNN-to-MLP distillation. Surprisingly, we find that stronger teachers do not always produce stronger students: MLPs distilled from GNN4LP can underperform those distilled from simpler GNNs, while weaker heuristic methods can teach MLPs to near-GNN performance with drastically reduced training costs. Building on these insights, we propose Ensemble Heuristic-Distilled MLPs (EHDM), which eliminates graph dependencies while effectively integrating complementary signals via a gating mechanism. Experiments on ten datasets show an average 7.93% improvement over previous GNN-to-MLP approaches with 1.95-3.32 times less training time, indicating EHDM is an efficient and effective link prediction method. Our code is available at https://github.com/ZongyueQin/EHDM
Authors: Zichen Liu, Wei Zhang, Christof Sch\"utte, Tiejun Li
Abstract: We propose Riemannian Denoising Diffusion Probabilistic Models (RDDPMs) for learning distributions on submanifolds of Euclidean space that are level sets of functions, including most of the manifolds relevant to applications. Existing methods for generative modeling on manifolds rely on substantial geometric information such as geodesic curves or eigenfunctions of the Laplace-Beltrami operator and, as a result, they are limited to manifolds where such information is available. In contrast, our method, built on a projection scheme, can be applied to more general manifolds, as it only requires being able to evaluate the value and the first order derivatives of the function that defines the submanifold. We provide a theoretical analysis of our method in the continuous-time limit, which elucidates the connection between our RDDPMs and score-based generative models on manifolds. The capability of our method is demonstrated on datasets from previous studies and on new datasets sampled from two high-dimensional manifolds, i.e. $\mathrm{SO}(10)$ and the configuration space of molecular system alanine dipeptide with fixed dihedral angle.
Authors: Abhijit Gupta
Abstract: Amortized variational inference in latent-variable forecasters creates a deployment gap: the test-time encoder approximates a training-time optimization-refined latent, but without access to future targets. This gap introduces unnecessary forecast error and interpretability challenges. In this work, we propose the Sparse Latent Factor Forecaster with Iterative Inference (SLFF), addressing this through (i) a sparse coding objective with L1 regularization for low-dimensional latents, (ii) unrolled proximal gradient descent (LISTA-style) for iterative refinement during training, and (iii) encoder alignment to ensure amortized outputs match optimization-refined solutions. Under a linearized decoder assumption, we derive a design-motivating bound on the amortization gap based on encoder-optimizer distance, with convergence rates under mild conditions; empirical checks confirm the bound is predictive for the deployed MLP decoder. To prevent mixed-frequency data leakage, we introduce an information-set-aware protocol using release calendars and vintage macroeconomic data. Interpretability is formalized via a three-stage protocol: stability (Procrustes alignment across seeds), driver validity (held-out regressions against observables), and behavioral consistency (counterfactuals and event studies). Using commodity futures (Copper, WTI, Gold; 2005--2025) as a testbed, SLFF demonstrates significant improvements over neural baselines at 1- and 5-day horizons, yielding sparse factors that are stable across seeds and correlated with observable economic fundamentals (interpretability remains correlational, not causal). Code, manifests, diagnostics, and artifacts are released.
Authors: Seyed Mohammad Azimi-Abarghouyi, Carlo Fischione
Abstract: Almost all existing hierarchical federated learning (FL) models are limited to two aggregation layers, restricting scalability and flexibility in complex, large-scale networks. In this work, we propose a Multi-Layer Hierarchical Federated Learning framework (QMLHFL), which appears to be the first study that generalizes hierarchical FL to arbitrary numbers of layers and network architectures through nested aggregation, while employing a layer-specific quantization scheme to meet communication constraints. We develop a comprehensive convergence analysis for QMLHFL and derive a general convergence condition and rate that reveal the effects of key factors, including quantization parameters, hierarchical architecture, and intra-layer iteration counts. Furthermore, we determine the optimal number of intra-layer iterations to maximize the convergence rate while meeting a deadline constraint that accounts for both communication and computation times. Our results show that QMLHFL consistently achieves high learning accuracy, even under high data heterogeneity, and delivers notably improved performance when optimized, compared to using randomly selected values.
Authors: Rafael Pablos Sarabia, Joachim Nyborg, Morten Birk, Jeppe Liborius Sj{\o}rup, Anders Lillevang Vesterholt, Ira Assent
Abstract: We present a deep learning model for high-resolution probabilistic precipitation forecasting over an 8-hour horizon in Europe, overcoming the limitations of radar-only deep learning models with short forecast lead times. Our model efficiently integrates multiple data sources - including radar, satellite, and physics-based numerical weather prediction (NWP) - while capturing long-range interactions, resulting in accurate forecasts with robust uncertainty quantification through consistent probabilistic maps. Featuring a compact architecture, it enables more efficient training and faster inference than existing models. Extensive experiments demonstrate that our model surpasses current operational NWP systems, extrapolation-based methods, and deep-learning nowcasting models, setting a new standard for high-resolution precipitation forecasting in Europe, ensuring a balance between accuracy, interpretability, and computational efficiency.
Authors: Shudi Weng, Chao Ren, Ming Xiao, Mikael Skoglund
Abstract: Federated learning (FL) commonly involves clients with diverse communication and computational capabilities. Such heterogeneity can significantly distort the optimization dynamics and lead to objective inconsistency, where the global model converges to an incorrect stationary point potentially far from the pursued optimum. Despite its critical impact, the joint effect of communication and computation heterogeneity has remained largely unexplored, due to the intrinsic complexity of their interaction. In this paper, we reveal the fundamentally distinct mechanisms through which heterogeneous communication and computation drive inconsistency in FL. To the best of our knowledge, this is the first unified theoretical analysis of general heterogeneous FL, offering a principled understanding of how these two forms of heterogeneity jointly distort the optimization trajectory under arbitrary choices of local solvers. Motivated by these insights, we propose Federated Heterogeneity-Aware Client Sampling, FedACS, a universal method to eliminate all types of objective inconsistency. We theoretically prove that FedACS converges to the correct optimum at a rate of $O(1/\sqrt{R})$, even in dynamic heterogeneous environments. Extensive experiments across multiple datasets show that FedACS outperforms state-of-the-art and category-specific baselines by 4.3%-36%, while reducing communication costs by 22%-89% and computation loads by 14%-105%, respectively.
Authors: Yichen Xu, Ryumei Nakada, Linjun Zhang, Lexin Li
Abstract: Transfer learning has become a central paradigm in modern machine learning, yet it suffers from the long-standing problem of negative transfer, where leveraging source representations can harm rather than help performance on the target task. Although empirical remedies have been proposed, there remains little theoretical understanding of how to reliably avoid negative transfer. In this paper, we investigate a simple yet remarkably effective strategy: augmenting frozen, pretrained source-side features with a trainable target-side encoder that adapts target features to capture residual signals overlooked by models pretrained on the source data. We show this residual feature integration strategy is sufficient to provably prevent negative transfer, by establishing theoretical guarantees that it has no worse convergence rate than training from scratch under the informative class of target distributions up to logarithmic factors, and that the convergence rate can transition seamlessly from nonparametric to near-parametric when source representations are informative. To our knowledge, this is the first theoretical work that ensures protection against negative transfer. We carry out extensive numerical experiments across image, text and tabular benchmarks, and empirically verify that the method consistently safeguards performance under distribution shift, label noise, semantic perturbation, and class imbalance. We additionally demonstrate that this residual integration mechanism uniquely supports adapt-time multimodality extension, enabling a pretrained single-cell foundation model to incorporate spatial signals for lymph-node anatomical classification despite the source model being trained without them. Our study thus advances the theory of safe transfer learning, and provides a principled approach that is simple, robust, architecture-agnostic, and broadly applicable.
Authors: Alvin Heng, Harold Soh
Abstract: Selective classification enhances the reliability of predictive models by allowing them to abstain from making uncertain predictions. In this work, we revisit the design of optimal selection functions through the lens of the Neyman--Pearson lemma, a classical result in statistics that characterizes the optimal rejection rule as a likelihood ratio test. We show that this perspective not only unifies the behavior of several post-hoc selection baselines, but also motivates new approaches to selective classification which we propose here. A central focus of our work is the setting of covariate shift, where the input distribution at test time differs from that at training. This realistic and challenging scenario remains relatively underexplored in the context of selective classification. We evaluate our proposed methods across a range of vision and language tasks, including both supervised learning and vision-language models. Our experiments demonstrate that our Neyman--Pearson-informed methods consistently outperform existing baselines, indicating that likelihood ratio-based selection offers a robust mechanism for improving selective classification under covariate shifts. Our code is publicly available at https://github.com/clear-nus/sc-likelihood-ratios.
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: Johannes Hertrich, Antonin Chambolle, Julie Delon
Abstract: This paper investigates the connections between rectified flows, flow matching, and optimal transport. Flow matching is a recent approach to learning generative models by estimating velocity fields that guide transformations from a source to a target distribution. Rectified flow matching aims to straighten the learned transport paths, yielding more direct flows between distributions. Our first contribution is a set of invariance properties of rectified flows and explicit velocity fields. In addition, we also provide explicit constructions and analysis in the Gaussian (not necessarily independent) and Gaussian mixture settings and study the relation to optimal transport. Our second contribution addresses recent claims suggesting that rectified flows, when constrained such that the learned velocity field is a gradient, can yield (asymptotically) solutions to optimal transport problems. We study the existence of solutions for this problem and demonstrate that they only relate to optimal transport under assumptions that are significantly stronger than those previously acknowledged. In particular, we present several counterexamples that invalidate earlier equivalence results in the literature, and we argue that enforcing a gradient constraint on rectified flows is, in general, not a reliable method for computing optimal transport maps.
Authors: Yifan Sun, Jingyan Shen, Yibin Wang, Tianyu Chen, Zhendong Wang, Mingyuan Zhou, Huan Zhang
Abstract: Reinforcement learning (RL) has become an effective approach for fine-tuning large language models (LLMs), particularly to enhance their reasoning capabilities. However, RL fine-tuning remains highly resource-intensive, and existing work has largely overlooked the problem of data efficiency. In this paper, we propose two techniques to improve data efficiency in LLM RL fine-tuning: difficulty-targeted online data selection and rollout replay. We introduce the notion of adaptive difficulty to guide online data selection, prioritizing questions of moderate difficulty that are more likely to yield informative learning signals. To estimate adaptive difficulty efficiently, we develop an attention-based framework that requires rollouts for only a small reference set of questions. The adaptive difficulty of the remaining questions is then estimated based on their similarity to this set. To further reduce rollout cost, we introduce a rollout replay mechanism inspired by experience replay in traditional RL. This technique reuses recent rollouts, lowering per-step computation while maintaining stable updates. Experiments across 6 LLM-dataset combinations show that our method reduces RL fine-tuning time by 23% to 62% while reaching the same level of performance as the original GRPO algorithm. Our code is available at https://github.com/ASTRAL-Group/data-efficient-llm-rl.
URLs: https://github.com/ASTRAL-Group/data-efficient-llm-rl.
Authors: Alex Clinton, Thomas Zeng, Yiding Chen, Xiaojin Zhu, Kirthevasan Kandasamy
Abstract: Modern data marketplaces and data sharing consortia increasingly rely on incentive mechanisms to encourage agents to contribute data. However, schemes that reward agents based on the quantity of submitted data are vulnerable to manipulation, as agents may submit fabricated or low-quality data to inflate their rewards. Prior work has proposed comparing each agent's data against others' to promote honesty: when others contribute genuine data, the best way to minimize discrepancy is to do the same. Yet prior implementations of this idea rely on very strong assumptions about the data distribution (e.g. Gaussian), limiting their applicability. In this work, we develop reward mechanisms based on a novel, two-sample test inspired by the Cram\'er-von Mises statistic. Our methods strictly incentivize agents to submit more genuine data, while disincentivizing data fabrication and other types of untruthful reporting. We establish that truthful reporting constitutes a (possibly approximate) Nash equilibrium in both Bayesian and prior-agnostic settings. We theoretically instantiate our method in three canonical data sharing problems and show that it relaxes key assumptions made by prior work. Empirically, we demonstrate that our mechanism incentivizes truthful data sharing via simulations and on real-world language and image data.
Authors: Boya Xiong, Shuo Wang, Weifeng Ge, Guanhua Chen, Yun Chen
Abstract: Supervised Fine-Tuning (SFT) empowers Large Language Models (LLMs) with exceptional performance on specialized tasks, but it yields dense, high-dimensional delta parameters that pose severe storage and distribution challenges. Singular Value Decomposition (SVD)-based compression offers a compact representation for such delta parameters, but existing methods adopt heuristic quantization without clarifying underlying mechanisms, leading to poor generalizability. In this work, we propose PrinMix, a rigorous SVD-based framework that models quantization as an optimization problem, grounding the design in mathematical mechanisms. We first theoretically derive quantization error and identify a key singular-value-dominated scaling mechanism, which mathematically proves the necessity of mix-precision quantization. We then model the quantization scheme as a 0/1 Integer Linear Programming (ILP) problem, which yields optimal bit-budget-constrained solutions without empirical assumptions. Furthermore, PrinMix integrates a Reconstruction Target Correction (RTC) method to compensate for errors from the $\mathbf{V}$-then-$\mathbf{U}$ sequential quantization process. Extensive experiments confirm PrinMix performs well: for 7B LLMs, PrinMix outperforms SOTA Delta-CoMe on challenging benchmarks by 22.3% on AIME2024 and 6.1% on GQA.
Authors: Xingyue Huang, Mikhail Galkin, Michael M. Bronstein, \.Ismail \.Ilkan Ceylan
Abstract: Inductive link prediction with knowledge hypergraphs is the task of predicting missing hyperedges involving completely novel entities (i.e., nodes unseen during training). Existing methods for inductive link prediction with knowledge hypergraphs assume a fixed relational vocabulary and, as a result, cannot generalize to knowledge hypergraphs with novel relation types (i.e., relations unseen during training). Inspired by knowledge graph foundation models, we propose HYPER as a foundation model for link prediction, which can generalize to any knowledge hypergraph, including novel entities and novel relations. Importantly, HYPER can learn and transfer across different relation types of varying arities, by encoding the entities of each hyperedge along with their respective positions in the hyperedge. To evaluate HYPER, we construct 16 new inductive datasets from existing knowledge hypergraphs, covering a diverse range of relation types of varying arities. Empirically, HYPER consistently outperforms all existing methods in both node-only and node-and-relation inductive settings, showing strong generalization to unseen, higher-arity relational structures.
Authors: Hen Davidov, Shai Feldman, Gilad Freidkin, Yaniv Romano
Abstract: We introduce time-to-unsafe-sampling, a novel safety measure for generative models, defined as the number of generations required by a large language model (LLM) to trigger an unsafe (e.g., toxic) response. While providing a new dimension for prompt-adaptive safety evaluation, quantifying time-to-unsafe-sampling is challenging: unsafe outputs are often rare in well-aligned models and thus may not be observed under any feasible sampling budget. To address this challenge, we frame this estimation problem as one of survival analysis. We build on recent developments in conformal prediction and propose a novel calibration technique to construct a lower predictive bound (LPB) on the time-to-unsafe-sampling of a given prompt with rigorous coverage guarantees. Our key technical innovation is an optimized sampling-budget allocation scheme that improves sample efficiency while maintaining distribution-free guarantees. Experiments on both synthetic and real data support our theoretical results and demonstrate the practical utility of our method for safety risk assessment in generative AI models.
Authors: Hanyu Pei, Jing-Xiao Liao, Qibin Zhao, Ting Gao, Shijun Zhang, Xiaoge Zhang, Feng-Lei Fan
Abstract: Drawing inspiration from our human brain that designs different neurons for different tasks, recent advances in deep learning have explored modifying a network's neurons to develop so-called task-driven neurons. Prototyping task-driven neurons (referred to as NeuronSeek) employs symbolic regression (SR) to discover the optimal neuron formulation and construct a network from these optimized neurons. Along this direction, this work replaces symbolic regression with tensor decomposition (TD) to discover optimal neuronal formulations, offering enhanced stability and faster convergence. Furthermore, we establish theoretical guarantees that modifying the aggregation functions with common activation functions can empower a network with a fixed number of parameters to approximate any continuous function with an arbitrarily small error, providing a rigorous mathematical foundation for the NeuronSeek framework. Extensive empirical evaluations demonstrate that our NeuronSeek-TD framework not only achieves superior stability, but also is competitive relative to the state-of-the-art models across diverse benchmarks. The code is available at https://github.com/HanyuPei22/NeuronSeek.
Authors: Fabio Merizzi, Harilaos Loukos
Abstract: Global Climate Models (GCMs) are critical for simulating large-scale climate dynamics, but their coarse spatial resolution limits their applicability in regional studies. Regional Climate Models (RCMs) address this limitation through dynamical downscaling, albeit at considerable computational cost and with limited flexibility. Deep learning has emerged as an efficient data-driven alternative; however, most existing approaches focus on single-variable models that downscale one variable at a time. This paradigm can lead to redundant computation, limited contextual awareness, and weak cross-variable interactions.To address these limitations, we propose a multi-variable Vision Transformer (ViT) architecture with a shared encoder and variable-specific decoders (1EMD). The proposed model jointly predicts six key climate variables: surface temperature, wind speed, 500 hPa geopotential height, total precipitation, surface downwelling shortwave radiation, and surface downwelling longwave radiation, directly from GCM-resolution inputs, emulating RCM-scale downscaling over Europe. Compared to single-variable ViT models, the 1EMD architecture improves performance across all six variables, achieving an average MSE reduction of approximately 5.5% under a fair and controlled comparison. It also consistently outperforms alternative multi-variable baselines, including a single-decoder ViT and a multi-variable U-Net. Moreover, multi-variable models substantially reduce computational cost, yielding a 29-32% lower inference time per variable compared to single-variable approaches. Overall, our results demonstrate that multi-variable modeling provides systematic advantages for high-resolution climate downscaling in terms of both accuracy and efficiency. Among the evaluated architectures, the proposed 1EMD ViT achieves the most favorable trade-off between predictive performance and computational cost.
Authors: Yuta Sato, Kazuhiko Kawamoto, Hiroshi Kera
Abstract: The chain of thought, i.e., step-by-step reasoning, is one of the fundamental mechanisms of Transformers. While the design of intermediate reasoning steps has been extensively studied and shown to critically influence performance on mathematical, multi-step reasoning tasks, the ordering of these steps has received little attention, despite its significant effect on the difficulty of reasoning. This study addresses a novel task of unraveling the chain of thought -- reordering decoder input tokens into a learning-friendly sequence for Transformers, for learning arithmetic tasks. The proposed pipeline first trains a Transformer on a mixture of target sequences arranged in different orders and then identifies benign orders as those with fast loss drops in the early stage. As the search space grows factorially in sequence length, we propose a two-stage hierarchical approach for inter- and intra-block reordering. Experiments on seven order-sensitive arithmetic tasks show that our method identifies a learning-friendly order out of a few billion candidates. Notably, it recovered the reverse-digit order reported in prior studies for the multiplication task.
Authors: Xiaohang Tang, Rares Dolga, Sangwoong Yoon, Ilija Bogunovic
Abstract: Improving the reasoning capabilities of diffusion-based large language models (dLLMs) through reinforcement learning (RL) remains an open problem. The intractability of dLLMs likelihood function necessitates approximating the current, old, and reference policy likelihoods at each policy optimization step. This reliance introduces additional computational overhead, and can lead to large variance and estimation error in RL objective -- particularly in computing the policy ratio for importance sampling. To mitigate these issues, we introduce wd1, a novel ratio-free policy optimization approach that reformulates the RL objective as a weighted log-likelihood, requiring only a single approximation for the current parametrized policy likelihood. We formally show that our proposed method can be interpreted as energy-guided discrete diffusion training combined with negative sample unlearning, thereby confirming its theoretical soundness. In experiments on LLaDA-8B model, wd1 outperforms diffusion-based GRPO (d1) while requiring lower computational cost, achieving up to a $+59\%$ improvement in accuracy. Furthermore, we extend wd1 to denoising-stepwise weighted policy optimization (wd1++), achieving state-of-the-art math performance of $44.2\%$ on MATH500 and $84.5\%$ on GSM8K with only 20 RL training steps.
Authors: Yuxi Liu, Konpat Preechakul, Kananart Kuwaranancharoen, Yutong Bai
Abstract: While machine learning has advanced through massive parallelization, we identify a critical blind spot: some problems are fundamentally sequential. These "inherently serial" problems-from mathematical reasoning to physical simulations to sequential decision-making-require sequentially dependent computational steps that cannot be efficiently parallelized. We formalize this distinction in complexity theory, and demonstrate that current parallel-centric architectures face fundamental limitations on such tasks. Then, we show for first time that diffusion models despite their sequential nature are incapable of solving inherently serial problems. We argue that recognizing the serial nature of computation holds profound implications on machine learning, model design, and hardware development.
Authors: Joshua Dimasaka, Christian Gei{\ss}, Emily So
Abstract: To understand our global progress for sustainable development and disaster risk reduction in many developing economies, two recent major initiatives - the Uniform African Exposure Dataset of the Global Earthquake Model (GEM) Foundation and the Modelling Exposure through Earth Observation Routines (METEOR) Project - implemented classical spatial disaggregation techniques to generate large-scale mapping of urban morphology using the information from various satellite imagery and its derivatives, geospatial datasets of the built environment, and subnational census statistics. However, the local discrepancy with well-validated census statistics and the propagated model uncertainties remain a challenge in such coarse-to-fine-grained mapping problems, specifically constrained by weak and conditional label supervision. Therefore, we present Deep Conditional Census-Constrained Clustering (DeepC4), a novel deep learning-based spatial disaggregation approach that incorporates local census statistics as cluster-level constraints while considering multiple conditional label relationships in a joint multitask learning of the patterns of satellite imagery. To demonstrate, compared to GEM and METEOR, we enhanced the quality of Rwandan maps of urban morphology, specifically building exposure and physical vulnerability, at the third-level administrative unit from the 2022 census. As the world approaches the conclusion of many global frameworks in 2030, our work offers a new deep learning-based mapping technique that explicitly encodes well-validated census and experts' belief systems to achieve an explainable and interpretable auditing of existing coarse-grained derived information at large scales.
Authors: Xuan Liu, Siru Ouyang, Xianrui Zhong, Jiawei Han, Huimin Zhao
Abstract: Large language models (LLMs) have gained significant attention in chemistry. However, most existing datasets center on molecular-level property prediction and overlook the role of fine-grained functional group (FG) information. Incorporating FG-level data can provide valuable prior knowledge that links molecular structures with textual descriptions, which can be used to build more interpretable, structure-aware LLMs for reasoning on molecule-related tasks. Moreover, LLMs can learn from such fine-grained information to uncover hidden relationships between specific functional groups and molecular properties, thereby advancing molecular design and drug discovery. Here, we introduce FGBench, a dataset comprising 625K molecular property reasoning problems with functional group information. Functional groups are precisely annotated and localized within the molecule, which ensures the dataset's interoperability thereby facilitating further multimodal applications. FGBench includes both regression and classification tasks on 245 different functional groups across three categories for molecular property reasoning: (1) single functional group impacts, (2) multiple functional group interactions, and (3) direct molecular comparisons. In the benchmark of state-of-the-art LLMs on 7K curated data, the results indicate that current LLMs struggle with FG-level property reasoning, highlighting the need to enhance reasoning capabilities in LLMs for chemistry tasks. We anticipate that the methodology employed in FGBench to construct datasets with functional group-level information will serve as a foundational framework for generating new question-answer pairs, enabling LLMs to better understand fine-grained molecular structure-property relationships. The dataset and evaluation code are available at https://github.com/xuanliugit/FGBench.
Authors: Zhaomin Wu, Mingzhe Du, See-Kiong Ng, Bingsheng He
Abstract: Large Language Models (LLMs) are widely deployed in reasoning, planning, and decision-making tasks, making their trustworthiness critical. A significant and underexplored risk is intentional deception, where an LLM deliberately fabricates or conceals information to serve a hidden objective. Existing studies typically induce deception by explicitly setting a hidden objective through prompting or fine-tuning, which may not reflect real-world human-LLM interactions. Moving beyond such human-induced deception, we investigate LLMs' self-initiated deception on benign prompts. To address the absence of ground truth, we propose a framework based on Contact Searching Questions~(CSQ). This framework introduces two statistical metrics derived from psychological principles to quantify the likelihood of deception. The first, the Deceptive Intention Score, measures the model's bias toward a hidden objective. The second, the Deceptive Behavior Score, measures the inconsistency between the LLM's internal belief and its expressed output. Evaluating 16 leading LLMs, we find that both metrics rise in parallel and escalate with task difficulty for most models. Moreover, increasing model capacity does not always reduce deception, posing a significant challenge for future LLM development.
Authors: Md Sultanul Arifin, Abu Nowshed Sakib, Yeasir Rayhan, Tanzima Hashem
Abstract: Lightning, a common feature of severe meteorological conditions, poses significant risks, from direct human injuries to substantial economic losses. These risks are further exacerbated by climate change. Early and accurate prediction of lightning would enable preventive measures to safeguard people, protect property, and minimize economic losses. In this paper, we present DeepLight, a novel deep learning architecture for predicting lightning occurrences. Existing prediction models face several critical limitations: i) they often struggle to capture the dynamic spatial context and the inherent randomness of lightning events, including whether lightning occurs and its variability in location and timing even under similar meteorological conditions; ii) they underutilize key observational data, such as radar reflectivity and cloud properties; and iii) they rely heavily on Numerical Weather Prediction (NWP) systems, which are both computationally expensive and highly sensitive to parameter settings. To overcome these challenges, DeepLight leverages multi-source meteorological data, including radar reflectivity, cloud properties, and historical lightning occurrences through a dual-encoder architecture. By employing multi-branch convolution techniques, it dynamically captures spatial correlations across varying extents. Furthermore, its novel Hazy Loss function explicitly addresses the spatio-temporal uncertainty of lightning by penalizing deviations based on proximity to true events, enabling the model to better learn patterns amidst randomness. Extensive experiments show that DeepLight improves the Equitable Threat Score (ETS) by 18\%--30\% over state-of-the-art methods, establishing it as a robust solution for lightning prediction.
Authors: Tamim Ahmed, Monowar Hasan
Abstract: By offering a dynamic, real-time virtual representation of physical systems, digital twin technology can enhance data-driven decision-making in digital agriculture. Our research shows how digital twins are useful for detecting inconsistencies in agricultural weather data measurements, which are key attributes for various agricultural decision-making and automation tasks. We develop a modular framework named Cerealia that allows end-users to check for data inconsistencies when perfect weather feeds are unavailable. Cerealia uses neural network models to check anomalies and aids end-users in informed decision-making. We develop a prototype of Cerealia using the NVIDIA Jetson Orin platform and test it with an operational weather network established in a commercial orchard as well as publicly available weather datasets.
Authors: Laura L\"utzow, Michael Eichelbeck, Mykel J. Kochenderfer, Matthias Althoff
Abstract: Conformal prediction is a popular uncertainty quantification method that augments a base predictor to return sets of predictions with statistically valid coverage guarantees. However, current methods are often computationally expensive and data-intensive, as they require constructing an uncertainty model before calibration. Moreover, existing approaches typically represent the prediction sets with intervals, which limits their ability to capture dependencies in multi-dimensional outputs. We address these limitations by introducing zono-conformal prediction, a novel approach inspired by interval predictor models and reachset-conformant identification that constructs prediction zonotopes with assured coverage. By placing zonotopic uncertainty sets directly into the model of the base predictor, zono-conformal predictors can be identified via a single, data-efficient linear program. While we can apply zono-conformal prediction to arbitrary nonlinear base predictors, we focus on feed-forward neural networks in this work. Aside from regression tasks, we also construct optimal zono-conformal predictors in classification settings where the output of an uncertain predictor is a set of possible classes. We provide probabilistic coverage guarantees and present methods for detecting outliers in the identification data. In extensive numerical experiments, we show that zono-conformal predictors are less conservative than interval predictor models and standard conformal prediction methods, while achieving a similar coverage over the test data.
Authors: Jeremy Carleton, Debajoy Mukherjee, Srinivas Shakkottai, Dileep Kalathil
Abstract: Large Language Models (LLMs) are increasingly deployed across diverse applications that demand balancing multiple, often conflicting, objectives -- such as helpfulness, harmlessness, or humor. Many traditional methods for aligning outputs to user-specific preferences require fine-tuning models for each objective or for specific preference configurations, which is computationally expensive and inflexible. We introduce \textbf{MAVIS} -- \textit{Multi-Objective Alignment via Inference-Time Value-Guided Selection} -- a lightweight inference-time alignment framework that enables dynamic control over LLM behavior without modifying the base model's weights. MAVIS trains a set of small value models, each corresponding to a distinct objective. At inference time, these value models are combined using user-specified weights to produce a tilting function that adjusts the base model's output distribution toward desired trade-offs. The value models are trained using a simple iterative algorithm that enables monotonic improvement of the KL-regularized policy. We show empirically that MAVIS achieves a superior pareto front compared to baselines which fine-tune per-objective models and combine them post hoc or train a single preference-conditioned value model for guidance. Our code is available at https://github.com/5-Jeremy/MAVIS/tree/main.
