Authors: Elbert Ho
Abstract: Recently, machine learning has made a significant impact on de novo drug design. However, current approaches to creating novel molecules conditioned on a target protein typically rely on generating molecules directly in the 3D conformational space, which are often slow and overly complex. In this work, we propose SOLD (SELFIES-based Objective-driven Latent Diffusion), a novel latent diffusion model that generates molecules in a latent space derived from 1D SELFIES strings and conditioned on a target protein. In the process, we also train an innovative SELFIES transformer and propose a new way to balance losses when training multi-task machine learning models.Our model generates high-affinity molecules for the target protein in a simple and efficient way, while also leaving room for future improvements through the addition of more data.
Authors: Zhuoning Xu, Xinyan Liu
Abstract: Jigsaw puzzle solving remains challenging in computer vision, requiring an understanding of both local fragment details and global spatial relationships. While most traditional approaches only focus on visual cues like edge matching and visual coherence, few methods explore natural language descriptions for semantic guidance in challenging scenarios, especially for eroded gap puzzles. We propose a vision-language framework that leverages textual context to enhance puzzle assembly performance. Our approach centers on the Vision-Language Hierarchical Semantic Alignment (VLHSA) module, which aligns visual patches with textual descriptions through multi-level semantic matching from local tokens to global context. Also, a multimodal architecture that combines dual visual encoders with language features for cross-modal reasoning is integrated into this module. Experiments demonstrate that our method significantly outperforms state-of-the-art models across various datasets, achieving substantial improvements, including a 14.2 percentage point gain in piece accuracy. Ablation studies confirm the critical role of the VLHSA module in driving improvements over vision-only approaches. Our work establishes a new paradigm for jigsaw puzzle solving by incorporating multimodal semantic insights.
Authors: Jin Li, Zhebo Wang, Tianliang Lu, Mohan Li, Wenpeng Xing, Meng Han
Abstract: Entropy-based inference methods have gained traction for improving the reliability of Large Language Models (LLMs). However, many existing approaches, such as entropy minimization techniques, suffer from high computational overhead and fail to leverage historical token context effectively. To address these limitations, we propose Spectral Logit Sculpting (SLS), a lightweight inference-time optimization method that dynamically modulates token distributions using spectral and entropic properties of recent logits. SLS maintains a sliding buffer of top-K logits, performs on-the-fly Singular Value Decomposition (SVD) to identify dominant spectral directions, and adaptively rescales logits based on both entropy and logit gap statistics--only activating when uncertainty is high. Without updating any model parameters, SLS effectively sharpens the output distribution while preserving contextual consistency. Experimental results on multiple public benchmarks demonstrate that SLS consistently outperforms existing baseline methods, achieving superior accuracy in mathematical, coding, and scientific reasoning tasks.
Authors: Daksh Pandey
Abstract: Self-supervised learning (SSL) has emerged as a powerful paradigm for learning representations on graph data without requiring manual labels. However, leading SSL methods like GRACE are fundamentally incompatible with privacy-preserving technologies such as Homomorphic Encryption (HE) due to their reliance on non-polynomial operations. This paper introduces Poly-GRACE, a novel framework for HE-compatible self-supervised learning on graphs. Our approach consists of a fully polynomial-friendly Graph Convolutional Network (GCN) encoder and a novel, polynomial-based contrastive loss function. Through experiments on three benchmark datasets -- Cora, CiteSeer, and PubMed -- we demonstrate that Poly-GRACE not only enables private pre-training but also achieves performance that is highly competitive with, and in the case of CiteSeer, superior to the standard non-private baseline. Our work represents a significant step towards practical and high-performance privacy-preserving graph representation learning.
Authors: Yanke Wang, Kyriakos Flouris
Abstract: This work explores optimization methods on hyperbolic manifolds. Building on Riemannian optimization principles, we extend the Hyperbolic Stochastic Gradient Descent (a specialization of Riemannian SGD) to a Hyperbolic Adam optimizer. While these methods are particularly relevant for learning on the Poincar\'e ball, they may also provide benefits in Euclidean and other non-Euclidean settings, as the chosen optimization encourages the learning of Poincar\'e embeddings. This representation, in turn, accelerates convergence in the early stages of training, when parameters are far from the optimum. As a case study, we train diffusion models using the hyperbolic optimization methods with hyperbolic time-discretization of the Langevin dynamics, and show that they achieve faster convergence on certain datasets without sacrificing generative quality.
Authors: Yebin Lim, Susik Yoon
Abstract: Recent advancements in large language models (LLMs) have shown promise in feature engineering for tabular data, but concerns about their reliability persist, especially due to variability in generated outputs. We introduce a multi-level diagnosis and evaluation framework to assess the robustness of LLMs in feature engineering across diverse domains, focusing on the three main factors: key variables, relationships, and decision boundary values for predicting target classes. We demonstrate that the robustness of LLMs varies significantly over different datasets, and that high-quality LLM-generated features can improve few-shot prediction performance by up to 10.52%. This work opens a new direction for assessing and enhancing the reliability of LLM-driven feature engineering in various domains.
Authors: Zenghui Huang, Ting Shu, Zhonglei Wang, Yang Lu, Yan Yan, Wei Zhong, Hanzi Wang
Abstract: Accurate and timely forecasting of heavy rainfall remains a critical challenge for modern society. Precipitation exhibits a highly imbalanced distribution: most observations record no or light rain, while heavy rainfall events are rare. Such an imbalanced distribution obstructs deep learning models from effectively predicting heavy rainfall events. To address this challenge, we treat rainfall forecasting explicitly as a long-tailed learning problem, identifying the insufficient representation of heavy rainfall events as the primary barrier to forecasting accuracy. Therefore, we introduce DPSformer, a long-tail-aware model that enriches representation of heavy rainfall events through a high-resolution branch. For heavy rainfall events $ \geq $ 50 mm/6 h, DPSformer lifts the Critical Success Index (CSI) of a baseline Numerical Weather Prediction (NWP) model from 0.012 to 0.067. For the top 1% coverage of heavy rainfall events, its Fraction Skill Score (FSS) exceeds 0.45, surpassing existing methods. Our work establishes an effective long-tailed paradigm for heavy rainfall prediction, offering a practical tool to enhance early warning systems and mitigate the societal impacts of extreme weather events.
Authors: Hao Chen, Tao Han, Jie Zhang, Song Guo, Lei Bai
Abstract: To gain finer regional forecasts, many works have explored the regional integration from the global atmosphere, e.g., by solving boundary equations in physics-based methods or cropping regions from global forecasts in data-driven methods. However, the effectiveness of these methods is often constrained by static and imprecise regional boundaries, resulting in poor generalization ability. To address this issue, we propose Spatial-Temporal Weather Forecasting (STCast), a novel AI-driven framework for adaptive regional boundary optimization and dynamic monthly forecast allocation. Specifically, our approach employs a Spatial-Aligned Attention (SAA) mechanism, which aligns global and regional spatial distributions to initialize boundaries and adaptively refines them based on attention-derived alignment patterns. Furthermore, we design a Temporal Mixture-of-Experts (TMoE) module, where atmospheric variables from distinct months are dynamically routed to specialized experts using a discrete Gaussian distribution, enhancing the model's ability to capture temporal patterns. Beyond global and regional forecasting, we evaluate our STCast on extreme event prediction and ensemble forecasting. Experimental results demonstrate consistent superiority over state-of-the-art methods across all four tasks.
Authors: Remi Genet, Hugo Inzirillo
Abstract: This paper introduces Large Execution Models (LEMs), a novel deep learning framework that extends transformer-based architectures to address complex execution problems with flexible time boundaries and multiple execution constraints. Building upon recent advances in neural VWAP execution strategies, LEMs generalize the approach from fixed-duration orders to scenarios where execution duration is bounded between minimum and maximum time horizons, similar to share buyback contract structures. The proposed architecture decouples market information processing from execution allocation decisions: a common feature extraction pipeline using Temporal Kolmogorov-Arnold Networks (TKANs), Variable Selection Networks (VSNs), and multi-head attention mechanisms processes market data to create informational context, while independent allocation networks handle the specific execution logic for different scenarios (fixed quantity vs. fixed notional, buy vs. sell orders). This architectural separation enables a unified model to handle diverse execution objectives while leveraging shared market understanding across scenarios. Through comprehensive empirical evaluation on intraday cryptocurrency markets and multi-day equity trading using DOW Jones constituents, we demonstrate that LEMs achieve superior execution performance compared to traditional benchmarks by dynamically optimizing execution paths within flexible time constraints. The unified model architecture enables deployment across different execution scenarios (buy/sell orders, varying duration boundaries, volume/notional targets) through a single framework, providing significant operational advantages over asset-specific approaches.
Authors: Sai Varun Kodathala
Abstract: The optimization of hyperparameters in convolutional neural networks (CNNs) remains a challenging and computationally expensive process, often requiring extensive trial-and-error approaches or exhaustive grid searches. This study introduces the application of Taguchi Design of Experiments methodology, a statistical optimization technique traditionally used in quality engineering, to systematically optimize CNN hyperparameters for professional boxing action recognition. Using an L12(211) orthogonal array, eight hyperparameters including image size, color mode, activation function, learning rate, rescaling, shuffling, vertical flip, and horizontal flip were systematically evaluated across twelve experimental configurations. To address the multi-objective nature of machine learning optimization, five different approaches were developed to simultaneously optimize training accuracy, validation accuracy, training loss, and validation loss using Signal-to-Noise ratio analysis. The study employed a novel logarithmic scaling technique to unify conflicting metrics and enable comprehensive multi-quality assessment within the Taguchi framework. Results demonstrate that Approach 3, combining weighted accuracy metrics with logarithmically transformed loss functions, achieved optimal performance with 98.84% training accuracy and 86.25% validation accuracy while maintaining minimal loss values. The Taguchi analysis revealed that learning rate emerged as the most influential parameter, followed by image size and activation function, providing clear guidance for hyperparameter prioritization in CNN optimization.
Authors: Rongguang Ye, Ming Tang, Edith C. H. Ngai
Abstract: As increasingly large pre-trained models are released, deploying them on edge devices for privacy-preserving applications requires effective compression. Recent works combine quantization with the fine-tuning of high-precision LoRA adapters, which can substantially reduce model size while mitigating the accuracy loss from quantization. However, edge devices have inherently heterogeneous capabilities, while performing configuration-wise fine-tuning for every quantization setting is computationally prohibitive. In this paper, we propose CoA-LoRA, a method that dynamically adjusts the LoRA adapter to arbitrary quantization configurations (i.e., the per-layer bit-width choices of a pre-trained model) without requiring repeated fine-tuning. This is accomplished via a configuration-aware model that maps each configuration to its low-rank adjustments. The effectiveness of this model critically depends on the training configuration set, a collection of configurations chosen to cover different total bit-width budgets. However, constructing a high-quality configuration set is non-trivial. We therefore design a Pareto-based configuration search that iteratively optimizes the training configuration set, yielding more precise low-rank adjustments. Our experiments demonstrate that, unlike the state-of-the-art methods that require fine-tuning a separate LoRA adapter for each configuration, CoA-LoRA incurs no additional time cost while achieving comparable or even superior performance to those methods.
Authors: Pierre Lotte (IRIT, IRIT-SIG, UT, UT3), Andr\'e P\'eninou (IRIT, IRIT-SIG, UT2J), Olivier Teste (IRIT-SIG, IRIT, UT2J, UT)
Abstract: In this article, we suggest a novel non-supervised partition based anomaly detection method for anomaly detection in multivariate time series called PARADISE. This methodology creates a partition of the variables of the time series while ensuring that the inter-variable relations remain untouched. This partitioning relies on the clustering of multiple correlation coefficients between variables to identify subsets of variables before executing anomaly detection algorithms locally for each of those subsets. Through multiple experimentations done on both synthetic and real datasets coming from the literature, we show the relevance of our approach with a significant improvement in anomaly detection performance.
Authors: Guillermo Comesa\~na Cimadevila
Abstract: Classical learning theory describes a well-characterised U-shaped relationship between model complexity and prediction error, reflecting a transition from underfitting in underparameterised regimes to overfitting as complexity grows. Recent work, however, has introduced the notion of a second descent in test error beyond the interpolation threshold-giving rise to the so-called double descent phenomenon. While double descent has been studied extensively in the context of deep learning, it has also been reported in simpler models, including decision trees and gradient boosting. In this work, we revisit these claims through the lens of classical machine learning applied to a biological classification task: predicting isoniazid resistance in Mycobacterium tuberculosis using whole-genome sequencing data. We systematically vary model complexity along two orthogonal axes-learner capacity (e.g., Pleaf, Pboost) and ensemble size (i.e., Pens)-and show that double descent consistently emerges only when complexity is scaled jointly across these axes. When either axis is held fixed, generalisation behaviour reverts to classical U- or L-shaped patterns. These results are replicated on a synthetic benchmark and support the unfolding hypothesis, which attributes double descent to the projection of distinct generalisation regimes onto a single complexity axis. Our findings underscore the importance of treating model complexity as a multidimensional construct when analysing generalisation behaviour. All code and reproducibility materials are available at: https://github.com/guillermocomesanacimadevila/Demystifying-Double-Descent-in-ML.
URLs: https://github.com/guillermocomesanacimadevila/Demystifying-Double-Descent-in-ML.
Authors: Brij Malhotra, Shivvrat Arya, Tahrima Rahman, Vibhav Giridhar Gogate
Abstract: We introduce learning to condition (L2C), a scalable, data-driven framework for accelerating Most Probable Explanation (MPE) inference in Probabilistic Graphical Models (PGMs), a fundamentally intractable problem. L2C trains a neural network to score variable-value assignments based on their utility for conditioning, given observed evidence. To facilitate supervised learning, we develop a scalable data generation pipeline that extracts training signals from the search traces of existing MPE solvers. The trained network serves as a heuristic that integrates with search algorithms, acting as a conditioning strategy prior to exact inference or as a branching and node selection policy within branch-and-bound solvers. We evaluate L2C on challenging MPE queries involving high-treewidth PGMs. Experiments show that our learned heuristic significantly reduces the search space while maintaining or improving solution quality over state-of-the-art methods.
Authors: Tobiasz Puslecki, Krzysztof Walkowiak
Abstract: The recent progress in TinyML technologies triggers the need to address the challenge of balancing inference time and classification quality. TinyML systems are defined by specific constraints in computation, memory and energy. These constraints emphasize the need for specialized optimization techniques when implementing Machine Learning (ML) applications on such platforms. While deep neural networks are widely used in TinyML, the exploration of Dynamic Ensemble Selection (DES) methods is also beneficial. This study examines a DES-Clustering approach for a multi-class computer vision task within TinyML systems. This method allows for adjusting classification accuracy, thereby affecting latency and energy consumption per inference. We implemented the TinyDES-Clustering library, optimized for embedded system limitations. Experiments have shown that a larger pool of classifiers for dynamic selection improves classification accuracy, and thus leads to an increase in average inference time on the TinyML device.
Authors: Yutong Liang, Chang Hou, Guy Y. Cornejo Maceda, Andrea Ianiro, Stefano Discetti, Andrea Meil\'an-Vila, Didier Sornette, Sandro Claudio Lera, Jialong Chen, Xiaozhou He, Bernd R. Noack
Abstract: We propose a physics-informed machine-learned framework for sensor-based flow estimation for drone trajectories in complex urban terrain. The input is a rich set of flow simulations at many wind conditions. The outputs are velocity and uncertainty estimates for a target domain and subsequent sensor optimization for minimal uncertainty. The framework has three innovations compared to traditional flow estimators. First, the algorithm scales proportionally to the domain complexity, making it suitable for flows that are too complex for any monolithic reduced-order representation. Second, the framework extrapolates beyond the training data, e.g., smaller and larger wind velocities. Last, and perhaps most importantly, the sensor location is a free input, significantly extending the vast majority of the literature. The key enablers are (1) a Reynolds number-based scaling of the flow variables, (2) a physics-based domain decomposition, (3) a cluster-based flow representation for each subdomain, (4) an information entropy correlating the subdomains, and (5) a multi-variate probability function relating sensor input and targeted velocity estimates. This framework is demonstrated using drone flight paths through a three-building cluster as a simple example. We anticipate adaptations and applications for estimating complete cities and incorporating weather input.