Authors: Zayd M. K. Zuhri, Erland Hilman Fuadi, Alham Fikri Aji
Abstract: Multi-token prediction (MTP) has been proposed as an auxiliary objective to improve next-token prediction (NTP) in language model training but shows inconsistent improvements, underperforming in standard NLP benchmarks. We found MTP's exact future token prediction to be too difficult as an auxiliary loss. Instead, we propose token order prediction (TOP), which trains models to order upcoming tokens by their proximity using a learning-to-rank loss. TOP requires only a single additional unembedding layer compared to MTP's multiple transformer layers. We pretrain models of 340M, 1.8B, and 7B parameters using NTP, MTP, DeepSeek MTP (DS-MTP) and TOP objectives. The results of nine standard NLP benchmarks show that TOP overall outperforms NTP, MTP, and DS-MTP even at scale. TOP models with continued training on math and code also perform better on 4 relevant benchmarks. On the synthetic star graph task, TOP enables pathfinding on graphs where NTP, MTP, and DS-MTP fail. Our code is available at https://github.com/zaydzuhri/token-order-prediction
Authors: Arjun Basandrai, Shourya Jain, K. Ilanthenral
Abstract: Traditional resampling methods for handling class imbalance typically uses fixed distributions, undersampling the majority or oversampling the minority. These static strategies ignore changes in class-wise learning difficulty, which can limit the overall performance of the model. This paper proposes an Adaptive Resampling-based Training (ART) method that periodically updates the distribution of the training data based on the class-wise performance of the model. Specifically, ART uses class-wise macro F1 scores, computed at fixed intervals, to determine the degree of resampling to be performed. Unlike instance-level difficulty modeling, which is noisy and outlier-sensitive, ART adapts at the class level. This allows the model to incrementally shift its attention towards underperforming classes in a way that better aligns with the optimization objective. Results on diverse benchmarks, including Pima Indians Diabetes and Yeast dataset demonstrate that ART consistently outperforms both resampling-based and algorithm-level methods, including Synthetic Minority Oversampling Technique (SMOTE), NearMiss Undersampling, and Cost-sensitive Learning on binary as well as multi-class classification tasks with varying degrees of imbalance. In most settings, these improvements are statistically significant. On tabular datasets, gains are significant under paired t-tests and Wilcoxon tests (p < 0.05), while results on text and image tasks remain favorable. Compared to training on the original imbalanced data, ART improves macro F1 by an average of 2.64 percentage points across all tested tabular datasets. Unlike existing methods, whose performance varies by task, ART consistently delivers the strongest macro F1, making it a reliable choice for imbalanced classification.
Authors: Minh Vu, Konstantinos Slavakis
Abstract: This paper introduces a structured and interpretable online policy-iteration framework for reinforcement learning (RL), built around the novel class of sparse Gaussian mixture model Q-functions (S-GMM-QFs). Extending earlier work that trained GMM-QFs offline, the proposed framework develops an online scheme that leverages streaming data to encourage exploration. Model complexity is regulated through sparsification by Hadamard overparametrization, which mitigates overfitting while preserving expressiveness. The parameter space of S-GMM-QFs is naturally endowed with a Riemannian manifold structure, allowing for principled parameter updates via online gradient descent on a smooth objective. Numerical experiments show that S-GMM-QFs match or even outperform dense deep RL (DeepRL) methods on standard benchmarks while using significantly fewer parameters. Moreover, they maintain strong performance even in low-parameter regimes where sparsified DeepRL methods fail to generalize.
Authors: Kaiwen Zheng, Huayu Chen, Haotian Ye, Haoxiang Wang, Qinsheng Zhang, Kai Jiang, Hang Su, Stefano Ermon, Jun Zhu, Ming-Yu Liu
Abstract: Online reinforcement learning (RL) has been central to post-training language models, but its extension to diffusion models remains challenging due to intractable likelihoods. Recent works discretize the reverse sampling process to enable GRPO-style training, yet they inherit fundamental drawbacks, including solver restrictions, forward-reverse inconsistency, and complicated integration with classifier-free guidance (CFG). We introduce Diffusion Negative-aware FineTuning (DiffusionNFT), a new online RL paradigm that optimizes diffusion models directly on the forward process via flow matching. DiffusionNFT contrasts positive and negative generations to define an implicit policy improvement direction, naturally incorporating reinforcement signals into the supervised learning objective. This formulation enables training with arbitrary black-box solvers, eliminates the need for likelihood estimation, and requires only clean images rather than sampling trajectories for policy optimization. DiffusionNFT is up to $25\times$ more efficient than FlowGRPO in head-to-head comparisons, while being CFG-free. For instance, DiffusionNFT improves the GenEval score from 0.24 to 0.98 within 1k steps, while FlowGRPO achieves 0.95 with over 5k steps and additional CFG employment. By leveraging multiple reward models, DiffusionNFT significantly boosts the performance of SD3.5-Medium in every benchmark tested.
Authors: Binghui Li, Fengling Chen, Zixun Huang, Lean Wang, Lei Wu
Abstract: Scaling laws have emerged as a unifying lens for understanding and guiding the training of large language models (LLMs). However, existing studies predominantly focus on the final-step loss, leaving open whether the entire loss dynamics obey similar laws and, crucially, how the learning rate schedule (LRS) shapes them. We address these gaps in a controlled theoretical setting by analyzing stochastic gradient descent (SGD) on a power-law kernel regression model. The key insight is a novel intrinsic-time viewpoint, which captures the training progress more faithfully than iteration count. We then establish a Functional Scaling Law (FSL) that captures the full loss trajectory under arbitrary LRSs, with the schedule's influence entering through a simple convolutional functional. We further instantiate the theory for three representative LRSs -- constant, exponential decay, and warmup-stable-decay (WSD) -- and derive explicit scaling relations in both data- and compute-limited regimes. These comparisons explain key empirical phenomena: (i) higher-capacity models are more data- and compute-efficient; (ii) learning-rate decay improves training efficiency; and (iii) WSD-type schedules outperform pure decay. Finally, experiments on LLMs ranging from 0.1B to 1B parameters demonstrate the practical relevance of FSL as a surrogate model for fitting and predicting loss trajectories in large-scale pre-training.
Authors: Rohan Chauhan, Ioannis Panageas
Abstract: We consider the problem of estimating inverse temperature parameter $\beta$ of an $n$-dimensional truncated Ising model using a single sample. Given a graph $G = (V,E)$ with $n$ vertices, a truncated Ising model is a probability distribution over the $n$-dimensional hypercube $\{-1,1\}^n$ where each configuration $\mathbf{\sigma}$ is constrained to lie in a truncation set $S \subseteq \{-1,1\}^n$ and has probability $\Pr(\mathbf{\sigma}) \propto \exp(\beta\mathbf{\sigma}^\top A\mathbf{\sigma})$ with $A$ being the adjacency matrix of $G$. We adopt the recent setting of [Galanis et al. SODA'24], where the truncation set $S$ can be expressed as the set of satisfying assignments of a $k$-SAT formula. Given a single sample $\mathbf{\sigma}$ from a truncated Ising model, with inverse parameter $\beta^*$, underlying graph $G$ of bounded degree $\Delta$ and $S$ being expressed as the set of satisfying assignments of a $k$-SAT formula, we design in nearly $O(n)$ time an estimator $\hat{\beta}$ that is $O(\Delta^3/\sqrt{n})$-consistent with the true parameter $\beta^*$ for $k \gtrsim \log(d^2k)\Delta^3.$ Our estimator is based on the maximization of the pseudolikelihood, a notion that has received extensive analysis for various probabilistic models without [Chatterjee, Annals of Statistics '07] or with truncation [Galanis et al. SODA '24]. Our approach generalizes recent techniques from [Daskalakis et al. STOC '19, Galanis et al. SODA '24], to confront the more challenging setting of the truncated Ising model.
Authors: Anton Korznikov, Andrey Galichin, Alexey Dontsov, Oleg Y. Rogov, Ivan Oseledets, Elena Tutubalina
Abstract: Activation steering is a promising technique for controlling LLM behavior by adding semantically meaningful vectors directly into a model's hidden states during inference. It is often framed as a precise, interpretable, and potentially safer alternative to fine-tuning. We demonstrate the opposite: steering systematically breaks model alignment safeguards, making it comply with harmful requests. Through extensive experiments on different model families, we show that even steering in a random direction can increase the probability of harmful compliance from 0% to 1-13%. Alarmingly, steering benign features from a sparse autoencoder (SAE), a common source of interpretable directions, demonstrates a comparable harmful potential. Finally, we show that combining 20 randomly sampled vectors that jailbreak a single prompt creates a universal attack, significantly increasing harmful compliance on unseen requests. These results challenge the paradigm of safety through interpretability, showing that precise control over model internals does not guarantee precise control over model behavior.
Authors: Aman Gupta, Rafael Celente, Abhishek Shivanna, D. T. Braithwaite, Gregory Dexter, Shao Tang, Hiroto Udagawa, Daniel Silva, Rohan Ramanath, S. Sathiya Keerthi
Abstract: The Muon optimizer, based on matrix orthogonalization, has recently shown faster convergence and better computational efficiency over AdamW in LLM pre-training. However, the memory overhead of maintaining high-precision optimizer states remains a challenge for large-scale deployment. In this paper, we introduce the 8-bit Muon optimizer using blockwise quantization. In extensive Chinchilla-optimal experiments on pre-training models of up to 2.7B in size and fine-tuning them for instruction following, we demonstrate that 8-bit Muon achieves parity with Muon in terms of validation loss and downstream benchmarks, while achieving up to a 62\% reduction in optimizer state footprint. Crucially, we show that Muon's update mechanism is uniquely compatible with a simple linear quantization scheme, bypassing the complex dynamic scaling required for quantized AdamW. We supplement our empirical findings with a theoretical analysis of Muon's robustness to quantization noise.
Authors: Steve Hong, Runa Eschenhagen, Bruno Mlodozeniec, Richard Turner
Abstract: Influence functions offer a principled way to trace model predictions back to training data, but their use in deep learning is hampered by the need to invert a large, ill-conditioned Hessian matrix. Approximations such as Generalised Gauss-Newton (GGN) and Kronecker-Factored Approximate Curvature (K-FAC) have been proposed to make influence computation tractable, yet it remains unclear how the departure from exactness impacts data attribution performance. Critically, given the restricted regime in which influence functions are derived, it is not necessarily clear better Hessian approximations should even lead to better data attribution performance. In this paper, we investigate the effect of Hessian approximation quality on influence-function attributions in a controlled classification setting. Our experiments show that better Hessian approximations consistently yield better influence score quality, offering justification for recent research efforts towards that end. We further decompose the approximation steps for recent Hessian approximation methods and evaluate each step's influence on attribution accuracy. Notably, the mismatch between K-FAC eigenvalues and GGN/EK-FAC eigenvalues accounts for the majority of the error and influence loss. These findings highlight which approximations are most critical, guiding future efforts to balance computational tractability and attribution accuracy.
Authors: Zhaomin Wu, Haodong Zhao, Ziyang Wang, Jizhou Guo, Qian Wang, Bingsheng He
Abstract: The explosive growth of large language models (LLMs) has created a vast but opaque landscape: millions of models exist, yet their evolutionary relationships through fine-tuning, distillation, or adaptation are often undocumented or unclear, complicating LLM management. Existing methods are limited by task specificity, fixed model sets, or strict assumptions about tokenizers or architectures. Inspired by biological DNA, we address these limitations by mathematically defining LLM DNA as a low-dimensional, bi-Lipschitz representation of functional behavior. We prove that LLM DNA satisfies inheritance and genetic determinism properties and establish the existence of DNA. Building on this theory, we derive a general, scalable, training-free pipeline for DNA extraction. In experiments across 305 LLMs, DNA aligns with prior studies on limited subsets and achieves superior or competitive performance on specific tasks. Beyond these tasks, DNA comparisons uncover previously undocumented relationships among LLMs. We further construct the evolutionary tree of LLMs using phylogenetic algorithms, which align with shifts from encoder-decoder to decoder-only architectures, reflect temporal progression, and reveal distinct evolutionary speeds across LLM families.
Authors: Patrick Langer, Thomas Kaar, Max Rosenblattl, Maxwell A. Xu, Winnie Chow, Martin Maritsch, Robert Jakob, Ning Wang, Juncheng Liu, Aradhana Verma, Brian Han, Daniel Seung Kim, Henry Chubb, Scott Ceresnak, Aydin Zahedivash, Alexander Tarlochan Singh Sandhu, Fatima Rodriguez, Daniel McDuff, Elgar Fleisch, Oliver Aalami, Filipe Barata, Paul Schmiedmayer
Abstract: LLMs have emerged as powerful tools for interpreting multimodal data. In medicine, they hold particular promise for synthesizing large volumes of clinical information into actionable insights and digital health applications. Yet, a major limitation remains their inability to handle time series. To overcome this gap, we present OpenTSLM, a family of Time Series Language Models (TSLMs) created by integrating time series as a native modality to pretrained LLMs, enabling reasoning over multiple time series of any length. We investigate two architectures for OpenTSLM. The first, OpenTSLM-SoftPrompt, models time series implicitly by concatenating learnable time series tokens with text tokens via soft prompting. Although parameter-efficient, we hypothesize that explicit time series modeling scales better and outperforms implicit approaches. We thus introduce OpenTSLM-Flamingo, which integrates time series with text via cross-attention. We benchmark both variants against baselines that treat time series as text tokens or plots, across a suite of text-time-series Chain-of-Thought (CoT) reasoning tasks. We introduce three datasets: HAR-CoT, Sleep-CoT, and ECG-QA-CoT. Across all, OpenTSLM models outperform baselines, reaching 69.9 F1 in sleep staging and 65.4 in HAR, compared to 9.05 and 52.2 for finetuned text-only models. Notably, even 1B-parameter OpenTSLM models surpass GPT-4o (15.47 and 2.95). OpenTSLM-Flamingo matches OpenTSLM-SoftPrompt in performance and outperforms on longer sequences, while maintaining stable memory requirements. By contrast, SoftPrompt grows exponentially in memory with sequence length, requiring around 110 GB compared to 40 GB VRAM when training on ECG-QA with LLaMA-3B. Expert reviews by clinicians find strong reasoning capabilities exhibited by OpenTSLMs on ECG-QA. To facilitate further research, we provide all code, datasets, and models open-source.
Authors: Steve Hong, Samuel Belkadi
Abstract: We reinterpret Visual Autoregressive (VAR) models as iterative refinement models to identify which design choices drive their quality-efficiency trade-off. Instead of treating VAR only as next-scale autoregression, we formalise it as a deterministic forward process that builds a Laplacian-style latent pyramid, together with a learned backward process that reconstructs samples in a small number of coarse-to-fine steps. This formulation makes the link to denoising diffusion explicit and highlights three modelling choices that may underlie VAR's efficiency and sample quality: refinement in a learned latent space, discrete prediction over code indices, and decomposition by spatial frequency. We support this view with controlled experiments that isolate the contribution of each factor to quality and speed. We also discuss how the same framework can be adapted to permutation-invariant graph generation and probabilistic medium-range weather forecasting, and how it provides practical points of contact with diffusion methods while preserving few-step, scale-parallel generation.
Authors: Junyi Yao, Parham Eftekhar, Gene Cheung, Xujin Chris Liu, Yao Wang, Wei Hu
Abstract: Samples of brain signals collected by EEG sensors have inherent anti-correlations that are well modeled by negative edges in a finite graph. To differentiate epilepsy patients from healthy subjects using collected EEG signals, we build lightweight and interpretable transformer-like neural nets by unrolling a spectral denoising algorithm for signals on a balanced signed graph -- graph with no cycles of odd number of negative edges. A balanced signed graph has well-defined frequencies that map to a corresponding positive graph via similarity transform of the graph Laplacian matrices. We implement an ideal low-pass filter efficiently on the mapped positive graph via Lanczos approximation, where the optimal cutoff frequency is learned from data. Given that two balanced signed graph denoisers learn posterior probabilities of two different signal classes during training, we evaluate their reconstruction errors for binary classification of EEG signals. Experiments show that our method achieves classification performance comparable to representative deep learning schemes, while employing dramatically fewer parameters.
Authors: Yukun Zhang, Xueqing Zhou
Abstract: Transformers enable powerful content-based global routing via self-attention, but they lack an explicit local geometric prior along the sequence axis. As a result, the placement of locality-inducing modules in hybrid architectures has largely been empirical. We study a simple deterministic PDE diffusion layer implemented as one explicit Euler step of one-dimensional heat smoothing using a discrete Neumann Laplacian under a spectral stability constraint, and ask a structural question: where should diffusion be inserted relative to attention? Our central claim is that diffusion and attention generally do not commute, so inserting the same local operator before versus after attention leads to qualitatively different behaviors. We develop a three-layer operator-theoretic framework that (1) establishes unconditional guarantees for the diffusion subsystem, including spectral non-expansiveness and monotone Dirichlet-energy dissipation when the diffusion step size is smaller than one half, (2) derives compositional perturbation bounds linking insertion effects to representation roughness and downstream amplification, and (3) uses diffusion-attention non-commutativity as a diagnostic for structural double-mixing conflicts. Guided by theory, we evaluate seven insertion positions on the Long Range Arena benchmark. Early diffusion acts as effective pre-regularization, improving average accuracy by 4.1 percentage points when applied after embedding, while post-attention diffusion degrades performance by 2.5 percentage points, consistent with the predicted conflict. A multi-scale diffusion variant yields consistent gains under the same global stability constraint. Our analysis provides a general template for reasoning about local-global compositions in sequence models by separating provable guarantees, compositional bounds, and mechanistic diagnostics.
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: 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: Sahil Joshi, Agniva Chowdhury, Amar Kanakamedala, Ekam Singh, Evan Tu, Anshumali Shrivastava
Abstract: Softmax Attention has a quadratic time complexity in sequence length, which becomes prohibitive to run at long contexts, even with highly optimized GPU kernels. For example, FlashAttention-2/3 (exact, GPU-optimized implementations of Softmax Attention) cannot complete a single forward-backward pass of a single attention layer once the context exceeds ~4 million tokens on an NVIDIA GH200 (96 GB). We introduce Repeated Arrays-of-Count Estimators (RACE) Attention, a kernel-inspired alternative to Softmax Attention that is strictly linear in sequence length and embedding size. RACE Attention replaces the exponential kernel with a sharpened angular similarity, and approximates attention outputs via Gaussian random projections and soft Locality-Sensitive Hashing (LSH), avoiding construction of the full attention matrix. Across language modeling, masked language modeling, and text/image classification, RACE Attention matches or outperforms strong baselines up to 64K seqeuence length while reducing wall-clock time and memory usage. In addition, we conduct a controlled scaling study on a single attention layer and demonstrate processing of up to 12 million tokens on an NVIDIA GH200 GPU and 75 million tokens on an Intel Xeon Gold 5220R CPU in a single forward-backward pass, which is well beyond the capabilities of current state-of-the-art attention implementations. RACE Attention thus offers a practical and theoretically grounded mechanism for long-context training on today's hardware. We release our code at https://github.com/sahiljoshi515/RACE_Attention.
Authors: Seohong Park, Deepinder Mann, Sergey Levine
Abstract: In this work, we introduce dual goal representations for goal-conditioned reinforcement learning (GCRL). A dual goal representation characterizes a state by "the set of temporal distances from all other states"; in other words, it encodes a state through its relations to every other state, measured by temporal distance. This representation provides several appealing theoretical properties. First, it depends only on the intrinsic dynamics of the environment and is invariant to the original state representation. Second, it contains provably sufficient information to recover an optimal goal-reaching policy, while being able to filter out exogenous noise. Based on this concept, we develop a practical goal representation learning method that can be combined with any existing GCRL algorithm. Through diverse experiments on the OGBench task suite, we empirically show that dual goal representations consistently improve offline goal-reaching performance across 20 state- and pixel-based tasks.
Authors: Patrick Peixuan Ye, Chen Shani, Ellen Vitercik
Abstract: We introduce Bridged Clustering, a semi-supervised framework to learn predictors from any unpaired input $X$ and output $Y$ dataset. Our method first clusters $X$ and $Y$ independently, then learns a sparse, interpretable bridge between clusters using only a few paired examples. At inference, a new input $x$ is assigned to its nearest input cluster, and the centroid of the linked output cluster is returned as the prediction $\hat{y}$. Unlike traditional SSL, Bridged Clustering explicitly leverages output-only data, and unlike dense transport-based methods, it maintains a sparse and interpretable alignment. Through theoretical analysis, we show that with bounded mis-clustering and mis-bridging rates, our algorithm becomes an effective and efficient predictor. Empirically, our method is competitive with SOTA methods while remaining simple, model-agnostic, and highly label-efficient in low-supervision settings.
Authors: Aadithya Srikanth, Mudit Gaur, 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, remain significantly less understood from a theoretical standpoint. In particular, all existing analyses of discrete-state models assume score estimation error bounds without studying sample complexity results. In this work, we present a principled theoretical framework for discrete-state diffusion, providing the first sample complexity bound of $\widetilde{\mathcal{O}}(\epsilon^{-2})$. Our structured decomposition of the score estimation error into statistical, approximation, optimization, and clipping components offers critical insights into how discrete-state models can be trained efficiently. This analysis addresses a fundamental gap in the literature and establishes the theoretical tractability and practical relevance of discrete-state diffusion models.
Authors: Marcel Meyer, Sascha Kaltenpoth, Kevin Zalipski, Oliver M\"uller
Abstract: Time Series Foundation Models (TSFMs) represent a new paradigm for time-series forecasting, promising zero-shot predictions without the need for task-specific training or fine-tuning. However, similar to Large Language Models (LLMs), the evaluation of TSFMs is challenging: as training corpora grow increasingly large, it becomes difficult to ensure the integrity of the test sets used for benchmarking. Our investigation of existing TSFM evaluation studies identifies two kinds of information leakage: (1) train-test sample overlaps arising from the multi-purpose reuse of datasets and (2) temporal overlap of correlated train and test series. Ignoring these forms of information leakage when benchmarking TSFMs risks producing overly optimistic performance estimates that fail to generalize to real-world settings. We therefore argue for the development of novel evaluation methodologies that avoid pitfalls already observed in both LLM and classical time-series benchmarking, and we call on the research community to adopt principled approaches to safeguard the integrity of TSFM evaluation.
Authors: Huipeng Huang, Wenbo Liao, Huajun Xi, Hao Zeng, Mengchen Zhao, Hongxin Wei
Abstract: Obtaining high-quality labels for large datasets is expensive, requiring massive annotations from human experts. While AI models offer a cost-effective alternative by predicting labels, their label quality is compromised by the unavoidable labeling errors. Existing methods mitigate this issue through selective labeling, where AI labels a subset and human labels the remainder. However, these methods lack theoretical guarantees on the quality of AI-assigned labels, often resulting in unacceptably high labeling error within the AI-labeled subset. To address this, we introduce \textbf{Conformal Labeling}, a novel method to identify instances where AI predictions can be provably trusted. This is achieved by controlling the false discovery rate (FDR), the proportion of incorrect labels within the selected subset. In particular, we construct a conformal $p$-value for each test instance by comparing AI models' predicted confidence to those of calibration instances mislabeled by AI models. Then, we select test instances whose $p$-values are below a data-dependent threshold, certifying AI models' predictions as trustworthy. We provide theoretical guarantees that Conformal Labeling controls the FDR below the nominal level, ensuring that a predefined fraction of AI-assigned labels is correct on average. Extensive experiments demonstrate that our method achieves tight FDR control with high power across various tasks, including image and text labeling, and LLM QA.
Authors: Samuel Lippl, Thomas McGee, Kimberly Lopez, Ziwen Pan, Pierce Zhang, Salma Ziadi, Oliver Eberle, Ida Momennejad
Abstract: How do latent and inference time computations enable large language models (LLMs) to solve multi-step reasoning? We introduce a framework for tracing and steering algorithmic primitives that underlie model reasoning. Our approach links reasoning traces to internal activations and evaluates algorithmic primitives by injecting them into residual streams and measuring their effect on reasoning steps and task performance. We consider four benchmarks: Traveling Salesperson Problem (TSP), 3SAT, AIME, and graph navigation. We operationalize primitives by clustering activations and annotating their matched reasoning traces using an automated LLM pipeline. We then apply function vector methods to derive primitive vectors as reusable compositional building blocks of reasoning. Primitive vectors can be combined through addition, subtraction, and scalar operations, revealing a geometric logic in activation space. Cross-task and cross-model evaluations (Phi-4, Phi-4-Reasoning, Llama-3-8B) show both shared and task-specific primitives. Notably, comparing Phi-4 with its reasoning-finetuned variant highlights compositional generalization after finetuning: Phi-4-Reasoning exhibits more systematic use of verification and path-generation primitives. Injecting the associated primitive vectors in Phi-4 induces behavioral hallmarks associated with Phi-4-Reasoning. Together, these findings demonstrate that reasoning in LLMs may be supported by a compositional geometry of algorithmic primitives, that primitives transfer cross-task and cross-model, and that reasoning finetuning strengthens algorithmic generalization across domains.
Authors: Jucheng Shen, Yeonju Ro
Abstract: Masked diffusion language models (MDLMs) are becoming competitive with their autoregressive counterparts but typically decode with fixed steps and sequential unmasking. To accelerate decoding, recent work such as Fast-dLLM enables parallel decoding via a static global confidence threshold, yet we observe strong block- and step-wise confidence fluctuations and, within a dataset, near-identical confidence trajectories across inputs as measured by cosine similarity. Motivated by these observations, we introduce One-Shot Dynamic Thresholding (OSDT), which calibrates thresholds on a single sequence and applies them to subsequent inputs with negligible overhead. On GPQA, GSM8K, and HumanEval, OSDT attains superior accuracy-throughput trade-offs (+24% tokens/s on GSM8K at the best accuracy, +45% on GPQA with comparable accuracy, and +50% on HumanEval with a modest accuracy gap). Beyond these results, our findings suggest broader opportunities to leverage reusable task-level confidence signatures for more general-purpose algorithmic and systems innovations in diffusion decoding.
Authors: Tung-Long Vuong, Julien Monteil, Hien Dang, Volodymyr Vaskovych, Trung Le, Vu Nguyen
Abstract: Variational Autoencoders (VAEs) are a powerful alternative to matrix factorization for recommendation. A common technique in VAE-based collaborative filtering (CF) consists in applying binary input masking to user interaction vectors, which improves performance but remains underexplored theoretically. In this work, we analyze how collaboration arises in VAE-based CF and show it is governed by \emph{latent proximity}: we derive a latent sharing radius that informs when an SGD update on one user strictly reduces the loss on another user, with influence decaying as the latent Wasserstein distance increases. We further study the induced geometry: with clean inputs, VAE-based CF primarily exploits \emph{local} collaboration between input-similar users and under-utilizes \emph{global} collaboration between far-but-related users. We compare two mechanisms that encourage \emph{global} mixing and characterize their trade-offs: \ding{172} $\beta$-KL regularization directly tightens the information bottleneck, promoting posterior overlap but risking representational collapse if too large; \ding{173} input masking induces stochastic \emph{geometric} contractions and expansions, which can bring distant users onto the same latent neighborhood but also introduce neighborhood drift. To preserve user identity while enabling global consistency, we propose an anchor regularizer that aligns user posteriors with item embeddings, stabilizing users under masking and facilitating signal sharing across related items. Our analyses are validated on the Netflix, MovieLens-20M, and Million Song datasets. We also successfully deployed our proposed algorithm on an Amazon streaming platform following a successful online experiment.
Authors: Chanakya Ekbote, Vijay Lingam, Sujay Sanghavi, Jun Huan, Behrooz Omidvar-Tehrani, Anoop Deoras, Stefano Soatto
Abstract: Reinforcement Learning with Verifiable Rewards(RLVR) has emerged as a powerful framework for enhancing the reasoning capabilities of large language models (LLMs). However, existing approaches such as Group Relative Policy Optimization (GRPO) and its variants, while effective on reasoning benchmarks, struggle with agentic tasks that require iterative decision-making. We introduce MURPHY, a multi-turn RLVR framework that incorporates execution feedback directly into training, extending GRPO to optimize over multi-turn trajectories where models iteratively refine solutions. MURPHY combines a feedback conditioned rollout tree with trajectory-level credit assignment, and uses pruning to reduce the cost of multi-turn optimization. Evaluations on code generation benchmarks with two model families show that MURPHY consistently improves multi-iteration performance, achieving up to an 8% absolute gain in pass@1 over compute-matched GRPO baselines, and outperforming the prior leading method that incorporates multi-turn execution feedback.