Authors: Xunhao Lai, Jialiang Kang, Jianqiao Lu, Tong Lin, Pengyu Zhao
Abstract: Linear attention offers a linear-time alternative to self-attention but often struggles to capture long-range patterns. We revisit linear attention through a prediction-correction lens and show that prevalent variants can be written as a combination of a historical prediction and a single-token correction, which creates an expressivity bottleneck. To address this bottleneck, we introduce Residual Linear Attention (RLA), a framework that equips linear attention with an explicit residual-fitting mechanism. RLA maintains an auxiliary recurrent state that learns to accumulate residual errors over time and correct the base prediction. We further instantiate a delta-rule version, Residual Delta Net (RDN), incorporating adaptive gating and residual clipping for enhanced correction control and stability. Our implementation leverages highly optimized linear attention kernels and preserves linear time and memory. Across language modeling and recall-intensive evaluations, RLA and RDN consistently outperform their respective baselines and other modern linear-attention methods, narrowing the gap to standard Transformers while retaining linear scaling.
Authors: Qichen Liao, Chengqiu Hu, Fangzheng Miao, Bao Li, Yiyang Liu, Junlong Lyu, Lirui Jiang, Jun Wang, Lingchao Zheng, Jun Li, Yuwei Fan
Abstract: Multi-head Latent Attention (MLA) significantly reduces KVCache memory usage in Large Language Models while introducing substantial computational overhead and intermediate variable expansion. This poses challenges for efficient hardware implementation -- especially during the decode phase. This paper introduces Ascend MLA (AMLA), a high-performance kernel specifically optimized for Huawei's Ascend NPUs. AMLA is built on two core innovations: (1) A novel FlashAttention-based algorithm that replaces floating-point multiplications with integer additions for output block rescaling, leveraging binary correspondence between FP32 and INT32 representations; (2) A Preload Pipeline strategy with hierarchical tiling that maximizes FLOPS utilization: the Preload Pipeline achieves Cube-bound performance, while hierarchical tiling overlaps data movement and computation within the Cube core. Experiments show that on Ascend 910 NPUs (integrated in CloudMatrix384), AMLA achieves up to 614 TFLOPS, reaching 86.8% of the theoretical maximum FLOPS, outperforming the state-of-the-art open-source FlashMLA implementation, whose FLOPS utilization is up to 66.7% on NVIDIA H800 SXM5. The AMLA kernel has been integrated into Huawei's CANN and will be released soon.
Authors: Long Xu, Yongcai Chen, Fengshuo Liu, Yuzhong Peng
Abstract: Structure-Based Drug Design (SBDD) is a powerful strategy in computational drug discovery, utilizing three-dimensional protein structures to guide the design of molecules with improved binding affinity. However, capturing complex protein-ligand interactions across multiple scales remains challenging, as current methods often overlook the hierarchical organization and intrinsic asymmetry of these interactions. To address these limitations, we propose MSCoD, a novel Bayesian updating-based generative framework for structure-based drug design. In our MSCoD, Multi-Scale Information Bottleneck (MSIB) was developed, which enables semantic compression at multiple abstraction levels for efficient hierarchical feature extraction. Furthermore, a multi-head cooperative attention (MHCA) mechanism was developed, which employs asymmetric protein-to-ligand attention to capture diverse interaction types while addressing the dimensionality disparity between proteins and ligands. Empirical studies showed that MSCoD outperforms state-of-the-art methods on the benchmark dataset. Case studies on challenging targets such as KRAS G12D further demonstrate its applicability in real-world scenarios. The code and data underlying this article are freely available at https://github.com/xulong0826/MSCoD.
Authors: Baoyi Xie, Shuiling Shi, Wenqi Liu
Abstract: Integrated wind-solar-wave marine energy systems hold broad promise for supplying clean electricity in offshore and coastal regions. By leveraging the spatiotemporal complementarity of multiple resources, such systems can effectively mitigate the intermittency and volatility of single-source outputs, thereby substantially improving overall power-generation efficiency and resource utilization. Accurate ultra-short-term forecasting is crucial for ensuring secure operation and optimizing proactive dispatch. However, most existing forecasting methods construct separate models for each energy source, insufficiently account for the complex couplings among multiple energies, struggle to capture the system's nonlinear and nonstationary dynamics, and typically depend on extensive manual parameter tuning-limitations that constrain both predictive performance and practicality. We address this issue using a Bayesian-optimized Multivariate Variational Mode Decomposition-Long Short-Term Memory (MVMD-LSTM) framework. The framework first applies MVMD to jointly decompose wind, solar and wave power series so as to preserve cross-source couplings; it uses Bayesian optimization to automatically search the number of modes and the penalty parameter in the MVMD process to obtain intrinsic mode functions (IMFs); finally, an LSTM models the resulting IMFs to achieve ultra-short-term power forecasting for the integrated system. Experiments based on field measurements from an offshore integrated energy platform in China show that the proposed framework significantly outperforms benchmark models in terms of MAPE, RMSE and MAE. The results demonstrate superior predictive accuracy, robustness, and degree of automation.
Authors: Ahmad Ayaz Amin
Abstract: We introduce Random Projection Flows (RPFs), a principled framework for injective normalizing flows that leverages tools from random matrix theory and the geometry of random projections. RPFs employ random semi-orthogonal matrices, drawn from Haar-distributed orthogonal ensembles via QR decomposition of Gaussian matrices, to project data into lower-dimensional latent spaces for the base distribution. Unlike PCA-based flows or learned injective maps, RPFs are plug-and-play, efficient, and yield closed-form expressions for the Riemannian volume correction term. We demonstrate that RPFs are both theoretically grounded and practically effective, providing a strong baseline for generative modeling and a bridge between random projection theory and normalizing flows.
Authors: Aaron Zweig, Mingxuan Zhang, Elham Azizi, David Knowles
Abstract: A useful inductive bias for temporal data is that trajectories should stay close to the data manifold. Traditional flow matching relies on straight conditional paths, and flow matching methods which learn geodesics rely on RBF kernels or nearest neighbor graphs that suffer from the curse of dimensionality. We propose to use score matching and annealed energy distillation to learn a metric tensor that faithfully captures the underlying data geometry and informs more accurate flows. We demonstrate the efficacy of this strategy on synthetic manifolds with analytic geodesics, and interpolation of cell
Authors: Xiaojian Wang, Chaoli Zhang, Zhonglong Zheng, Yunliang Jiang
Abstract: Time series forecasting has various applications, such as meteorological rainfall prediction, traffic flow analysis, financial forecasting, and operational load monitoring for various systems. Due to the sparsity of time series data, relying solely on time-domain or frequency-domain modeling limits the model's ability to fully leverage multi-domain information. Moreover, when applied to time series forecasting tasks, traditional attention mechanisms tend to over-focus on irrelevant historical information, which may introduce noise into the prediction process, leading to biased results. We proposed WDformer, a wavelet-based differential Transformer model. This study employs the wavelet transform to conduct a multi-resolution analysis of time series data. By leveraging the advantages of joint representation in the time-frequency domain, it accurately extracts the key information components that reflect the essential characteristics of the data. Furthermore, we apply attention mechanisms on inverted dimensions, allowing the attention mechanism to capture relationships between multiple variables. When performing attention calculations, we introduced the differential attention mechanism, which computes the attention score by taking the difference between two separate softmax attention matrices. This approach enables the model to focus more on important information and reduce noise. WDformer has achieved state-of-the-art (SOTA) results on multiple challenging real-world datasets, demonstrating its accuracy and effectiveness. Code is available at https://github.com/xiaowangbc/WDformer.
Authors: Yongchao Huang
Abstract: We present a family of \textit{Gaussian Mixture Approximation} (GMA) samplers for sampling unnormalised target densities, encompassing \textit{weights-only GMA} (W-GMA), \textit{Laplace Mixture Approximation} (LMA), \textit{expectation-maximization GMA} (EM-GMA), and further variants. GMA adopts a simple two-stage paradigm: (i) initialise a finite set of Gaussian components and draw samples from a proposal mixture; (ii) fit the mixture to the target by optimising either only the component weights or also the means and variances, via a sample-based KL divergence objective that requires only evaluations of the unnormalised density, followed by stratified resampling. The method is gradient-free, and computationally efficient: it leverages the ease of sampling from Gaussians, efficient optimisation methods (projected gradient descent, mirror descent, and EM), and the robustness of stratified resampling to produce samples faithful to the target. We show that this optimisation-resampling scheme yields consistent approximations under mild conditions, and we validate this methodology with empirical results demonstrating accuracy and speed across diverse densities.
Authors: Kasun Eranda Wijethilake, Adnan Mahmood, Quan Z. Sheng
Abstract: Federated Learning (FL) is a distributed machine learning technique that preserves data privacy by sharing only the trained parameters instead of the client data. This makes FL ideal for highly dynamic, heterogeneous, and time-critical applications, in particular, the Internet of Vehicles (IoV) networks. However, FL encounters considerable challenges in such networks owing to the high data and device heterogeneity. To address these challenges, we propose FedCLF, i.e., FL with Calibrated Loss and Feedback control, which introduces calibrated loss as a utility in the participant selection process and a feedback control mechanism to dynamically adjust the sampling frequency of the clients. The envisaged approach (a) enhances the overall model accuracy in case of highly heterogeneous data and (b) optimizes the resource utilization for resource constrained IoV networks, thereby leading to increased efficiency in the FL process. We evaluated FedCLF vis-\`a-vis baseline models, i.e., FedAvg, Newt, and Oort, using CIFAR-10 dataset with varying data heterogeneity. Our results depict that FedCLF significantly outperforms the baseline models by up to a 16% improvement in high data heterogeneity-related scenarios with improved efficiency via reduced sampling frequency.
Authors: Nikola Prianikov, Evelyne Janssen-van Dam, Marcin Pietrasik, Charalampos S. Kouzinopoulos
Abstract: Correct identification of failure mechanisms is essential for manufacturers to ensure the quality of their products. Certain failures of printheads developed by Canon Production Printing can be identified from the behavior of individual nozzles, the states of which are constantly recorded and can form distinct patterns in terms of the number of failed nozzles over time, and in space in the nozzle grid. In our work, we investigate the problem of printhead failure classification based on a multifaceted dataset of nozzle logging and propose a Machine Learning classification approach for this problem. We follow the feature-based framework of time-series classification, where a set of time-based and spatial features was selected with the guidance of domain experts. Several traditional ML classifiers were evaluated, and the One-vs-Rest Random Forest was found to have the best performance. The proposed model outperformed an in-house rule-based baseline in terms of a weighted F1 score for several failure mechanisms.
Authors: Sri Vatsa Vuddanti, Aarav Shah, Satwik Kumar Chittiprolu, Tony Song, Sunishchal Dev, Kevin Zhu, Maheep Chaudhary
Abstract: Tool-augmented language agents frequently fail in real-world deployment due to tool malfunctions--timeouts, API exceptions, or inconsistent outputs--triggering cascading reasoning errors and task abandonment. Existing agent training pipelines optimize only for success trajectories, failing to expose models to the tool failures that dominate real-world usage. We propose \textbf{PALADIN}, a generalizable framework for equipping language agents with robust failure recovery capabilities. PALADIN trains on 50,000+ recovery-annotated trajectories constructed via systematic failure injection and expert demonstrations on an enhanced ToolBench dataset. Training uses LoRA-based fine-tuning to retain base capabilities while injecting recovery competence. At inference, PALADIN detects execution-time errors and retrieves the most similar case from a curated bank of 55+ failure exemplars aligned with ToolScan's taxonomy, then executes the corresponding recovery action. This approach generalizes to novel failures beyond the training distribution, retaining 95.2\% recovery performance on unseen tool APIs. Evaluation across PaladinEval and ToolReflectEval demonstrates consistent improvements in Recovery Rate (RR), Task Success Rate (TSR), Catastrophic Success Rate (CSR), and Efficiency Score (ES). PALADIN improves RR from 32.76% to 89.68% (+57% relative) over ToolBench and outperforms the strongest baseline CRITIC (76.34%) by +13.3%. Against vanilla agents, PALADIN achieves 89.86\% RR (+66% relative improvement from 23.75%). These results establish PALADIN as an effective method for building fault-tolerant agents capable of robust recovery in real-world tool environments.
Authors: Ming Yang, Xiaofan Li, Zhiyuan Ma, Dengliang Shi, Jintao Du, Yu Cheng, Weiguo Zheng
Abstract: Recent curriculum reinforcement learning for large language models (LLMs) typically rely on difficulty-based annotations for data filtering and ordering. However, such methods suffer from local optimization, where continual training on simple samples in the early steps can cause the policy to lose its exploration. We propose a novel schema, namely Hamiltonian curiosity augmented large language model reinforcement (HAMMER), that transfers diversity metrics, commonly used in dataset evaluation, into the dynamic reinforcement learning procedure, where training samples are ordered via a minimum-semantic Hamiltonian path making the initial training retrain more exploration. From a theoretical perspective of generalization bounds, diversity-driven ordering facilitates stable convergence. Empirical evaluations indicate that HAMMER stimulates model "curiosity" and consistently achieves a 3% to 4% average accuracy gain across diverse inference benchmark.
Authors: Yuan Huang
Abstract: Recent advancements in training paradigms for Large Language Models (LLMs) have unlocked their remarkable capabilities in natural language processing and cross-domain generalization. While LLMs excel in tasks like programming and mathematical problem-solving, their zero-shot performance in specialized domains requiring expert knowledge, such as cybersecurity, is often suboptimal. This limitation arises because foundational LLMs are designed for general-purpose applications, constraining their ability to encapsulate domain-specific expertise within their parameter space. To address this, we explore fine-tuning strategies to embed cybersecurity knowledge into LLMs, enhancing their performance in cybersecurity question-answering (Q\&A) tasks while prioritizing computational efficiency. Specifically, we investigate Supervised Fine-Tuning (SFT), Low-Rank Adaptation (LoRA), and Quantized Low-Rank Adaptation (QLoRA) using a cybersecurity Q\&A dataset. Our results demonstrate that these fine-tuning approaches significantly outperform the foundational model in cybersecurity Q\&A tasks. Moreover, LoRA and QLoRA achieve comparable performance to SFT with substantially lower computational costs, offering an efficient pathway for adapting LLMs to specialized domains. Our work highlights the potential of low-rank fine-tuning strategies to bridge the gap between general-purpose LLMs and domain-specific applications.
Authors: Prajjwal Bhattarai, Mohammad Amjad, Dmytro Zhylko, Tuka Alhanai
Abstract: Knowledge distillation is a common paradigm for transferring capabilities from larger models to smaller ones. While traditional distillation methods leverage a probabilistic divergence over the output of the teacher and student models, feature-based distillation methods often minimize variants of Euclidean norms between the hidden layer representations. The main goal is for the student to mimic the structure of the feature space of the teacher. In this work, we theoretically show that existing feature distillation methods, such as projection based mean squared loss or Centered Kernel Alignment (CKA), cannot capture the feature structure, even under zero loss. We then motivate the use of Procrustes distance and the Frobenius norm of Feature Gram Matrix, distances already common in the context of measuring representational alignment, as distillation losses. We show that feature distillation through our method showcases statistically significant improvement in distillation performance across language models families (BERT and OPT) in classification and instruction-following tasks by up to 2 percentage points, showcasing the potential of integrating feature geometry into existing distillation methods.
Authors: Xianyang Deng, Wenshuai Liu, Yaru FuB, Qi Zhu
Abstract: Unmanned aerial vehicles (UAVs)-assisted mobile crowdsensing (MCS) has emerged as a promising paradigm for data collection. However, challenges such as spectrum scarcity, device heterogeneity, and user mobility hinder efficient coordination of sensing, communication, and computation. To tackle these issues, we propose a joint optimization framework that integrates time slot partition for sensing, communication, and computation phases, resource allocation, and UAV 3D trajectory planning, aiming to maximize the amount of processed sensing data. The problem is formulated as a non-convex stochastic optimization and further modeled as a partially observable Markov decision process (POMDP) that can be solved by multi-agent deep reinforcement learning (MADRL) algorithm. To overcome the limitations of conventional multi-layer perceptron (MLP) networks, we design a novel MADRL algorithm with hybrid actor network. The newly developed method is based on heterogeneous agent proximal policy optimization (HAPPO), empowered by convolutional neural networks (CNN) for feature extraction and Kolmogorov-Arnold networks (KAN) to capture structured state-action dependencies. Extensive numerical results demonstrate that our proposed method achieves significant improvements in the amount of processed sensing data when compared with other benchmarks.
Authors: Yifang Zhang, Pengfei Duan, Henan Wang, Shengwu Xiong
Abstract: Rainfall nowcasting, which aims to predict precipitation within the next 0 to 3 hours, is critical for disaster mitigation and real-time response planning. However, most time series forecasting benchmarks in meteorology are evaluated on variables with strong periodicity, such as temperature and humidity, which fail to reflect model capabilities in more complex and practically meteorology scenarios like rainfall nowcasting. To address this gap, we propose RainfallBench, a benchmark designed for rainfall nowcasting, a highly challenging and practically relevant task characterized by zero inflation, temporal decay, and non-stationarity, focused on predicting precipitation within the next 0 to 3 hours. The dataset is derived from five years of meteorological observations, recorded at 15-minute intervals across six essential variables, and collected from more than 12,000 GNSS stations globally. In particular, it incorporates precipitable water vapor (PWV), a crucial indicator of rainfall that is absent in other datasets. We further design specialized evaluation strategies to assess model performance on key meteorological challenges, such as multi-scale prediction and extreme rainfall events, and evaluate over 20 state-of-the-art models across six major architectures on RainfallBench. Additionally, to address the zero-inflation and temporal decay issues overlooked by existing models, we introduce Bi-Focus Precipitation Forecaster (BFPF), a plug-and-play module that incorporates domain-specific priors to enhance rainfall time series forecasting. Statistical analysis and ablation studies validate the comprehensiveness of our dataset as well as the superiority of our methodology. Code and datasets are available at https://anonymous.4open.science/r/RainfallBench-A710.
Authors: Jiexi Xu
Abstract: The performance of Large Language Models (LLMs) depends heavily on the chosen prompting strategy, yet static approaches such as Zero-Shot, Few-Shot, or Chain-of-Thought (CoT) impose a rigid efficiency-accuracy trade-off. Highly accurate strategies like Self-Consistency (SC) incur substantial computational waste on simple tasks, while lightweight methods often fail on complex inputs. This paper introduces the Prompt Policy Network (PPN), a lightweight reinforcement learning framework that formalizes adaptive strategy selection as a single-step Markov Decision Process (MDP). The PPN, trained with Proximal Policy Optimization (PPO) and guided by a resource-explicit reward function, learns to allocate costly reasoning strategies only when necessary. Experiments on arithmetic reasoning benchmarks demonstrate that PPN achieves superior performance on the efficiency-accuracy Pareto front, delivering up to 61.5% token cost reduction compared to Self-Consistency while maintaining competitive accuracy. This work contributes a systematic, adaptive framework for cost-efficient LLM deployment, advancing the design of lightweight optimization techniques for scalable and sustainable language model applications.
Authors: Cristian Bodnar, Rapha\"el Rousseau-Rizzi, Nikhil Shankar, James Merleau, Stylianos Flampouris, Guillem Candille, Slavica Antic, Fran\c{c}ois Miralles, Jayesh K. Gupta
Abstract: Weather foundation models (WFMs) have recently set new benchmarks in global forecast skill, yet their concrete value for the weather-sensitive infrastructure that powers modern society remains largely unexplored. In this study, we fine-tune Silurian AI's 1.5B-parameter WFM, Generative Forecasting Transformer (GFT), on a rich archive of Hydro-Qu\'ebec asset observations--including transmission-line weather stations, wind-farm met-mast streams, and icing sensors--to deliver hyper-local, asset-level forecasts for five grid-critical variables: surface temperature, precipitation, hub-height wind speed, wind-turbine icing risk, and rime-ice accretion on overhead conductors. Across 6-72 h lead times, the tailored model surpasses state-of-the-art NWP benchmarks, trimming temperature mean absolute error (MAE) by 15%, total-precipitation MAE by 35%, and lowering wind speed MAE by 15%. Most importantly, it attains an average precision score of 0.72 for day-ahead rime-ice detection, a capability absent from existing operational systems, which affords several hours of actionable warning for potentially catastrophic outage events. These results show that WFMs, when post-trained with small amounts of high-fidelity, can serve as a practical foundation for next-generation grid-resilience intelligence.
Authors: Liangjian Wen, Qun Dai, Jianzhuang Liu, Jiangtao Zheng, Yong Dai, Dongkai Wang, Zhao Kang, Jun Wang, Zenglin Xu, Jiang Duan
Abstract: In multimodal representation learning, synergistic interactions between modalities not only provide complementary information but also create unique outcomes through specific interaction patterns that no single modality could achieve alone. Existing methods may struggle to effectively capture the full spectrum of synergistic information, leading to suboptimal performance in tasks where such interactions are critical. This is particularly problematic because synergistic information constitutes the fundamental value proposition of multimodal representation. To address this challenge, we introduce InfMasking, a contrastive synergistic information extraction method designed to enhance synergistic information through an \textbf{Inf}inite \textbf{Masking} strategy. InfMasking stochastically occludes most features from each modality during fusion, preserving only partial information to create representations with varied synergistic patterns. Unmasked fused representations are then aligned with masked ones through mutual information maximization to encode comprehensive synergistic information. This infinite masking strategy enables capturing richer interactions by exposing the model to diverse partial modality combinations during training. As computing mutual information estimates with infinite masking is computationally prohibitive, we derive an InfMasking loss to approximate this calculation. Through controlled experiments, we demonstrate that InfMasking effectively enhances synergistic information between modalities. In evaluations on large-scale real-world datasets, InfMasking achieves state-of-the-art performance across seven benchmarks. Code is released at https://github.com/brightest66/InfMasking.
Authors: Payal Mohapatra, Yueyuan Sui, Akash Pandey, Stephen Xia, Qi Zhu
Abstract: From clinical healthcare to daily living, continuous sensor monitoring across multiple modalities has shown great promise for real-world intelligent decision-making but also faces various challenges. In this work, we introduce MAESTRO, a novel framework that overcomes key limitations of existing multimodal learning approaches: (1) reliance on a single primary modality for alignment, (2) pairwise modeling of modalities, and (3) assumption of complete modality observations. These limitations hinder the applicability of these approaches in real-world multimodal time-series settings, where primary modality priors are often unclear, the number of modalities can be large (making pairwise modeling impractical), and sensor failures often result in arbitrary missing observations. At its core, MAESTRO facilitates dynamic intra- and cross-modal interactions based on task relevance, and leverages symbolic tokenization and adaptive attention budgeting to construct long multimodal sequences, which are processed via sparse cross-modal attention. The resulting cross-modal tokens are routed through a sparse Mixture-of-Experts (MoE) mechanism, enabling black-box specialization under varying modality combinations. We evaluate MAESTRO against 10 baselines on four diverse datasets spanning three applications, and observe average relative improvements of 4% and 8% over the best existing multimodal and multivariate approaches, respectively, under complete observations. Under partial observations -- with up to 40% of missing modalities -- MAESTRO achieves an average 9% improvement. Further analysis also demonstrates the robustness and efficiency of MAESTRO's sparse, modality-aware design for learning from dynamic time series.
Authors: Oluwaseyi Giwa, Jonathan Shock, Jaco Du Toit, Tobi Awodumila
Abstract: Dynamic resource allocation in heterogeneous wireless networks (HetNets) is challenging for traditional methods under varying user loads and channel conditions. We propose a deep reinforcement learning (DRL) framework that jointly optimises transmit power, bandwidth, and scheduling via a multi-objective reward balancing throughput, energy efficiency, and fairness. Using real base station coordinates, we compare Proximal Policy Optimisation (PPO) and Twin Delayed Deep Deterministic Policy Gradient (TD3) against three heuristic algorithms in multiple network scenarios. Our results show that DRL frameworks outperform heuristic algorithms in optimising resource allocation in dynamic networks. These findings highlight key trade-offs in DRL design for future HetNets.
Authors: Mohammadreza Bakhtyari, Bogdan Mazoure, Renato Cordeiro de Amorim, Guillaume Rabusseau, Vladimir Makarenkov
Abstract: We introduce ClustRecNet - a novel deep learning (DL)-based recommendation framework for determining the most suitable clustering algorithms for a given dataset, addressing the long-standing challenge of clustering algorithm selection in unsupervised learning. To enable supervised learning in this context, we construct a comprehensive data repository comprising 34,000 synthetic datasets with diverse structural properties. Each of them was processed using 10 popular clustering algorithms. The resulting clusterings were assessed via the Adjusted Rand Index (ARI) to establish ground truth labels, used for training and evaluation of our DL model. The proposed network architecture integrates convolutional, residual, and attention mechanisms to capture both local and global structural patterns from the input data. This design supports end-to-end training to learn compact representations of datasets and enables direct recommendation of the most suitable clustering algorithm, reducing reliance on handcrafted meta-features and traditional Cluster Validity Indices (CVIs). Comprehensive experiments across synthetic and real-world benchmarks demonstrate that our DL model consistently outperforms conventional CVIs (e.g. Silhouette, Calinski-Harabasz, Davies-Bouldin, and Dunn) as well as state-of-the-art AutoML clustering recommendation approaches (e.g. ML2DAC, AutoCluster, and AutoML4Clust). Notably, the proposed model achieves a 0.497 ARI improvement over the Calinski-Harabasz index on synthetic data and a 15.3% ARI gain over the best-performing AutoML approach on real-world data.
Authors: Zelin Tan, Hejia Geng, Mulei Zhang, Xiaohang Yu, Guancheng Wan, Yifan Zhou, Qiang He, Xiangyuan Xue, Heng Zhou, Yutao Fan, Zhongzhi Li, Zaibin Zhang, Guibin Zhang, Chen Zhang, Zhenfei Yin, Lei Bai
Abstract: While scaling laws for large language models (LLMs) during pre-training have been extensively studied, their behavior under reinforcement learning (RL) post-training remains largely unexplored. This paper presents a systematic empirical investigation of scaling behaviors in RL-based post-training, with a particular focus on mathematical reasoning. Based on 54 experiments across diverse model sizes and training settings, we characterize how model scale, data volume, and computational budget interact to shape performance. Our analysis leads to four key findings: (1). Under a fixed computational budget, larger models trained for fewer steps consistently outperform smaller models trained for more steps. (2). Given a fixed amount of training data, larger models achieve superior sample efficiency, yielding lower loss. (3). In data-constrained regimes, repeated reuse of high-quality data proves highly effective, as final performance is primarily governed by the total number of optimization steps rather than the uniqueness of samples. (4). These scaling behaviors are robust across both base and instruction-tuned models, which share similar learning dynamics (e.g., larger models show faster convergence) even while differing in absolute accuracy. Collectively, these results provide a principled foundation and practical guidelines for efficiently scaling the reasoning capabilities of LLMs through RL post-training.
Authors: Amirhossein Zare (SeyedAbolfazl), Amirhessam Zare (SeyedAbolfazl), Parmida Sadat Pezeshki (SeyedAbolfazl), Herlock (SeyedAbolfazl), Rahimi, Ali Ebrahimi, Ignacio V\'azquez-Garc\'ia, Leo Anthony Celi
Abstract: Class imbalance remains a major challenge in machine learning, especially for high-dimensional biomedical data where nonlinear manifold structures dominate. Traditional oversampling methods such as SMOTE rely on local linear interpolation, often producing implausible synthetic samples. Deep generative models like Conditional Variational Autoencoders (CVAEs) better capture nonlinear distributions, but standard variants treat all minority samples equally, neglecting the importance of uncertain, boundary-region examples emphasized by heuristic methods like Borderline-SMOTE and ADASYN. We propose Local Entropy-Guided Oversampling with a CVAE (LEO-CVAE), a generative oversampling framework that explicitly incorporates local uncertainty into both representation learning and data generation. To quantify uncertainty, we compute Shannon entropy over the class distribution in a sample's neighborhood: high entropy indicates greater class overlap, serving as a proxy for uncertainty. LEO-CVAE leverages this signal through two mechanisms: (i) a Local Entropy-Weighted Loss (LEWL) that emphasizes robust learning in uncertain regions, and (ii) an entropy-guided sampling strategy that concentrates generation in these informative, class-overlapping areas. Applied to clinical genomics datasets (ADNI and TCGA lung cancer), LEO-CVAE consistently improves classifier performance, outperforming both traditional oversampling and generative baselines. These results highlight the value of uncertainty-aware generative oversampling for imbalanced learning in domains governed by complex nonlinear structures, such as omics data.
Authors: Shuang Liang, Guido Mont\'ufar
Abstract: We examine gradient descent in matrix factorization and show that under large step sizes the parameter space develops a fractal structure. We derive the exact critical step size for convergence in scalar-vector factorization and show that near criticality the selected minimizer depends sensitively on the initialization. Moreover, we show that adding regularization amplifies this sensitivity, generating a fractal boundary between initializations that converge and those that diverge. The analysis extends to general matrix factorization with orthogonal initialization. Our findings reveal that near-critical step sizes induce a chaotic regime of gradient descent where the long-term dynamics are unpredictable and there are no simple implicit biases, such as towards balancedness, minimum norm, or flatness.
Authors: Linus Aronsson, Han Wu, Morteza Haghir Chehreghani
Abstract: We study active correlation clustering where pairwise similarities are not provided upfront and must be queried in a cost-efficient manner through active learning. Specifically, we focus on the cold-start scenario, where no true initial pairwise similarities are available for active learning. To address this challenge, we propose a coverage-aware method that encourages diversity early in the process. We demonstrate the effectiveness of our approach through several synthetic and real-world experiments.
Authors: Yogesh Verma, Markus Heinonen, Vikas Garg
Abstract: Protein structure prediction and folding are fundamental to understanding biology, with recent deep learning advances reshaping the field. Diffusion-based generative models have revolutionized protein design, enabling the creation of novel proteins. However, these methods often neglect the intrinsic physical realism of proteins, driven by noising dynamics that lack grounding in physical principles. To address this, we first introduce a physically motivated non-linear noising process, grounded in classical physics, that unfolds proteins into secondary structures (e.g., alpha helices, linear beta sheets) while preserving topological integrity--maintaining bonds, and preventing collisions. We then integrate this process with the flow-matching paradigm on SE(3) to model the invariant distribution of protein backbones with high fidelity, incorporating sequence information to enable sequence-conditioned folding and expand the generative capabilities of our model. Experimental results demonstrate that the proposed method achieves state-of-the-art performance in unconditional protein generation, producing more designable and novel protein structures while accurately folding monomer sequences into precise protein conformations.