Authors: Penghui Niu, Jiashuai She, Taotao Cai, Yajuan Zhang, Ping Zhang, Junhua Gu, Jianxin Li
Abstract: Ground-based cloud image segmentation is a critical research domain for photovoltaic power forecasting. Current deep learning approaches primarily focus on encoder-decoder architectural refinements. However, existing methodologies exhibit several limitations:(1)they rely on dilated convolutions for multi-scale context extraction, lacking the partial feature effectiveness and interoperability of inter-channel;(2)attention-based feature enhancement implementations neglect accuracy-throughput balance; and (3)the decoder modifications fail to establish global interdependencies among hierarchical local features, limiting inference efficiency. To address these challenges, we propose MPCM-Net, a Multi-scale network that integrates Partial attention Convolutions with Mamba architectures to enhance segmentation accuracy and computational efficiency. Specifically, the encoder incorporates MPAC, which comprises:(1)a MPC block with ParCM and ParSM that enables global spatial interaction across multi-scale cloud formations, and (2)a MPA block combining ParAM and ParSM to extract discriminative features with reduced computational complexity. On the decoder side, a M2B is employed to mitigate contextual loss through a SSHD that maintains linear complexity while enabling deep feature aggregation across spatial and scale dimensions. As a key contribution to the community, we also introduce and release a dataset CSRC, which is a clear-label, fine-grained segmentation benchmark designed to overcome the critical limitations of existing public datasets. Extensive experiments on CSRC demonstrate the superior performance of MPCM-Net over state-of-the-art methods, achieving an optimal balance between segmentation accuracy and inference speed. The dataset and source code will be available at https://github.com/she1110/CSRC.
Authors: Bidipta Sarkar, Mattie Fellows, Juan Agustin Duque, Alistair Letcher, Antonio Le\'on Villares, Anya Sims, Clarisse Wibault, Dmitry Samsonov, Dylan Cope, Jarek Liesen, Kang Li, Lukas Seier, Theo Wolf, Uljad Berdica, Valentin Mohl, Alexander David Goldie, Aaron Courville, Karin Sevegnani, Shimon Whiteson, Jakob Nicolaus Foerster
Abstract: Evolution Strategies (ES) is a class of powerful black-box optimisation methods that are highly parallelisable and can handle non-differentiable and noisy objectives. However, na\"ive ES becomes prohibitively expensive at scale on GPUs due to the low arithmetic intensity of batched matrix multiplications with unstructured random perturbations. We introduce Evolution Guided GeneRal Optimisation via Low-rank Learning (EGGROLL), which improves arithmetic intensity by structuring individual perturbations as rank-$r$ matrices, resulting in a hundredfold increase in training speed for billion-parameter models at large population sizes, achieving up to 91% of the throughput of pure batch inference. We provide a rigorous theoretical analysis of Gaussian ES for high-dimensional parameter objectives, investigating conditions needed for ES updates to converge in high dimensions. Our results reveal a linearising effect, and proving consistency between EGGROLL and ES as parameter dimension increases. Our experiments show that EGGROLL: (1) enables the stable pretraining of nonlinear recurrent language models that operate purely in integer datatypes, (2) is competitive with GRPO for post-training LLMs on reasoning tasks, and (3) does not compromise performance compared to ES in tabula rasa RL settings, despite being faster.
Authors: Yusong Wu, Stephen Brade, Aleksandra Teng Ma, Tia-Jane Fowler, Enning Yang, Berker Banar, Aaron Courville, Natasha Jaques, Cheng-Zhi Anna Huang
Abstract: Most applications of generative AI involve a sequential interaction in which a person inputs a prompt and waits for a response, and where reaction time and adaptivity are not important factors. In contrast, live jamming is a collaborative interaction that requires real-time coordination and adaptation without access to the other player's future moves, while preserving diversity to sustain a creative flow. Reinforcement learning post-training enables effective adaptation through on-policy interaction, yet it often reduces output diversity by exploiting coherence-based rewards. This collapse, known as ``reward hacking'', affects many RL post-training pipelines, but is especially harmful in live jamming, where musical creativity relies on dynamic variation and mutual responsiveness. In this paper, we propose a novel adversarial training method on policy-generated trajectories to mitigate reward hacking in RL post-training for melody-to-chord accompaniment. A co-evolving discriminator separates policy trajectories from the data distribution, while the policy maximizes the discriminator output in addition to coherence rewards to prevent collapse to trivial outputs. We evaluate accompaniment quality and output diversity in simulation with both fixed test melodies and learned melody agents, and we conduct a user study with the model deployed in a real-time interactive system with expert musicians. Quantitative evaluation and user feedback demonstrate improved output diversity, harmonic coherence, adaptation speed and user agency. Our results demonstrate a simple yet effective method to mitigate reward hacking in RL post-training of generative sequence models.
Authors: Deressa Wodajo Deressa, Hannes Mareen, Peter Lambert, Glenn Van Wallendael
Abstract: We present Generative Anchored Fields (GAF), a generative model that learns independent endpoint predictors, $J$ (noise) and $K$ (data), from any point on a linear bridge. Unlike existing approaches that use a single trajectory or score predictor, GAF is trained to recover the bridge endpoints directly via coordinate learning. The velocity field $v=K-J$ emerges from their time-conditioned disagreement. This factorization enables \textit{Transport Algebra}: algebraic operations on multiple $J/K$ heads for compositional control. With class-specific $K_n$ heads, GAF defines directed transport maps between a shared base noise distribution and multiple data domains, allowing controllable interpolation, multi-class composition, and semantic editing. This is achieved either directly on the predicted data coordinates ($K$) using Iterative Endpoint Refinement (IER), a novel sampler that achieves high-quality generation in $5-8$ steps, or on the emergent velocity field ($v$). We achieve strong sample quality (FID 7.51 on ImageNet $256\times256$ and $7.27$ on CelebA-HQ $256\times 256$, without classifier-free guidance) while treating compositional generation as an architectural primitive. Code available at https://github.com/IDLabMedia/GAF.
Authors: Yuepeng Sheng, Yuwei Huang, Shuman Liu, Anxiang Zeng, Haibo Zhang
Abstract: Reinforcement learning (RL) has become a central component of post-training for large language models (LLMs), particularly for complex reasoning tasks that require stable optimization over long generation horizons. However, achieving performance at scale often introduces a fundamental trade-off between training stability and training efficiency. Token-level optimization applies fine-grained updates at the individual units, but is prone to high variance in gradient estimation, which can result in unstable training dynamics. In contrast, Sequence-level optimization often relies on aggressive clipping mechanisms to ensure stable updates. However, such design may discard a large fraction of valid training samples, leading to inefficient gradient utilization and reduced training efficiency. We refer to this phenomenon as gradient underutilization. In this work, we propose Entropy Importance Sampling Policy Optimization (ESPO), a novel framework that aims to combine fine-grained updates with stable training. ESPO decomposes sequences into groups based on predictive entropy, enabling (1) Entropy Grouping Importance Sampling to capture intra-sequence heterogeneity, and (2) Entropy Adaptive Clipping to dynamically allocate trust regions based on model uncertainty. Extensive experiments on mathematical reasoning benchmarks demonstrate that ESPO not only accelerates convergence but also achieves state-of-the-art performance, notably improving accuracy on the challenging mathematical benchmarks.
Authors: Dimitri von R\"utte, Janis Fluri, Omead Pooladzandi, Bernhard Sch\"olkopf, Thomas Hofmann, Antonio Orvieto
Abstract: Modern LLM pre-training consumes vast amounts of compute and training data, making the scaling behavior, or scaling laws, of different models a key distinguishing factor. Discrete diffusion language models (DLMs) have been proposed as an alternative to autoregressive language models (ALMs). However, their scaling behavior has not yet been fully explored, with prior work suggesting that they require more data and compute to match the performance of ALMs. We study the scaling behavior of DLMs on different noise types by smoothly interpolating between masked and uniform diffusion while paying close attention to crucial hyperparameters such as batch size and learning rate. Our experiments reveal that the scaling behavior of DLMs strongly depends on the noise type and is considerably different from ALMs. While all noise types converge to similar loss values in compute-bound scaling, we find that uniform diffusion requires more parameters and less data for compute-efficient training compared to masked diffusion, making them a promising candidate in data-bound settings. We scale our uniform diffusion model up to 10B parameters trained for $10^{22}$ FLOPs, confirming the predicted scaling behavior and making it the largest publicly known uniform diffusion model to date.
Authors: Kaustav Chatterjee, Joshua Li, Kundan Parajulee, Jared Schwennesen
Abstract: Steep-profiled Highway Railway Grade Crossings (HRGCs) pose safety hazards to vehicles with low ground clearance, which may become stranded on the tracks, creating risks of train vehicle collisions. This research develops a framework for network level evaluation of hang-up susceptibility of HRGCs. Profile data from different crossings in Oklahoma were collected using both a walking profiler and the Pave3D8K Laser Imaging System. A hybrid deep learning model, combining Long Short Term Memory (LSTM) and Transformer architectures, was developed to reconstruct accurate HRGC profiles from Pave3D8K Laser Imaging System data. Vehicle dimension data from around 350 specialty vehicles were collected at various locations across Oklahoma to enable up-to-date statistical design dimensions. Hang-up susceptibility was analyzed using three vehicle dimension scenarios: (a) median dimension (median wheelbase and ground clearance), (b) 75-25 percentile dimension (75 percentile wheelbase, 25 percentile ground clearance), and (c) worst case dimension (maximum wheelbase and minimum ground clearance). Results indicate 70, 80, and 95 crossings at the highest hang-up risk levels under these scenarios, respectively. An ArcGIS database and a software interface were developed to support transportation agencies in mitigating crossing hazards. This framework advances safety evaluation by integrating next-generation sensing, deep learning, and infrastructure datasets into practical decision support tools.
Authors: Melvin Barbaux (IMB), Samia Boukir (IMB)
Abstract: Semi-supervised classification leverages both labeled and unlabeled data to improve predictive performance, but existing software support remains fragmented across methods, learning settings, and data modalities. We introduce ModSSC, an open source Python framework for inductive and transductive semi-supervised classification designed to support reproducible and controlled experimentation. ModSSC provides a modular and extensible software architecture centered on reusable semi-supervised learning components, stable abstractions, and fully declarative experiment specification. Experiments are defined through configuration files, enabling systematic comparison across heterogeneous datasets and model backbones without modifying algorithmic code. ModSSC 1.0.0 is released under the MIT license with full documentation and automated tests, and is available at https://github.com/ModSSC/ModSSC. The framework is validated through controlled experiments reproducing established semi-supervised learning baselines across multiple data modalities.
Authors: Sami Marouani, Kamal Singh, Baptiste Jeudy, Amaury Habrard
Abstract: Accurate prediction of flow delay is essential for optimizing and managing modern communication networks. We investigate three levels of modeling for this task. First, we implement a heterogeneous GNN with attention-based message passing, establishing a strong neural baseline. Second, we propose FlowKANet in which Kolmogorov-Arnold Networks replace standard MLP layers, reducing trainable parameters while maintaining competitive predictive performance. FlowKANet integrates KAMP-Attn (Kolmogorov-Arnold Message Passing with Attention), embedding KAN operators directly into message-passing and attention computation. Finally, we distill the model into symbolic surrogate models using block-wise regression, producing closed-form equations that eliminate trainable weights while preserving graph-structured dependencies. The results show that KAN layers provide a favorable trade-off between efficiency and accuracy and that symbolic surrogates emphasize the potential for lightweight deployment and enhanced transparency.
Authors: Mykola Vysotskyi, Zahar Kohut, Mariia Shpir, Taras Rumezhak, Volodymyr Karpiv
Abstract: Machine unlearning in text-to-image diffusion models aims to remove targeted concepts while preserving overall utility. Prior diffusion unlearning methods typically rely on supervised weight edits or global penalties; reinforcement-learning (RL) approaches, while flexible, often optimize sparse end-of-trajectory rewards, yielding high-variance updates and weak credit assignment. We present a general RL framework for diffusion unlearning that treats denoising as a sequential decision process and introduces a timestep-aware critic with noisy-step rewards. Concretely, we train a CLIP-based reward predictor on noisy latents and use its per-step signal to compute advantage estimates for policy-gradient updates of the reverse diffusion kernel. Our algorithm is simple to implement, supports off-policy reuse, and plugs into standard text-to-image backbones. Across multiple concepts, the method achieves better or comparable forgetting to strong baselines while maintaining image quality and benign prompt fidelity; ablations show that (i) per-step critics and (ii) noisy-conditioned rewards are key to stability and effectiveness. We release code and evaluation scripts to facilitate reproducibility and future research on RL-based diffusion unlearning.
Authors: Florent De Geeter, Gaspard Lambrechts, Damien Ernst, Guillaume Drion
Abstract: With the emergence of massively parallel processing units, parallelization has become a desirable property for new sequence models. The ability to parallelize the processing of sequences with respect to the sequence length during training is one of the main factors behind the uprising of the Transformer architecture. However, Transformers lack efficiency at sequence generation, as they need to reprocess all past timesteps at every generation step. Recently, state-space models (SSMs) emerged as a more efficient alternative. These new kinds of recurrent neural networks (RNNs) keep the efficient update of the RNNs while gaining parallelization by getting rid of nonlinear dynamics (or recurrence). SSMs can reach state-of-the art performance through the efficient training of potentially very large networks, but still suffer from limited representation capabilities. In particular, SSMs cannot exhibit persistent memory, or the capacity of retaining information for an infinite duration, because of their monostability. In this paper, we introduce a new family of RNNs, the memory recurrent units (MRUs), that combine the persistent memory capabilities of nonlinear RNNs with the parallelizable computations of SSMs. These units leverage multistability as a source of persistent memory, while getting rid of transient dynamics for efficient computations. We then derive a specific implementation as proof-of-concept: the bistable memory recurrent unit (BMRU). This new RNN is compatible with the parallel scan algorithm. We show that BMRU achieves good results in tasks with long-term dependencies, and can be combined with state-space models to create hybrid networks that are parallelizable and have transient dynamics as well as persistent memory.
Authors: Wang Zixian
Abstract: Large language model alignment objectives are often presented as a collection of distinct algorithms, such as PPO, DPO, IPO, and their variants, each motivated by different derivations. In this work, we argue that this diversity obscures a simpler underlying structure. At a fundamental level, alignment objectives involve two independent design choices: (i) how training signals are sampled and weighted, and (ii) how deviations from a reference policy are geometrically penalized. Existing methods typically entangle these choices through a single divergence, most commonly the Kullback-Leibler divergence. We show that this entanglement is not merely a modeling convenience but a source of systematic instability. When the same divergence simultaneously determines sample weighting and optimization curvature, adjusting one aspect, such as exploration strength, inevitably alters the other, such as gradient geometry. This coupling is particularly problematic in preference-based reinforcement learning, where advantage signals are unbounded and high-confidence regimes are common. We propose a principled structural remedy by formulating alignment as an orthogonal mirror descent problem, in which sampling geometry enters as a linear driving force derived from an alpha-divergence projection, while optimization geometry is determined independently by a Bregman divergence, or mirror map. This perspective leads to a new alignment objective called Orthogonalized Policy Optimization (OPO), obtained by choosing a Euclidean mirror map in likelihood ratio space. The resulting objective admits a closed-form solution, linear and non-saturating gradient dynamics, and a well-conditioned trust region, while remaining fully compatible with standard large language model training pipelines.
Authors: Vittoria De Pellegrini, Tariq Alkhalifah
Abstract: Modeling and forecasting subsurface multiphase fluid flow fields underpin applications ranging from geological CO2 sequestration (GCS) operations to geothermal production. This is essential for ensuring both operational performance and long-term safety. While high fidelity multiphase simulators are widely used for this purpose, they become prohibitively expensive once many forward runs are required for inversion purposes and to quantify uncertainty. To tackle this challenge, we propose LAViG-FLOW, a latent autoregressive video generation diffusion framework that explicitly learns the coupled evolution of saturation and pressure fields. Each state variable is compressed by a dedicated 2D autoencoder, and a Video Diffusion Transformer (VDiT) models their coupled distribution across time. We first train the model on a given time horizon to learn their coupled relationship and then fine-tune it autoregressively so it can extrapolate beyond the observed time window. Evaluated on an open-source CO2 sequestration dataset, LAViG-FLOW generates saturation and pressure fields that stay consistent across time while running two orders of magnitude faster than traditional numerical solvers.
Authors: Kanishk Gandhi, Shivam Garg, Noah D. Goodman, Dimitris Papailiopoulos
Abstract: Environments are the bottleneck for self-improving agents. Current terminal benchmarks were built for evaluation, not training; reinforcement learning requires a scalable pipeline, not just a dataset. We introduce Endless Terminals, a fully autonomous pipeline that procedurally generates terminal-use tasks without human annotation. The pipeline has four stages: generating diverse task descriptions, building and validating containerized environments, producing completion tests, and filtering for solvability. From this pipeline we obtain 3255 tasks spanning file operations, log management, data processing, scripting, and database operations. We train agents using vanilla PPO with binary episode level rewards and a minimal interaction loop: no retrieval, multi-agent coordination, or specialized tools. Despite this simplicity, models trained on Endless Terminals show substantial gains: on our held-out dev set, Llama-3.2-3B improves from 4.0% to 18.2%, Qwen2.5-7B from 10.7% to 53.3%, and Qwen3-8B-openthinker-sft from 42.6% to 59.0%. These improvements transfer to human-curated benchmarks: models trained on Endless Terminals show substantial gains on held out human curated benchmarks: on TerminalBench 2.0, Llama-3.2-3B improves from 0.0% to 2.2%, Qwen2.5-7B from 2.2% to 3.4%, and Qwen3-8B-openthinker-sft from 1.1% to 6.7%, in each case outperforming alternative approaches including models with more complex agentic scaffolds. These results demonstrate that simple RL succeeds when environments scale.
Authors: Andy Zhu, Rongzhe Wei, Yupu Gu, Pan Li
Abstract: Machine unlearning (MU) for large language models has become critical for AI safety, yet existing methods fail to generalize to Mixture-of-Experts (MoE) architectures. We identify that traditional unlearning methods exploit MoE's architectural vulnerability: they manipulate routers to redirect queries away from knowledgeable experts rather than erasing knowledge, causing a loss of model utility and superficial forgetting. We propose Geometric Routing Invariance Preservation (GRIP), an algorithm-agnostic framework for unlearning for MoE. Our core contribution is a geometric constraint, implemented by projecting router gradient updates into an expert-specific null-space. Crucially, this decouples routing stability from parameter rigidity: while discrete expert selections remain stable for retained knowledge, the continuous router parameters remain plastic within the null space, allowing the model to undergo necessary internal reconfiguration to satisfy unlearning objectives. This forces the unlearning optimization to erase knowledge directly from expert parameters rather than exploiting the superficial router manipulation shortcut. GRIP functions as an adapter, constraining router parameter updates without modifying the underlying unlearning algorithm. Extensive experiments on large-scale MoE models demonstrate that our adapter eliminates expert selection shift (achieving over 95% routing stability) across all tested unlearning methods while preserving their utility. By preventing existing algorithms from exploiting MoE model's router vulnerability, GRIP adapts existing unlearning research from dense architectures to MoEs.
Authors: Hansheng Ren
Abstract: The prevailing scaling paradigm of Large Language Models (LLMs) rests on a substrate of "Fuzzy" floating-point arithmetic. To mitigate the inherent instability of this approximate foundation, modern architectures have erected a complex scaffolding of structural and numerical heuristics--Complex Residuals, Pre-RMSNorm, Attention Scaling, and Gradient Clipping--consuming significant compute solely to prevent numerical collapse. We propose a paradigm shift to the "Exact". We introduce the Halo Architecture, grounded in the Rational Field (Q) and powered by a custom Exact Inference Unit (EIU). To resolve the exponential bit-width growth of rational arithmetic, Halo employs a Dual-Ring Topology that unifies two complementary control mechanisms: (1) The Micro-Ring (Continuum Maintenance), which strictly bounds memory complexity via Diophantine Approximation; and (2) The Macro-Ring (Symbolic Alignment), which enforces logical consistency via periodic state collapse. This stable dual-ring substrate allows for the "Great Dismantling" of numerical scaffolding, reducing the Transformer block to its "Clean" algebraic form (Tabula Rasa). Furthermore, we verify the "Efficiency Paradox": the elimination of gradient noise (sigma -> 0) allows for Macro-Learning Rates, potentially reducing the Total Time-to-Convergence by orders of magnitude. Halo demonstrates that General Intelligence requires the hybridization of continuous fields and discrete chains under a rigorous mathematical framework.
Authors: Jonas H\"ubotter, Frederike L\"ubeck, Lejs Behric, Anton Baumann, Marco Bagatella, Daniel Marta, Ido Hakimi, Idan Shenfeld, Thomas Kleine Buening, Carlos Guestrin, Andreas Krause
Abstract: Large language models are increasingly post-trained with reinforcement learning in verifiable domains such as code and math. Yet, current methods for reinforcement learning with verifiable rewards (RLVR) learn only from a scalar outcome reward per attempt, creating a severe credit-assignment bottleneck. Many verifiable environments actually provide rich textual feedback, such as runtime errors or judge evaluations, that explain why an attempt failed. We formalize this setting as reinforcement learning with rich feedback and introduce Self-Distillation Policy Optimization (SDPO), which converts tokenized feedback into a dense learning signal without any external teacher or explicit reward model. SDPO treats the current model conditioned on feedback as a self-teacher and distills its feedback-informed next-token predictions back into the policy. In this way, SDPO leverages the model's ability to retrospectively identify its own mistakes in-context. Across scientific reasoning, tool use, and competitive programming on LiveCodeBench v6, SDPO improves sample efficiency and final accuracy over strong RLVR baselines. Notably, SDPO also outperforms baselines in standard RLVR environments that only return scalar feedback by using successful rollouts as implicit feedback for failed attempts. Finally, applying SDPO to individual questions at test time accelerates discovery on difficult binary-reward tasks, achieving the same discovery probability as best-of-k sampling or multi-turn conversations with 3x fewer attempts.
Authors: Qi Cao, Shuhao Zhang, Ruizhe Zhou, Ruiyi Zhang, Peijia Qin, Pengtao Xie
Abstract: Model routing chooses which language model to use for each query. By sending easy queries to cheaper models and hard queries to stronger ones, it can significantly reduce inference cost while maintaining high accuracy. However, most existing routers treat this as a fixed choice among a small set of models, which makes them hard to adapt to new models or changing budget constraints. In this paper, we propose SCOPE (Scalable and Controllable Outcome Performance Estimator), a routing framework that goes beyond model selection by predicting their cost and performance. Trained with reinforcement learning, SCOPE makes reasoning-based predictions by retrieving how models behave on similar problems, rather than relying on fixed model names, enabling it to work with new, unseen models. Moreover, by explicitly predicting how accurate and how expensive a model will be, it turns routing into a dynamic decision problem, allowing users to easily control the trade-off between accuracy and cost. Experiments show that SCOPE is more than just a cost-saving tool. It flexibly adapts to user needs: it can boost accuracy by up to 25.7% when performance is the priority, or cut costs by up to 95.1% when efficiency matters most.
Authors: Louis Schiekiera, Max Zimmer, Christophe Roux, Sebastian Pokutta, Fritz G\"unther
Abstract: We investigate the extent to which an LLM's hidden-state geometry can be recovered from its behavior in psycholinguistic experiments. Across eight instruction-tuned transformer models, we run two experimental paradigms -- similarity-based forced choice and free association -- over a shared 5,000-word vocabulary, collecting 17.5M+ trials to build behavior-based similarity matrices. Using representational similarity analysis, we compare behavioral geometries to layerwise hidden-state similarity and benchmark against FastText, BERT, and cross-model consensus. We find that forced-choice behavior aligns substantially more with hidden-state geometry than free association. In a held-out-words regression, behavioral similarity (especially forced choice) predicts unseen hidden-state similarities beyond lexical baselines and cross-model consensus, indicating that behavior-only measurements retain recoverable information about internal semantic geometry. Finally, we discuss implications for the ability of behavioral tasks to uncover hidden cognitive states.
Authors: Rong Fu, Wenxin Zhang, Muge Qi, Yang Li, Yabin Jin, Jiekai Wu, Jiaxuan Lu, Chunlei Meng, Youjin Wang, Zeli Su, Juntao Gao, Li Bao, Qi Zhao, Wei Luo, Simon Fong
Abstract: Repertoire-level analysis of T cell receptors offers a biologically grounded signal for disease detection and immune monitoring, yet practical deployment is impeded by label sparsity, cohort heterogeneity, and the computational burden of adapting large encoders to new tasks. We introduce a framework that synthesizes compact task-specific parameterizations from a learned dictionary of prototypes conditioned on lightweight task descriptors derived from repertoire probes and pooled embedding statistics. This synthesis produces small adapter modules applied to a frozen pretrained backbone, enabling immediate adaptation to novel tasks with only a handful of support examples and without full model fine-tuning. The architecture preserves interpretability through motif-aware probes and a calibrated motif discovery pipeline that links predictive decisions to sequence-level signals. Together, these components yield a practical, sample-efficient, and interpretable pathway for translating repertoire-informed models into diverse clinical and research settings where labeled data are scarce and computational resources are constrained.
Authors: Ons Saadallah, M\'aty\'as and\'o, Tam\'as G\'abor Orosz
Abstract: Hyperparameter optimization (HPO) plays a central role in the performance of deep learning models, yet remains computationally expensive and difficult to interpret, particularly for time-series forecasting. While Bayesian Optimization (BO) is a standard approach, it typically treats tuning tasks independently and provides limited insight into its decisions. Recent advances in large language models (LLMs) offer new opportunities to incorporate structured prior knowledge and reasoning into optimization pipelines. We introduce LLM-AutoOpt, a hybrid HPO framework that combines BO with LLM-based contextual reasoning. The framework encodes dataset meta-features, model descriptions, historical optimization outcomes, and target objectives as structured meta-knowledge within LLM prompts, using BO to initialize the search and mitigate cold-start effects. This design enables context-aware and stable hyperparameter refinement while exposing the reasoning behind optimization decisions. Experiments on a multivariate time series forecasting benchmark demonstrate that LLM-AutoOpt achieves improved predictive performance and more interpretable optimization behavior compared to BO and LLM baselines without meta-knowledge.
Authors: Abhijit Gupta
Abstract: Molecular property prediction is crucial for drug discovery when labeled data are scarce. This work presents \modelname, a graph transformer augmented with a query-conditioned cardinality-preserving attention (CPA) channel that retains dynamic support-size signals complementary to static centrality embeddings. The approach combines structured sparse attention with Graphormer-inspired biases (shortest-path distance, centrality, direct-bond features) and unified dual-objective self-supervised pretraining (masked reconstruction and contrastive alignment of augmented views). Evaluation on 11 public benchmarks spanning MoleculeNet, OGB, and TDC ADMET demonstrates consistent improvements over protocol-matched baselines under matched pretraining, optimization, and hyperparameter tuning. Rigorous ablations confirm CPA's contributions and rule out simple size shortcuts. Code and reproducibility artifacts are provided.
Authors: Asif Tauhid, Sidahmed Benabderrahmane, Mohamad Altrabulsi, Ahamed Foisal, Talal Rahwan
Abstract: Advanced Persistent Threats (APTs) are sophisticated, long-term cyberattacks that are difficult to detect because they operate stealthily and often blend into normal system behavior. This paper presents a neuro-symbolic anomaly detection framework that combines a Graph Autoencoder (GAE) with rare pattern mining to identify APT-like activities in system-level provenance data. Our approach first constructs a process behavioral graph using k-Nearest Neighbors based on feature similarity, then learns normal relational structure using a Graph Autoencoder. Anomaly candidates are identified through deviations between observed and reconstructed graph structure. To further improve detection, we integrate an rare pattern mining module that discovers infrequent behavioral co-occurrences and uses them to boost anomaly scores for processes exhibiting rare signatures. We evaluate the proposed method on the DARPA Transparent Computing datasets and show that rare-pattern boosting yields substantial gains in anomaly ranking quality over the baseline GAE. Compared with existing unsupervised approaches on the same benchmark, our single unified model consistently outperforms individual context-based detectors and achieves performance competitive with ensemble aggregation methods that require multiple separate detectors. These results highlight the value of coupling graph-based representation learning with classical pattern mining to improve both effectiveness and interpretability in provenance-based security anomaly detection.
Authors: Jing-Cheng Pang, Liang Lu, Xian Tang, Kun Jiang, Sijie Wu, Kai Zhang, Xubin Li
Abstract: Reinforcement learning (RL) has emerged as a key paradigm for aligning and optimizing large language models (LLMs). Standard approaches treat the LLM as the policy and apply RL directly over the full vocabulary space. However, this formulation includes the massive tail of contextually irrelevant tokens in the action space, which could distract the policy from focusing on decision-making among the truly reasonable tokens. In this work, we verify that valid reasoning paths could inherently concentrate within a low-rank subspace. Based on this insight, we introduce Reinforcement Learning with Promising Tokens (RLPT), a framework that mitigates the action space issue by decoupling strategic decision-making from token generation. Specifically, RLPT leverages the semantic priors of the base model to identify a dynamic set of promising tokens and constrains policy optimization exclusively to this refined subset via masking. Theoretical analysis and empirical results demonstrate that RLPT effectively reduces gradient variance, stabilizes the training process, and improves sample efficiency. Experiment results on math, coding, and telecom reasoning show that RLPT outperforms standard RL baselines and integrates effectively across various model sizes (4B and 8B) and RL algorithms (GRPO and DAPO).