Authors: Shane Bergsma, Nolan Dey, Joel Hestness
Abstract: Data curriculums have become central to successful LLM training, yet principles governing optimal data placement remain unclear. We introduce the *training re-evaluation curve (TREC)*, a diagnostic that retrospectively evaluates training batches *using the final model weights*. The TREC characterizes how well a trained model retains training data as a function of *when* the data was encountered during training. Analyzing TRECs for models from 111M to 3.9B parameters, we show that placing high-quality data at low points on the TREC significantly improves performance. Importantly, while a TREC is initially observable only after training, we demonstrate it can be *predicted in advance* from AdamW's implicit EMA coefficients, enabling proactive curriculum design. By predicting TRECs for published training recipes, we explain prior ablations and reveal suboptimal data placements. We also align high-quality data with TREC minima in order to improve continual pre-training of a 3.9B-parameter LLM trained on 900B tokens.
Authors: Penglei Gao, Yan Zou, Abhijit Duggal, Shuaiqi Huang, Faming Liang, Xiaofeng Wang
Abstract: We introduce the Functional Competing Risk Net (FCRN), a unified deep-learning framework for discrete-time survival analysis under competing risks, which seamlessly integrates functional covariates and handles missing data within an end-to-end model. By combining a micro-network Basis Layer for functional data representation with a gradient-based imputation module, FCRN simultaneously learns to impute missing values and predict event-specific hazards. Evaluated on multiple simulated datasets and a real-world ICU case study using the MIMIC-IV and Cleveland Clinic datasets, FCRN demonstrates substantial improvements in prediction accuracy over random survival forests and traditional competing risks models. This approach advances prognostic modeling in critical care by more effectively capturing dynamic risk factors and static predictors while accommodating irregular and incomplete data.
Authors: Fernanda Zapata Bascu\~n\'an
Abstract: In this work, we explore the latent space of a denoising variational autoencoder with a mixture-of-Gaussians prior (VAE-MoG), trained on gravitational wave data from event GW150914. To evaluate how well the model captures the underlying structure, we use Hamiltonian Monte Carlo (HMC) to draw posterior samples conditioned on clean inputs, and compare them to the encoder's outputs from noisy data. Although the model reconstructs signals accurately, statistical comparisons reveal a clear mismatch in the latent space. This shows that strong denoising performance doesn't necessarily mean the latent representations are reliable highlighting the importance of using posterior-based validation when evaluating generative models.
Authors: Jordyn E. A. Lorentz, Katharine M. Clark
Abstract: This paper introduces mixsemble, an ensemble method that adapts the Dawid-Skene model to aggregate predictions from multiple model-based clustering algorithms. Unlike traditional crowdsourcing, which relies on human labels, the framework models the outputs of clustering algorithms as noisy annotations. Experiments on both simulated and real-world datasets show that, although the mixsemble is not always the single top performer, it consistently approaches the best result and avoids poor outcomes. This robustness makes it a practical alternative when the true data structure is unknown, especially for non-expert users.
Authors: S C Bee, N Dervilis, K Worden, L A Bull
Abstract: Equation discovery provides a grey-box approach to system identification by uncovering governing dynamics directly from observed data. However, a persistent challenge lies in ensuring that identified models generalise across operating conditions rather than over-fitting to specific datasets. This work investigates this issue by applying a Bayesian relevance vector machine (RVM) within a multi-task learning (MTL) framework for simultaneous parameter identification across multiple datasets. In this formulation, responses from the same structure under different excitation levels are treated as related tasks that share model parameters but retain task-specific noise characteristics. A simulated single degree-of-freedom oscillator with linear and cubic stiffness provided the case study, with datasets generated under three excitation regimes. Standard single-task RVM models were able to reproduce system responses but often failed to recover the true governing terms when excitations insufficiently stimulated non-linear dynamics. By contrast, the MTL-RVM combined information across tasks, improving parameter recovery for weakly and moderately excited datasets, while maintaining strong performance under high excitation. These findings demonstrate that multi-task Bayesian inference can mitigate over-fitting and promote generalisation in equation discovery. The approach is particularly relevant to structural health monitoring, where varying load conditions reveal complementary aspects of system physics.
Authors: Liang Qiao, Yue Dai, Yeqi Huang, Hongyu Kan, Jun Shi, Hong An
Abstract: Multi-Modal Diffusion Transformers (DiTs) demonstrate exceptional capabilities in visual synthesis, yet their deployment remains constrained by substantial computational demands. To alleviate this bottleneck, many sparsity-based acceleration methods have been proposed. However, their diverse sparsity patterns often require customized kernels for high-performance inference, limiting universality. We propose FlashOmni, a unified sparse attention engine compatible with arbitrary DiT architectures. FlashOmni introduces flexible sparse symbols to standardize the representation of a wide range of sparsity strategies, such as feature caching and block-sparse skipping. This unified abstraction enables the execution of diverse sparse computations within a single attention kernel. In addition, FlashOmni designs optimized sparse GEMMs for attention blocks, leveraging sparse symbols to eliminate redundant computations and further improve efficiency. Experiments demonstrate that FlashOmni delivers near-linear, closely matching the sparsity ratio speedup (1:1) in attention and GEMM-$Q$, and achieves 2.5$\times$-3.8$\times$ acceleration in GEMM-$O$ (max peaking at about 87.5% of the theoretical limit). Applied with a multi-granularity sparsity strategy, it enables the Hunyuan model (33K) to achieve about 1.5$\times$ end-to-end acceleration without degrading visual quality.
Authors: Hao Ban, Kaiyi Ji
Abstract: Large language models are often adapted using parameter-efficient techniques such as Low-Rank Adaptation (LoRA), formulated as $y = W_0x + BAx$, where $W_0$ is the pre-trained parameters and $x$ is the input to the adapted layer. While multi-adapter extensions often employ multiple LoRAs, prior studies suggest that the inner $A$ matrices are highly similar during training and thus suitable for sharing. We revisit this phenomenon and find that this similarity is largely attributable to the identical initialization rather than shared knowledge, with $B$ playing a more critical role in knowledge encoding and transfer. Motivated by these insights, we propose \textbf{ALoRA}, an asymmetric multi-LoRA design with multiple $A$ matrices and a single shared $B$ in multi-task fine-tuning, and \textbf{Fed-ALoRA}, which shares $B$ across clients in federated fine-tuning under both homogeneous and heterogeneous settings, through a novel matrix decomposition strategy to accommodate heterogeneous ranks across clients. Experiments on commonsense reasoning, math reasoning, multi-task NLP dataset, and federated NLP dataset demonstrate that our methods achieve more balanced performance across tasks with comparable or superior average accuracy relative to existing multi-LoRA approaches. Codes are available at https://github.com/OptMN-Lab/ALoRA.
Authors: Dong Hyun Jeon, Lijing Zhu, Haifang Li, Pengze Li, Jingna Feng, Tiehang Duan, Houbing Herbert Song, Cui Tao, Shuteng Niu
Abstract: Temporal Graph Neural Networks (TGNNs) have become indispensable for analyzing dynamic graphs in critical applications such as social networks, communication systems, and financial networks. However, the robustness of TGNNs against adversarial attacks, particularly sophisticated attacks that exploit the temporal dimension, remains a significant challenge. Existing attack methods for Spatio-Temporal Dynamic Graphs (STDGs) often rely on simplistic, easily detectable perturbations (e.g., random edge additions/deletions) and fail to strategically target the most influential nodes and edges for maximum impact. We introduce the High Impact Attack (HIA), a novel restricted black-box attack framework specifically designed to overcome these limitations and expose critical vulnerabilities in TGNNs. HIA leverages a data-driven surrogate model to identify structurally important nodes (central to network connectivity) and dynamically important nodes (critical for the graph's temporal evolution). It then employs a hybrid perturbation strategy, combining strategic edge injection (to create misleading connections) and targeted edge deletion (to disrupt essential pathways), maximizing TGNN performance degradation. Importantly, HIA minimizes the number of perturbations to enhance stealth, making it more challenging to detect. Comprehensive experiments on five real-world datasets and four representative TGNN architectures (TGN, JODIE, DySAT, and TGAT) demonstrate that HIA significantly reduces TGNN accuracy on the link prediction task, achieving up to a 35.55% decrease in Mean Reciprocal Rank (MRR) - a substantial improvement over state-of-the-art baselines. These results highlight fundamental vulnerabilities in current STDG models and underscore the urgent need for robust defenses that account for both structural and temporal dynamics.
Authors: Jubayer Ibn Hamid, Ifdita Hasan Orney, Ellen Xu, Chelsea Finn, Dorsa Sadigh
Abstract: Reinforcement learning fine-tuning (RLFT) is a dominant paradigm for improving pretrained policies for downstream tasks. These pretrained policies, trained on large datasets, produce generations with a broad range of promising but unrefined behaviors. Often, a critical failure mode of RLFT arises when policies lose this diversity and collapse into a handful of easily exploitable outputs. This convergence hinders exploration, which is essential for expanding the capabilities of the pretrained policy and for amplifying the benefits of test-time compute scaling. To address this, we introduce an objective for policy gradient methods that explicitly enforces the exploration and refinement of diverse generations, which we call a polychromic objective. We then show how proximal policy optimization (PPO) can be adapted to optimize this objective. Our method (1) employs vine sampling to collect on-policy rollouts and (2) modifies the advantage function to reflect the advantage under our new objective. Experiments on BabyAI, Minigrid, and Algorithmic Creativity show that our method improves success rates by reliably solving a larger set of environment configurations and generalizes better under large perturbations. Moreover, when given multiple attempts in pass@$k$ experiments, the policy achieves substantially higher coverage, demonstrating its ability to maintain and exploit a diverse repertoire of strategies.
Authors: Sreeja Apparaju, Yichuan Niu, Xixi Qi
Abstract: Budget pacing is critical in online advertising to align spend with campaign goals under dynamic auctions. Existing pacing methods often rely on ad-hoc parameter tuning, which can be unstable and inefficient. We propose a principled controller that combines bucketized hysteresis with proportional feedback to provide stable and adaptive spend control. Our method provides a framework and analysis for parameter selection that enables accurate tracking of desired spend rates across campaigns. Experiments in real-world auctions demonstrate significant improvements in pacing accuracy and delivery consistency, reducing pacing error by 13% and $\lambda$-volatility by 54% compared to baseline method. By bridging control theory with advertising systems, our approach offers a scalable and reliable solution for budget pacing, with particular benefits for small-budget campaigns.
Authors: Zhibo Hou, Zhiyu An, Wan Du
Abstract: When there exists an unlearnable source of randomness (noisy-TV) in the environment, a naively intrinsic reward driven exploring agent gets stuck at that source of randomness and fails at exploration. Intrinsic reward based on uncertainty estimation or distribution similarity, while eventually escapes noisy-TVs as time unfolds, suffers from poor sample efficiency and high computational cost. Inspired by recent findings from neuroscience that humans monitor their improvements during exploration, we propose a novel method for intrinsically-motivated exploration, named Learning Progress Monitoring (LPM). During exploration, LPM rewards model improvements instead of prediction error or novelty, effectively rewards the agent for observing learnable transitions rather than the unlearnable transitions. We introduce a dual-network design that uses an error model to predict the expected prediction error of the dynamics model in its previous iteration, and use the difference between the model errors of the current iteration and previous iteration to guide exploration. We theoretically show that the intrinsic reward of LPM is zero-equivariant and a monotone indicator of Information Gain (IG), and that the error model is necessary to achieve monotonicity correspondence with IG. We empirically compared LPM against state-of-the-art baselines in noisy environments based on MNIST, 3D maze with 160x120 RGB inputs, and Atari. Results show that LPM's intrinsic reward converges faster, explores more states in the maze experiment, and achieves higher extrinsic reward in Atari. This conceptually simple approach marks a shift-of-paradigm of noise-robust exploration. For code to reproduce our experiments, see https://github.com/Akuna23Matata/LPM_exploration
Authors: Hanyuan Gao, Xiaoxuan Yang
Abstract: Hidden Markov models (HMM) are commonly used in generation tasks and have demonstrated strong capabilities in neuro-symbolic applications for the Markov property. These applications leverage the strengths of neural networks and symbolic reasoning to create robust and interpretable AI systems. However, they may inherit and amplify the shortcomings of both approaches. Both components require dense computation and data transfer, and their communication further hinders performance. This paper proposes Norm-Q, a normalized linear quantization approach for compressing probabilistic symbolic models, such as HMMs. We reduce the bit width of the data with minimal impact, thereby alleviating memory and bandwidth stress and enabling deployment on potential custom hardware. Our method introduces a normalized quantization-aware expectation maximization process for probabilistic model training. The experimental results show that Norm-Q achieves a higher compression rate with reasonable score loss compared to traditional quantization methods. In the case of the constrained generation task of large language models, we successfully quantize an HMM of 4096 hidden states to 8 bits without loss and, at most, 3 bits with acceptable loss. Notably, the Norm-Q method can achieve a compression rate of 99% for the weights of the HMM. The code is open source at https://github.com/superstarghy/Norm-Q.
Authors: Sofiane Ennadir, Siavash Golkar, Leopoldo Sarra
Abstract: Self-supervised learning has seen great success recently in unsupervised representation learning, enabling breakthroughs in natural language and image processing. However, these methods often rely on autoregressive and masked modeling, which aim to reproduce masked information in the input, which can be vulnerable to the presence of noise or confounding variables. To address this problem, Joint-Embedding Predictive Architectures (JEPA) has been introduced with the aim to perform self-supervised learning in the latent space. To leverage these advancements in the domain of time series, we introduce Time Series JEPA (TS-JEPA), an architecture specifically adapted for time series representation learning. We validate TS-JEPA on both classification and forecasting, showing that it can match or surpass current state-of-the-art baselines on different standard datasets. Notably, our approach demonstrates a strong performance balance across diverse tasks, indicating its potential as a robust foundation for learning general representations. Thus, this work lays the groundwork for developing future time series foundation models based on Joint Embedding.
Authors: Haotian Fu, Ran Gong, Xiaohan Zhang, Maria Vittoria Minniti, Jigarkumar Patel, Karl Schmeckpeper
Abstract: Generalist robot policies that can perform many tasks typically require extensive expert data or simulations for training. In this work, we propose a novel Data-Efficient multitask DAgger framework that distills a single multitask policy from multiple task-specific expert policies. Our approach significantly increases the overall task success rate by actively focusing on tasks where the multitask policy underperforms. The core of our method is a performance-aware scheduling strategy that tracks how much each task's learning process benefits from the amount of data, using a Kalman filter-based estimator to robustly decide how to allocate additional demonstrations across tasks. We validate our approach on MetaWorld, as well as a suite of diverse drawer-opening tasks in IsaacLab. The resulting policy attains high performance across all tasks while using substantially fewer expert demonstrations, and the visual policy learned with our method in simulation shows better performance than naive DAgger and Behavior Cloning when transferring zero-shot to a real robot without using real data.
Authors: Danyang Li, Yixuan Wang, Matthew Cleaveland, Mingyu Cai, Roberto Tron
Abstract: Signal Temporal Logic (STL) inference seeks to extract human-interpretable rules from time-series data, but existing methods lack formal confidence guarantees for the inferred rules. Conformal prediction (CP) is a technique that can provide statistical correctness guarantees, but is typically applied as a post-training wrapper without improving model learning. Instead, we introduce an end-to-end differentiable CP framework for STL inference that enhances both reliability and interpretability of the resulting formulas. We introduce a robustness-based nonconformity score, embed a smooth CP layer directly into training, and employ a new loss function that simultaneously optimizes inference accuracy and CP prediction sets with a single term. Following training, an exact CP procedure delivers statistical guarantees for the learned STL formulas. Experiments on benchmark time-series tasks show that our approach reduces uncertainty in predictions (i.e., it achieves high coverage while reducing prediction set size), and improves accuracy (i.e., the number of misclassifications when using a fixed threshold) over state-of-the-art baselines.
Authors: Hui Ji, Wei Gao, Pengfei Zhou
Abstract: The 12-lead electrocardiogram (ECG) is the gold standard for cardiovascular monitoring, offering superior diagnostic granularity and specificity compared to photoplethysmography (PPG). However, existing 12-lead ECG systems rely on cumbersome multi-electrode setups, limiting sustained monitoring in ambulatory settings, while current PPG-based methods fail to reconstruct multi-lead ECG due to the absence of inter-lead constraints and insufficient modeling of spatial-temporal dependencies across leads. To bridge this gap, we introduce P2Es, an innovative demographic-aware diffusion framework designed to generate clinically valid 12-lead ECG from PPG signals via three key innovations. Specifically, in the forward process, we introduce frequency-domain blurring followed by temporal noise interference to simulate real-world signal distortions. In the reverse process, we design a temporal multi-scale generation module followed by frequency deblurring. In particular, we leverage KNN-based clustering combined with contrastive learning to assign affinity matrices for the reverse process, enabling demographic-specific ECG translation. Extensive experimental results show that P2Es outperforms baseline models in 12-lead ECG reconstruction.
Authors: Dingyi Kang, Dongming Jiang, Hanshen Yang, Hang Liu, Bingzhe Li
Abstract: Approximate Nearest Neighbor Search (ANNS), as the core of vector databases (VectorDBs), has become widely used in modern AI and ML systems, powering applications from information retrieval to bio-informatics. While graph-based ANNS methods achieve high query efficiency, their scalability is constrained by the available host memory. Recent disk-based ANNS approaches mitigate memory usage by offloading data to Solid-State Drives (SSDs). However, they still suffer from issues such as long I/O traversal path, misalignment with storage I/O granularity, and high in-memory indexing overhead, leading to significant I/O latency and ultimately limiting scalability for large-scale vector search. In this paper, we propose PageANN, a disk-based approximate nearest neighbor search (ANNS) framework designed for high performance and scalability. PageANN introduces a page-node graph structure that aligns logical graph nodes with physical SSD pages, thereby shortening I/O traversal paths and reducing I/O operations. Specifically, similar vectors are clustered into page nodes, and a co-designed disk data layout leverages this structure with a merging technique to store only representative vectors and topology information, avoiding unnecessary reads. To further improve efficiency, we design a memory management strategy that combines lightweight indexing with coordinated memory-disk data allocation, maximizing host memory utilization while minimizing query latency and storage overhead. Experimental results show that PageANN significantly outperforms state-of-the-art (SOTA) disk-based ANNS methods, achieving 1.85x-10.83x higher throughput and 51.7%-91.9% lower latency across different datasets and memory budgets, while maintaining comparable high recall accuracy.
Authors: Langzhou He, Junyou Zhu, Fangxin Wang, Junhua Liu, Haoyan Xu, Yue Zhao, Philip S. Yu, Qitian Wu
Abstract: Molecular foundation models are rapidly advancing scientific discovery, but their unreliability on out-of-distribution (OOD) samples severely limits their application in high-stakes domains such as drug discovery and protein design. A critical failure mode is chemical hallucination, where models make high-confidence yet entirely incorrect predictions for unknown molecules. To address this challenge, we introduce Molecular Preference-Aligned Instance Ranking (Mole-PAIR), a simple, plug-and-play module that can be flexibly integrated with existing foundation models to improve their reliability on OOD data through cost-effective post-training. Specifically, our method formulates the OOD detection problem as a preference optimization over the estimated OOD affinity between in-distribution (ID) and OOD samples, achieving this goal through a pairwise learning objective. We show that this objective essentially optimizes AUROC, which measures how consistently ID and OOD samples are ranked by the model. Extensive experiments across five real-world molecular datasets demonstrate that our approach significantly improves the OOD detection capabilities of existing molecular foundation models, achieving up to 45.8%, 43.9%, and 24.3% improvements in AUROC under distribution shifts of size, scaffold, and assay, respectively.
Authors: Chang Liu, Danial Chitnis
Abstract: The design of Analog and Mixed-Signal (AMS) integrated circuits (ICs) often involves significant manual effort, especially during the transistor sizing process. While Machine Learning techniques in Electronic Design Automation (EDA) have shown promise in reducing complexity and minimizing human intervention, they still face challenges such as numerous iterations and a lack of knowledge about AMS circuit design. Recently, Large Language Models (LLMs) have demonstrated significant potential across various fields, showing a certain level of knowledge in circuit design and indicating their potential to automate the transistor sizing process. In this work, we propose EEsizer, an LLM-based AI agent that integrates large language models with circuit simulators and custom data analysis functions, enabling fully automated, closed-loop transistor sizing without relying on external knowledge. By employing prompt engineering and Chain-of-Thought reasoning, the agent iteratively explores design directions, evaluates performance, and refines solutions with minimal human intervention. We first benchmarked 8 LLMs on six basic circuits and selected three high-performing models to optimize a 20-transistor CMOS operational amplifier, targeting multiple performance metrics, including rail-to-rail operation from 180 nm to 90 nm technology nodes. Notably, OpenAI o3 successfully achieved the user-intended target at 90 nm across three different test groups, with a maximum of 20 iterations, demonstrating adaptability and robustness at advanced nodes. To assess design robustness, we manually designed a bias circuit and performed a variation analysis using Gaussian-distributed variations on transistor dimensions and threshold voltages.
Authors: Harry Robertshaw, Han-Ru Wu, Alejandro Granados, Thomas C Booth
Abstract: Autonomous navigation for mechanical thrombectomy (MT) remains a critical challenge due to the complexity of vascular anatomy and the need for precise, real-time decision-making. Reinforcement learning (RL)-based approaches have demonstrated potential in automating endovascular navigation, but current methods often struggle with generalization across multiple patient vasculatures and long-horizon tasks. We propose a world model for autonomous endovascular navigation using TD-MPC2, a model-based RL algorithm. We trained a single RL agent across multiple endovascular navigation tasks in ten real patient vasculatures, comparing performance against the state-of-the-art Soft Actor-Critic (SAC) method. Results indicate that TD-MPC2 significantly outperforms SAC in multi-task learning, achieving a 65% mean success rate compared to SAC's 37%, with notable improvements in path ratio. TD-MPC2 exhibited increased procedure times, suggesting a trade-off between success rate and execution speed. These findings highlight the potential of world models for improving autonomous endovascular navigation and lay the foundation for future research in generalizable AI-driven robotic interventions.
Authors: Alireza Mousavi-Hosseini, Stephen Y. Zhang, Michal Klein, Marco Cuturi
Abstract: Flow models parameterized as time-dependent velocity fields can generate data from noise by integrating an ODE. These models are often trained using flow matching, i.e. by sampling random pairs of noise and target points $(\mathbf{x}_0,\mathbf{x}_1)$ and ensuring that the velocity field is aligned, on average, with $\mathbf{x}_1-\mathbf{x}_0$ when evaluated along a segment linking $\mathbf{x}_0$ to $\mathbf{x}_1$. While these pairs are sampled independently by default, they can also be selected more carefully by matching batches of $n$ noise to $n$ target points using an optimal transport (OT) solver. Although promising in theory, the OT flow matching (OT-FM) approach is not widely used in practice. Zhang et al. (2025) pointed out recently that OT-FM truly starts paying off when the batch size $n$ grows significantly, which only a multi-GPU implementation of the Sinkhorn algorithm can handle. Unfortunately, the costs of running Sinkhorn can quickly balloon, requiring $O(n^2/\varepsilon^2)$ operations for every $n$ pairs used to fit the velocity field, where $\varepsilon$ is a regularization parameter that should be typically small to yield better results. To fulfill the theoretical promises of OT-FM, we propose to move away from batch-OT and rely instead on a semidiscrete formulation that leverages the fact that the target dataset distribution is usually of finite size $N$. The SD-OT problem is solved by estimating a dual potential vector using SGD; using that vector, freshly sampled noise vectors at train time can then be matched with data points at the cost of a maximum inner product search (MIPS). Semidiscrete FM (SD-FM) removes the quadratic dependency on $n/\varepsilon$ that bottlenecks OT-FM. SD-FM beats both FM and OT-FM on all training metrics and inference budget constraints, across multiple datasets, on unconditional/conditional generation, or when using mean-flow models.
Authors: Yichi Zhang, Fangzheng Xie, Shu Yang, Chong Wu
Abstract: In language tasks that require extensive human--model interaction, deploying a single "best" model for every query can be expensive. To reduce inference cost while preserving the quality of the responses, a large language model (LLM) router selects the most appropriate model from a pool of candidates for each query. A central challenge to training a high-quality router is the scarcity of reliable supervision. Gold-standard data (e.g., expert-verified labels or rubric-based scores) provide accurate quality evaluations of LLM responses but are costly and difficult to scale. In contrast, preference-based data, collected via crowdsourcing or LLM-as-a-judge systems, are cheaper and more scalable, yet often biased in reflecting the true quality of responses. We cast the problem of LLM router training with combined gold-standard and preference-based data into a causal inference framework by viewing the response evaluation mechanism as the treatment assignment. This perspective further reveals that the bias in preference-based data corresponds to the well-known causal estimand: the conditional average treatment effect. Based on this new perspective, we develop an integrative causal router training framework that corrects preference-data bias, address imbalances between two data sources, and improve routing robustness and efficiency. Numerical experiments demonstrate that our approach delivers more accurate routing and improves the trade-off between cost and quality.
Authors: Marcus Schwarting, Logan Ward, Nathaniel Hudson, Xiaoli Yan, Ben Blaiszik, Santanu Chaudhuri, Eliu Huerta, Ian Foster
Abstract: Generative AI poses both opportunities and risks for solving inverse design problems in the sciences. Generative tools provide the ability to expand and refine a search space autonomously, but do so at the cost of exploring low-quality regions until sufficiently fine tuned. Here, we propose a queue prioritization algorithm that combines generative modeling and active learning in the context of a distributed workflow for exploring complex design spaces. We find that incorporating an active learning model to prioritize top design candidates can prevent a generative AI workflow from expending resources on nonsensical candidates and halt potential generative model decay. For an existing generative AI workflow for discovering novel molecular structure candidates for carbon capture, our active learning approach significantly increases the number of high-quality candidates identified by the generative model. We find that, out of 1000 novel candidates, our workflow without active learning can generate an average of 281 high-performing candidates, while our proposed prioritization with active learning can generate an average 604 high-performing candidates.
Authors: Guojun Tang, Jiayu Zhou, Mohammad Mamun, Steve Drew
Abstract: Federated learning (FL) faces persistent robustness challenges due to non-IID data distributions and adversarial client behavior. A promising mitigation strategy is contribution evaluation, which enables adaptive aggregation by quantifying each client's utility to the global model. However, state-of-the-art Shapley-value-based approaches incur high computational overhead due to repeated model reweighting and inference, which limits their scalability. We propose FedIF, a novel FL aggregation framework that leverages trajectory-based influence estimation to efficiently compute client contributions. FedIF adapts decentralized FL by introducing normalized and smoothed influence scores computed from lightweight gradient operations on client updates and a public validation set. Theoretical analysis demonstrates that FedIF yields a tighter bound on one-step global loss change under noisy conditions. Extensive experiments on CIFAR-10 and Fashion-MNIST show that FedIF achieves robustness comparable to or exceeding SV-based methods in the presence of label noise, gradient noise, and adversarial samples, while reducing aggregation overhead by up to 450x. Ablation studies confirm the effectiveness of FedIF's design choices, including local weight normalization and influence smoothing. Our results establish FedIF as a practical, theoretically grounded, and scalable alternative to Shapley-value-based approaches for efficient and robust FL in real-world deployments.
Authors: Amir Moeini, Minjae Kwon, Alper Kamil Bozkurt, Yuichi Motai, Rohan Chandra, Lu Feng, Shangtong Zhang
Abstract: In-context reinforcement learning (ICRL) is an emerging RL paradigm where the agent, after some pretraining procedure, is able to adapt to out-of-distribution test tasks without any parameter updates. The agent achieves this by continually expanding the input (i.e., the context) to its policy neural networks. For example, the input could be all the history experience that the agent has access to until the current time step. The agent's performance improves as the input grows, without any parameter updates. In this work, we propose the first method that promotes the safety of ICRL's adaptation process in the framework of constrained Markov Decision Processes. In other words, during the parameter-update-free adaptation process, the agent not only maximizes the reward but also minimizes an additional cost function. We also demonstrate that our agent actively reacts to the threshold (i.e., budget) of the cost tolerance. With a higher cost budget, the agent behaves more aggressively, and with a lower cost budget, the agent behaves more conservatively.
Authors: Morteza Kimiaei, Vyacheslav Kungurtsev
Abstract: Black-box optimization (BBO) addresses problems where objectives are accessible only through costly queries without gradients or explicit structure. Classical derivative-free methods -- line search, direct search, and model-based solvers such as Bayesian optimization -- form the backbone of BBO, yet often struggle in high-dimensional, noisy, or mixed-integer settings. Recent advances use machine learning (ML) and reinforcement learning (RL) to enhance BBO: ML provides expressive surrogates, adaptive updates, meta-learning portfolios, and generative models, while RL enables dynamic operator configuration, robustness, and meta-optimization across tasks. This paper surveys these developments, covering representative algorithms such as NNs with the modular model-based optimization framework (mlrMBO), zeroth-order adaptive momentum methods (ZO-AdaMM), automated BBO (ABBO), distributed block-wise optimization (DiBB), partition-based Bayesian optimization (SPBOpt), the transformer-based optimizer (B2Opt), diffusion-model-based BBO, surrogate-assisted RL for differential evolution (Surr-RLDE), robust BBO (RBO), coordinate-ascent model-based optimization with relative entropy (CAS-MORE), log-barrier stochastic gradient descent (LB-SGD), policy improvement with black-box (PIBB), and offline Q-learning with Mamba backbones (Q-Mamba). We also review benchmark efforts such as the NeurIPS 2020 BBO Challenge and the MetaBox framework. Overall, we highlight how ML and RL transform classical inexact solvers into more scalable, robust, and adaptive frameworks for real-world optimization.
Authors: Lucia Quirke, Stepan Shabalin, Nora Belrose
Abstract: Sparse autoencoders (SAEs) are used to decompose neural network activations into sparsely activating features, but many SAE features are only interpretable at high activation strengths. To address this issue we propose to use binary sparse autoencoders (BAEs) and binary transcoders (BTCs), which constrain all activations to be zero or one. We find that binarisation significantly improves the interpretability and monosemanticity of the discovered features, while increasing reconstruction error. By eliminating the distinction between high and low activation strengths, we prevent uninterpretable information from being smuggled in through the continuous variation in feature activations. However, we also find that binarisation increases the number of uninterpretable ultra-high frequency features, and when interpretability scores are frequency-adjusted, the scores for continuous sparse coders are slightly better than those of binary ones. This suggests that polysemanticity may be an ineliminable property of neural activations.
Authors: Yixuan Wang, Dan Guralnik, Saiedeh Akbari, Warren Dixon
Abstract: We introduce Effective Model Pruning (EMP), a context-agnostic, parameter-free rule addressing a fundamental question about pruning: how many entries to keep. EMP does not prescribe how to score the parameters or prune the models; instead, it supplies a universal adaptive threshold that can be applied to any pruning criterion: weight magnitude, attention score, KAN importance score, or even feature-level signals such as image pixel, and used on structural parts or weights of the models. Given any score vector s, EMP maps s to a built-in effective number N_eff which is inspired by the Inverse Simpson index of contributors. Retaining the N_eff highest scoring entries and zeroing the remainder yields sparse models with performance comparable to the original dense networks across MLPs, CNNs, Transformers/LLMs, and KAN, in our experiments. By leveraging the geometry of the simplex, we derive a tight lower bound on the preserved mass s_eff (the sum of retained scores) over the corresponding ordered probability simplex associated with the score vector s. We further verify the effectiveness of N_eff by pruning the model with a scaled threshold \b{eta}*N_eff across a variety of criteria and models. Experiments suggest that the default \b{eta} = 1 yields a robust threshold for model pruning while \b{eta} not equal to 1 still serves as an optional adjustment to meet specific sparsity requirements.