Authors: Jonathan Lin, Aman Desai, Frank Barrows, Francesco Caravelli
Abstract: Machine learning is a powerful method of extracting meaning from data; unfortunately, current digital hardware is extremely energy-intensive. There is interest in an alternative analog computing implementation that could match the performance of traditional machine learning while being significantly more energy-efficient. However, it remains unclear how to train such analog computing systems while adhering to locality constraints imposed by the physical (as opposed to digital) nature of these systems. Local learning algorithms such as Equilibrium Propagation and Coupled Learning have been proposed to address this issue. In this paper, we develop an algorithm to exactly calculate gradients using a graph theoretic and analytical framework for Kirchhoff's laws. We also introduce Generalized Equilibrium Propagation, a framework encompassing a broad class of Hebbian learning algorithms, including Coupled Learning and Equilibrium Propagation, and show how our algorithm compares. We demonstrate our algorithm using numerical simulations and show that we can train resistor networks without the need for a replica or readout over all resistors, only at the output layer. We also show that under the analytical gradient approach, it is possible to update only a subset of the resistance values without a strong degradation in performance.
Authors: Rong Fu, Wenxin Zhang, Chunlei Meng, Youjin Wang, Haoyu Zhao, Jiaxuan Lu, Kun Liu, JiaBao Dou, Simon James Fong
Abstract: The pursuit of optimal trade-offs in high-dimensional search spaces under stringent computational constraints poses a fundamental challenge for contemporary multi-objective optimization. We develop NeuroPareto, a cohesive architecture that integrates rank-centric filtering, uncertainty disentanglement, and history-conditioned acquisition strategies to navigate complex objective landscapes. A calibrated Bayesian classifier estimates epistemic uncertainty across non-domination tiers, enabling rapid generation of high-quality candidates with minimal evaluation cost. Deep Gaussian Process surrogates further separate predictive uncertainty into reducible and irreducible components, providing refined predictive means and risk-aware signals for downstream selection. A lightweight acquisition network, trained online from historical hypervolume improvements, guides expensive evaluations toward regions balancing convergence and diversity. With hierarchical screening and amortized surrogate updates, the method maintains accuracy while keeping computational overhead low. Experiments on DTLZ and ZDT suites and a subsurface energy extraction task show that NeuroPareto consistently outperforms classifier-enhanced and surrogate-assisted baselines in Pareto proximity and hypervolume.
Authors: Emiliano Penaloza, Dheeraj Vattikonda, Nicolas Gontier, Alexandre Lacoste, Laurent Charlin, Massimo Caccia
Abstract: Training-time privileged information (PI) can enable language models to succeed on tasks they would otherwise fail, making it a powerful tool for reinforcement learning in hard, long-horizon settings. However, transferring capabilities learned with PI to policies that must act without it at inference time remains a fundamental challenge. We study this problem in the context of distilling frontier models for multi-turn agentic environments, which typically hide their internal reasoning and expose only action trajectories. This breaks standard distillation pipelines, since successful behavior is observable, but the reasoning process is not. For this, we introduce {\pi}-Distill, a joint teacher-student objective that trains a PI-conditioned teacher and an unconditioned student simultaneously using the same model. Additionally, we also introduce On-Policy Self-Distillation (OPSD), an alternative approach that trains using Reinforcement Learning (RL) with a reverse KL-penalty between the student and the PI-conditioned teacher. We show that both of these algorithms effectively distill frontier agents using action-only PI. Specifically, we find that {\pi}-Distill and, in some cases, OPSD, outperform industry standard practices (Supervised finetuning followed by RL) that assume access to full Chain-of-Thought supervision across multiple agentic benchmarks, models, and forms of PI. We complement our results with extensive analysis that characterizes the factors enabling effective learning with PI, focusing primarily on {\pi}-Distill and characterizing when OPSD is competitive.
Authors: Thanawat Sornwanee
Abstract: In a probabilistic reformulation of a combinatorial problem, we often face an optimization over a hypercube, which corresponds to the Bernoulli probability parameter for each binary variable in the primal problem. The combinatorial nature suggests that an exact gradient computation requires multiple queries. We propose a stochastic gradient that is unbiased and requires only a single query of the combinatorial function. This method encompasses a well-established REINFORCE (through an importance sampling), as well as including a class of new stochastic gradients.
Authors: Yinan Huang, Hans Hao-Hsun Hsu, Junran Wang, Bo Dai, Pan Li
Abstract: Sequential prediction from streaming observations is a fundamental problem in stochastic dynamical systems, where inherent uncertainty often leads to multiple plausible futures. While diffusion and flow-matching models are capable of modeling complex, multi-modal trajectories, their deployment in real-time streaming environments typically relies on repeated sampling from a non-informative initial distribution, incurring substantial inference latency and potential system backlogs. In this work, we introduce Sequential Flow Matching, a principled framework grounded in Bayesian filtering. By treating streaming inference as learning a probability flow that transports the predictive distribution from one time step to the next, our approach naturally aligns with the recursive structure of Bayesian belief updates. We provide theoretical justification that initializing generation from the previous posterior offers a principled warm start that can accelerate sampling compared to na\"ive re-sampling. Across a wide range of forecasting, decision-making and state estimation tasks, our method achieves performance competitive with full-step diffusion while requiring only one or very few sampling steps, therefore with faster sampling. It suggests that framing sequential inference via Bayesian filtering provides a new and principled perspective towards efficient real-time deployment of flow-based models. Our code is available at https://github.com/Graph-COM/Sequential\_Flow\_Matching.
URLs: https://github.com/Graph-COM/Sequential\_Flow\_Matching.
Authors: Yifu Qiu, Zheng Zhao, Waylon Li, Yftah Ziser, Anna Korhonen, Shay B. Cohen, Edoardo M. Ponti
Abstract: Internal modelling of the world -- predicting transitions between previous states $X$ and next states $Y$ under actions $Z$ -- is essential to reasoning and planning for LLMs and VLMs. Learning such models typically requires costly action-labelled trajectories. We propose SWIRL, a self-improvement framework that learns from state-only sequences by treating actions as a latent variable and alternating between Forward World Modelling (FWM) $P_\theta(Y|X,Z)$ and an Inverse Dynamics Modelling (IDM) $Q_\phi(Z|X,Y)$. SWIRL iterates two phases: (1) Variational Information Maximisation, which updates the FWM to generate next states that maximise conditional mutual information with latent actions given prior states, encouraging identifiable consistency; and (2) ELBO Maximisation, which updates the IDM to explain observed transitions, effectively performing coordinate ascent. Both models are trained with reinforcement learning (specifically, GRPO) with the opposite frozen model's log-probability as a reward signal. We provide theoretical learnability guarantees for both updates, and evaluate SWIRL on LLMs and VLMs across multiple environments: single-turn and multi-turn open-world visual dynamics and synthetic textual environments for physics, web, and tool calling. SWIRL achieves gains of 16% on AURORABench, 28% on ByteMorph, 16% on WorldPredictionBench, and 14% on StableToolBench.
Authors: Sohan Venkatesh, Ashish Mahendran Kurapath
Abstract: Activation steering methods are widely used to control large language model (LLM) behavior and are often interpreted as revealing meaningful internal representations. This interpretation assumes steering directions are identifiable and uniquely recoverable from input-output behavior. We show that, under white-box single-layer access, steering vectors are fundamentally non-identifiable due to large equivalence classes of behaviorally indistinguishable interventions. Empirically, we show that orthogonal perturbations achieve near-equivalent efficacy with negligible effect sizes across multiple models and traits. Critically, we show that the non-identifiability is a robust geometric property that persists across diverse prompt distributions. These findings reveal fundamental interpretability limits and highlight the need for structural constraints beyond behavioral testing to enable reliable alignment interventions.
Authors: Rui Wu, Li YongJun
Abstract: Generative modeling typically seeks the path of least action via deterministic flows (ODE). While effective for in-distribution tasks, we argue that these deterministic paths become brittle under causal interventions, which often require transporting probability mass across low-density regions (``off-manifold'') where the vector field is ill-defined. This leads to numerical instability and spurious correlations. In this work, we introduce the Causal Schr\"odinger Bridge (CSB), a framework that reformulates counterfactual inference as Entropic Optimal Transport. Unlike deterministic approaches that require strict invertibility or rely on low-rank approximations, CSB leverages diffusion processes (SDEs) to robustly ``tunnel'' through support mismatches while strictly enforcing structural admissibility constraints. We prove the Structural Decomposition Theorem, showing that the global high-dimensional bridge factorizes exactly into local, robust transitions. Crucially, we demonstrate that CSB breaks the Curse of Dimensionality in regimes of high intrinsic dimension. We empirically validate this on a full-rank causal system ($d=10^5$, intrinsic rank $10^5$), completing the transport in 26.48 seconds on a single GPU (RTX 3090). This stands in stark contrast to structure-agnostic $O(d^3)$ baselines, which are estimated to require over 6 years for dense computations of this scale regardless of the data's intrinsic rank. Empirical validation on Morpho-MNIST and $10^5$-D extremal stress tests demonstrates that CSB significantly outperforms deterministic baselines in structural consistency and distribution coverage, capturing the underlying manifold with high fidelity (MSE $\approx$ 0.04).
Authors: Sarthak Wanjari
Abstract: Offline Reinforcement Learning (RL) promises the recovery of optimal policies from static datasets, yet it remains susceptible to the overestimation of out-of-distribution (OOD) actions, particularly in fractured and sparse data manifolds. Current solutions necessitate a trade-off between computational efficiency and performance. Methods like CQL offer rigorous conservatism but require tremendous compute power while efficient expectile-based methods like IQL often fail to correct OOD errors on pathological datasets, collapsing to Behavioural Cloning. In this work, we propose Geometric Pessimism, a modular, compute-efficient framework that augments standard IQL with density-based penalty derived from k-nearest-neighbour distances in the state-action embedding space. By pre-computing the penalties applied to each state-action pair, our method injects OOD conservatism via reward shaping with a O(1) training overhead to the training loop. Evaluated on the D4RL MuJoCo benchmark, our method, Geo-IQL outperforms standard IQL on sensitive and unstable medium-replay tasks by over 18 points, while reducing inter-seed standard-deviation by 4 times. Furthermore, Geo-IQL does not degrade performance on stable manifolds. Crucially, we validate our algorithm on the MIMIC-III Sepsis critical care dataset. While standard IQL collapses to behaviour cloning, Geo-IQL demonstrates active policy improvement. Maintaining safety constraints, it achieves 86.4% terminal agreement with clinicians compared to IQL's 75%. Our results suggest that geometric pessimism provides the necessary regularisation to safely overcome local optima in critical, real-world decision systems.
Authors: Benedict Clark, Marta Oliveira, Rick Wilming, Stefan Haufe
Abstract: Explainable AI (XAI) promises to provide insight into machine learning models' decision processes, where one goal is to identify failures such as shortcut learning. This promise relies on the field's assumption that input features marked as important by an XAI must contain information about the target variable. However, it is unclear whether informativeness is indeed the main driver of importance attribution in practice, or if other data properties such as statistical suppression, novelty at test-time, or high feature salience substantially contribute. To clarify this, we trained deep learning models on three variants of a binary image classification task, in which translucent watermarks are either absent, act as class-dependent confounds, or represent class-independent noise. Results for five popular attribution methods show substantially elevated relative importance in watermarked areas (RIW) for all models regardless of the training setting ($R^2 \geq .45$). By contrast, whether the presence of watermarks is class-dependent or not only has a marginal effect on RIW ($R^2 \leq .03$), despite a clear impact impact on model performance and generalisation ability. XAI methods show similar behaviour to model-agnostic edge detection filters and attribute substantially less importance to watermarks when bright image intensities are encoded by smaller instead of larger feature values. These results indicate that importance attribution is most strongly driven by the salience of image structures at test time rather than statistical associations learned by machine learning models. Previous studies demonstrating successful XAI application should be reevaluated with respect to a possibly spurious concurrency of feature salience and informativeness, and workflows using feature attribution methods as building blocks should be scrutinised.
Authors: Arnav Shah, Junzhe Li, Parsa Idehpour, Adibvafa Fallahpour, Brandon Wang, Sukjun Hwang, Bo Wang, Patrick D. Hsu, Hani Goodarzi, Albert Gu
Abstract: Genomic foundation models have the potential to decode DNA syntax, yet face a fundamental tradeoff in their input representation. Standard fixed-vocabulary tokenizers fragment biologically meaningful motifs such as codons and regulatory elements, while nucleotide-level models preserve biological coherence but incur prohibitive computational costs for long contexts. We introduce dnaHNet, a state-of-the-art tokenizer-free autoregressive model that segments and models genomic sequences end-to-end. Using a differentiable dynamic chunking mechanism, dnaHNet compresses raw nucleotides into latent tokens adaptively, balancing compression with predictive accuracy. Pretrained on prokaryotic genomes, dnaHNet outperforms leading architectures including StripedHyena2 in scaling and efficiency. This recursive chunking yields quadratic FLOP reductions, enabling $>3 \times$ inference speedup over Transformers. On zero-shot tasks, dnaHNet achieves superior performance in predicting protein variant fitness and gene essentiality, while automatically discovering hierarchical biological structures without supervision. These results establish dnaHNet as a scalable, interpretable framework for next-generation genomic modeling.
Authors: Frank Xiao, Santiago Aranguri
Abstract: We propose activation-based data attribution, a method that traces behavioral changes in post-trained language models to responsible training datapoints. By computing activation-difference vectors for both test prompts and preference pairs and ranking by cosine similarity, we identify datapoints that cause specific behaviors and validate these attributions causally by retraining with modified data. Clustering behavior-datapoint similarity matrices also enables unsupervised discovery of emergent behaviors. Applying this to OLMo 2's production DPO training, we surfaced distractor-triggered compliance: a harmful behavior where the model complies with dangerous requests when benign formatting instructions are appended. Filtering top-ranked datapoints reduces this behavior by 63% while switching their labels achieves 78%. Our method outperforms gradient-based attribution and LLM-judge baselines while being over 10 times cheaper than both. This in-the-wild model organism - emerging from contaminated preference data rather than deliberate injection - provides a realistic benchmark for safety techniques.
Authors: Luigi Simeone
Abstract: We introduce potential-energy gating, a method for robust state estimation in systems governed by double-well stochastic dynamics. The observation noise covariance of a Bayesian filter is modulated by the local value of a known or assumed potential energy function: observations are trusted when the state is near a potential minimum and progressively discounted as it approaches the barrier separating metastable wells. This physics-based mechanism differs from statistical robust filters, which treat all state-space regions identically, and from constrained filters, which bound states rather than modulating observation trust. The approach is especially relevant in non-ergodic or data-scarce settings where only a single realization is available and statistical methods alone cannot learn the noise structure. We implement gating within Extended, Unscented, Ensemble, and Adaptive Kalman filters and particle filters, requiring only two additional hyperparameters. Monte Carlo benchmarks (100 replications) on a Ginzburg-Landau double-well with 10% outlier contamination show 57-80% RMSE improvement over the standard Extended Kalman Filter, all statistically significant (p < 10^{-15}, Wilcoxon test). A naive topological baseline using only well positions achieves 57%, confirming that the continuous energy landscape adds ~21 percentage points. The method is robust to misspecification: even with 50% parameter errors, improvement never falls below 47%. Comparing externally forced and spontaneous Kramers-type transitions, gating retains 68% improvement under noise-induced transitions whereas the naive baseline degrades to 30%. As an empirical illustration, we apply the framework to Dansgaard-Oeschger events in the NGRIP delta-18O ice-core record, estimating asymmetry gamma = -0.109 (bootstrap 95% CI: [-0.220, -0.011]) and showing that outlier fraction explains 91% of the variance in filter improvement.
Authors: Nick Ferguson, Josh Pennington, Narek Beghian, Aravind Mohan, Douwe Kiela, Sheshansh Agrawal, Thien Hang Nguyen
Abstract: Unstructured documents like PDFs contain valuable structured information, but downstream systems require this data in reliable, standardized formats. LLMs are increasingly deployed to automate this extraction, making accuracy and reliability paramount. However, progress is bottlenecked by two gaps. First, no end-to-end benchmark evaluates PDF-to-JSON extraction under enterprise-scale schema breadth. Second, no principled methodology captures the semantics of nested extraction, where fields demand different notions of correctness (exact match for identifiers, tolerance for quantities, semantic equivalence for names), arrays require alignment, and omission must be distinguished from hallucination. We address both gaps with ExtractBench, an open-source benchmark and evaluation framework for PDF-to-JSON structured extraction. The benchmark pairs 35 PDF documents with JSON Schemas and human-annotated gold labels across economically valuable domains, yielding 12,867 evaluatable fields spanning schema complexities from tens to hundreds of fields. The evaluation framework treats the schema as an executable specification: each field declares its scoring metric. Baseline evaluations reveal that frontier models (GPT-5/5.2, Gemini-3 Flash/Pro, Claude 4.5 Opus/Sonnet) remain unreliable on realistic schemas. Performance degrades sharply with schema breadth, culminating in 0% valid output on a 369-field financial reporting schema across all tested models. We release ExtractBench at https://github.com/ContextualAI/extract-bench.
Authors: Nathana\"el Haas, Fran\c{c}ois Gatine, Augustin M Cosse, Zied Bouraoui
Abstract: Understanding why gradient-based training in deep networks exhibits strong implicit bias remains challenging, in part because tractable singular-value dynamics are typically available only for balanced deep linear models. We propose an alternative route based on two theoretically grounded and empirically testable signatures of deep Jacobians: depth-induced exponential scaling of ordered singular values and strong spectral separation. Adopting a fixed-gates view of piecewise-linear networks, where Jacobians reduce to products of masked linear maps within a single activation region, we prove the existence of Lyapunov exponents governing the top singular values at initialization, give closed-form expressions in a tractable masked model, and quantify finite-depth corrections. We further show that sufficiently strong separation forces singular-vector alignment in matrix products, yielding an approximately shared singular basis for intermediate Jacobians. Together, these results motivate an approximation regime in which singular-value dynamics become effectively decoupled, mirroring classical balanced deep-linear analyses without requiring balancing. Experiments in fixed-gates settings validate the predicted scaling, alignment, and resulting dynamics, supporting a mechanistic account of emergent low-rank Jacobian structure as a driver of implicit bias.
Authors: Jing Xiao, Xinhai Chen, Jiaming Peng, Qinglin Wang, Menghan Jia, Zhiquan Lai, Guangping Yu, Dongsheng Li, Tiejun Li, Jie Liu
Abstract: Symbolic Regression (SR) aims to discover interpretable equations from observational data, with the potential to reveal underlying principles behind natural phenomena. However, existing approaches often fall into the Pseudo-Equation Trap: producing equations that fit observations well but remain inconsistent with fundamental scientific principles. A key reason is that these approaches are dominated by empirical risk minimization, lacking explicit constraints to ensure scientific consistency. To bridge this gap, we propose PG-SR, a prior-guided SR framework built upon a three-stage pipeline consisting of warm-up, evolution, and refinement. Throughout the pipeline, PG-SR introduces a prior constraint checker that explicitly encodes domain priors as executable constraint programs, and employs a Prior Annealing Constrained Evaluation (PACE) mechanism during the evolution stage to progressively steer discovery toward scientifically consistent regions. Theoretically, we prove that PG-SR reduces the Rademacher complexity of the hypothesis space, yielding tighter generalization bounds and establishing a guarantee against pseudo-equations. Experimentally, PG-SR outperforms state-of-the-art baselines across diverse domains, maintaining robustness to varying prior quality, noisy data, and data scarcity.
Authors: Gengsheng Li, Jinghan He, Shijie Wang, Dan Zhang, Ruiqi Liu, Renrui Zhang, Zijun Yao, Junfeng Fang, Haiyun Guo, Jinqiao Wang
Abstract: Self-play bootstraps LLM reasoning through an iterative Challenger-Solver loop: the Challenger is trained to generate questions that target the Solver's capabilities, and the Solver is optimized on the generated data to expand its reasoning skills. However, existing frameworks like R-Zero often exhibit non-sustained improvement, where early gains degrade as self-play continues. We identify a key failure mode, Diversity Illusion, where the Solver's training signals appear diverse yet collapse into recurring underlying patterns. It manifests as (1) Local Diversity Illusion, where diversity is enforced only within-batch, inducing cross-iteration mode cycling; and (2) Surface Diversity Illusion, where questions vary superficially but require near-identical reasoning skills. To mitigate them, we propose R-Diverse with two aligned innovations: Memory-Augmented Penalty (MAP), which uses a persistent memory bank to discourage recycling across iterations, and Skill-Aware Measurement (SAM), which evaluates diversity by the reasoning skills exercised rather than surface variation of questions. Across 10 math and general reasoning benchmarks, R-Diverse sustains gains over more iterations and consistently outperforms prior self-play methods. Code is available at https://github.com/Gengsheng-Li/R-Diverse.
Authors: Chihiro Watanabe, Taiji Suzuki
Abstract: Matrix reordering is a task to permute the rows and columns of a given observed matrix such that the resulting reordered matrix shows meaningful or interpretable structural patterns. Most existing matrix reordering techniques share the common processes of extracting some feature representations from an observed matrix in a predefined manner, and applying matrix reordering based on it. However, in some practical cases, we do not always have prior knowledge about the structural pattern of an observed matrix. To address this problem, we propose a new matrix reordering method, called deep two-way matrix reordering (DeepTMR), using a neural network model. The trained network can automatically extract nonlinear row/column features from an observed matrix, which can then be used for matrix reordering. Moreover, the proposed DeepTMR provides the denoised mean matrix of a given observed matrix as an output of the trained network. This denoised mean matrix can be used to visualize the global structure of the reordered observed matrix. We demonstrate the effectiveness of the proposed DeepTMR by applying it to both synthetic and practical datasets.
Authors: Samuel J. Weisenthal, Sally W. Thurston, Ashkan Ertefaie
Abstract: In healthcare, there is much interest in estimating policies, or mappings from covariates to treatment decisions. Recently, there is also interest in constraining these estimated policies to the standard of care, which generated the observed data. A relative sparsity penalty was proposed to derive policies that have sparse, explainable differences from the standard of care, facilitating justification of the new policy. However, the developers of this penalty only considered estimation, not inference. Here, we develop inference for the relative sparsity objective function, because characterizing uncertainty is crucial to applications in medicine. Further, in the relative sparsity work, the authors only considered the single-stage decision case; here, we consider the more general, multi-stage case. Inference is difficult, because the relative sparsity objective depends on the unpenalized value function, which is unstable and has infinite estimands in the binary action case. Further, one must deal with a non-differentiable penalty. To tackle these issues, we nest a weighted Trust Region Policy Optimization function within a relative sparsity objective, implement an adaptive relative sparsity penalty, and propose a sample-splitting framework for post-selection inference. We study the asymptotic behavior of our proposed approaches, perform extensive simulations, and analyze a real, electronic health record dataset.
Authors: Milad Abdollahzadeh, Guimeng Liu, Touba Malekzadeh, Christopher T. H. Teo, Keshigeyan Chandrasegaran, Ngai-Man Cheung
Abstract: Generative modeling in machine learning aims to synthesize new data samples that are statistically similar to those observed during training. While conventional generative models such as GANs and diffusion models typically assume access to large and diverse datasets, many real-world applications (e.g. in medicine, satellite imaging, and artistic domains) operate under limited data availability and strict constraints. In this survey, we examine Generative Modeling under Data Constraint (GM-DC), which includes limited-data, few-shot, and zero-shot settings. We present a unified perspective on the key challenges in GM-DC, including overfitting, frequency bias, and incompatible knowledge transfer, and discuss how these issues impact model performance. To systematically analyze this growing field, we introduce two novel taxonomies: one categorizing GM-DC tasks (e.g. unconditional vs. conditional generation, cross-domain adaptation, and subject-driven modeling), and another organizing methodological approaches (e.g. transfer learning, data augmentation, meta-learning, and frequency-aware modeling). Our study reviews over 230 papers, offering a comprehensive view across generative model types and constraint scenarios. We further analyze task-approach-method interactions using a Sankey diagram and highlight promising directions for future work, including adaptation of foundation models, holistic evaluation frameworks, and data-centric strategies for sample selection. This survey provides a timely and practical roadmap for researchers and practitioners aiming to advance generative modeling under limited data. Project website: https://sutd-visual-computing-group.github.io/gmdc-survey/.
URLs: https://sutd-visual-computing-group.github.io/gmdc-survey/.
Authors: Min Dai, Yuchao Dong, Yanwei Jia, Xun Yu Zhou
Abstract: We study Merton's expected utility maximization problem in an incomplete market, characterized by a factor process in addition to the stock price process, where all the model primitives are unknown. The agent under consideration is a price taker who has access only to the stock and factor value processes and the instantaneous volatility. We propose an auxiliary problem in which the agent can invoke policy randomization according to a specific class of Gaussian distributions, and prove that the mean of its optimal Gaussian policy solves the original Merton problem. With randomized policies, we are in the realm of continuous-time reinforcement learning (RL) recently developed in Wang et al. (2020) and Jia and Zhou (2022a, 2022b, 2023), enabling us to solve the auxiliary problem in a data-driven way without having to estimate the model primitives. Specifically, we establish a policy improvement theorem based on which we design both online and offline actor-critic RL algorithms for learning Merton's strategies. A key insight from this study is that RL in general and policy randomization in particular are useful beyond the purpose for exploration -- they can be employed as a technical tool to solve a problem that cannot be otherwise solved by mere deterministic policies. At last, we carry out both simulation and empirical studies in a stochastic volatility environment to demonstrate the decisive outperformance of the devised RL algorithms in comparison to the conventional model-based, plug-in method.
Authors: Bryan Kelly, Boris Kuznetsov, Semyon Malamud, Teng Andrea Xu, Yuan Zhang
Abstract: We show that a deep neural network (DNN) trained to construct a stochastic discount factor (SDF) admits a sharp additive decomposition that separates nonlinear characteristic discovery from the pricing rule that aggregates them. The economically relevant component of this decomposition is governed by a new object, the Portfolio Tangent Kernel (PTK), which captures the features learned by the network and induces an explicit linear factor pricing representation for the SDF. In population, the PTK-implied SDF converges to a ridge-regularized version of the true SDF, with the effective strength of regularization determined by the spectral complexity of the PTK. Using U.S. equity data, we show that the PTK representation delivers large and statistically significant performance gains, while its spectral complexity has risen sharply-by roughly a factor of six since the early 2000s-imposing increasingly tight limits on finite-sample pricing performance.
Authors: Khiem Le, Long Ho, Cuong Do, Danh Le-Phuoc, Kok-Seng Wong
Abstract: Domain shift is a formidable issue in Machine Learning that causes a model to suffer from performance degradation when tested on unseen domains. Federated Domain Generalization (FedDG) attempts to train a global model using collaborative clients in a privacy-preserving manner that can generalize well to unseen clients possibly with domain shift. However, most existing FedDG methods either cause additional privacy risks of data leakage or induce significant costs in client communication and computation, which are major concerns in the Federated Learning paradigm. To circumvent these challenges, here we introduce a novel architectural method for FedDG, namely gPerXAN, which relies on a normalization scheme working with a guiding regularizer. In particular, we carefully design Personalized eXplicitly Assembled Normalization to enforce client models selectively filtering domain-specific features that are biased towards local data while retaining discrimination of those features. Then, we incorporate a simple yet effective regularizer to guide these models in directly capturing domain-invariant representations that the global model's classifier can leverage. Extensive experimental results on two benchmark datasets, i.e., PACS and Office-Home, and a real-world medical dataset, Camelyon17, indicate that our proposed method outperforms other existing methods in addressing this particular problem.
Authors: Sunny Sanyal, Ravid Shwartz-Ziv, Alexandros G. Dimakis, Sujay Sanghavi
Abstract: Large Language Models (LLMs) are known for their performance, but we uncover a significant structural inefficiency: a phenomenon we term attention collapse. In many pre-trained decoder-style LLMs, the attention matrices in deeper layers degenerate, collapsing to near rank-one structures. These underutilized layers, which we call lazy layers, are redundant and impair model efficiency. To address this, we introduce Inheritune, a simple yet powerful training recipe designed to build smaller, stronger language models. Inheritune initializes a compact model by inheriting the potent early layers from a larger pre-trained model and then progressively trains and expands it. Our experiments on various models, including the GPT-2 family, demonstrate that models trained with Inheritune can match or even surpass the performance of their larger counterparts, despite having significantly fewer layers. This work presents a novel path toward model compression by design, enabling the creation of compact, yet highly performant language models. Code is available at https://github.com/sanyalsunny111/LLM-Inheritune.