Authors: Muhammad Imran Hossain, Jignesh Solanki, Sarika Khushlani Solanki
Abstract: Ensuring power grid resilience requires the timely and unsupervised detection of anomalies in synchrophasor data streams. We introduce T-BiGAN, a novel framework that integrates window-attention Transformers within a bidirectional Generative Adversarial Network (BiGAN) to address this challenge. Its self-attention encoder-decoder architecture captures complex spatio-temporal dependencies across the grid, while a joint discriminator enforces cycle consistency to align the learned latent space with the true data distribution. Anomalies are flagged in real-time using an adaptive score that combines reconstruction error, latent space drift, and discriminator confidence. Evaluated on a realistic hardware-in-the-loop PMU benchmark, T-BiGAN achieves an ROC-AUC of 0.95 and an average precision of 0.996, significantly outperforming leading supervised and unsupervised methods. It shows particular strength in detecting subtle frequency and voltage deviations, demonstrating its practical value for live, wide-area monitoring without relying on manually labeled fault data.
Authors: Zhendong Mi, Bian Sun, Grace Li Zhang, Shaoyi Huang
Abstract: Large language models (LLMs) have rapidly scaled in size, bringing severe memory and computational challenges that hinder their deployment. Singular Value Decomposition (SVD)-based compression has emerged as an appealing post-training compression technique for LLMs, yet most existing methods apply a uniform compression ratio across all layers, implicitly assuming homogeneous information included in various layers. This overlooks the substantial intra-layer heterogeneity observed in LLMs, where middle layers tend to encode richer information while early and late layers are more redundant. In this work, we revisit the existing SVD-based compression method and propose D-Rank, a framework with layer-wise balanced Dynamic Rank allocation for LLMs compression. We first introduce effective rank as a principled metric to measure the information density of weight matrices, and then allocate ranks via a Lagrange multiplier-based optimization scheme to adaptively assign more capacity to groups with higher information density under a fixed compression ratio. Moreover, we rebalance the allocated ranks across attention layers to account for their varying importance and extend D-Rank to latest LLMs with grouped-query attention. Extensive experiments on various LLMs with different scales across multiple compression ratios demonstrate that D-Rank consistently outperforms SVD-LLM, ASVD, and Basis Sharing, achieving more than 15 lower perplexity with LLaMA-3-8B model on C4 datasets at 20% compression ratio and up to 5% higher zero-shot reasoning accuracy with LLaMA-7B model at 40% compression ratio while achieving even higher throughput.
Authors: Jason Stock, Troy Arcomano, Rao Kotamarthi
Abstract: Diffusion models offer a physically grounded framework for probabilistic weather forecasting, but their typical reliance on slow, iterative solvers during inference makes them impractical for subseasonal-to-seasonal (S2S) applications where long lead-times and domain-driven calibration are essential. To address this, we introduce Swift, a single-step consistency model that, for the first time, enables autoregressive finetuning of a probability flow model with a continuous ranked probability score (CRPS) objective. This eliminates the need for multi-model ensembling or parameter perturbations. Results show that Swift produces skillful 6-hourly forecasts that remain stable for up to 75 days, running $39\times$ faster than state-of-the-art diffusion baselines while achieving forecast skill competitive with the numerical-based, operational IFS ENS. This marks a step toward efficient and reliable ensemble forecasting from medium-range to seasonal-scales.
Authors: Kotaro Yoshida, Atsushi Nitanda
Abstract: Preconditioning is widely used in machine learning to accelerate convergence on the empirical risk, yet its role on the expected risk remains underexplored. In this work, we investigate how preconditioning affects feature learning and generalization performance. We first show that the input information available to the model is conveyed solely through the Gram matrix defined by the preconditioner's metric, thereby inducing a controllable spectral bias on feature learning. Concretely, instantiating the preconditioner as the $p$-th power of the input covariance matrix and within a single-index teacher model, we prove that in generalization, the exponent $p$ and the alignment between the teacher and the input spectrum are crucial factors. We further investigate how the interplay between these factors influences feature learning from three complementary perspectives: (i) Robustness to noise, (ii) Out-of-distribution generalization, and (iii) Forward knowledge transfer. Our results indicate that the learned feature representations closely mirror the spectral bias introduced by the preconditioner -- favoring components that are emphasized and exhibiting reduced sensitivity to those that are suppressed. Crucially, we demonstrate that generalization is significantly enhanced when this spectral bias is aligned with that of the teacher.
Authors: Lei Ma, Ling Guo, Hao Wu, Tao Zhou
Abstract: Learning operators from data is central to scientific machine learning. While DeepONets are widely used for their ability to handle complex domains, they require fixed sensor numbers and locations, lack mechanisms for uncertainty quantification (UQ), and are thus limited in practical applicability. Recent permutationinvariant extensions, such as the Variable-Input Deep Operator Network (VIDON), relax these sensor constraints but still rely on sufficiently dense observations and cannot capture uncertainties arising from incomplete measurements or from operators with inherent randomness. To address these challenges, we propose UQ-SONet, a permutation-invariant operator learning framework with built-in UQ. Our model integrates a set transformer embedding to handle sparse and variable sensor locations, and employs a conditional variational autoencoder (cVAE) to approximate the conditional distribution of the solution operator. By minimizing the negative ELBO, UQ-SONet provides principled uncertainty estimation while maintaining predictive accuracy. Numerical experiments on deterministic and stochastic PDEs, including the Navier-Stokes equation, demonstrate the robustness and effectiveness of the proposed framework.
Authors: Fangji Wang, Panagiotis Tsiotras
Abstract: Branch-and-bound with preactivation splitting has been shown highly effective for deterministic verification of neural networks. In this paper, we extend this framework to the probabilistic setting. We propose BaB-prob that iteratively divides the original problem into subproblems by splitting preactivations and leverages linear bounds computed by linear bound propagation to bound the probability for each subproblem. We prove soundness and completeness of BaB-prob for feedforward-ReLU neural networks. Furthermore, we introduce the notion of uncertainty level and design two efficient strategies for preactivation splitting, yielding BaB-prob-ordered and BaB+BaBSR-prob. We evaluate BaB-prob on untrained networks, MNIST and CIFAR-10 models, respectively, and VNN-COMP 2025 benchmarks. Across these settings, our approach consistently outperforms state-of-the-art approaches in medium- to high-dimensional input problems.
Authors: Qihang Yao, Constantine Dovrolis
Abstract: The lottery ticket hypothesis suggests that dense networks contain sparse subnetworks that can be trained in isolation to match full-model performance. Existing approaches-iterative pruning, dynamic sparse training, and pruning at initialization-either incur heavy retraining costs or assume the target density is fixed in advance. We introduce Path Weight Magnitude Product-biased Random growth (PWMPR), a constructive sparse-to-dense training paradigm that grows networks rather than pruning them, while automatically discovering their operating density. Starting from a sparse seed, PWMPR adds edges guided by path-kernel-inspired scores, mitigates bottlenecks via randomization, and stops when a logistic-fit rule detects plateauing accuracy. Experiments on CIFAR, TinyImageNet, and ImageNet show that PWMPR approaches the performance of IMP-derived lottery tickets-though at higher density-at substantially lower cost (~1.5x dense vs. 3-4x for IMP). These results establish growth-based density discovery as a promising paradigm that complements pruning and dynamic sparsity.
Authors: Justin Chih-Yao Chen, Becky Xiangyu Peng, Prafulla Kumar Choubey, Kung-Hsiang Huang, Jiaxin Zhang, Mohit Bansal, Chien-Sheng Wu
Abstract: Current online reinforcement learning (RL) algorithms like GRPO share a key limitation in LLM reasoning: they cannot learn from problems that are "unsolvable" to the model. In other words, they can only improve performance on problems where the model is capable of exploring the correct answer. Consequently, the model's "upper limit" remains unchanged after RL training, even though the likelihood of solving easier, solvable problems may increase. These hard samples cannot contribute to training, as no rollouts yield rewards and thus no gradients are produced. To unlock learning from these hard samples, we propose NuRL, a "nudging" method that aims to push the upper bound of LLM reasoning using self-generated hints, i.e., abstract cues that help reduce the problem difficulty for the model. Given a question and its gold answer, the model generates a CoT and then produces a hint containing the core knowledge needed to solve the problem. During training, we generate G rollouts from the base policy and use the pass rate to decide whether the hint should be injected. For hard samples with a 0% pass rate, we inject the hint and regenerate a new batch of trajectories. This yields two benefits: (1) the hint boosts pass rates (from 0% to non-zero), thereby introducing training signals for previously unsolvable samples, and (2) the hints are self-generated, avoiding distributional shift and do not rely on external models. NuRL achieves consistent improvements across 6 benchmarks and 3 models, while remaining complementary to test-time scaling. Notably, NuRL can raise the model's upper limit, whereas GRPO leaves pass@1024 unchanged from the base model. Furthermore, we present a systematic study of what makes an effective hint and when hints are most useful. Interestingly, the best hints are abstract and high-level, and are most beneficial when applied necessarily and after GRPO has converged.
Authors: Bipul Thapa, Biplov Paneru, Bishwash Paneru, Khem Narayan Poudyal
Abstract: This paper presents an Artificial Intelligence (AI) integrated novel approach to Brain-Computer Interface (BCI)-based wheelchair development, utilizing a motor imagery right-left-hand movement mechanism for control. The system is designed to simulate wheelchair navigation based on motor imagery right and left-hand movements using electroencephalogram (EEG) data. A pre-filtered dataset, obtained from an open-source EEG repository, was segmented into arrays of 19x200 to capture the onset of hand movements. The data was acquired at a sampling frequency of 200Hz. The system integrates a Tkinter-based interface for simulating wheelchair movements, offering users a functional and intuitive control system. We propose a BiLSTM-BiGRU model that shows a superior test accuracy of 92.26% as compared with various machine learning baseline models, including XGBoost, EEGNet, and a transformer-based model. The Bi-LSTM-BiGRU attention-based model achieved a mean accuracy of 90.13% through cross-validation, showcasing the potential of attention mechanisms in BCI applications.
Authors: Xing Han, Hsing-Huan Chung, Joydeep Ghosh, Paul Pu Liang, Suchi Saria
Abstract: Mixture-of-Experts (MoE) architectures have become pivotal for large-scale multimodal models. However, their routing mechanisms typically overlook the informative, time-varying interaction dynamics between modalities. This limitation hinders expert specialization, as the model cannot explicitly leverage intrinsic modality relationships for effective reasoning. To address this, we propose a novel framework that guides MoE routing using quantified temporal interaction. A multimodal interaction-aware router learns to dispatch tokens to experts based on the nature of their interactions. This dynamic routing encourages experts to acquire generalizable interaction-processing skills rather than merely learning task-specific features. Our framework builds on a new formulation of temporal multimodal interaction dynamics, which are used to guide expert routing. We first demonstrate that these temporal multimodal interactions reveal meaningful patterns across applications, and then show how they can be leveraged to improve both the design and performance of MoE-based models. Comprehensive experiments on challenging multimodal benchmarks validate our approach, demonstrating both enhanced performance and improved interpretability.
Authors: Pirzada Suhail, Aditya Anand, Amit Sethi
Abstract: Machine learning models, by virtue of training, learn a large repertoire of decision rules for any given input, and any one of these may suffice to justify a prediction. However, in high-dimensional input spaces, such rules are difficult to identify and interpret. In this paper, we introduce an activation-matching based approach to generate minimal and faithful explanations for the decisions of pre-trained image classifiers. We aim to identify minimal explanations that not only preserve the model's decision but are also concise and human-readable. To achieve this, we train a lightweight autoencoder to produce binary masks that learns to highlight the decision-wise critical regions of an image while discarding irrelevant background. The training objective integrates activation alignment across multiple layers, consistency at the output label, priors that encourage sparsity, and compactness, along with a robustness constraint that enforces faithfulness. The minimal explanations so generated also lead us to mechanistically interpreting the model internals. In this regard we also introduce a circuit readout procedure wherein using the explanation's forward pass and gradients, we identify active channels and construct a channel-level graph, scoring inter-layer edges by ingress weight magnitude times source activation and feature-to-class links by classifier weight magnitude times feature activation. Together, these contributions provide a practical bridge between minimal input-level explanations and a mechanistic understanding of the internal computations driving model decisions.
Authors: Zihui Zhao, Yuanbo Tang, Jieyu Ren, Xiaoping Zhang, Yang Li
Abstract: Dictionary learning is traditionally formulated as an $L_1$-regularized signal reconstruction problem. While recent developments have incorporated discriminative, hierarchical, or generative structures, most approaches rely on encouraging representation sparsity over individual samples that overlook how atoms are shared across samples, resulting in redundant and sub-optimal dictionaries. We introduce a parsimony promoting regularizer based on the row-wise $L_\infty$ norm of the coefficient matrix. This additional penalty encourages entire rows of the coefficient matrix to vanish, thereby reducing the number of dictionary atoms activated across the dataset. We derive the formulation from a probabilistic model with Beta-Bernoulli priors, which provides a Bayesian interpretation linking the regularization parameters to prior distributions. We further establish theoretical calculation for optimal hyperparameter selection and connect our formulation to both Minimum Description Length, Bayesian model selection and pathlet learning. Extensive experiments on benchmark datasets demonstrate that our method achieves substantially improved reconstruction quality (with a 20\% reduction in RMSE) and enhanced representation sparsity, utilizing fewer than one-tenth of the available dictionary atoms, while empirically validating our theoretical analysis.
Authors: Tingyu Shi, Fan Lyu, Shaoliang Peng
Abstract: Active Test-Time Adaptation (ATTA) improves model robustness under domain shift by selectively querying human annotations at deployment, but existing methods use heuristic uncertainty measures and suffer from low data selection efficiency, wasting human annotation budget. We propose Conformal Prediction Active TTA (CPATTA), which first brings principled, coverage-guaranteed uncertainty into ATTA. CPATTA employs smoothed conformal scores with a top-K certainty measure, an online weight-update algorithm driven by pseudo coverage, a domain-shift detector that adapts human supervision, and a staged update scheme balances human-labeled and model-labeled data. Extensive experiments demonstrate that CPATTA consistently outperforms the state-of-the-art ATTA methods by around 5% in accuracy. Our code and datasets are available at https://github.com/tingyushi/CPATTA.
Authors: Takuya Fujimura, Kota Dohi, Natsuo Yamashita, Yohei Kawaguchi
Abstract: Time-series question answering (TSQA) tasks face significant challenges due to the lack of labeled data. Alternatively, with recent advancements in large-scale models, vision-language models (VLMs) have demonstrated the potential to analyze time-series signals in a zero-shot manner. In this paper, we propose a training approach that uses pseudo labels generated by a VLM. Although VLMs can produce incorrect labels, TSQA models can still be effectively trained based on the property that deep neural networks are inherently robust to such noisy labels. Our experimental results demonstrate that TSQA models are not only successfully trained with pseudo labels, but also surpass the performance of the VLM itself by leveraging a large amount of unlabeled data.
Authors: Cheng Guo, Giuseppe L'Erario, Giulio Romualdi, Mattia Leonori, Marta Lorenzini, Arash Ajoudani, Daniele Pucci
Abstract: Accurate and physically feasible human motion prediction is crucial for safe and seamless human-robot collaboration. While recent advancements in human motion capture enable real-time pose estimation, the practical value of many existing approaches is limited by the lack of fu- ture predictions and consideration of physical constraints. Conventional motion prediction schemes rely heavily on past poses, which are not always available in real-world scenarios. To address these limitations, we present a physics-informed learning framework that integrates domain knowledge into both training and inference to predict human motion using inertial measurements from only 5 IMUs. We propose a network that accounts for the spatial characteristics of human movements. During training, we incorporate forward and differential kinematics functions as additional loss components to regularize the learned joint predictions. At the inference stage, we refine the prediction from the previous iteration to update a joint state buffer, which is used as extra inputs to the network. Experimental results demonstrate that our approach achieves high accuracy, smooth transitions between motions, and generalizes well to unseen subjects
Authors: Rostyslav Olshevskyi, Madeline Navarro, Santiago Segarra
Abstract: We propose an adaptive graph coarsening method to jointly learn graph neural network (GNN) parameters and merge nodes via K-means clustering during training. As real-world graphs grow larger, processing them directly becomes increasingly challenging and sometimes infeasible. Tailoring algorithms to large-scale data may sacrifice performance, so we instead consider graph reduction to decrease the amount of data used during training. In particular, we propose a method to simultaneously train a GNN and coarsen its graph by partitioning nodes via K-means clustering based on their embeddings. Unlike past graph coarsening works, our approach allows us to merge nodes during training. Not only does this preclude coarsening as a preprocessing step, but our node clusters can adapt to the learning task instead of relying solely on graph connectivity and features. Thus, our method is amenable to scenarios that are challenging for other methods, such as heterophilic data. We validate our approach on both homophilic and heterophilic node classification datasets. We further visualize relationships between node embeddings and their corresponding clusters to illustrate that our coarsened graph adapts to the learning task during training.
Authors: Dengming Zhang, Xiaowen Ma, Zhenliang Ni, Zhenkai Wu, Han Shu, Xin Jiang, Xinghao Chen
Abstract: Model merging, which combines multiple domain-specialized experts into a single model, offers a practical path to endow Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) with broad capabilities without the cost of joint training or serving many models. However, training-free methods rely on hand-tuned coefficients, whereas training-based methods primarily align parameters rather than downstream task behavior and typically treat all layers uniformly, ignoring inter-layer heterogeneity. We introduce Expert Merging, a training-light method that learns a small set of layer-wise coefficients using only unlabeled calibration data. The coefficients are optimized to explicitly align the merged model's hidden states and logits with those of the corresponding experts, with a coefficient regularizer for stability and task-weighted losses for controllable trade-offs. To capture inter-layer variation, Expert Merging++ augments this design with importance-guided chunking: a normalized layer-importance metric, derived from learned coefficients, task-vector magnitudes, and parameter counts, allocates more chunk-wise coefficients to high-importance layers while keeping low-importance layers lightweight. The result is a label-free, parameter-efficient, and scalable approach to multi-expert model merging across LLMs and MLLMs. Across MLLM backbones (InternVL and Qwen2-VL) and the LLM backbone (Mistral), our method surpasses strong training-free and training-based merging baselines, with Expert Merging++ delivering further gains and, in some cases, even exceeding supervised Mixture Training. The source code is available at https://github.com/Littleor/ExpertMerging.
Authors: Hyunsoo Song, Minjung Gim, Jaewoong Choi
Abstract: Flow matching has recently emerged as a powerful framework for continuous-time generative modeling. However, when applied to long-tailed distributions, standard flow matching suffers from majority bias, producing minority modes with low fidelity and failing to match the true class proportions. In this work, we propose Unbalanced Optimal Transport Reweighted Flow Matching (UOT-RFM), a novel framework for generative modeling under class-imbalanced (long-tailed) distributions that operates without any class label information. Our method constructs the conditional vector field using mini-batch Unbalanced Optimal Transport (UOT) and mitigates majority bias through a principled inverse reweighting strategy. The reweighting relies on a label-free majority score, defined as the density ratio between the target distribution and the UOT marginal. This score quantifies the degree of majority based on the geometric structure of the data, without requiring class labels. By incorporating this score into the training objective, UOT-RFM theoretically recovers the target distribution with first-order correction ($k=1$) and empirically improves tail-class generation through higher-order corrections ($k > 1$). Our model outperforms existing flow matching baselines on long-tailed benchmarks, while maintaining competitive performance on balanced datasets.
Authors: Hanghui Guo, Shimin Di, Pasquale De Meo, Zhangze Chen, Jia Zhu
Abstract: As a critical task in data quality control, claim verification aims to curb the spread of misinformation by assessing the truthfulness of claims based on a wide range of evidence. However, traditional methods often overlook the complex interactions between evidence, leading to unreliable verification results. A straightforward solution represents the claim and evidence as a fully connected graph, which we define as the Claim-Evidence Graph (C-E Graph). Nevertheless, claim verification methods based on fully connected graphs face two primary confounding challenges, Data Noise and Data Biases. To address these challenges, we propose a novel framework, Multi-Path Causal Optimization (MuPlon). MuPlon integrates a dual causal intervention strategy, consisting of the back-door path and front-door path. In the back-door path, MuPlon dilutes noisy node interference by optimizing node probability weights, while simultaneously strengthening the connections between relevant evidence nodes. In the front-door path, MuPlon extracts highly relevant subgraphs and constructs reasoning paths, further applying counterfactual reasoning to eliminate data biases within these paths. The experimental results demonstrate that MuPlon outperforms existing methods and achieves state-of-the-art performance.
Authors: Haozhe Lei, Hao Guo, Tommy Svensson, Sundeep Rangan
Abstract: Modern wireless systems require not only position estimates, but also quantified uncertainty to support planning, control, and radio resource management. We formulate localization as posterior inference of an unknown transmitter location from receiver measurements. We propose Monte Carlo Candidate-Likelihood Estimation (MC-CLE), which trains a neural scoring network using Monte Carlo sampling to compare true and candidate transmitter locations. We show that in line-of-sight simulations with a multi-antenna receiver, MC-CLE learns critical properties including angular ambiguity and front-to-back antenna patterns. MC-CLE also achieves lower cross-entropy loss relative to a uniform baseline and Gaussian posteriors. alternatives under a uniform-loss metric.
Authors: Huikang Su, Dengyun Peng, Zifeng Zhuang, YuHan Liu, Qiguang Chen, Donglin Wang, Qinghe Liu
Abstract: Offline safe reinforcement learning aims to learn policies that satisfy predefined safety constraints from static datasets. Existing sequence-model-based methods condition action generation on symmetric input tokens for return-to-go and cost-to-go, neglecting their intrinsic asymmetry: return-to-go (RTG) serves as a flexible performance target, while cost-to-go (CTG) should represent a rigid safety boundary. This symmetric conditioning leads to unreliable constraint satisfaction, especially when encountering out-of-distribution cost trajectories. To address this, we propose Boundary-to-Region (B2R), a framework that enables asymmetric conditioning through cost signal realignment . B2R redefines CTG as a boundary constraint under a fixed safety budget, unifying the cost distribution of all feasible trajectories while preserving reward structures. Combined with rotary positional embeddings , it enhances exploration within the safe region. Experimental results show that B2R satisfies safety constraints in 35 out of 38 safety-critical tasks while achieving superior reward performance over baseline methods. This work highlights the limitations of symmetric token conditioning and establishes a new theoretical and practical approach for applying sequence models to safe RL. Our code is available at https://github.com/HuikangSu/B2R.
Authors: Indu Kant Deo, Akash Venkateshwaran, Rajeev K. Jaiman
Abstract: Ship traffic is an increasing source of underwater radiated noise in coastal waters, motivating real-time digital twins of ocean acoustics for operational noise mitigation. We present a physics-guided probabilistic framework to predict three-dimensional transmission loss in realistic ocean environments. As a case study, we consider the Salish Sea along shipping routes from the Pacific Ocean to the Port of Vancouver. A dataset of over 30 million source-receiver pairs was generated with a Gaussian beam solver across seasonal sound speed profiles and one-third-octave frequency bands spanning 12.5 Hz to 8 kHz. We first assess sparse variational Gaussian processes (SVGP) and then incorporate physics-based mean functions combining spherical spreading with frequency-dependent absorption. To capture nonlinear effects, we examine deep sigma-point processes and stochastic variational deep kernel learning. The final framework integrates four components: (i) a learnable physics-informed mean that represents dominant propagation trends, (ii) a convolutional encoder for bathymetry along the source-receiver track, (iii) a neural encoder for source, receiver, and frequency coordinates, and (iv) a residual SVGP layer that provides calibrated predictive uncertainty. This probabilistic digital twin facilitates the construction of sound-exposure bounds and worst-case scenarios for received levels. We further demonstrate the application of the framework to ship speed optimization, where predicted transmission loss combined with near-field source models provides sound exposure level estimates for minimizing acoustic impacts on marine mammals. The proposed framework advances uncertainty-aware digital twins for ocean acoustics and illustrates how physics-guided machine learning can support sustainable maritime operations.
Authors: Yanan Zhao, Feng Ji, Jingyang Dai, Jiaze Ma, Wee Peng Tay
Abstract: Graph Contrastive Learning (GCL) has shown strong promise for unsupervised graph representation learning, yet its effectiveness on heterophilic graphs, where connected nodes often belong to different classes, remains limited. Most existing methods rely on complex augmentation schemes, intricate encoders, or negative sampling, which raises the question of whether such complexity is truly necessary in this challenging setting. In this work, we revisit the foundations of supervised and unsupervised learning on graphs and uncover a simple yet effective principle for GCL: mitigating node feature noise by aggregating it with structural features derived from the graph topology. This observation suggests that the original node features and the graph structure naturally provide two complementary views for contrastive learning. Building on this insight, we propose an embarrassingly simple GCL model that uses a GCN encoder to capture structural features and an MLP encoder to isolate node feature noise. Our design requires neither data augmentation nor negative sampling, yet achieves state-of-the-art results on heterophilic benchmarks with minimal computational and memory overhead, while also offering advantages in homophilic graphs in terms of complexity, scalability, and robustness. We provide theoretical justification for our approach and validate its effectiveness through extensive experiments, including robustness evaluations against both black-box and white-box adversarial attacks.
Authors: Xiang Zhang, Kun Wei, Xu Yang, Chenghao Xu, Su Yan, Cheng Deng
Abstract: As Large Language Models (LLMs) become increasingly prevalent, their security vulnerabilities have already drawn attention. Machine unlearning is introduced to seek to mitigate these risks by removing the influence of undesirable data. However, existing methods not only rely on the retained dataset to preserve model utility, but also suffer from cumulative catastrophic utility loss under continuous unlearning requests. To solve this dilemma, we propose a novel method, called Rotation Control Unlearning (RCU), which leverages the rotational salience weight of RCU to quantify and control the unlearning degree in the continuous unlearning process. The skew symmetric loss is designed to construct the existence of the cognitive rotation space, where the changes of rotational angle can simulate the continuous unlearning process. Furthermore, we design an orthogonal rotation axes regularization to enforce mutually perpendicular rotation directions for continuous unlearning requests, effectively minimizing interference and addressing cumulative catastrophic utility loss. Experiments on multiple datasets confirm that our method without retained dataset achieves SOTA performance.
Authors: Kaizhuo Yan (University of Illinois Urbana-Champaign), Yingjie Yu (University of Illinois Urbana-Champaign), Yifan Yu (University of Illinois Urbana-Champaign), Haizhong Zheng (Carnegie Mellon University), Fan Lai (University of Illinois Urbana-Champaign)
Abstract: Proximal Policy Optimization (PPO)-based reinforcement learning from human feedback (RLHF) is a widely adopted paradigm for aligning large language models (LLMs) with human preferences. However, its training pipeline suffers from substantial inefficiencies due to sequential multi-model dependencies (e.g., reward model depends on actor outputs) and long-tail response lengths, where a few long responses straggle the stage completion. We present OPPO, a novel, lightweight, and model-agnostic PPO-based RLHF framework that improves training efficiency by overlapping pipeline execution. OPPO introduces two novel techniques: (1) Intra-step overlap, which streams upstream model outputs (e.g., actor model) in right-sized chunks, enabling the downstream model (e.g., reward) to begin prefill while the upstream continues decoding; and (2) Inter-step overlap, which adaptively overcommits a few prompts and defers long generations to future steps, mitigating tail latency without discarding partial work. OPPO integrates easily with existing PPO implementations with a few lines of code change. Extensive evaluations show that OPPO accelerates PPO-based RLHF training by $1.8 \times-2.8 \times$ and improves GPU utilization by $1.4 \times-2.1 \times$ without compromising training convergence.
Authors: Amber Srivastava, Salar Basiri, Srinivasa Salapaka
Abstract: Clustering arises in a wide range of problem formulations, yet most existing approaches assume that the entities under clustering are passive and strictly conform to their assigned groups. In reality, entities often exhibit local autonomy, overriding prescribed associations in ways not fully captured by feature representations. Such autonomy can substantially reshape clustering outcomes -- altering cluster compositions, geometry, and cardinality -- with significant downstream effects on inference and decision-making. We introduce autonomy-aware clustering, a reinforcement (RL) learning framework that learns and accounts for the influence of local autonomy without requiring prior knowledge of its form. Our approach integrates RL with a deterministic annealing (DA) procedure, where, to determine underlying clusters, DA naturally promotes exploration in early stages of annealing and transitions to exploitation later. We also show that the annealing procedure exhibits phase transitions that enable design of efficient annealing schedules. To further enhance adaptability, we propose the Adaptive Distance Estimation Network (ADEN), a transformer-based attention model that learns dependencies between entities and cluster representatives within the RL loop, accommodates variable-sized inputs and outputs, and enables knowledge transfer across diverse problem instances. Empirical results show that our framework closely aligns with underlying data dynamics: even without explicit autonomy models, it achieves solutions close to the ground truth (gap ~3-4%), whereas ignoring autonomy leads to substantially larger gaps (~35-40%). The code and data are publicly available at https://github.com/salar96/AutonomyAwareClustering.
Authors: Jianyu Xu, Vidhi Jain, Bryan Wilder, Aarti Singh
Abstract: With advances in generative AI, decision-making agents can now dynamically create new actions during online learning, but action generation typically incurs costs that must be balanced against potential benefits. We study an online learning problem where an agent can generate new actions at any time step by paying a one-time cost, with these actions becoming permanently available for future use. The challenge lies in learning the optimal sequence of two-fold decisions: which action to take and when to generate new ones, further complicated by the triangular tradeoffs among exploitation, exploration and $\textit{creation}$. To solve this problem, we propose a doubly-optimistic algorithm that employs Lower Confidence Bounds (LCB) for action selection and Upper Confidence Bounds (UCB) for action generation. Empirical evaluation on healthcare question-answering datasets demonstrates that our approach achieves favorable generation-quality tradeoffs compared to baseline strategies. From theoretical perspectives, we prove that our algorithm achieves the optimal regret of $O(T^{\frac{d}{d+2}}d^{\frac{d}{d+2}} + d\sqrt{T\log T})$, providing the first sublinear regret bound for online learning with expanding action spaces.
Authors: Prosper Rosaire Mama Assandje, Teumsa Aboubakar, Dongho Joseph, Takemi Nakamura
Abstract: Bridging information geometry with machine learning, this paper presents a method for constructing neural networks intrinsically on statistical manifolds. We demonstrate this approach by formulating a neural network architecture directly on the lognormal statistical manifold. The construction is driven by the Hamiltonian system that is equivalent to the gradient flow on this manifold. First, we define the network's input values using the coordinate system of this Hamiltonian dynamics, naturally embedded in the Poincare disk. The core of our contribution lies in the derivation of the network's components from geometric principles: the rotation component of the synaptic weight matrix is determined by the Lie group action of SU(1,1) on the disk, while the activation function emerges from the symplectic structure of the system. We subsequently obtain the complete weight matrix, including its translation vector, and the resulting output values. This work shows that the lognormal manifold can be seamlessly viewed as a neural manifold, with its geometric properties dictating a unique and interpretable neural network structure. The proposed method offers a new paradigm for building learning systems grounded in the differential geometry of their underlying parameter spaces.