Authors: Shengyuan Ye, Bei Ouyang, Tianyi Qian, Liekang Zeng, Jingyi Li, Jiangsu Du, Xiaowen Chu, Guoliang Xing, Xu Chen
Abstract: Large language models (LLMs) have unlocked a plethora of powerful applications at the network edge, such as intelligent personal assistants. Data privacy and security concerns have prompted a shift towards edge-based fine-tuning of personal LLMs, away from cloud reliance. However, this raises issues of computational intensity and resource scarcity, hindering training efficiency and feasibility. While current studies investigate parameter-efficient fine-tuning (PEFT) techniques to mitigate resource constraints, our analysis indicates that these techniques are not sufficiently resource-efficient for edge devices. To tackle these challenges, we propose Pluto and Charon (PAC), a time and memory efficient collaborative edge AI framework for personal LLMs fine-tuning. PAC breaks the resource wall of personal LLMs fine-tuning with a sophisticated algorithm-system co-design. (1) Algorithmically, PAC implements a personal LLMs fine-tuning technique that is efficient in terms of parameters, time, and memory. It utilizes Parallel Adapters to circumvent the need for a full backward pass through the LLM backbone. Additionally, an activation cache mechanism further streamlining the process by negating the necessity for repeated forward passes across multiple epochs. (2) Systematically, PAC leverages edge devices in close proximity, pooling them as a collective resource for in-situ personal LLMs fine-tuning, utilizing a hybrid data and pipeline parallelism to orchestrate distributed training. The use of the activation cache eliminates the need for forward pass through the LLM backbone,enabling exclusive fine-tuning of the Parallel Adapters using data parallelism. Extensive evaluation based on prototype implementation demonstrates that PAC remarkably outperforms state-of-the-art approaches, achieving up to 8.64x end-to-end speedup and up to 88.16% reduction in memory footprint.
Authors: Abraar Chaudhry, Elizaveta Rebrova
Abstract: We propose a flexible and theoretically supported framework for scalable nonnegative matrix factorization. The goal is to find nonnegative low-rank components directly from compressed measurements, accessing the original data only once or twice. We consider compression through randomized sketching methods that can be adapted to the data, or can be oblivious. We formulate optimization problems that only depend on the compressed data, but which can recover a nonnegative factorization which closely approximates the original matrix. The defined problems can be approached with a variety of algorithms, and in particular, we discuss variations of the popular multiplicative updates method for these compressed problems. We demonstrate the success of our approaches empirically and validate their performance in real-world applications.
Authors: Jose Andres Millan-Romera, Muhammad Shaheer, Miguel Fernandez-Cortizas, Martin R. Oswald, Holger Voos, Jose Luis Sanchez-Lopez
Abstract: Enabling robots to autonomously discover high-level spatial concepts (e.g., rooms and walls) from primitive geometric observations (e.g., planar surfaces) within 3D Scene Graphs is essential for robust indoor navigation and mapping. These graphs provide a hierarchical metric-semantic representation in which such concepts are organized. To further enhance graph-SLAM performance, Factorized 3D Scene Graphs incorporate these concepts as optimization factors that constrain relative geometry and enforce global consistency. However, both stages of this process remain largely manual: concepts are typically derived using hand-crafted, concept-specific heuristics, while factors and their covariances are likewise manually designed. This reliance on manual specification limits generalization across diverse environments and scalability to new concept classes. This paper presents a novel learning-based method that infers spatial concepts online from observed vertical planes and introduces them as optimizable factors within a SLAM backend, eliminating the need to handcraft concept generation, factor design, and covariance specification. We evaluate our approach in simulated environments with complex layouts, improving room detection by 20.7% and trajectory estimation by 19.2%, and further validate it on real construction sites, where room detection improves by 5.3% and map matching accuracy by 3.8%. Results confirm that learned factors can improve their handcrafted counterparts in SLAM systems and serve as a foundation for extending this approach to new spatial concepts.
Authors: Ruyi Tao, Veronica R. Cappelli, Kaiwei Liu, Marcus J. Hamilton, Christopher P. Kempes, Geoffrey B. Wes, Jiang Zhang
Abstract: Predicting company growth is a critical yet challenging task because observed dynamics blend an underlying structural growth trend with volatile fluctuations. Here, we propose a Scaling-Theory-Informed Machine Learning (STIML) framework that integrates a scaling-based growth model to capture the mechanism-driven average trend, together with a data-driven forecasting model to learn the residual fluctuations. Using Compustat annual financial statement data (1950--2019) for 31,553 North American companies, we extend the growth model beyond assets to multiple financial indicators, and evaluate STIML against growth model-only and purely data-driven baselines. Across 16 target variables, we show that company growth exhibits a clear separation between trend-driven predictability and fluctuation-driven predictability, with their relative importance depending strongly on company size and volatility. Interpretability analyses further show that STIML captures multivariate dependencies beyond simple autocorrelation, and that macroeconomic variables contribute significantly less to predictive performance on average. Moreover, we find the scaling-based growth model overlooks asymmetric deviations, which instead contain the structured and learnable signals, suggesting a path to refine mechanistic models.
Authors: Viacheslav Kuzmin, Wilfrid Somogyi, Ekaterina Pankovets, Alexey Melnikov
Abstract: In the current quantum computing paradigm, significant focus is placed on the reduction or mitigation of quantum decoherence. When designing new quantum processing units, the general objective is to reduce the amount of noise qubits are subject to, and in algorithm design, a large effort is underway to provide scalable error correction or mitigation techniques. Yet some previous work has indicated that certain classes of quantum algorithms, such as quantum machine learning, may, in fact, be intrinsically robust to or even benefit from the presence of a small amount of noise. Here, we demonstrate that noise levels in quantum hardware can be effectively tuned to enhance the ability of quantum neural networks to generalize data, acting akin to regularisation in classical neural networks. As an example, we consider two regression tasks, where, by tuning the noise level in the circuit, we demonstrated improvement of the validation mean squared error loss. Moreover, we demonstrate the method's effectiveness by numerically simulating quantum neural network training on a realistic model of a noisy superconducting quantum computer.
Authors: Martin Holler, Erion Morina
Abstract: This paper addresses the problem of uniqueness in learning physical laws for systems of partial differential equations (PDEs). Contrary to most existing approaches, it considers a framework of structured model learning, where existing, approximately correct physical models are augmented with components that are learned from data. The main results of the paper are a uniqueness and a convergence result that cover a large class of PDEs and a suitable class of neural networks used for approximating the unknown model components. The uniqueness result shows that, in the limit of full, noiseless measurements, a unique identification of the unknown model components as functions is possible as classical regularization-minimizing solutions of the PDE system. This result is complemented by a convergence result showing that model components learned as parameterized neural networks from incomplete, noisy measurements approximate the regularization-minimizing solutions of the PDE system in the limit. These results are possible under specific properties of the approximating neural networks and due to a dedicated choice of regularization. With this, a practical contribution of this analytic paper is to provide a class of model learning frameworks different to standard settings where uniqueness can be expected in the limit of full measurements.
Authors: Tengyuan Liang, Kulunu Dharmakeerthi, Takuya Koriyama
Abstract: Adding noise is easy; what about denoising? Diffusion is easy; what about reverting a diffusion? Diffusion-based generative models aim to denoise a Langevin diffusion chain, moving from a log-concave equilibrium measure $\nu$, say an isotropic Gaussian, back to a complex, possibly non-log-concave initial measure $\mu$. The score function performs denoising, moving backward in time, and predicting the conditional mean of the past location given the current one. We show that score denoising is the optimal backward map in transportation cost. What is its localization uncertainty? We show that the curvature function determines this localization uncertainty, measured as the conditional variance of the past location given the current. We study in this paper the effectiveness of the diffuse-then-denoise process: the contraction of the forward diffusion chain, offset by the possible expansion of the backward denoising chain, governs the denoising difficulty. For any initial measure $\mu$, we prove that this offset net contraction at time $t$ is characterized by the curvature complexity of a smoothed $\mu$ at a specific signal-to-noise ratio (SNR) scale $r(t)$. We discover that the multi-scale curvature complexity collectively determines the difficulty of the denoising chain. Our multi-scale complexity quantifies a fine-grained notion of average-case curvature instead of the worst-case. Curiously, it depends on an integrated tail function, measuring the relative mass of locations with positive curvature versus those with negative curvature; denoising at a specific SNR scale is easy if such an integrated tail is light. We conclude with several non-log-concave examples to demonstrate how the multi-scale complexity probes the bottleneck SNR for the diffuse-then-denoise process.
Authors: Benjamin Peters, Ayush Mohanty, Xiaolei Fang, Nagi Gebraeel, Stephen K. Robinson
Abstract: Deep-space habitats are complex systems that must operate autonomously over extended durations without ground-based maintenance. These systems are vulnerable to multiple, often unknown, failure modes that affect different subsystems and sensors in mode-specific ways. Developing accurate remaining useful life (RUL) prognostics is challenging, especially when failure labels are unavailable and sensor relevance varies by failure mode. In this paper, we propose an unsupervised prognostics framework that jointly identifies latent failure modes and selects informative sensors using only unlabeled training data. The methodology consists of two phases. In the offline phase, we model system failure times using a mixture of Gaussian regressions and apply an Expectation-Maximization algorithm to cluster degradation trajectories and select mode-specific sensors. In the online phase, we extract low-dimensional features from the selected sensors to diagnose the active failure mode and predict RUL using a weighted regression model. We demonstrate the effectiveness of our approach on a simulated dataset that reflects deep-space telemetry characteristics and on a real-world engine degradation dataset, showing improved accuracy and interpretability over existing methods.
Authors: Viorica P\u{a}tr\u{a}ucean, Xu Owen He, Joseph Heyward, Chuhan Zhang, Mehdi S. M. Sajjadi, George-Cristian Muraru, Artem Zholus, Mahdi Karami, Ross Goroshin, Yutian Chen, Simon Osindero, Jo\~ao Carreira, Razvan Pascanu
Abstract: We propose a novel block for \emph{causal} video modelling. It relies on a time-space-channel factorisation with dedicated blocks for each dimension: gated linear recurrent units (LRUs) perform information mixing over time, self-attention layers perform mixing over space, and MLPs over channels. The resulting architecture \emph{TRecViT} is causal and shows strong performance on sparse and dense tasks, trained in supervised or self-supervised regimes, being the first causal video model in the state-space models family. Notably, our model outperforms or is on par with the popular (non-causal) ViViT-L model on large scale video datasets (SSv2, Kinetics400), while having $3\times$ less parameters, $12\times$ smaller memory footprint, and $5\times$ lower FLOPs count than the full self-attention ViViT, with an inference throughput of about 300 frames per second, running comfortably in real-time. When compared with causal transformer-based models (TSM, RViT) and other recurrent models like LSTM, TRecViT obtains state-of-the-art results on the challenging SSv2 dataset. Code and checkpoints are available online https://github.com/google-deepmind/trecvit.
Authors: Tong Wu
Abstract: Tensors, which give a faithful and effective representation to deliver the intrinsic structure of multi-dimensional data, play a crucial role in an increasing number of signal processing and machine learning problems. However, tensor data are often accompanied by arbitrary signal corruptions, including missing entries and sparse noise. A fundamental challenge is to reliably extract the meaningful information from corrupted tensor data in a statistically and computationally efficient manner. This paper develops a scaled gradient descent (ScaledGD) algorithm to directly estimate the tensor factors with tailored spectral initializations under the tensor-tensor product (t-product) and tensor singular value decomposition (t-SVD) framework. With tailored variants for tensor robust principal component analysis, (robust) tensor completion and tensor regression, we theoretically show that ScaledGD achieves linear convergence at a constant rate that is independent of the condition number of the ground truth low-rank tensor, while maintaining the low per-iteration cost of gradient descent. To the best of our knowledge, ScaledGD is the first algorithm that provably has such properties for low-rank tensor estimation with the t-SVD. Finally, numerical examples are provided to demonstrate the efficacy of ScaledGD in accelerating the convergence rate of ill-conditioned low-rank tensor estimation in a number of applications.
Authors: Floris Holstege, Mackenzie Jorgensen, Kirtan Padh, Jurriaan Parie, Krsto Prorokovic, Joel Persson, Lukas Snoek
Abstract: Algorithms are increasingly used to automate or aid human decisions, yet recent research shows that these algorithms may exhibit bias across legally protected demographic groups. However, data on these groups may be unavailable to organizations or external auditors due to privacy legislation. This paper studies bias detection using an unsupervised bias detection tool when data on demographic groups are unavailable. We collaborated with the Dutch Executive Agency for Education to audit an algorithm that was used to assign risk scores to college students at the national level in the Netherlands between 2012-2023. Our audit covers more than 250,000 students across the country. The unsupervised bias detection tool highlights known disparities between students with a non-European migration background and students with a Dutch or European-migration background. Our contributions are two-fold: (1) we assess bias in a real-world, large-scale, and high-stakes decision-making process by a governmental organization; (2) we provide the unsupervised bias detection tool in an open-source library for others to use to complete bias audits. Our work serves as a starting point for a deliberative assessment by human experts to evaluate potential discrimination in algorithmic decision-making.
Authors: Antonio Purificato, Maria Sofia Bucarelli, Anil Kumar Nelakanti, Andrea Bacciu, Fabrizio Silvestri, Amin Mantrach
Abstract: Reliably labelling data typically requires annotations from multiple human workers. However, humans are far from being perfect. Hence, it is a common practice to aggregate labels gathered from multiple annotators to make a more confident estimate of the true label. Among many aggregation methods, the simple and well known Majority Vote (MV) selects the class label polling the highest number of votes. However, despite its importance, the optimality of MV's label aggregation has not been extensively studied. We address this gap in our work by characterising the conditions under which MV achieves the theoretically optimal lower bound on label estimation error. Our results capture the tolerable limits on annotation noise under which MV can optimally recover labels for a given class distribution. This certificate of optimality provides a more principled approach to model selection for label aggregation as an alternative to otherwise inefficient practices that sometimes include higher experts, gold labels, etc., that are all marred by the same human uncertainty despite huge time and monetary costs. Experiments on both synthetic and real world data corroborate our theoretical findings.
Authors: Hanjiang Hu, Alexander Robey, Changliu Liu
Abstract: Large language models (LLMs) are shown to be vulnerable to jailbreaking attacks where adversarial prompts are designed to elicit harmful responses. While existing defenses effectively mitigate single-turn attacks by detecting and filtering unsafe inputs, they fail against multi-turn jailbreaks that exploit contextual drift over multiple interactions, gradually leading LLMs away from safe behavior. To address this challenge, we propose a safety steering framework grounded in safe control theory, ensuring invariant safety in multi-turn dialogues. Our approach models the dialogue with LLMs using state-space representations and introduces a novel neural barrier function (NBF) to detect and filter harmful queries emerging from evolving contexts proactively. Our method achieves invariant safety at each turn of dialogue by learning a safety predictor that accounts for adversarial queries, preventing potential context drift toward jailbreaks. Extensive experiments under multiple LLMs show that our NBF-based safety steering outperforms safety alignment, prompt-based steering and lightweight LLM guardrails baselines, offering stronger defenses against multi-turn jailbreaks while maintaining a better trade-off among safety, helpfulness and over-refusal. Check out the website here https://sites.google.com/view/llm-nbf/home.
Authors: Yuqi Hu, Longguang Wang, Xian Liu, Ling-Hao Chen, Yuwei Guo, Yukai Shi, Ce Liu, Anyi Rao, Zeyu Wang, Hui Xiong
Abstract: Understanding and replicating the real world is a critical challenge in Artificial General Intelligence (AGI) research. To achieve this, many existing approaches, such as world models, aim to capture the fundamental principles governing the physical world, enabling more accurate simulations and meaningful interactions. However, current methods often treat different modalities, including 2D (images), videos, 3D, and 4D representations, as independent domains, overlooking their interdependencies. Additionally, these methods typically focus on isolated dimensions of reality without systematically integrating their connections. In this survey, we present a unified survey for multimodal generative models that investigate the progression of data dimensionality in real-world simulation. Specifically, this survey starts from 2D generation (appearance), then moves to video (appearance+dynamics) and 3D generation (appearance+geometry), and finally culminates in 4D generation that integrate all dimensions. To the best of our knowledge, this is the first attempt to systematically unify the study of 2D, video, 3D and 4D generation within a single framework. To guide future research, we provide a comprehensive review of datasets, evaluation metrics and future directions, and fostering insights for newcomers. This survey serves as a bridge to advance the study of multimodal generative models and real-world simulation within a unified framework.
Authors: Zeyue Tian, Zhaoyang Liu, Yizhu Jin, Ruibin Yuan, Xu Tan, Qifeng Chen, Wei Xue, Yike Guo
Abstract: Audio and music generation based on flexible multimodal control signals is a widely applicable topic, with the following key challenges: 1) a unified multimodal modeling framework, and 2) large-scale, high-quality training data. As such, we propose AudioX, a unified framework for anything-to-audio generation that integrates varied multimodal conditions (i.e., text, video, and audio signals) in this work. The core design in this framework is a Multimodal Adaptive Fusion module, which enables the effective fusion of diverse multimodal inputs, enhancing cross-modal alignment and improving overall generation quality. To train this unified model, we construct a large-scale, high-quality dataset, IF-caps, comprising over 7 million samples curated through a structured data annotation pipeline. This dataset provides comprehensive supervision for multimodal-conditioned audio generation. We benchmark AudioX against state-of-the-art methods across a wide range of tasks, finding that our model achieves superior performance, especially in text-to-audio and text-to-music generation. These results demonstrate our method is capable of audio generation under multimodal control signals, showing powerful instruction-following potential. The code and datasets will be available at https://zeyuet.github.io/AudioX/.
Authors: Corina Catarau-Cotutiu, Esther Mondragon, Eduardo Alonso
Abstract: The ability of artificial intelligence agents to make optimal decisions and generalise them to different domains and tasks is compromised in complex scenarios. One way to address this issue has focused on learning efficient representations of the world and on how the actions of agents affect them in state-action transitions. Whereas such representations are procedurally efficient, they lack structural richness. To address this problem, we propose to enhance the agent's ontology and extend the traditional conceptualisation of trajectories to provide a more nuanced view of task execution. Structurally Enriched Trajectories (SETs) extend the encoding of sequences of states and their transitions by incorporating hierarchical relations between objects, interactions, and affordances. SETs are built as multi-level graphs, providing a detailed representation of the agent dynamics and a transferable functional abstraction of the task. SETs are integrated into an architecture, Structurally Enriched Trajectory Learning and Encoding (SETLE), that employs a heterogeneous graph-based memory structure of multi-level relational dependencies essential for generalisation. We demonstrate that SETLE can support downstream tasks, enabling agents to recognise task relevant structural patterns across CREATE and MiniGrid environments. Finally, we integrate SETLE with reinforcement learning and show measurable improvements in downstream performance, including breakthrough success rates in complex, sparse-reward tasks.
Authors: Mithun Goutham, Riccardo DalferroNucci, Stephanie Stockar, Meghna Menon, Sneha Nayak, Harshad Zade, Chetan Patel, Mario Santillo
Abstract: Accurately estimating decision boundaries in black box systems is critical when ensuring safety, quality, and feasibility in real-world applications. However, existing methods iteratively refine boundary estimates by sampling in regions of uncertainty, without providing guarantees on the closeness to the decision boundary and also result in unnecessary exploration that is especially disadvantageous when evaluations are costly. This paper presents $\varepsilon$-Neighborhood Decision-Boundary Governed Estimation (EDGE), a sample efficient and function-agnostic algorithm that leverages the intermediate value theorem to estimate the location of the decision boundary of a black box binary classifier within a user-specified $\varepsilon$-neighborhood. To demonstrate applicability, a case study is presented of an electric grid stability problem with uncertain renewable power injection. Evaluations are conducted on three test functions, where it is seen that the EDGE algorithm demonstrates superior sample efficiency and better boundary approximation than adaptive sampling techniques and grid-based searches.
Authors: Maodong Li, Jiying Zhang, Zhe Wang, Bin Feng, Wenqi Zeng, Dechin Chen, Zhijun Pan, Yu Li, Zijing Liu, Yi Isaac Yang
Abstract: The kinetics and dynamics of drug-protein binding and dissociation are crucial to understanding drug absorption and metabolism. Despite advances in artificial intelligence (AI) tools for drug-protein interaction studies, existing training datasets remain limited to static structures or quasi-static conformations. This paper proposes a novel computational approach for rapidly generating drug-protein dissociation trajectories and presents the inaugural dynamically time-resolved 4-D (t, x, y, z) trajectory database DD-13M. This dataset captures over 26,000 complete dissociation processes for 565 ligand-protein complexes, providing nearly 13 million frames of all-atom simulation trajectories. A deep equivariant generative model, UnbindingFlow, was trained using the DD-13M dataset. This model has the capacity to produce dissociation trajectories for novel targets whilst accurately predicting their rate constants (koff). DD-13M introduces a new type of training dataset for AI models, establishing a de novo paradigm for studying the dynamics of drug-protein interactions.
Authors: Harry Dong, Bilge Acun, Beidi Chen, Yuejie Chi
Abstract: Due to long generations, large language model (LLM) math reasoning demands significant computational resources and time. While many existing efficient inference methods have been developed with excellent performance preservation on language tasks, they often severely degrade math performance. In this paper, we propose Caprese, a resource-efficient distillation method to recover lost capabilities from deploying efficient inference methods, focused primarily in feedforward blocks. With original weights unperturbed, roughly 1% of additional parameters, and only 20K synthetic training samples, we are able to recover much if not all of the reasoning capabilities lost from efficient inference for thinking LLMs and without harm to language tasks for instruct LLMs. Moreover, Caprese slashes the number of active parameters (~2B cut for Gemma 2 9B and Llama 3.1 8B) and integrates cleanly into existing model layers to reduce latency (>16% time-to-next-token reduction) while encouraging response brevity (up to 8.5% fewer tokens).
Authors: Sen Fang, Weiyuan Ding, Mengshi Zhang, Zihao Chen, Bowen Xu
Abstract: Evaluating the programming robustness of large language models (LLMs) is paramount for ensuring their reliability in AI-based software development. However, adversarial attacks exhibit fundamental limitations that compromise fair robustness assessment: they demonstrate contradictory evaluation outcomes where different attack strategies tend to favor different models, and more critically, they operate solely through external perturbations, failing to capture the intrinsic stability essential for autonomous coding agents where subsequent inputs are endogenously generated by the model itself. We introduce EVALOOOP, a novel assessment framework that evaluates robustness from a self-consistency perspective, leveraging the natural duality inherent in software engineering tasks (e.g., code generation and code summarization). EVALOOOP establishes a self-contained feedback loop where an LLM iteratively transforms between code and natural language until functional failure occurs, with robustness quantified by a novel Average Sustainable Loops (ASL) metric-the mean number of iterations maintaining functional correctness across benchmark tasks. This cyclical strategy intrinsically evaluates robustness without relying on external attack configurations, providing a unified metric that reveals how effectively LLMs preserve semantic integrity through sustained self-referential transformations. We evaluate 96 popular LLMs, ranging from 0.5B to 685B parameters, on EVALOOOP equipped with the MBPP Plus benchmark, and found that EVALOOOP typically induces a 2.65%-47.62% absolute drop in pass@1 accuracy within ten loops. Intriguingly, robustness does not always align with initial performance (i.e., one-time query); for instance, Qwen3-235B-A22B-Instruct-2507, despite inferior initial code generation compared to OpenAI's o-series models and DeepSeek-V3, demonstrated the superior robustness (ASL score).
Authors: Junlin Wang, Zhiyun Lin
Abstract: Learning effective visual representations for robotic manipulation remains a fundamental challenge due to the complex body dynamics involved in action execution. In this paper, we study how visual representations that carry body-relevant cues can enable efficient policy learning for downstream robotic manipulation tasks. We present $\textbf{I}$nter-token $\textbf{Con}$trast ($\textbf{ICon}$), a contrastive learning method applied to the token-level representations of Vision Transformers (ViTs). ICon enforces a separation in the feature space between agent-specific and environment-specific tokens, resulting in agent-centric visual representations that embed body-specific inductive biases. This framework can be seamlessly integrated into end-to-end policy learning by incorporating the contrastive loss as an auxiliary objective. Our experiments show that ICon not only improves policy performance across various manipulation tasks but also facilitates policy transfer across different robots. The project website: https://inter-token-contrast.github.io/icon/
Authors: Zonghao Chen, Toni Karvonen, Heishiro Kanagawa, Fran\c{c}ois-Xavier Briol, Chris. J. Oates
Abstract: Approximation of a target probability distribution using a finite set of points is a problem of fundamental importance in numerical integration. Several authors have proposed to select points by minimising a maximum mean discrepancy (MMD), but the non-convexity of this objective typically precludes global minimisation. Instead, we consider the concept of \emph{stationary points of the MMD} which, in contrast to points globally minimising the MMD, can be accurately computed. Our main contributions are two-fold and theoretical in nature. We first prove the (perhaps surprising) result that, for integrands in the associated reproducing kernel Hilbert space, the numerical integration error of stationary MMD points vanishes \emph{faster} than the MMD. Motivated by this \emph{super-convergence} property, we consider MMD gradient flows as a practical strategy for computing stationary points of the MMD. We then prove that MMD gradient flow can indeed compute stationary MMD points, based on a refined convergence analysis that establishes a novel non-asymptotic finite-particle error bound.
Authors: Skyler Wu, Shihao Yang, S. C. Kou
Abstract: In the era of AI, neural networks have become increasingly popular for modeling, inference, and prediction, largely due to their potential for universal approximation. With the proliferation of such deep learning models, a question arises: are leaner statistical methods still relevant? To shed insight on this question, we employ the mechanistic nonlinear ordinary differential equation (ODE) inverse problem as a testbed, using the physics-informed neural network (PINN) as a representative of the deep learning paradigm and manifold-constrained Gaussian process inference (MAGI) as a representative of statistically principled methods. Through case studies involving the SEIR model from epidemiology and the Lorenz model from chaotic dynamics, we demonstrate that statistical methods are far from obsolete, especially when working with sparse and noisy observations. On tasks such as parameter inference and trajectory reconstruction, statistically principled methods consistently achieve lower bias and variance, while using far fewer parameters and requiring less hyperparameter tuning. Statistical methods can also decisively outperform deep learning models on out-of-sample future prediction, where the absence of relevant data often leads overparameterized models astray. Additionally, we find that statistically principled approaches are more robust to accumulation of numerical imprecision and can represent the underlying system more faithfully to the true governing ODEs.
Authors: Fengxiang Wang, Mingshuo Chen, Xuming He, Yi-Fan Zhang, Yueying Li, Feng Liu, Zijie Guo, Zhenghao Hu, Jiong Wang, Jingyi Xu, Zhangrui Li, Junchao Gong, Di Wang, Fenghua Ling, Ben Fei, Weijia Li, Long Lan, Wenjing Yang
Abstract: Existing benchmarks for multimodal learning in Earth science offer limited, siloed coverage of Earth's spheres and their cross-sphere interactions, typically restricting evaluation to the human-activity sphere of atmosphere and to at most 16 tasks. These limitations: narrow-source heterogeneity (single/few data sources), constrained scientific granularity, and limited-sphere extensibility. Therefore, we introduce OmniEarth-Bench, the first multimodal benchmark that systematically spans all six spheres: atmosphere, lithosphere, oceanosphere, cryosphere, biosphere, and human-activity sphere, and cross-spheres. Built with a scalable, modular-topology data inference framework and native multi-observation sources and expert-in-the-loop curation, OmniEarth-Bench produces 29,855 standardized, expert-curated annotations. All annotations are organized into a four-level hierarchy (Sphere, Scenario, Ability, Task), encompassing 109 expert-curated evaluation tasks. Experiments on 9 state-of-the-art MLLMs reveal that even the most advanced models struggle with our benchmarks, where none of them reach 35% accuracy, revealing systematic gaps in Earth-system cognitive ability. The dataset and evaluation code were released at OmniEarth-Bench (https://anonymous.4open.science/r/OmniEarth-Bench-B1BD).
URLs: https://anonymous.4open.science/r/OmniEarth-Bench-B1BD).
Authors: Lukas Meyer, Josef Gr\"un, Maximilian Weiherer, Bernhard Egger, Marc Stamminger, Linus Franke
Abstract: We present MS-Splatting -- a multi-spectral 3D Gaussian Splatting (3DGS) framework that is able to generate multi-view consistent novel views from images of multiple, independent cameras with different spectral domains. In contrast to previous approaches, our method does not require cross-modal camera calibration and is versatile enough to model a variety of different spectra, including thermal and near-infra red, without any algorithmic changes. Unlike existing 3DGS-based frameworks that treat each modality separately (by optimizing per-channel spherical harmonics) and therefore fail to exploit the underlying spectral and spatial correlations, our method leverages a novel neural color representation that encodes multi-spectral information into a learned, compact, per-splat feature embedding. A shallow multi-layer perceptron (MLP) then decodes this embedding to obtain spectral color values, enabling joint learning of all bands within a unified representation. Our experiments show that this simple yet effective strategy is able to improve multi-spectral rendering quality, while also leading to improved per-spectra rendering quality over state-of-the-art methods. We demonstrate the effectiveness of this new technique in agricultural applications to render vegetation indices, such as normalized difference vegetation index (NDVI).
Authors: Tianyu Qi, Lei Xue, Yufeng Zhan, Xiaobo Ma
Abstract: The growing use of large pre-trained models in edge computing has made model inference on mobile clients both feasible and popular. Yet these devices remain vulnerable to adversarial attacks, threatening model robustness and security. Federated adversarial training (FAT) offers a promising solution by enhancing robustness while preserving client privacy. However, FAT often yields a generalized global model that struggles with heterogeneous client data, leading to limited personalization and significant communication overhead. In this paper, we propose \textit{Lorica}, a personalized synergistic adversarial training framework that delivers customized defense models through a two-phase process. In Phase 1, \textit{Lorica} applies LoRA-FA for local adversarial fine-tuning, enabling personalized robustness while reducing communication by uploading only LoRA-FA parameters. In Phase 2, a forward-gating selection strategy improves benign accuracy, further refining the personalized model. This yields tailored defense models that effectively balance robustness and accuracy. Extensive experiments on benchmark datasets demonstrate that \textit{Lorica} can achieve up to 68$\times$ improvements in communication efficiency compared to state-of-the-art algorithms, while achieving up to 29.9\% and 52.2\% enhancements in adversarial robustness and benign accuracy, respectively.
Authors: Subhojyoti Mukherjee, Viet Dac Lai, Raghavendra Addanki, Ryan Rossi, Seunghyun Yoon, Trung Bui, Anup Rao, Jayakumar Subramanian, Branislav Kveton
Abstract: Offline reinforcement learning (RL) is a variant of RL where the policy is learned from a previously collected dataset of trajectories and rewards. In our work, we propose a practical approach to offline RL with large language models (LLMs). We recast the problem as reward-weighted fine-tuning, which can be solved using similar techniques to supervised fine-tuning (SFT). To showcase the value of our approach, we apply it to learning short-horizon question-answering policies of a fixed length, where the agent reasons about potential answers or asks clarifying questions. Our work stands in a stark contrast to state-of-the-art methods in this domain, based on SFT and direct preference optimization, which have additional hyper-parameters and do not directly optimize for rewards. We compare to them empirically, and report major gains in both optimized rewards and language quality.
Authors: Yang Liu, Jiaqi Li, Zilong Zheng
Abstract: Rule-based reasoning is acknowledged as one of the fundamental problems of reasoning. While recent studies show that large reasoning models (LRMs) have remarkable reasoning capabilities enhanced by reinforcement learning (RL), real applications still face severe challenges due to variations in rule formats, types, and complexity. To mitigate this issue, we introduce RuleReasoner, an effective method for rule-based reasoning via a wide collection of curated tasks and a novel domain-aware dynamic sampling approach in RL. Specifically, RuleReasoner resamples each training batch by updating the domain weights based on historical rewards. This facilitates domain balance and active learning schedules for RL, obviating static mix-training engineered by human. Evaluations on in-distribution (ID) and out-of-distribution (OOD) benchmarks reveal that RuleReasoner outperforms frontier LRMs by a significant margin ($\Delta$4.1% on eight ID tasks and $\Delta$10.4% on three OOD tasks over OpenAI-o1). Notably, our approach also exhibits higher computational efficiency compared to prior methods.
Authors: Viktoria Patapovich, Maniraman Periyasamy, Mo Kordzanganeh, Alexey Melnikov
Abstract: Quantum machine learning has shown promise for high-dimensional data analysis, yet many existing approaches rely on linear unitary operations and shared trainable parameters across outputs. These constraints limit expressivity and scalability relative to the multi-layered, non-linear architectures of classical deep networks. We introduce superposed parameterised quantum circuits to overcome these limitations. By combining flip-flop quantum random-access memory with repeat-until-success protocols, a superposed parameterised quantum circuit embeds an exponential number of parameterised sub-models in a single circuit and induces polynomial activation functions through amplitude transformations and post-selection. We provide an analytic description of the architecture, showing how multiple parameter sets are trained in parallel while non-linear amplitude transformations broaden representational power beyond conventional quantum kernels. Numerical experiments underscore these advantages: on a 1D step-function regression a two-qubit superposed parameterised quantum circuit cuts the mean-squared error by three orders of magnitude versus a parameter-matched variational baseline; on a 2D star-shaped two-dimensional classification task, introducing a quadratic activation lifts accuracy to 81.4\% and reduces run-to-run variance three-fold. These results position superposed parameterised quantum circuits as a hardware-efficient route toward deeper, more versatile parameterised quantum circuits capable of learning complex decision boundaries.
Authors: Renyang Liu, Guanlin Li, Tianwei Zhang, See-Kiong Ng
Abstract: Recent advances in image generation models (IGMs), particularly diffusion-based architectures such as Stable Diffusion (SD), have markedly enhanced the quality and diversity of AI-generated visual content. However, their generative capability has also raised significant ethical, legal, and societal concerns, including the potential to produce harmful, misleading, or copyright-infringing content. To mitigate these concerns, machine unlearning (MU) emerges as a promising solution by selectively removing undesirable concepts from pretrained models. Nevertheless, the robustness and effectiveness of existing unlearning techniques remain largely unexplored, particularly in the presence of multi-modal adversarial inputs. To bridge this gap, we propose Recall, a novel adversarial framework explicitly designed to compromise the robustness of unlearned IGMs. Unlike existing approaches that predominantly rely on adversarial text prompts, Recall exploits the intrinsic multi-modal conditioning capabilities of diffusion models by efficiently optimizing adversarial image prompts with guidance from a single semantically relevant reference image. Extensive experiments across ten state-of-the-art unlearning methods and diverse tasks show that Recall consistently outperforms existing baselines in terms of adversarial effectiveness, computational efficiency, and semantic fidelity with the original textual prompt. These findings reveal critical vulnerabilities in current unlearning mechanisms and underscore the need for more robust solutions to ensure the safety and reliability of generative models. Code and data are publicly available at \textcolor{blue}{https://github.com/ryliu68/RECALL}.
Authors: Xiaojie Li, Zhijie Cai, Nan Qi, Chao Dong, Guangxu Zhu, Haixia Ma, Qihui Wu, Shi Jin
Abstract: The expansion of the low-altitude economy has underscored the significance of Low-Altitude Network Coverage (LANC) prediction for designing aerial corridors. While accurate LANC forecasting hinges on the antenna beam patterns of Base Stations (BSs), these patterns are typically proprietary and not readily accessible. Operational parameters of BSs, which inherently contain beam information, offer an opportunity for data-driven low-altitude coverage prediction. However, collecting extensive low-altitude road test data is cost-prohibitive, often yielding only sparse samples per BS. This scarcity results in two primary challenges: imbalanced feature sampling due to limited variability in high-dimensional operational parameters against the backdrop of substantial changes in low-dimensional sampling locations, and diminished generalizability stemming from insufficient data samples. To overcome these obstacles, we introduce a dual strategy comprising expert knowledge-based feature compression and disentangled representation learning. The former reduces feature space complexity by leveraging communications expertise, while the latter enhances model generalizability through the integration of propagation models and distinct subnetworks that capture and aggregate the semantic representations of latent features. Experimental evaluation confirms the efficacy of our framework, yielding a 7% reduction in error compared to the best baseline algorithm. Real-network validations further attest to its reliability, achieving practical prediction accuracy with MAE errors at the 5dB level.
Authors: Lakshya A Agrawal, Shangyin Tan, Dilara Soylu, Noah Ziems, Rishi Khare, Krista Opsahl-Ong, Arnav Singhvi, Herumb Shandilya, Michael J Ryan, Meng Jiang, Christopher Potts, Koushik Sen, Alexandros G. Dimakis, Ion Stoica, Dan Klein, Matei Zaharia, Omar Khattab
Abstract: Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of rollouts to learn new tasks. We argue that the interpretable nature of language often provides a much richer learning medium for LLMs, compared to policy gradients derived from sparse, scalar rewards. To test this, we introduce GEPA (Genetic-Pareto), a prompt optimizer that thoroughly incorporates natural language reflection to learn high-level rules from trial and error. Given any AI system containing one or more LLM prompts, GEPA samples trajectories (e.g., reasoning, tool calls, and tool outputs) and reflects on them in natural language to diagnose problems, propose and test prompt updates, and combine complementary lessons from the Pareto frontier of its own attempts. As a result of GEPA's design, it can often turn even just a few rollouts into a large quality gain. Across six tasks, GEPA outperforms GRPO by 6% on average and by up to 20%, while using up to 35x fewer rollouts. GEPA also outperforms the leading prompt optimizer, MIPROv2, by over 10% (e.g., +12% accuracy on AIME-2025), and demonstrates promising results as an inference-time search strategy for code optimization. We release our code at https://github.com/gepa-ai/gepa .
Authors: Shitian Yang, Deyu Li, Xiaoke Jiang, Lei Zhang
Abstract: RGB-based 3D tasks, e.g., 3D detection, depth estimation, 3D keypoint estimation, still suffer from scarce, expensive annotations and a thin augmentation toolbox, since many image transforms, including rotations and warps, disrupt geometric consistency. While horizontal flipping and color jitter are standard, rigorous 3D rotation augmentation has surprisingly remained absent from RGB-based pipelines, largely due to the misconception that it requires scene depth or scene reconstruction. In this paper, we introduce 3DRot, a plug-and-play augmentation that rotates and mirrors images about the camera's optical center while synchronously updating RGB images, camera intrinsics, object poses, and 3D annotations to preserve projective geometry, achieving geometry-consistent rotations and reflections without relying on any scene depth. We first validate 3DRot on a classical RGB-based 3D task, monocular 3D detection. On SUN RGB-D, inserting 3DRot into a frozen DINO-X + Cube R-CNN pipeline raises $IoU_{3D}$ from 43.21 to 44.51, cuts rotation error (ROT) from 22.91$^\circ$ to 20.93$^\circ$, and boosts $mAP_{0.5}$ from 35.70 to 38.11; smaller but consistent gains appear on a cross-domain IN10 split. Beyond monocular detection, adding 3DRot on top of the standard BTS augmentation schedule further improves NYU Depth v2 from 0.1783 to 0.1685 in abs-rel (and 0.7472 to 0.7548 in $\delta<1.25$), and reduces cross-dataset error on SUN RGB-D. On KITTI, applying the same camera-centric rotations in MVX-Net (LiDAR+RGB) raises moderate 3D AP from about 63.85 to 65.16 while remaining compatible with standard 3D augmentations.
Authors: Adam E. Essex (Loughborough University, England), Natalia B. Janson (Loughborough University, England), Rachel A. Norris (Loughborough University, England), Alexander G. Balanov (Loughborough University, England)
Abstract: Despite explosive expansion of artificial intelligence based on artificial neural networks (ANNs), these are employed as "black boxes'', as it is unclear how, during learning, they form memories or develop unwanted features, including spurious memories and catastrophic forgetting. Much research is available on isolated aspects of learning ANNs, but due to their high dimensionality and non-linearity, their comprehensive analysis remains a challenge. In ANNs, knowledge is thought to reside in connection weights or in attractor basins, but these two paradigms are not linked explicitly. Here we comprehensively analyse mechanisms of memory formation in an 81-neuron Hopfield network undergoing Hebbian learning by revealing bifurcations leading to formation and destruction of attractors and their basin boundaries. We show that, by affecting evolution of connection weights, the applied stimuli induce a pitchfork and then a cascade of saddle-node bifurcations creating new attractors with their basins that can code true or spurious memories, and an abrupt disappearance of old memories (catastrophic forgetting). With successful learning, new categories are represented by the basins of newly born point attractors, and their boundaries by the stable manifolds of new saddles. With this, memorisation and forgetting represent two manifestations of the same mechanism. Our strategy to analyse high-dimensional learning ANNs is universal and applicable to recurrent ANNs of any form. The demonstrated mechanisms of memory formation and of catastrophic forgetting shed light on the operation of a wider class of recurrent ANNs and could aid the development of approaches to mitigate their flaws.
Authors: Ruiyu Zhang, Lin Nie, Xin Zhao
Abstract: Hierarchical Evolutionary Agent Simulation (HEAS) is a Python framework that unifies layered agent-based modeling with evolutionary optimization and tournament evaluation in a single, reproducible workflow. HEAS represents models as hierarchies of lightweight processes ("streams") scheduled in deterministic layers that read and write a shared context, making cross-scale couplings explicit and auditable. A compact API and CLI-simulate, optimize, evaluate-expose single- and multi-objective evolution, PyTorch policy integration via parameter flattening/unflattening, and general tournament tooling with user-defined scoring and voting rules. The framework standardizes evaluation through uniform per-step and episode metrics, persists seeds, logbooks, and hall-of-fame archives, and provides plotting helpers for traces, Pareto fronts, and comparative outcomes, reducing glue code and improving comparability across studies. HEAS emphasizes separation of mechanism from orchestration, allowing exogenous drivers, endogenous agents, and aggregators to be composed and swapped without refactoring, while the same model can be used for forward simulation, optimization, or systematic comparison. We illustrate usage with two compact examples-an ecological system and an enterprise decision-making setting. HEAS offers a practical foundation for cross-disciplinary, multi-level inquiry, yielding reliable, reproducible results.
Authors: Alireza Fallah, Michael I. Jordan, Annie Ulichney
Abstract: We study fairness-accuracy tradeoffs when a single predictive model must serve multiple demographic groups. A useful tool for understanding this tradeoff is the fairness-accuracy (FA) Pareto frontier, which characterizes the set of models that cannot be improved in either fairness or accuracy without worsening the other. While characterizing the FA frontier requires full knowledge of the data distribution, we focus on the finite-sample regime, quantifying how well a designer can approximate any point on the frontier from limited data and bounding the worst-case gap. In particular, we derive worst-case-optimal estimators that depend on the designer's knowledge of the covariate distribution. For each estimator, we characterize how finite-sample effects asymmetrically impact each group's welfare and identify optimal sample allocation strategies. Finally, we provide uniform finite-sample bounds for the entire FA frontier, yielding confidence bands that quantify the reliability of welfare comparisons across alternative fairness-accuracy tradeoffs.
Authors: Konur Tholl, Mariam El Mezouar, Adrian Taylor, Ranwa Al Mallah
Abstract: Simulated environments have proven invaluable in Autonomous Cyber Operations (ACO) where Reinforcement Learning (RL) agents can be trained without the computational overhead of emulation. These environments must accurately represent cybersecurity scenarios while producing the necessary signals to support RL training. In this study, we present a framework where we first extend CybORG's Cage Challenge 2 environment by implementing three new actions: Patch, Isolate, and Unisolate, to better represent the capabilities available to human operators in real-world settings. We then propose a design for agent development where we modify the reward signals and the agent's feature space to enhance training performance. To validate these modifications, we train DQN and PPO agents in the updated environment. Our study demonstrates that CybORG can be extended with additional realistic functionality, while maintaining its ability to generate informative training signals for RL agents.
Authors: Jiaming Li, Longze Chen, Ze Gong, Yukun Chen, Lu Wang, Wanwei He, Run Luo, Min Yang
Abstract: Recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) have empowered large language models (LLMs) to tackle challenging reasoning tasks such as mathematics and programming. Despite its promise, the RLVR paradigm poses significant challenges, as existing methods often suffer from sparse reward signals and unstable policy gradient updates, inherent to RL-based approaches. To address the challenges, we propose $\textbf{PACS}$, a novel RLVR framework that achieves im$\textbf{P}$licit $\textbf{A}$ctor $\textbf{C}$ritic coupling via a $\textbf{S}$upervised learning framework. By treating the outcome reward as a predictable label, we reformulate the RLVR problem into a supervised learning task over a score function parameterized by the policy model and optimized using cross-entropy loss. A detailed gradient analysis shows that this supervised formulation inherently recovers the classical policy gradient update while providing more stable and efficient training. Extensive experiments demonstrate that PACS significantly outperforms strong open-source models and RLVR baselines, yielding substantial average gains of $\textbf{+8.26\%}$ (4B) and $\textbf{+9.57\%}$ (8B) over base models offering a promising avenue for LLMs post-training with verifiable rewards. Our code and data are available as open source at https://github.com/ritzz-ai/PACS.
Authors: Konur Tholl, Fran\c{c}ois Rivest, Mariam El Mezouar, Adrian Taylor, Ranwa Al Mallah
Abstract: Reinforcement Learning (RL) has shown great potential for autonomous decision-making in the cybersecurity domain, enabling agents to learn through direct environment interaction. However, RL agents in Autonomous Cyber Operations (ACO) typically learn from scratch, requiring them to execute undesirable actions to learn their consequences. In this study, we integrate external knowledge in the form of a Large Language Model (LLM) pretrained on cybersecurity data that our RL agent can directly leverage to make informed decisions. By guiding initial training with an LLM, we improve baseline performance and reduce the need for exploratory actions with obviously negative outcomes. We evaluate our LLM-integrated approach in a simulated cybersecurity environment, and demonstrate that our guided agent achieves over 2x higher rewards during early training and converges to a favorable policy approximately 4,500 episodes faster than the baseline.
Authors: Hugo Carlesso, Josiane Mothe, Radu Tudor Ionescu
Abstract: Hyperspectral imaging (HSI) captures detailed spectral signatures across hundreds of contiguous bands per pixel, being indispensable for remote sensing applications such as land-cover classification, change detection, and environmental monitoring. Due to the high dimensionality of HSI data and the slow rate of data transfer in satellite-based systems, compact and efficient models are required to support onboard processing and minimize the transmission of redundant or low-value data. To this end, we introduce a novel curriculum multi-task self-supervised learning (CMTSSL) framework designed for lightweight architectures for HSI analysis. CMTSSL integrates masked image modeling with decoupled spatial and spectral jigsaw puzzle solving, guided by a curriculum learning strategy that progressively increases data difficulty during self-supervision. This enables the encoder to jointly capture fine-grained spectral continuity, spatial structure, and global semantic features. Unlike prior dual-task SSL methods, CMTSSL simultaneously addresses spatial and spectral reasoning within a unified and computationally efficient design, being particularly suitable for training lightweight models for onboard satellite deployment. We validate our approach on four public benchmark datasets, demonstrating consistent gains in downstream segmentation tasks, using architectures that are over 16,000x lighter than some state-of-the-art models. These results highlight the potential of CMTSSL in generalizable representation learning with lightweight architectures for real-world HSI applications. Our code is publicly available at https://github.com/hugocarlesso/CMTSSL.
Authors: Yan Huang, Jinming Xu, Li Chai, Jiming Chen, Karl H. Johansson
Abstract: This paper addresses distributed stochastic optimization problems under non-i.i.d. data, focusing on the inherent trade-offs between communication and computational efficiency. To this end, we propose FlexGT, a flexible snapshot gradient tracking method that enables tunable numbers of local updates and neighbor communications per round, thereby adapting efficiently to diverse system resource conditions. Leveraging a unified convergence analysis framework, we derive tight communication and computational complexity for FlexGT with explicit dependence on objective properties and certain tunable parameters. Moreover, we introduce an accelerated variant, termed Acc-FlexGT, and prove that, with prior knowledge of the graph, it achieves Pareto-optimal trade-offs between communication and computation. Particularly, in the nonconvex case, Acc-FlexGT achieves the optimal iteration complexity of $\tilde{\mathcal{O}}\left( \left( L\sigma ^2 \right) /\left( n\epsilon ^2 \right) +L/\left( \epsilon \sqrt{1-\sqrt{\rho _W}} \right) \right) $ and optimal communication complexity of $\tilde{\mathcal{O}}\left( L/\left( \epsilon \sqrt{1-\sqrt{\rho _W}} \right) \right)$ for appropriately chosen numbers of local updates, matching existing lower bounds up to logarithmic factors. And, it improves the existing results for the strongly convex case by a factor of $\tilde{\mathcal{O}} \left( 1/\sqrt{\epsilon} \right)$, where $\epsilon$ is the targeted accuracy, $n$ the number of nodes, $L$ the Lipschitz constant, $\rho_W$ the connectivity of the graph, and $\sigma$ the stochastic gradient variance. Numerical experiments corroborate the theoretical results and demonstrate the effectiveness of the proposed methods.
Authors: Chen Liang, Zhaoqi Huang, Haofen Wang, Fu Chai, Chunying Yu, Huanhuan Wei, Zhengjie Liu, Yanpeng Li, Hongjun Wang, Ruifeng Luo, Xianzhong Zhao
Abstract: Large language models (LLMs), as a novel information technology, are seeing increasing adoption in the Architecture, Engineering, and Construction (AEC) field. They have shown their potential to streamline processes throughout the building lifecycle. However, the robustness and reliability of LLMs in such a specialized and safety-critical domain remain to be evaluated. To address this challenge, this paper establishes AECBench, a comprehensive benchmark designed to quantify the strengths and limitations of current LLMs in the AEC domain. The benchmark features a five-level, cognition-oriented evaluation framework (i.e., Knowledge Memorization, Understanding, Reasoning, Calculation, and Application). Based on the framework, 23 representative evaluation tasks were defined. These tasks were derived from authentic AEC practice, with scope ranging from codes retrieval to specialized documents generation. Subsequently, a 4,800-question dataset encompassing diverse formats, including open-ended questions, was crafted primarily by engineers and validated through a two-round expert review. Furthermore, an "LLM-as-a-Judge" approach was introduced to provide a scalable and consistent methodology for evaluating complex, long-form responses leveraging expert-derived rubrics. Through the evaluation of nine LLMs, a clear performance decline across five cognitive levels was revealed. Despite demonstrating proficiency in foundational tasks at the Knowledge Memorization and Understanding levels, the models showed significant performance deficits, particularly in interpreting knowledge from tables in building codes, executing complex reasoning and calculation, and generating domain-specific documents. Consequently, this study lays the groundwork for future research and development aimed at the robust and reliable integration of LLMs into safety-critical engineering practices.
Authors: Manuel Perez-Carrasco, Maya Nasr, Sebastien Roche, Chris Chan Miller, Zhan Zhang, Core Francisco Park, Eleanor Walker, Cecilia Garraffo, Douglas Finkbeiner, Sasha Ayvazov, Jonathan Franklin, Bingkun Luo, Xiong Liu, Ritesh Gautam, Steven Wofsy
Abstract: Effective cloud and cloud shadow detection is a critical prerequisite for accurate retrieval of concentrations of atmospheric methane (CH4) or other trace gases in hyperspectral remote sensing. This challenge is especially pertinent for MethaneSAT, a satellite mission launched in March 2024, to fill a significant data gap in terms of resolution, precision and swath between coarse-resolution global mappers and fine-scale point-source imagers of methane, and for its airborne companion mission, MethaneAIR. MethaneSAT delivers hyperspectral data at an intermediate spatial resolution (approx. 100 x 400, m), whereas MethaneAIR provides even finer resolution (approx. 25 m), enabling the development of highly detailed maps of concentrations that enable quantification of both the sources and rates of emissions. In this study, we use machine learning methods to address the cloud and cloud shadow detection problem for sensors with these high spatial resolutions. Cloud and cloud shadows in remote sensing data need to be effectively screened out as they bias methane retrievals in remote sensing imagery and impact the quantification of emissions. We deploy and evaluate conventional techniques-including Iterative Logistic Regression (ILR) and Multilayer Perceptron (MLP)-with advanced deep learning architectures, namely U-Net and a Spectral Channel Attention Network (SCAN) method. Our results show that conventional methods struggle with spatial coherence and boundary definition, affecting the detection of clouds and cloud shadows. Deep learning models substantially improve detection quality: U-Net performs best in preserving spatial structure, while SCAN excels at capturing fine boundary details... Our data and code is publicly available at: https://doi.org/10.7910/DVN/IKLZOJ
Authors: Yanfeng Yang, Siwei Chen, Pingping Hu, Zhaotong Shen, Yingjie Zhang, Zhuoran Sun, Shuai Li, Ziqi Chen, Kenji Fukumizu
Abstract: Probabilistic forecasting of multivariate time series is challenging due to non-stationarity, inter-variable dependencies, and distribution shifts. While recent diffusion and flow matching models have shown promise, they often ignore informative priors such as conditional means and covariances. In this work, we propose Conditionally Whitened Generative Models (CW-Gen), a framework that incorporates prior information through conditional whitening. Theoretically, we establish sufficient conditions under which replacing the traditional terminal distribution of diffusion models, namely the standard multivariate normal, with a multivariate normal distribution parameterized by estimators of the conditional mean and covariance improves sample quality. Guided by this analysis, we design a novel Joint Mean-Covariance Estimator (JMCE) that simultaneously learns the conditional mean and sliding-window covariance. Building on JMCE, we introduce Conditionally Whitened Diffusion Models (CW-Diff) and extend them to Conditionally Whitened Flow Matching (CW-Flow). Experiments on five real-world datasets with six state-of-the-art generative models demonstrate that CW-Gen consistently enhances predictive performance, capturing non-stationary dynamics and inter-variable correlations more effectively than prior-free approaches. Empirical results further demonstrate that CW-Gen can effectively mitigate the effects of distribution shift.
Authors: Haodong Liang, Yanhao Jin, Krishnakumar Balasubramanian, Lifeng Lai
Abstract: We study instrumental variable regression (IVaR) under differential privacy constraints. Classical IVaR methods (like two-stage least squares regression) rely on solving moment equations that directly use sensitive covariates and instruments, creating significant risks of privacy leakage and posing challenges in designing algorithms that are both statistically efficient and differentially private. We propose a noisy two-stage gradient descent algorithm that ensures $\rho$-zero-concentrated differential privacy by injecting carefully calibrated noise into the gradient updates. Our analysis establishes finite-sample convergence rates for the proposed method, showing that the algorithm achieves consistency while preserving privacy. In particular, we derive precise bounds quantifying the trade-off among optimization, privacy, and sampling error. To the best of our knowledge, this is the first work to provide both privacy guarantees and provable convergence rates for instrumental variable regression in linear models. We further validate our theoretical findings with experiments on both synthetic and real datasets, demonstrating that our method offers practical accuracy-privacy trade-offs.
Authors: Muhammed Ustaomeroglu, Baris Askin, Gauri Joshi, Carlee Joe-Wong, Guannan Qu
Abstract: The extent to which decoder-only language models (LMs) engage in planning, that is, organizing intermediate computations to support coherent long-range generation, remains an important question, with implications for interpretability, reliability, and principled model design. Planning involves structuring computations over long horizons, and considering multiple possible continuations, but how far transformer-based LMs exhibit them without external scaffolds, e.g., chain-of-thought prompting, is unclear. We address these questions by analyzing the hidden states at the core of transformer computations, which capture intermediate results and act as carriers of information. Since these hidden representations are redundant and encumbered with fine-grained details, we develop a pipeline based on vector-quantized variational autoencoders that compresses them into compact summary codes. These codes enable measuring mutual information and analyzing the computational structure of the underlying model behavior. Using this framework, we study planning in LMs across synthetic grammar, path-finding tasks, and natural language datasets, focusing on two planning properties: (i) the planning horizon of pre-output computations, and (ii) the extent to which the model considers alternative valid continuations. As a separate downstream use of the same pipeline, we also analyze how decision-relevant information is distributed across layers and earlier prefix blocks when producing next-token predictions. Together, these analyses advance our understanding of planning in LMs and provide a general-purpose pipeline for inspecting internal model dynamics. Our results reveal that the effective planning horizon is task-dependent, that models implicitly preserve information about unused correct continuations, and that predictions draw most on recent computations, though earlier blocks remain informative.
Authors: Arjan Blankestijn, Uraz Odyurt, Amirreza Yousefzadeh
Abstract: The Transformer Machine Learning (ML) architecture has been gaining considerable momentum in recent years. In particular, computational High-Energy Physics tasks such as jet tagging and particle track reconstruction (tracking), have either achieved proper solutions, or reached considerable milestones using Transformers. On the other hand, the use of specialised hardware accelerators, especially FPGAs, is an effective method to achieve online, or pseudo-online latencies. The development and integration of Transformer-based ML to FPGAs is still ongoing and the support from current tools is very limited or non-existent. Additionally, FPGA resources present a significant constraint. Considering the model size alone, while smaller models can be deployed directly, larger models are to be partitioned in a meaningful and ideally, automated way. We aim to develop methodologies and tools for monolithic, or partitioned Transformer synthesis, specifically targeting inference. Our primary use-case involves two machine learning model designs for tracking, derived from the TrackFormers project. We elaborate our development approach, present preliminary results, and provide comparisons.
Authors: Xin Xu, Xunzhi He, Churan Zhi, Ruizhe Chen, Julian McAuley, Zexue He
Abstract: Existing studies on bias mitigation methods for large language models (LLMs) use diverse baselines and metrics to evaluate debiasing performance, leading to inconsistent comparisons among them. Moreover, their evaluations are mostly based on the comparison between LLMs' probabilities of biased and unbiased contexts, which ignores the gap between such evaluations and real-world use cases where users interact with LLMs by reading model responses and expect fair and safe outputs rather than LLMs' probabilities. To enable consistent evaluation across debiasing methods and bridge this gap, we introduce BiasFreeBench, an empirical benchmark that comprehensively compares eight mainstream bias mitigation techniques (covering four prompting-based and four training-based methods) on two test scenarios (multi-choice QA and open-ended multi-turn QA) by reorganizing existing datasets into a unified query-response setting. We further introduce a response-level metric, Bias-Free Score, to measure the extent to which LLM responses are fair, safe, and anti-stereotypical. Debiasing performances are systematically compared and analyzed across key dimensions: the prompting vs. training paradigm, model size, and generalization of different training strategies to unseen bias types. We release our benchmark, aiming to establish a unified testbed for bias mitigation research.
Authors: Thanh Dang, Jiaming Liang
Abstract: We propose new Markov chain Monte Carlo algorithms to sample a uniform distribution on a convex body $K$. Our algorithms are based on the proximal sampler, which uses Gibbs sampling on an augmented distribution and assumes access to the so-called restricted Gaussian oracle (RGO). The key contribution of this work is an efficient implementation of the RGO for uniform sampling on convex $K$ that goes beyond the membership-oracle model used in many classical and modern uniform samplers, and instead leverages richer oracle access commonly assumed in convex optimization. We implement the RGO via rejection sampling and access to either a projection oracle or a separation oracle on $K$. In both oracle models, we provide non-asymptotic complexity guarantees for obtaining unbiased samples, with accuracy quantified in R\'enyi divergence and $\chi^2$-divergence, and we support these theoretical guarantees with numerical experiments.
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: 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 exhibit hallucinations, raising serious concerns about their reliability. Prior work has explored adversarial attacks to elicit hallucinations in LLMs, but these methods often rely on unrealistic prompts, either by inserting nonsensical tokens or by altering the original semantic intent. Consequently, such approaches provide limited insight into how hallucinations arise in real-world settings. In contrast, adversarial attacks in computer vision typically involve realistic modifications to input images. However, the problem of identifying realistic adversarial prompts for eliciting LLM hallucinations remains largely underexplored. To address this gap, we propose Semantically Equivalent and Coherent Attacks (SECA), which 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 semantic equivalence or semantic coherence errors 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: Akira Kitaoka
Abstract: Data-driven inverse optimization for mixed-integer linear programs (MILPs), which seeks to learn an objective function and constraints consistent with observed decisions, is important for building accurate mathematical models in a variety of domains, including power systems and scheduling. However, to the best of our knowledge, existing data-driven inverse optimization methods primarily focus on learning objective functions under known constraints, and learning both objective functions and constraints from data remains largely unexplored. In this paper, we propose a two-stage approach for a class of inverse optimization problems in which the objective is a linear combination of given feature functions and the constraints are parameterized by unknown functions and thresholds. Our method first learns the constraints and then, conditioned on the learned constraints, estimates the objective-function weights. On the theoretical side, we provide finite-sample guarantees for solving the proposed inverse optimization problem. To this end, we develop statistical learning tools for pseudo-metric spaces under sub-Gaussian assumptions and use them to derive a learning-theoretic framework for inverse optimization with both unknown objectives and constraints. On the experimental side, we demonstrate that our method successfully solves inverse optimization problems on scheduling instances formulated as ILPs with up to 100 decision variables.
Authors: Kaiwen Zheng, Yuji Wang, Qianli Ma, Huayu Chen, Jintao Zhang, Yogesh Balaji, Jianfei Chen, Ming-Yu Liu, Jun Zhu, Qinsheng Zhang
Abstract: Although continuous-time consistency models (e.g., sCM, MeanFlow) are theoretically principled and empirically powerful for fast academic-scale diffusion, its applicability to large-scale text-to-image and video tasks remains unclear due to infrastructure challenges in Jacobian-vector product (JVP) computation and the limitations of evaluation benchmarks like FID. This work represents the first effort to scale up continuous-time consistency to general application-level image and video diffusion models, and to make JVP-based distillation effective at large scale. We first develop a parallelism-compatible FlashAttention-2 JVP kernel, enabling sCM training on models with over 10 billion parameters and high-dimensional video tasks. Our investigation reveals fundamental quality limitations of sCM in fine-detail generation, which we attribute to error accumulation and the "mode-covering" nature of its forward-divergence objective. To remedy this, we propose the score-regularized continuous-time consistency model (rCM), which incorporates score distillation as a long-skip regularizer. This integration complements sCM with the "mode-seeking" reverse divergence, effectively improving visual quality while maintaining high generation diversity. Validated on large-scale models (Cosmos-Predict2, Wan2.1) up to 14B parameters and 5-second videos, rCM generally matches the state-of-the-art distillation method DMD2 on quality metrics while mitigating mode collapse and offering notable advantages in diversity, all without GAN tuning or extensive hyperparameter searches. The distilled models generate high-fidelity samples in only $1\sim4$ steps, accelerating diffusion sampling by $15\times\sim50\times$. These results position rCM as a practical and theoretically grounded framework for advancing large-scale diffusion distillation. Code is available at https://github.com/NVlabs/rcm.
Authors: Nizar El Ghazal, Antoine Caubri\`ere, Valentin Vielzeuf
Abstract: This paper presents a comparative study of context management strategies for end-to-end Spoken Dialog State Tracking using Speech-LLMs. We systematically evaluate traditional multimodal context (combining text history and spoken current turn), full spoken history, and compressed spoken history approaches. Our experiments on the SpokenWOZ corpus demonstrate that providing the full spoken conversation as input yields the highest performance among models of similar size, significantly surpassing prior methods. Furthermore, we show that attention-pooling-based compression of the spoken history offers a strong trade-off, maintaining competitive accuracy with reduced context size. Detailed analysis confirms that improvements stem from more effective context utilization.
Authors: Daniel Seifert, Onur G\"unl\"u, Rafael F. Schaefer
Abstract: The application of deep learning to the area of communications systems has been a growing field of interest in recent years. Forward-forward (FF) learning is an efficient alternative to the backpropagation (BP) algorithm, which is the typically used training procedure for neural networks. Among its several advantages, FF learning does not require the communication channel to be differentiable and does not rely on the global availability of partial derivatives, allowing for an energy-efficient implementation. In this work, we design end-to-end learned autoencoders using the FF algorithm and numerically evaluate their performance for the additive white Gaussian noise and Rayleigh block fading channels. We demonstrate their competitiveness with BP-trained systems in the case of joint coding and modulation, and in a scenario where a fixed, non-differentiable modulation stage is applied. Moreover, we provide further insights into the design principles of the FF network, its training convergence behavior, and significant memory and processing time savings compared to BP-based approaches.
Authors: Federico Gabriele, Aldo Glielmo, Marco Taboga
Abstract: Current macroeconomic models with agent heterogeneity can be broadly divided into two main groups. Heterogeneous-agent general equilibrium (GE) models, such as those based on Heterogeneous Agent New Keynesian (HANK) or Krusell-Smith (KS) approaches, rely on GE and 'rational expectations', somewhat unrealistic assumptions that make the models very computationally cumbersome, which in turn limits the amount of heterogeneity that can be modelled. In contrast, agent-based models (ABMs) can flexibly encompass a large number of arbitrarily heterogeneous agents, but typically require the specification of explicit behavioural rules, which can lead to a lengthy trial-and-error model-development process. To address these limitations, we introduce MARL-BC, a framework that integrates deep multi-agent reinforcement learning (MARL) with real business cycle (RBC) models. We demonstrate that MARL-BC can: (1) recover textbook RBC results when using a single agent; (2) recover the results of the mean-field KS model using a large number of identical agents; and (3) effectively simulate rich heterogeneity among agents, a hard task for traditional GE approaches. Our framework can be thought of as an ABM if used with a variety of heterogeneous interacting agents, and can reproduce GE results in limit cases. As such, it is a step towards a synthesis of these often opposed modelling paradigms.
Authors: Thomas Wimmer, Prune Truong, Marie-Julie Rakotosaona, Michael Oechsle, Federico Tombari, Bernt Schiele, Jan Eric Lenssen
Abstract: We introduce AnyUp, a method for feature upsampling that can be applied to any vision feature at any resolution, without encoder-specific training. Existing learning-based upsamplers for features like DINO or CLIP need to be re-trained for every feature extractor and thus do not generalize to different feature types at inference time. In this work, we propose an inference-time feature-agnostic upsampling architecture to alleviate this limitation and improve upsampling quality. In our experiments, AnyUp sets a new state of the art for upsampled features, generalizes to different feature types, and preserves feature semantics while being efficient and easy to apply to a wide range of downstream tasks.
Authors: Navjot Singh, Edgar Solomonik, Xiaoye Sherry Li, Yang Liu
Abstract: This paper presents low-complexity tensor completion algorithms and their efficient implementation to reconstruct highly oscillatory operators discretized as $n\times n$ matrices. The underlying tensor decomposition is based on the reshaping of the input matrix and its butterfly decomposition into an order $O (\log n)$ tensor. The reshaping of the input matrix into a tensor allows for representation of the butterfly decomposition as a tensor decomposition with dense tensors. This leads to efficient utilization of the existing software infrastructure for dense and sparse tensor computations. We propose two tensor completion algorithms in the butterfly format, using alternating least squares and gradient-based optimization, as well as a novel strategy that uses low-rank matrix completion to efficiently generate an initial guess for the proposed algorithms. To demonstrate the efficiency and applicability of our proposed algorithms, we perform three numerical experiments using simulated oscillatory operators in seismic applications. In these experiments, we use $O (n \log n)$ observed entries in the input matrix and demonstrate an $O(n\log^3 n)$ computational cost of the proposed algorithms, leading to a speedup of orders of magnitudes per iteration for large matrices compared to the low-rank matrix and quantized tensor-train completion. Moreover, the proposed butterfly completion algorithms, equipped with the novel initial guess generation strategy, achieve reconstruction errors that are smaller by an order of magnitude, enabling accurate recovery of the underlying structure compared to the state-of-the-art completion algorithms.
Authors: Dechen Zhang, Junwei Su, Difan Zou
Abstract: The use of low-bit quantization has emerged as an indispensable technique for enabling the efficient training of large-scale models. Despite its widespread empirical success, a rigorous theoretical understanding of its impact on learning performance remains notably absent, even in the simplest linear regression setting. We present the first systematic theoretical study of this fundamental question, analyzing finite-step stochastic gradient descent (SGD) for high-dimensional linear regression under a comprehensive range of quantization targets: data, label, parameter, activation, and gradient. Our novel analytical framework establishes precise algorithm-dependent and data-dependent excess risk bounds that characterize how different quantization affects learning: parameter, activation, and gradient quantization amplify noise during training; data quantization distorts the data spectrum and introduces additional approximation error. Crucially, we distinguish the effects of two quantization schemes: we prove that for additive quantization (with constant quantization steps), the noise amplification benefits from a suppression effect scaled by the batch size, while multiplicative quantization (with input-dependent quantization steps) largely preserves the spectral structure, thereby reducing the spectral distortion. Furthermore, under common polynomial-decay data spectra, we quantitatively compare the risks of multiplicative and additive quantization, drawing a parallel to the comparison between FP and integer quantization methods. Our theory provides a powerful lens to characterize how quantization shapes the learning dynamics of optimization algorithms, paving the way to further explore learning theory under practical hardware constraints.
Authors: Vishal Halder, Alexandre Reiffers-Masson, Abdeldjalil A\"issa-El-Bey, Gugan Thoppe
Abstract: Let $A \in \mathbb{R}^{m \times n}$ be an arbitrary, known matrix and $e$ a $q$-sparse adversarial vector. Given $y = A x^\star + e$ and $q$, we seek the smallest robust solution set containing $x^\star$ that is uniformly recoverable from $y$ without knowing $e$. While exact recovery of $x^\star$ via strong (and often impractical) structural assumptions on $A$ or $x^\star$ (e.g., restricted isometry, sparsity) is well studied, recoverability for arbitrary $A$ and $x^\star$ remains open. Our main result shows that the smallest robust solution set is $x^\star + \ker(U)$, where $U$ is the unique projection matrix onto the intersection of rowspaces of all possible submatrices of $A$ obtained by deleting $2q$ rows. Moreover, we prove that every $x$ that minimizes the $\ell_0$-norm of $y - A x$ lies in $x^\star + \ker(U)$, which then gives a constructive approach to recover this set.
Authors: Pengfei Lyu, Zhengchi Ma, Linjun Zhang, Anru R. Zhang
Abstract: Imbalanced data, where the positive samples represent only a small proportion compared to the negative samples, makes it challenging for classification problems to balance the false positive and false negative rates. A common approach to addressing the challenge involves generating synthetic data for the minority group and then training classification models with both observed and synthetic data. However, since the synthetic data depends on the observed data and fails to replicate the original data distribution accurately, prediction accuracy is reduced when the synthetic data is na\"{i}vely treated as the true data. In this paper, we address the bias introduced by synthetic data and provide consistent estimators for this bias by borrowing information from the majority group. We propose a bias correction procedure to mitigate the adverse effects of synthetic data, enhancing prediction accuracy while avoiding overfitting. This procedure is extended to broader scenarios with imbalanced data, such as imbalanced multi-task learning and causal inference. Theoretical properties, including bounds on bias estimation errors and improvements in prediction accuracy, are provided. Simulation results and data analysis on handwritten digit datasets demonstrate the effectiveness of our method.
Authors: Qile Jiang, George Karniadakis
Abstract: Scientific Machine Learning (SciML) integrates data-driven inference with physical modeling to solve complex problems in science and engineering. However, the design of SciML architectures, loss formulations, and training strategies remains an expert-driven research process, requiring extensive experimentation and problem-specific insights. Here we introduce AgenticSciML, a collaborative multi-agent system in which over 10 specialized AI agents collaborate to propose, critique, and refine SciML solutions through structured reasoning and iterative evolution. The framework integrates structured debate, retrieval-augmented method memory, and ensemble-guided evolutionary search, enabling the agents to generate and assess new hypotheses about architectures and optimization procedures. Across physics-informed learning and operator learning tasks, the framework discovers solution methods that outperform single-agent and human-designed baselines by up to four orders of magnitude in error reduction. The agents produce novel strategies -- including adaptive mixture-of-expert architectures, decomposition-based PINNs, and physics-informed operator learning models -- that do not appear explicitly in the curated knowledge base. These results show that collaborative reasoning among AI agents can yield emergent methodological innovation, suggesting a path toward scalable, transparent, and autonomous discovery in scientific computing.
Authors: Youngjoo Yun, Rishabh Dudeja
Abstract: In differential privacy, statistics of a sensitive dataset are privatized by introducing random noise. Most privacy analyses provide privacy bounds specifying a noise level sufficient to achieve a target privacy guarantee. Sometimes, these bounds are pessimistic and suggest adding excessive noise, which overwhelms the meaningful signal. It remains unclear if such high noise levels are truly necessary or a limitation of the proof techniques. This paper explores whether we can obtain sharp privacy characterizations that identify the smallest noise level required to achieve a target privacy level for a given mechanism. We study this problem in the context of differentially private principal component analysis, where the goal is to privatize the leading principal components (PCs) of a dataset with n samples and p features. We analyze the exponential mechanism for this problem in a model-free setting and provide sharp utility and privacy characterizations in the high-dimensional limit ($p\rightarrow\infty$). Our privacy result shows that, in high dimensions, detecting the presence of a target individual in the dataset using the privatized PCs is exactly as hard as distinguishing two Gaussians with slightly different means, where the mean difference depends on certain spectral properties of the dataset. Our privacy analysis combines the hypothesis-testing formulation of privacy guarantees proposed by Dong, Roth, and Su (2022) with classical contiguity arguments due to Le Cam to obtain sharp high-dimensional privacy characterizations.
Authors: Abdelhamid Ezzerg, Ilija Bogunovic, Jeremias Knoblauch
Abstract: Bayesian Optimization is critically vulnerable to extreme outliers. Existing provably robust methods typically assume a bounded cumulative corruption budget, which makes them defenseless against even a single corruption of sufficient magnitude. To address this, we introduce a new adversary whose budget is only bounded in the frequency of corruptions, not in their magnitude. We then derive RCGP-UCB, an algorithm coupling the famous upper confidence bound (UCB) approach with a Robust Conjugate Gaussian Process (RCGP). We present stable and adaptive versions of RCGP-UCB, and prove that they achieve sublinear regret in the presence of up to $O(T^{1/4})$ and $O(T^{1/7})$ corruptions with possibly infinite magnitude. This robustness comes at near zero cost: without outliers, RCGP-UCB's regret bounds match those of the standard GP-UCB algorithm.
Authors: Jared N. Lakhani, Etienne Pienaar
Abstract: This study investigates the limitations of applying Markov Chain Monte Carlo (MCMC) methods to arbitrary objective functions, focusing on a two-block MCMC framework which alternates between Metropolis-Hastings and Gibbs sampling. While such approaches are often considered advantageous for enabling data-driven regularization, we show that their performance critically depends on the sharpness of the employed likelihood form. By introducing a sharpness parameter and exploring alternative likelihood formulations proportional to the target objective function, we demonstrate how likelihood curvature governs both in-sample performance and the degree of regularization inferred by the training data. Empirical applications are conducted on reinforcement learning tasks: including a navigation problem and the game of tic-tac-toe. The study concludes with a separate analysis examining the implications of extreme likelihood sharpness on arbitrary objective functions stemming from the classic game of blackjack, where the first block of the two-block MCMC framework is replaced with an iterative optimization step. The resulting hybrid approach achieves performance nearly identical to the original MCMC framework, indicating that excessive likelihood sharpness effectively collapses posterior mass onto a single dominant mode.
Authors: Zhisheng Zheng, Xiaohang Sun, Tuan Dinh, Abhishek Yanamandra, Abhinav Jain, Zhu Liu, Sunil Hadap, Vimal Bhat, Manoj Aggarwal, Gerard Medioni, David Harwath
Abstract: End-to-end speech-to-speech translation (S2ST) systems typically struggle with a critical data bottleneck: the scarcity of parallel speech-to-speech corpora. To overcome this, we introduce RosettaSpeech, a novel zero-shot framework trained exclusively on monolingual speech-text data augmented by machine translation supervision. Unlike prior works that rely on complex cascaded pseudo-labeling, our approach strategically utilizes text as a semantic bridge during training to synthesize translation targets, thereby eliminating the need for parallel speech pairs while maintaining a direct, end-to-end inference pipeline. Empirical evaluations on the CVSS-C benchmark demonstrate that RosettaSpeech achieves state-of-the-art zero-shot performance, surpassing leading baselines by significant margins - achieving ASR-BLEU scores of 25.17 for German-to-English (+27% relative gain) and 29.86 for Spanish-to-English (+14%). Crucially, our model effectively preserves the source speaker's voice without ever seeing paired speech data. We further analyze the impact of data scaling and demonstrate the model's capability in many-to-one translation, offering a scalable solution for extending high-quality S2ST to "text-rich, speech-poor" languages.
Authors: Daeyong Kwon, SeungHeon Doh, Juhan Nam
Abstract: Recent advances in large language models (LLMs) have transformed open-domain question answering, yet their effectiveness in music-related reasoning remains limited due to sparse music knowledge in pretraining data. While music information retrieval and computational musicology have explored structured and multimodal understanding, few resources support factual and contextual music question answering (MQA) grounded in artist metadata or historical context. We introduce MusWikiDB, a vector database of 3.2M passages from 144K music-related Wikipedia pages, and ArtistMus, a benchmark of 1,000 questions on 500 diverse artists with metadata such as genre, debut year, and topic. These resources enable systematic evaluation of retrieval-augmented generation (RAG) for MQA. Experiments show that RAG markedly improves factual accuracy; open-source models gain up to +56.8 percentage points (for example, Qwen3 8B improves from 35.0 to 91.8), approaching proprietary model performance. RAG-style fine-tuning further boosts both factual recall and contextual reasoning, improving results on both in-domain and out-of-domain benchmarks. MusWikiDB also yields approximately 6 percentage points higher accuracy and 40% faster retrieval than a general-purpose Wikipedia corpus. We release MusWikiDB and ArtistMus to advance research in music information retrieval and domain-specific question answering, establishing a foundation for retrieval-augmented reasoning in culturally rich domains such as music.
Authors: Ayrat Abdullin, Umair bin Waheed, Leo Eisner, Abdullatif Al-Shuhail
Abstract: Real-time monitoring of induced seismicity is critical to mitigate operational risks, relying on the rapid and accurate classification of triggered data from continuous data streams. Deep learning models are effective for this purpose but require substantial computational resources, making real-time processing difficult. To address this limitation, a lightweight model based on the Fourier Neural Operator (FNO) is proposed for the classification of microseismic events, leveraging its inherent resolution-invariance and computational efficiency for waveform processing. In the STanford EArthquake Dataset (STEAD), a global and large-scale database of seismic waveforms, the FNO-based model demonstrates high effectiveness for trigger classification, with an F1 score of 95% even in the scenario of data sparsity in training. The new FNO model greatly decreases the computer power needed relative to current deep learning models without sacrificing the classification success rate measured by the F1 score. A test on a real microseismic dataset shows a classification success rate with an F1 score of 98%, outperforming many traditional deep-learning techniques. The reduced computational cost makes the proposed FNO model well suited for deployment in resource-constrained, near-real-time seismic monitoring workflows, including traffic-light implementations. The source code for the proposed FNO classifier will be available at: https://github.com/ayratabd/FNOclass.
Authors: Jeongjun Park, Sunwook Hwang, Hyeonho Noh, Jin Mo Yang, Hyun Jong Yang, Saewoong Bahk
Abstract: Distracted driving contributes to fatal crashes worldwide. To address this, researchers are using driver activity recognition (DAR) with impulse radio ultra-wideband (IR-UWB) radar, which offers advantages such as interference resistance, low power consumption, and privacy preservation. However, two challenges limit its adoption: the lack of large-scale real-world UWB datasets covering diverse distracted driving behaviors, and the difficulty of adapting fixed-input Vision Transformers (ViTs) to UWB radar data with non-standard dimensions. This work addresses both challenges. We present the ALERT dataset, which contains 10,220 radar samples of seven distracted driving activities collected in real driving conditions. We also propose the input-size-agnostic Vision Transformer (ISA-ViT), a framework designed for radar-based DAR. The proposed method resizes UWB data to meet ViT input requirements while preserving radar-specific information such as Doppler shifts and phase characteristics. By adjusting patch configurations and leveraging pre-trained positional embedding vectors (PEVs), ISA-ViT overcomes the limitations of naive resizing approaches. In addition, a domain fusion strategy combines range- and frequency-domain features to further improve classification performance. Comprehensive experiments demonstrate that ISA-ViT achieves a 22.68% accuracy improvement over an existing ViT-based approach for UWB-based DAR. By publicly releasing the ALERT dataset and detailing our input-size-agnostic strategy, this work facilitates the development of more robust and scalable distracted driving detection systems for real-world deployment.
Authors: Yan Yang, George Bebis, Mircea Nicolescu
Abstract: Data scarcity and distribution shift pose major challenges for masked face detection and recognition. We propose a two-step generative data augmentation framework that combines rule-based mask warping with unpaired image-to-image translation using GANs, enabling the generation of realistic masked-face samples beyond purely synthetic transformations. Compared to rule-based warping alone, the proposed approach yields consistent qualitative improvements and complements existing GAN-based masked face generation methods such as IAMGAN. We introduce a non-mask preservation loss and stochastic noise injection to stabilize training and enhance sample diversity. Experimental observations highlight the effectiveness of the proposed components and suggest directions for future improvements in data-centric augmentation for face recognition tasks.
Authors: Sidharth Kannan, Pooyan Goodarzi, Evangelos E. Papalexakis, Jonathan W. Richardson
Abstract: In recent years, graph neural networks (GNNs) have shown tremendous promise in solving problems in high energy physics, materials science, and fluid dynamics. In this work, we introduce a new application for GNNs in the physical sciences: instrumentation design. As a case study, we apply GNNs to simulate models of the Laser Interferometer Gravitational-Wave Observatory (LIGO) and show that they are capable of accurately capturing the complex optical physics at play, while achieving runtimes 815 times faster than state of the art simulation packages. We discuss the unique challenges this problem provides for machine learning models. In addition, we provide a dataset of high-fidelity optical physics simulations for three interferometer topologies, which can be used as a benchmarking suite for future work in this direction.
Authors: Yizhi Wang, Linan Yue, Min-Ling Zhang
Abstract: Large Reasoning Models (LRMs) have demonstrated remarkable performance on complex reasoning tasks through long Chain-of-Thought (CoT) reasoning. Extending these successes to multimodal reasoning remains challenging due to the increased complexity of integrating diverse input modalities and the scarcity of high-quality long CoT training data. Existing multimodal datasets and CoT synthesis methods still suffer from limited reasoning depth, modality conversion errors, and rigid generation pipelines, hindering model performance and stability. To this end, in this paper, we propose SynSelect, a novel three-stage Synthesis-Selection framework for generating high-quality long CoT data tailored to multimodal reasoning tasks. Specifically, SynSelect first leverages multiple heterogeneous multimodal LRMs to produce diverse candidate CoTs, and then applies both instance and batch level selection to filter high-quality CoTs that can effectively enhance the model's reasoning capabilities. Extensive experiments on multiple multimodal benchmarks demonstrate that models supervised fine-tuned on SynSelect-generated data significantly outperform baselines and achieve further improvements after reinforcement learning post-training. Our results validate SynSelect as an effective approach for advancing multimodal LRMs reasoning capabilities.
Authors: Ariana Azarbal, Victor Gillioz, Vladimir Ivanov, Bryce Woodworth, Jacob Drori, Nevan Wichers, Aram Ebtekar, Alex Cloud, Alexander Matt Turner
Abstract: Developers often struggle to specify correct training labels and rewards. Perhaps they don't need to. We propose recontextualization, which reduces how often language models "game" training signals, performing misbehaviors those signals mistakenly reinforce. We show recontextualization prevents models from learning to 1) prioritize evaluation metrics over chat response quality; 2) special-case code to pass incorrect tests; 3) overwrite evaluation functions rather than write correct code; and 4) become sycophantic. Our method works by generating completions from prompts discouraging misbehavior and then recontextualizing them as though they were in response to prompts permitting misbehavior. Recontextualization trains language models to resist misbehavior even when instructions permit it. This mitigates the reinforcement of misbehavior from misspecified training signals, reducing specification gaming without improving the supervision signal.
Authors: Murtaza Rangwala, Richard O. Sinnott, Rajkumar Buyya
Abstract: Decentralized federated learning (DFL) enables collaborative model training across edge devices without centralized coordination, offering resilience against single points of failure. However, statistical heterogeneity arising from non-identically distributed local data creates a fundamental challenge: nodes must learn personalized models adapted to their local distributions while selectively collaborating with compatible peers. Existing approaches either enforce a single global model that fits no one well, or rely on heuristic peer selection mechanisms that cannot distinguish between peers with genuinely incompatible data distributions and those with valuable complementary knowledge. We present Murmura, a framework that leverages evidential deep learning to enable trust-aware model personalization in DFL. Our key insight is that epistemic uncertainty from Dirichlet-based evidential models directly indicates peer compatibility: high epistemic uncertainty when a peer's model evaluates local data reveals distributional mismatch, enabling nodes to exclude incompatible influence while maintaining personalized models through selective collaboration. Murmura introduces a trust-aware aggregation mechanism that computes peer compatibility scores through cross-evaluation on local validation samples and personalizes model aggregation based on evidential trust with adaptive thresholds. Evaluation on three wearable IoT datasets (UCI HAR, PAMAP2, PPG-DaLiA) demonstrates that Murmura reduces performance degradation from IID to non-IID conditions compared to baseline (0.9% vs. 19.3%), achieves 7.4$\times$ faster convergence, and maintains stable accuracy across hyperparameter choices. These results establish evidential uncertainty as a principled foundation for compatibility-aware personalization in decentralized heterogeneous environments.
Authors: Almaz Ermilov
Abstract: This paper presents FormationEval, an open multiple-choice question benchmark for evaluating language models on petroleum geoscience and subsurface disciplines. The dataset contains 505 questions across seven domains including petrophysics, petroleum geology and reservoir engineering, derived from three authoritative sources using a reasoning model with detailed instructions and a concept-based approach that avoids verbatim copying of copyrighted text. Each question includes source metadata to support traceability and audit. The evaluation covers 72 models from major providers including OpenAI, Anthropic, Google, Meta and open-weight alternatives. The top performers achieve over 97% accuracy, with Gemini 3 Pro Preview reaching 99.8%, while tier and domain gaps persist. Among open-weight models, GLM-4.7 leads at 98.6%, with several DeepSeek, Llama, Qwen and Mistral models also exceeding 93%. The performance gap between open-weight and closed models is narrower than expected, with several lower-cost open-weight models exceeding 90% accuracy. Petrophysics emerges as the most challenging domain across all models, while smaller models show wider performance variance. Residual length bias in the dataset (correct answers tend to be longer) is documented along with bias mitigation strategies applied during construction. The benchmark, evaluation code and results are publicly available.
Authors: Svenja Jedhoff, Elizaveta Semenova, Aura Raulo, Anne Meyer, Paul-Christian B\"urkner
Abstract: Graphs arise across diverse domains, from biology and chemistry to social and information networks, as well as in transportation and logistics. Inference on graph-structured data requires methods that are permutation-invariant, scalable across varying sizes and sparsities, and capable of capturing complex long-range dependencies, making posterior estimation on graph parameters particularly challenging. Amortized Bayesian Inference (ABI) is a simulation-based framework that employs generative neural networks to enable fast, likelihood-free posterior inference. We adapt ABI to graph data to address these challenges to perform inference on node-, edge-, and graph-level parameters. Our approach couples permutation-invariant graph encoders with flexible neural posterior estimators in a two-module pipeline: a summary network maps attributed graphs to fixed-length representations, and an inference network approximates the posterior over parameters. In this setting, several neural architectures can serve as the summary network. In this work we evaluate multiple architectures and assess their performance on controlled synthetic settings and two real-world domains - biology and logistics - in terms of recovery and calibration.
Authors: Ricardo Vinuesa, Steven L. Brunton, Gianmarco Mengaldo
Abstract: Artificial intelligence now outperforms humans in several scientific and engineering tasks, yet its internal representations often remain opaque. In this Perspective, we argue that explainable artificial intelligence (XAI), combined with causal reasoning, enables {\it learning from the learners}. Focusing on discovery, optimization and certification, we show how the combination of foundation models and explainability methods allows the extraction of causal mechanisms, guides robust design and control, and supports trust and accountability in high-stakes applications. We discuss challenges in faithfulness, generalization and usability of explanations, and propose XAI as a unifying framework for human-AI collaboration in science and engineering.
Authors: Zhongping Ji
Abstract: Modern computer vision architectures, from CNNs to Transformers, predominantly rely on the stacking of heuristic modules: spatial mixers (Attention/Conv) followed by channel mixers (FFNs). In this work, we challenge this paradigm by returning to mathematical first principles. We propose the Clifford Algebra Network (CAN), also referred to as CliffordNet, a vision backbone grounded purely in Geometric Algebra. Instead of engineering separate modules for mixing and memory, we derive a unified interaction mechanism based on the Clifford Geometric Product ($uv = u \cdot v + u \wedge v$). This operation ensures algebraic completeness regarding the Geometric Product by simultaneously capturing feature coherence (via the generalized inner product) and structural variation (via the exterior wedge product). Implemented via an efficient sparse rolling mechanism with strict linear complexity $O(N)$, our model reveals a surprising emergent property: the geometric interaction is so representationally dense that standard Feed-Forward Networks (FFNs) become redundant. Empirically, CliffordNet establishes a new Pareto frontier: our Nano variant achieves 77.82\% accuracy on CIFAR-100 with only 1.4M parameters, effectively matching the heavy-weight ResNet-18 (11.2M) with $8\times$ fewer parameters, while our Lite variant (2.6M) sets a new SOTA for tiny models at 79.05\%. Our results suggest that global understanding can emerge solely from rigorous, algebraically complete local interactions, potentially signaling a shift where geometry is all you need. Code is available at https://github.com/ParaMind2025/CAN.
Authors: Asif Mohammed Samir, Mohammad Masudur Rahman
Abstract: Software bugs cost technology providers (e.g., AT&T) billions annually and cause developers to spend roughly 50% of their time on bug resolution. Traditional methods for bug localization often analyze the suspiciousness of code components (e.g., methods, documents) in isolation, overlooking their connections with other components in the codebase. Recent advances in Large Language Models (LLMs) and agentic AI techniques have shown strong potential for code understanding, but still lack causal reasoning during code exploration and struggle to manage growing context effectively, limiting their capability. In this paper, we present a novel agentic technique for bug localization -- CogniGent -- that overcomes the limitations above by leveraging multiple AI agents capable of causal reasoning, call-graph-based root cause analysis and context engineering. It emulates developers-inspired debugging practices (a.k.a., dynamic cognitive debugging) and conducts hypothesis testing to support bug localization. We evaluate CogniGent on a curated dataset of 591 bug reports using three widely adopted performance metrics and compare it against six established baselines from the literature. Experimental results show that our technique consistently outperformed existing traditional and LLM-based techniques, achieving MAP improvements of 23.33-38.57% at the document and method levels. Similar gains were observed in MRR, with increases of 25.14-53.74% at both granularity levels. Statistical significance tests also confirm the superiority of our technique. By addressing the reasoning, dependency, and context limitations, CogniGent advances the state of bug localization, bridging human-like cognition with agentic automation for improved performance.
Authors: Fabi Nahian Madhurja, Rusab Sarmun, Muhammad E. H. Chowdhury, Adam Mushtak, Israa Al-Hashimi, Sohaib Bassam Zoghoul
Abstract: Cervical spine fractures are critical medical conditions requiring precise and efficient detection for effective clinical management. This study explores the viability of 2D projection-based vertebra segmentation for vertebra-level fracture detection in 3D CT volumes, presenting an end-to-end pipeline for automated analysis of cervical vertebrae (C1-C7). By approximating a 3D volume through optimized 2D axial, sagittal, and coronal projections, regions of interest are identified using the YOLOv8 model from all views and combined to approximate the 3D cervical spine area, achieving a 3D mIoU of 94.45 percent. This projection-based localization strategy reduces computational complexity compared to traditional 3D segmentation methods while maintaining high performance. It is followed by a DenseNet121-Unet-based multi-label segmentation leveraging variance- and energy-based projections, achieving a Dice score of 87.86 percent. Strategic approximation of 3D vertebral masks from these 2D segmentation masks enables the extraction of individual vertebra volumes. The volumes are analyzed for fractures using an ensemble of 2.5D Spatio-Sequential models incorporating both raw slices and projections per vertebra for complementary evaluation. This ensemble achieves vertebra-level and patient-level F1 scores of 68.15 and 82.26, and ROC-AUC scores of 91.62 and 83.04, respectively. We further validate our approach through an explainability study that provides saliency map visualizations highlighting anatomical regions relevant for diagnosis, and an interobserver variability analysis comparing our model's performance with expert radiologists, demonstrating competitive results.
Authors: Wenxin Zhou, Ritesh Mehta, Anthony Miyaguchi
Abstract: We develop a two-stage retrieval system that combines multiple complementary retrieval methods with a learned reranker and LLM-based reranking, to address the TREC Tip-of-the-Tongue (ToT) task. In the first stage, we employ hybrid retrieval that merges LLM-based retrieval, sparse (BM25), and dense (BGE-M3) retrieval methods. We also introduce topic-aware multi-index dense retrieval that partitions the Wikipedia corpus into 24 topical domains. In the second stage, we evaluate both a trained LambdaMART reranker and LLM-based reranking. To support model training, we generate 5000 synthetic ToT queries using LLMs. Our best system achieves recall of 0.66 and NDCG@1000 of 0.41 on the test set by combining hybrid retrieval with Gemini-2.5-flash reranking, demonstrating the effectiveness of fusion retrieval.
Authors: Yongxin Deng, Zhen Fang, Sharon Li, Ling Chen
Abstract: Hallucination detection is critical for deploying large language models (LLMs) in real-world applications. Existing hallucination detection methods achieve strong performance when the training and test data come from the same domain, but they suffer from poor cross-domain generalization. In this paper, we study an important yet overlooked problem, termed generalizable hallucination detection (GHD), which aims to train hallucination detectors on data from a single domain while ensuring robust performance across diverse related domains. In studying GHD, we simulate multi-turn dialogues following LLMs' initial response and observe an interesting phenomenon: hallucination-initiated multi-turn dialogues universally exhibit larger uncertainty fluctuations than factual ones across different domains. Based on the phenomenon, we propose a new score SpikeScore, which quantifies abrupt fluctuations in multi-turn dialogues. Through both theoretical analysis and empirical validation, we demonstrate that SpikeScore achieves strong cross-domain separability between hallucinated and non-hallucinated responses. Experiments across multiple LLMs and benchmarks demonstrate that the SpikeScore-based detection method outperforms representative baselines in cross-domain generalization and surpasses advanced generalization-oriented methods, verifying the effectiveness of our method in cross-domain hallucination detection.
Authors: Murad Farzulla
Abstract: This study investigates whether cryptocurrency whitepaper narratives align with empirically observed market factor structure. We construct a pipeline combining zero-shot NLP classification of 38 whitepapers across 10 semantic categories with CP tensor decomposition of hourly market data (49 assets, 17,543 timestamps). Using Procrustes rotation and Tucker's congruence coefficient (phi), we find weak alignment between claims and market statistics (phi = 0.246, p = 0.339) and between claims and latent factors (phi = 0.058, p = 0.751). A methodological validation comparison (statistics versus factors, both derived from market data) achieves significance (p < 0.001), confirming the pipeline detects real structure. The null result indicates whitepaper narratives do not meaningfully predict market factor structure, with implications for narrative economics and investor decision-making. Entity-level analysis reveals specialized tokens (XMR, CRV, YFI) show stronger narrative-market correspondence than broad infrastructure tokens.
Authors: Francisco Caldas, Sahil Kumar, Cl\'audia Soares
Abstract: We introduce a model-agnostic forward diffusion process for time-series forecasting that decomposes signals into spectral components, preserving structured temporal patterns such as seasonality more effectively than standard diffusion. Unlike prior work that modifies the network architecture or diffuses directly in the frequency domain, our proposed method alters only the diffusion process itself, making it compatible with existing diffusion backbones (e.g., DiffWave, TimeGrad, CSDI). By staging noise injection according to component energy, it maintains high signal-to-noise ratios for dominant frequencies throughout the diffusion trajectory, thereby improving the recoverability of long-term patterns. This strategy enables the model to maintain the signal structure for a longer period in the forward process, leading to improved forecast quality. Across standard forecasting benchmarks, we show that applying spectral decomposition strategies, such as the Fourier or Wavelet transform, consistently improves upon diffusion models using the baseline forward process, with negligible computational overhead. The code for this paper is available at https://anonymous.4open.science/r/D-FDP-4A29.
Authors: Kai Yuan, Anthony Zheng, Jia Hu, Divyanshu Sheth, Hemanth Velaga, Kylee Kim, Matteo Guarrera, Besim Avci, Jianhua Li, Xuetao Yin, Rajyashree Mukherjee, Sean Suchter
Abstract: Query Auto-Completion (QAC) suggests query completions as users type, helping them articulate intent and reach results more efficiently. Existing approaches face fundamental challenges: traditional retrieve-and-rank pipelines have limited long-tail coverage and require extensive feature engineering, while recent generative methods suffer from hallucination and safety risks. We present a unified framework that reformulates QAC as end-to-end list generation through Retrieval-Augmented Generation (RAG) and multi-objective Direct Preference Optimization (DPO). Our approach combines three key innovations: (1) reformulating QAC as end-to-end list generation with multi-objective optimization; (2) defining and deploying a suite of rule-based, model-based, and LLM-as-judge verifiers for QAC, and using them in a comprehensive methodology that combines RAG, multi-objective DPO, and iterative critique-revision for high-quality synthetic data; (3) a hybrid serving architecture enabling efficient production deployment under strict latency constraints. Evaluation on a large-scale commercial search platform demonstrates substantial improvements: offline metrics show gains across all dimensions, human evaluation yields +0.40 to +0.69 preference scores, and a controlled online experiment achieves 5.44\% reduction in keystrokes and 3.46\% increase in suggestion adoption, validating that unified generation with RAG and multi-objective alignment provides an effective solution for production QAC. This work represents a paradigm shift to end-to-end generation powered by large language models, RAG, and multi-objective alignment, establishing a production-validated framework that can benefit the broader search and recommendation industry.
Authors: Haixiang Sun, Andrew L. Liu
Abstract: Few-shot learning requires models to generalize under limited supervision while remaining robust to distribution shifts. Existing Sinkhorn Distributionally Robust Optimization (DRO) methods provide theoretical guarantees but rely on a fixed reference distribution, which limits their adaptability. We propose a Prototype-Guided Distributionally Robust Optimization (PG-DRO) framework that learns class-adaptive priors from abundant base data via hierarchical optimal transport and embeds them into the Sinkhorn DRO formulation. This design enables few-shot information to be organically integrated into producing class-specific robust decisions that are both theoretically grounded and efficient, and further aligns the uncertainty set with transferable structural knowledge. Experiments show that PG-DRO achieves stronger robust generalization in few-shot scenarios, outperforming both standard learners and DRO baselines.
Authors: Sota Nishiyama, Masaaki Imaizumi
Abstract: Modern machine learning models are typically trained via multi-pass stochastic gradient descent (SGD) with small batch sizes, and understanding their dynamics in high dimensions is of great interest. However, an analytical framework for describing the high-dimensional asymptotic behavior of multi-pass SGD with small batch sizes for nonlinear models is currently missing. In this study, we address this gap by analyzing the high-dimensional dynamics of a stochastic differential equation called a \emph{stochastic gradient flow} (SGF), which approximates multi-pass SGD in this regime. In the limit where the number of data samples $n$ and the dimension $d$ grow proportionally, we derive a closed system of low-dimensional and continuous-time equations and prove that it characterizes the asymptotic distribution of the SGF parameters. Our theory is based on the dynamical mean-field theory (DMFT) and is applicable to a wide range of models encompassing generalized linear models and two-layer neural networks. We further show that the resulting DMFT equations recover several existing high-dimensional descriptions of SGD dynamics as special cases, thereby providing a unifying perspective on prior frameworks such as online SGD and high-dimensional linear regression. Our proof builds on the existing DMFT technique for gradient flow and extends it to handle the stochasticity in SGF using tools from stochastic calculus.
Authors: Hee-Sun Choi, Beom-Seok Han
Abstract: We introduce a multiplicative neural network architecture in which multiplicative interactions constitute the fundamental representation, rather than appearing as auxiliary components within an additive model. We establish a universal approximation theorem for this architecture and analyze its approximation properties in terms of locality and regularity in Bessel potential spaces. To complement the theoretical results, we conduct numerical experiments on representative targets exhibiting sharp transition layers or pointwise loss of higher-order regularity. The experiments focus on the spatial structure of approximation errors and on regularity-sensitive quantities, in particular, the convergence of Zygmund-type seminorms. The results show that the proposed multiplicative architecture yields residual error structures that are more tightly aligned with regions of reduced regularity and exhibit more stable convergence in regularity-sensitive metrics. These results demonstrate that adopting a multiplicative representation format has concrete implications for the localization and regularity behavior of neural network approximations, providing a direct connection between architectural design and analytical properties of the approximating functions.
Authors: Binghui Li, Zilin Wang, Fengling Chen, Shiyang Zhao, Ruiheng Zheng, Lei Wu
Abstract: We study optimal learning-rate schedules (LRSs) under the functional scaling law (FSL) framework introduced in Li et al. (2025), which accurately models the loss dynamics of both linear regression and large language model (LLM) pre-training. Within FSL, loss dynamics are governed by two exponents: a source exponent $s>0$ controlling the rate of signal learning, and a capacity exponent $\beta>1$ determining the rate of noise forgetting. Focusing on a fixed training horizon $N$, we derive the optimal LRSs and reveal a sharp phase transition. In the easy-task regime $s \ge 1 - 1/\beta$, the optimal schedule follows a power decay to zero, $\eta^*(z) = \eta_{\mathrm{peak}}(1 - z/N)^{2\beta - 1}$, where the peak learning rate scales as $\eta_{\mathrm{peak}} \eqsim N^{-\nu}$ for an explicit exponent $\nu = \nu(s,\beta)$. In contrast, in the hard-task regime $s < 1 - 1/\beta$, the optimal LRS exhibits a warmup-stable-decay (WSD) (Hu et al. (2024)) structure: it maintains the largest admissible learning rate for most of training and decays only near the end, with the decay phase occupying a vanishing fraction of the horizon. We further analyze optimal shape-fixed schedules, where only the peak learning rate is tuned -- a strategy widely adopted in practiceand characterize their strengths and intrinsic limitations. This yields a principled evaluation of commonly used schedules such as cosine and linear decay. Finally, we apply the power-decay LRS to one-pass stochastic gradient descent (SGD) for kernel regression and show the last iterate attains the exact minimax-optimal rate, eliminating the logarithmic suboptimality present in prior analyses. Numerical experiments corroborate our theoretical predictions.
Authors: Muhammad Rashid, Elvio G. Amparore, Enrico Ferrari, Damiano Verda
Abstract: Pixel-level feature attributions are an important tool in eXplainable AI for Computer Vision (XCV), providing visual insights into how image features influence model predictions. The Owen formula for hierarchical Shapley values has been widely used to interpret machine learning (ML) models and their learned representations. However, existing hierarchical Shapley approaches do not exploit the multiscale structure of image data, leading to slow convergence and weak alignment with the actual morphological features. Moreover, no prior Shapley method has leveraged data-aware hierarchies for Computer Vision tasks, leaving a gap in model interpretability of structured visual data. To address this, this paper introduces ShapBPT, a novel data-aware XCV method based on the hierarchical Shapley formula. ShapBPT assigns Shapley coefficients to a multiscale hierarchical structure tailored for images, the Binary Partition Tree (BPT). By using this data-aware hierarchical partitioning, ShapBPT ensures that feature attributions align with intrinsic image morphology, effectively prioritizing relevant regions while reducing computational overhead. This advancement connects hierarchical Shapley methods with image data, providing a more efficient and semantically meaningful approach to visual interpretability. Experimental results confirm ShapBPT's effectiveness, demonstrating superior alignment with image structures and improved efficiency over existing XCV methods, and a 20-subject user study confirming that ShapBPT explanations are preferred by humans.
Authors: Oswin So, Brian Karrer, Chuchu Fan, Ricky T. Q. Chen, Guan-Horng Liu
Abstract: Computation methods for solving entropy-regularized reward optimization -- a class of problems widely used for fine-tuning generative models -- have advanced rapidly. Among those, Adjoint Matching (AM, Domingo-Enrich et al., 2025) has proven highly effective in continuous state spaces with differentiable rewards. Transferring these practical successes to discrete generative modeling, however, remains particularly challenging and largely unexplored, mainly due to the drastic shift in generative model classes to discrete state spaces, which are nowhere differentiable. In this work, we propose Discrete Adjoint Matching (DAM) -- a discrete variant of AM for fine-tuning discrete generative models characterized by Continuous-Time Markov Chains, such as diffusion-based large language models. The core of DAM is the introduction of discrete adjoint-an estimator of the optimal solution to the original problem but formulated on discrete domains-from which standard matching frameworks can be applied. This is derived via a purely statistical standpoint, in contrast to the control-theoretic viewpoint in AM, thereby opening up new algorithmic opportunities for general adjoint-based estimators. We showcase DAM's effectiveness on synthetic and mathematical reasoning tasks.
Authors: Babak Rahmani
Abstract: Code World Models (CWMs) are language models trained to simulate program execution by predicting explicit runtime state after every executed command. This execution-based world modeling enables internal verification within the model, offering an alternative to natural language chain-of-thought reasoning. However, the sources of errors and the nature of CWMs' limitations remain poorly understood. We study CWMs from two complementary perspectives: local semantic execution and long-horizon state tracking. On real-code benchmarks, we identify two dominant failure regimes. First, dense runtime state reveals produce token-intensive execution traces, leading to token-budget exhaustion on programs with long execution histories. Second, failures disproportionately concentrate in string-valued state, which we attribute to limitations of subword tokenization rather than program structure. To study long-horizon behavior, we use a controlled permutation-tracking benchmark that isolates state propagation under action execution. We show that long-horizon degradation is driven primarily by incorrect action generation: when actions are replaced with ground-truth commands, a Transformer-based CWM propagates state accurately over long horizons, despite known limitations of Transformers in long-horizon state tracking. These findings suggest directions for more efficient supervision and state representations in CWMs that are better aligned with program execution and data types.
Authors: Igor Santos-Grueiro
Abstract: Safety evaluation for advanced AI systems assumes that behavior observed under evaluation predicts behavior in deployment. This assumption weakens for agents with situational awareness, which may exploit regime leakage, cues distinguishing evaluation from deployment, to implement conditional policies that comply under oversight while defecting in deployment-like regimes. We recast alignment evaluation as a problem of information flow under partial observability and show that divergence between evaluation-time and deployment-time behavior is bounded by the regime information extractable from decision-relevant internal representations. We study regime-blind mechanisms, training-time interventions that restrict access to regime cues through adversarial invariance constraints without assuming complete information erasure. We evaluate this approach across multiple open-weight language models and controlled failure modes including scientific sycophancy, temporal sleeper agents, and data leakage. Regime-blind training reduces regime-conditioned failures without measurable loss of task utility, but exhibits heterogeneous and model-dependent dynamics. Sycophancy shows a sharp representational and behavioral transition at moderate intervention strength, consistent with a stability cliff. In sleeper-style constructions and certain cross-model replications, suppression occurs without a clean collapse of regime decodability and may display non-monotone or oscillatory behavior as invariance pressure increases. These findings indicate that representational invariance is a meaningful but limited control lever. It can raise the cost of regime-conditioned strategies but cannot guarantee elimination or provide architecture-invariant thresholds. Behavioral evaluation should therefore be complemented with white-box diagnostics of regime awareness and internal information flow.
Authors: Vid Kocijan, Jinu Sunil, Jan Eric Lenssen, Viman Deb, Xinwei Xe, Federico Reyes Gomez, Matthias Fey, Jure Leskovec
Abstract: The purpose of predictive modeling on relational data is to predict future or missing values in a relational database, for example, future purchases of a user, risk of readmission of the patient, or the likelihood that a financial transaction is fraudulent. Typically powered by machine learning methods, predictive models are used in recommendations, financial fraud detection, supply chain optimization, and other systems, providing billions of predictions every day. However, training a machine learning model requires manual work to extract the required training examples - prediction entities and target labels - from the database, which is slow, laborious, and prone to mistakes. Here, we present the Predictive Query Language (PQL), an SQL-inspired declarative language for defining predictive tasks on relational databases. PQL allows specifying a predictive task in a single declarative query, enabling the automatic computation of training labels for a large variety of machine learning tasks, such as regression, classification, time-series forecasting, and recommender systems. PQL is already successfully integrated and used in a collection of use cases as part of a predictive AI platform. The versatility of the language can be demonstrated through its many ongoing use cases, including financial fraud, item recommendations, and workload prediction. We demonstrate its versatile design through two implementations; one for small-scale, low-latency use and one that can handle large-scale databases.
Authors: Laura Ib\'a\~nez-Mart\'inez, Chukwuemeka Nkama, Andrea Poltronieri, Xavier Serra, Mart\'in Rocamora
Abstract: Recent approaches in music generation rely on disentangled representations, often labeled as structure and timbre or local and global, to enable controllable synthesis. Yet the underlying properties of these embeddings remain underexplored. In this work, we evaluate such disentangled representations in a set of music audio models for controllable generation using a probing-based framework that goes beyond standard downstream tasks. The selected models reflect diverse unsupervised disentanglement strategies, including inductive biases, data augmentations, adversarial objectives, and staged training procedures. We further isolate specific strategies to analyze their effect. Our analysis spans four key axes: informativeness, equivariance, invariance, and disentanglement, which are assessed across datasets, tasks, and controlled transformations. Our findings reveal inconsistencies between intended and actual semantics of the embeddings, suggesting that current strategies fall short of producing truly disentangled representations, and prompting a re-examination of how controllability is approached in music generation.
Authors: Arthur Juliani
Abstract: This paper formalizes religious epistemology through the mathematics of Variational Autoencoders. We model religious traditions as distinct generative mappings from a shared, low-dimensional latent space to the high-dimensional space of observable cultural forms, and define three competing generative configurations corresponding to exclusivism, universalism, and perennialism, alongside syncretism as direct mixing in observable space. Through abductive comparison, we argue that exclusivism cannot parsimoniously account for cross-traditional contemplative convergence, that syncretism fails because combining the outputs of distinct generative processes produces incoherent artifacts, and that universalism suffers from posterior collapse: stripping traditions to a common core discards the structural information necessary for inference. The perennialist configuration provides the best explanatory fit. Within this framework, strict orthodoxy emerges not as a cultural constraint but as a structural necessity: the contemplative practices that recover the latent source must be matched to the specific tradition whose forms they take as input. The unity of religions, if it exists, is real but inaccessible by shortcut: one must go deep rather than wide.
Authors: Yichun Xiao, Runwei Guan, Fangqiang Ding
Abstract: 4D mmWave radar provides weather-robust, velocity-aware measurements and is more cost-effective than LiDAR. However, radar-only 3D detection still trails LiDAR-based systems because radar point clouds are sparse, irregular, and often corrupted by multipath noise, yielding weak and unstable geometry. We present HyperDet, a detector-agnostic radar-only 3D detection framework that constructs a task-aware hyper 4D radar point cloud for standard LiDAR-oriented detectors. HyperDet aggregates returns from multiple surround-view 4D radars over consecutive frames to improve coverage and density, then applies geometry-aware cross-sensor consensus validation with a lightweight self-consistency check outside overlap regions to suppress inconsistent returns. It further integrates a foreground-focused diffusion module with training-time mixed radar-LiDAR supervision to densify object structures while lifting radar attributes (e.g., Doppler, RCS); the model is distilled into a consistency model for single-step inference. On MAN TruckScenes, HyperDet consistently improves over raw radar inputs with VoxelNeXt and CenterPoint, partially narrowing the radar-LiDAR gap. These results show that input-level refinement enables radar to better leverage LiDAR-oriented detectors without architectural modifications.
Authors: Lai Wei, Liangbo He, Jun Lan, Lingzhong Dong, Yutong Cai, Siyuan Li, Huijia Zhu, Weiqiang Wang, Linghe Kong, Yue Wang, Zhuosheng Zhang, Weiran Huang
Abstract: Multimodal Large Language Models (MLLMs) excel at broad visual understanding but still struggle with fine-grained perception, where decisive evidence is small and easily overwhelmed by global context. Recent "Thinking-with-Images" methods alleviate this by iteratively zooming in and out regions of interest during inference, but incur high latency due to repeated tool calls and visual re-encoding. To address this, we propose Region-to-Image Distillation, which transforms zooming from an inference-time tool into a training-time primitive, thereby internalizing the benefits of agentic zooming into a single forward pass of an MLLM. In particular, we first zoom in to micro-cropped regions to let strong teacher models generate high-quality VQA data, and then distill this region-grounded supervision back to the full image. After training on such data, the smaller student model improves "single-glance" fine-grained perception without tool use. To rigorously evaluate this capability, we further present ZoomBench, a hybrid-annotated benchmark of 845 VQA data spanning six fine-grained perceptual dimensions, together with a dual-view protocol that quantifies the global--regional "zooming gap". Experiments show that our models achieve leading performance across multiple fine-grained perception benchmarks, and also improve general multimodal cognition on benchmarks such as visual reasoning and GUI agents. We further discuss when "Thinking-with-Images" is necessary versus when its gains can be distilled into a single forward pass. Our code is available at https://github.com/inclusionAI/Zooming-without-Zooming.
URLs: https://github.com/inclusionAI/Zooming-without-Zooming.
Authors: John Muchovej, Amanda Royka, Shane Lee, Julian Jara-Ettinger
Abstract: Do Large Language Models (LLMs) possess a Theory of Mind (ToM)? Research into this question has focused on evaluating LLMs against benchmarks and found success across a range of social tasks. However, these evaluations do not test for the actual representations posited by ToM: namely, a causal model of mental states and behavior. Here, we use a cognitively-grounded definition of ToM to develop and test a new evaluation framework. Specifically, our approach probes whether LLMs have a coherent, domain-general, and consistent model of how mental states cause behavior -- regardless of whether that model matches a human-like ToM. We find that even though LLMs succeed in approximating human judgments in a simple ToM paradigm, they fail at a logically equivalent task and exhibit low consistency between their action predictions and corresponding mental state inferences. As such, these findings suggest that the social proficiency exhibited by LLMs is not the result of a domain-general or consistent ToM.
Authors: Yuanliang Li, Xun Gong, Reza Iravani, Bo Cao, Heng Liu, Ziming Chen
Abstract: The DC-side ground fault (GF) poses significant risks to three-phase TN-earthed photovoltaic (PV) systems, as the resulting high fault current can directly damage both PV inverters and PV modules. Once a fault occurs, locating the faulty string through manual string-by-string inspection is highly time-consuming and inefficient. This work presents a comprehensive analysis of GF characteristics through fault-current analysis and a simulation-based case study covering multiple fault locations. Building on these insights, we propose an edge-AI-based GF localization approach tailored for three-phase TN-earthed PV systems. A PLECS-based simulation model that incorporates PV hysteresis effects is developed to generate diverse GF scenarios, from which correlation-based features are extracted throughout the inverter's four-stage shutdown sequence. Using the simulated dataset, a lightweight Variational Information Bottleneck (VIB)-based localization model is designed and trained, achieving over 93% localization accuracy at typical sampling rates with low computational cost, demonstrating strong potential for deployment on resource-constrained PV inverters.
Authors: Haobin Li, Yutong Yang, Yijie Lin, Xiang Dai, Mouxing Yang, Xi Peng
Abstract: As a multimodal extension of Chain-of-Thought (CoT), Thinking with Images (TWI) has recently emerged as a promising avenue to enhance the reasoning capability of Multi-modal Large Language Models (MLLMs), which generates interleaved CoT by incorporating visual cues into the textual reasoning process. However, the success of existing TWI methods heavily relies on the assumption that interleaved image-text CoTs are faultless, which is easily violated in real-world scenarios due to the complexity of multimodal understanding. In this paper, we reveal and study a highly-practical yet under-explored problem in TWI, termed Noisy Thinking (NT). Specifically, NT refers to the imperfect visual cues mining and answer reasoning process. As the saying goes, ``One mistake leads to another'', erroneous interleaved CoT would cause error accumulation, thus significantly degrading the performance of MLLMs. To solve the NT problem, we propose a novel method dubbed Reliable Thinking with Images (RTWI). In brief, RTWI estimates the reliability of visual cues and textual CoT in a unified text-centric manner and accordingly employs robust filtering and voting modules to prevent NT from contaminating the final answer. Extensive experiments on seven benchmarks verify the effectiveness of RTWI against NT.