Authors: Zhizhou Zhang, Youjia Wu, Kaixuan Zhang, Yanjia Wang
Abstract: Industrial design evaluation often relies on high-fidelity simulations of governing partial differential equations (PDEs). While accurate, these simulations are computationally expensive, making dense exploration of design spaces impractical. Operator learning has emerged as a promising approach to accelerate PDE solution prediction; however, its effectiveness is often limited by the scarcity of labeled physics-based data. At the same time, large numbers of geometry-only candidate designs are readily available but remain largely untapped. We propose a two-stage framework to better exploit this abundant, physics-agnostic resource and improve supervised operator learning under limited labeled data. In Stage 1, we pretrain an autoencoder on a geometry reconstruction task to learn an expressive latent representation without PDE labels. In Stage 2, the neural operator is trained in a standard supervised manner to predict PDE solutions, using the pretrained latent embeddings as inputs instead of raw point clouds. Transformer-based architectures are adopted for both the autoencoder and the neural operator to handle point cloud data and integrate both stages seamlessly. Across four PDE datasets and three state-of-the-art transformer-based neural operators, our approach consistently improves prediction accuracy compared to models trained directly on raw point cloud inputs. These results demonstrate that representations from physics-agnostic pretraining provide a powerful foundation for data-efficient operator learning.
Authors: Gongxu Luo, Loka Li, Guangyi Chen, Haoyue Dai, Kun Zhang
Abstract: Interventional causal discovery seeks to identify causal relations by leveraging distributional changes introduced by interventions, even in the presence of latent confounders. Beyond the spurious dependencies induced by latent confounders, we highlight a common yet often overlooked challenge in the problem due to post-treatment selection, in which samples are selectively included in datasets after interventions. This fundamental challenge widely exists in biological studies; for example, in gene expression analysis, both observational and interventional samples are retained only if they meet quality control criteria (e.g., highly active cells). Neglecting post-treatment selection may introduce spurious dependencies and distributional changes under interventions, which can mimic causal responses, thereby distorting causal discovery results and challenging existing causal formulations. To address this, we introduce a novel causal formulation that explicitly models post-treatment selection and reveals how its differential reactions to interventions can distinguish causal relations from selection patterns, allowing us to go beyond traditional equivalence classes toward the underlying true causal structure. We then characterize its Markov properties and propose a Fine-grained Interventional equivalence class, named FI-Markov equivalence, represented by a new graphical diagram, F-PAG. Finally, we develop a provably sound and complete algorithm, F-FCI, to identify causal relations, latent confounders, and post-treatment selection up to $\mathcal{FI}$-Markov equivalence, using both observational and interventional data. Experimental results on synthetic and real-world datasets demonstrate that our method recovers causal relations despite the presence of both selection and latent confounders.
Authors: Vaskar Chakma, Ju Xiaolin, Heling Cao, Xue Feng, Ji Xiaodong, Pan Haiyan, Gao Zhan
Abstract: This study aims to develop and evaluate an ensemble machine learning-based framework for the automatic detection of Wide QRS Complex Tachycardia (WCT) from ECG signals, emphasizing diagnostic accuracy and interpretability using Explainable AI. The proposed system integrates ensemble learning techniques, i.e., an optimized Random Forest known as CardioForest, and models like XGBoost and LightGBM. The models were trained and tested on ECG data from the publicly available MIMIC-IV dataset. The testing was carried out with the assistance of accuracy, balanced accuracy, precision, recall, F1 score, ROC-AUC, and error rate (RMSE, MAE) measures. In addition, SHAP (SHapley Additive exPlanations) was used to ascertain model explainability and clinical relevance. The CardioForest model performed best on all metrics, achieving a test accuracy of 94.95%, a balanced accuracy of 88.31%, and high precision and recall metrics. SHAP analysis confirmed the model's ability to rank the most relevant ECG features, such as QRS duration, in accordance with clinical intuitions, thereby fostering trust and usability in clinical practice. The findings recognize CardioForest as an extremely dependable and interpretable WCT detection model. Being able to offer accurate predictions and transparency through explainability makes it a valuable tool to help cardiologists make timely and well-informed diagnoses, especially for high-stakes and emergency scenarios.
Authors: Yuheng Zhang, Wenlin Yao, Changlong Yu, Yao Liu, Qingyu Yin, Bing Yin, Hyokun Yun, Lihong Li
Abstract: Large language models (LLMs) have achieved impressive reasoning performance, with reinforcement learning with verifiable rewards (RLVR) emerging as a standard paradigm for post-training. A representative algorithm, group relative policy optimization (GRPO) (Shao et al., 2024), computes advantages by normalizing outcome rewards within response groups, but suffers from a vanishing advantage issue when all responses in a group receive identical rewards. To address this issue, we propose Adaptive Rollout and Response Reuse Policy Optimization (AR3PO), a sampling efficient RLVR algorithm that introduces two novel techniques: adaptive rollout, which dynamically allocates more responses to difficult prompts while saving computation on easier ones, and response reuse, which leverages previously generated correct responses to provide useful training signals. We compare AR3PO with strong RLVR baselines on multiple representative benchmarks using two different families of base models. Across the 7B and 8B models, AR3PO consistently outperforms GRPO and matches or surpasses DAPO (Yu et al., 2025), reducing rollout cost by up to 4.2x. On the larger 32B model, AR3PO achieves comparable performance to DAPO at similar training steps while maintaining substantially lower rollout cost.
Authors: Shenao Zhang, Donghan Yu, Yihao Feng, Bowen Jin, Zhaoran Wang, John Peebles, Zirui Wang
Abstract: Large language models excel with reinforcement learning (RL), but fully unlocking this potential requires a mid-training stage. An effective mid-training phase should identify a compact set of useful actions and enable fast selection among them through online RL. We formalize this intuition by presenting the first theoretical result on how mid-training shapes post-training: it characterizes an action subspace that minimizes both the value approximation error from pruning and the RL error during subsequent planning. Our analysis reveals two key determinants of mid-training effectiveness: pruning efficiency, which shapes the prior of the initial RL policy, and its impact on RL convergence, which governs the extent to which that policy can be improved via online interactions. These results suggest that mid-training is most effective when the decision space is compact and the effective horizon is short, highlighting the importance of operating in the space of action abstractions rather than primitive actions. Building on these insights, we propose Reasoning as Action Abstractions (RA3), a scalable mid-training algorithm. Specifically, we derive a sequential variational lower bound and optimize it by iteratively discovering temporally-consistent latent structures via RL, followed by fine-tuning on the bootstrapped data. Experiments on code generation tasks demonstrate the effectiveness of our approach. Across multiple base models, RA3 improves the average performance on HumanEval and MBPP by 8 and 4 points over the base model and the next-token prediction baseline. Furthermore, RA3 achieves faster convergence and higher asymptotic performance in RLVR on HumanEval+, MBPP+, LiveCodeBench, and Codeforces.
Authors: Jingqi Fan, Canzhe Zhao, Shuai Li, Siwei Wang
Abstract: In recent years, multi-player multi-armed bandits (MP-MAB) have been extensively studied due to their wide applications in cognitive radio networks and Internet of Things systems. While most existing research on MP-MAB focuses on synchronized settings, real-world systems are often decentralized and asynchronous, where players may enter or leave the system at arbitrary times, and do not have a global clock. This decentralized asynchronous setting introduces two major challenges. First, without a global time, players cannot implicitly coordinate their actions through time, making it difficult to avoid collisions. Second, it is important to detect how many players are in the system, but doing so may cost a lot. In this paper, we address the challenges posed by such a fully asynchronous setting in a decentralized environment. We develop a novel algorithm in which players adaptively change between exploration and exploitation. During exploration, players uniformly pull their arms, reducing the probability of collisions and effectively mitigating the first challenge. Meanwhile, players continue pulling arms currently exploited by others with a small probability, enabling them to detect when a player has left, thereby addressing the second challenge. We prove that our algorithm achieves a regret of $\mathcal{O}(\sqrt{T \log T} + {\log T}/{\Delta^2})$, where $\Delta$ is the minimum expected reward gap between any two arms. To the best of our knowledge, this is the first efficient MP-MAB algorithm in the asynchronous and decentralized environment. Extensive experiments further validate the effectiveness and robustness of our algorithm, demonstrating its applicability to real-world scenarios.
Authors: Kun Feng, Shaocheng Lan, Yuchen Fang, Wenchao He, Lintao Ma, Xingyu Lu, Kan Ren
Abstract: Time series foundation models (TSFMs) have emerged as a powerful paradigm for time series analysis, driven by large-scale pretraining on diverse data corpora. However, time series inherently exhibit heterogeneous information density over time, influenced by system states and signal complexity, presenting significant modeling challenges especially in a zero-shot scenario. Current TSFMs rely on non-adaptive processing pipelines that fail to capture this dynamic nature. For example, common tokenization strategies such as fixed-size patching enforce rigid observational granularity, limiting their ability to adapt to varying information densities. Similarly, conventional positional encodings impose a uniform temporal scale, making it difficult to model diverse periodicities and trends across series. To overcome these limitations, we propose Kairos, a flexible TSFM framework that integrates a dynamic patching tokenizer and an instance-adaptive positional embedding. Kairos adaptively selects tokenization granularity and tailors positional encodings to the unique characteristics of each time series instance. Trained on a large-scale Predictability-Stratified Time Series (PreSTS) corpus comprising over 300 billion time points and adopting a multi-patch prediction strategy in the inference stage, Kairos achieves superior performance with much fewer parameters on two common zero-shot benchmarks, GIFT-Eval and the Time-Series-Library benchmark, consistently outperforming established methods across diverse tasks. The project page is at https://foundation-model-research.github.io/Kairos .
Authors: Seong-Hyeon Hwang, Soyoung Choi, Steven Euijong Whang
Abstract: Multimodal models often over-rely on dominant modalities, failing to achieve optimal performance. While prior work focuses on modifying training objectives or optimization procedures, data-centric solutions remain underexplored. We propose MIDAS, a novel data augmentation strategy that generates misaligned samples with semantically inconsistent cross-modal information, labeled using unimodal confidence scores to compel learning from contradictory signals. However, this confidence-based labeling can still favor the more confident modality. To address this within our misaligned samples, we introduce weak-modality weighting, which dynamically increases the loss weight of the least confident modality, thereby helping the model fully utilize weaker modality. Furthermore, when misaligned features exhibit greater similarity to the aligned features, these misaligned samples pose a greater challenge, thereby enabling the model to better distinguish between classes. To leverage this, we propose hard-sample weighting, which prioritizes such semantically ambiguous misaligned samples. Experiments on multiple multimodal classification benchmarks demonstrate that MIDAS significantly outperforms related baselines in addressing modality imbalance.
Authors: Yeongmin Kim, Donghyeok Shin, Mina Kang, Byeonghu Na, Il-Chul Moon
Abstract: Large language models (LLMs) deliver remarkable performance but are costly to deploy, motivating knowledge distillation (KD) for efficient inference. Existing KD objectives typically match student and teacher probabilities via softmax, which blurs valuable logit information. While direct logit distillation (DLD) mitigates softmax smoothing, it fails to account for logit shift invariance, thereby restricting the solution space. We propose Concrete Score Distillation (CSD), a discrete score-matching objective that overcomes both softmax-induced smoothing and restrictions on the optimal solution set. We resolve the training instability and quadratic complexity of discrete score-matching in autoregressive LLMs, and the resulting CSD objective aligns relative logit differences across all vocabulary pairs between student and teacher with flexible weighting. We provide both mode-seeking and mode-covering instances within our framework and evaluate CSD on task-agnostic instruction-following and task-specific distillation using GPT-2-1.5B, OpenLLaMA-7B, and GEMMA-7B-IT. Experiments show that CSD consistently surpasses recent KD objectives, achieves favorable fidelity-diversity trade-offs, and yields complementary gains when combined with on-policy techniques, demonstrating its scalability and effectiveness for LLM distillation.
Authors: Suping Xu, Chuyi Dai, Ye Liu, Lin Shang, Xibei Yang, Witold Pedrycz
Abstract: Feature selection is crucial for fuzzy decision systems (FDSs), as it identifies informative features and eliminates rule redundancy, thereby enhancing predictive performance and interpretability. Most existing methods either fail to directly align evaluation criteria with learning performance or rely solely on non-directional Euclidean distances to capture relationships among decision classes, which limits their ability to clarify decision boundaries. However, the spatial distribution of instances has a potential impact on the clarity of such boundaries. Motivated by this, we propose Spatially-aware Separability-driven Feature Selection (S$^2$FS), a novel framework for FDSs guided by a spatially-aware separability criterion. This criterion jointly considers within-class compactness and between-class separation by integrating scalar-distances with spatial directional information, providing a more comprehensive characterization of class structures. S$^2$FS employs a forward greedy strategy to iteratively select the most discriminative features. Extensive experiments on ten real-world datasets demonstrate that S$^2$FS consistently outperforms eight state-of-the-art feature selection algorithms in both classification accuracy and clustering performance, while feature visualizations further confirm the interpretability of the selected features.
Authors: Ziniu Li, Congliang Chen, Tianyun Yang, Tian Ding, Ruoyu Sun, Ge Zhang, Wenhao Huang, Zhi-Quan Luo
Abstract: Large Language Models (LLMs) can self-improve through reinforcement learning, where they generate trajectories to explore and discover better solutions. However, this exploration process is computationally expensive, often forcing current methods to assign limited exploration budgets to each task. This uniform allocation creates problematic edge cases: easy tasks consistently succeed while difficult tasks consistently fail, both producing zero gradients during training updates for the widely used Group Relative Policy Optimization (GRPO). We address this problem from the lens of exploration budget allocation. Viewing each task's exploration as an "item" with a distinct "value" and "cost", we establish a connection to the classical knapsack problem. This formulation allows us to derive an optimal assignment rule that adaptively distributes resources based on the model's current learning status. When applied to GRPO, our method increases the effective ratio of non-zero policy gradients by 20-40% during training. Acting as a computational "free lunch", our approach could reallocate exploration budgets from tasks where learning is saturated to those where it is most impactful. This enables significantly larger budgets (e.g., 93 rollouts) for especially challenging problems, which would be computationally prohibitive under a uniform allocation. These improvements translate to meaningful gains on mathematical reasoning benchmarks, with average improvements of 2-4 points and peak gains of 9 points on specific tasks. Notably, achieving comparable performance with traditional homogeneous allocation would require about 2x the computational resources.
Authors: Animesh Jha, Harshit Gupta, Ananjan Nandi
Abstract: Data selection for finetuning Large Language Models (LLMs) can be framed as a budget-constrained optimization problem: maximizing a model's downstream performance under a strict training data budget. Solving this problem is generally intractable, and existing approximate approaches are pretraining-oriented and transfer poorly to the fine-tuning setting. We reformulate this problem as a tractable Markov Decision Process (MDP) and train agents using various Reinforcement Learning (RL) methods to learn optimal data selection policies, guided by an efficient, proxy-model-based reward signal. Across four datasets, training on a $5\%$ subset selected by our approach matches or outperforms fine-tuning on the full dataset by up to $10.8$ accuracy points, while cutting wall-clock training time by up to $2 \times$, highlighting the promise of RL-guided data selection.
Authors: Xinyu Zhang, Aishik Deb, Klaus Mueller
Abstract: Policy-gradient methods such as Proximal Policy Optimization (PPO) are typically updated along a single stochastic gradient direction, leaving the rich local structure of the parameter space unexplored. Previous work has shown that the surrogate gradient is often poorly correlated with the true reward landscape. Building on this insight, we visualize the parameter space spanned by policy checkpoints within an iteration and reveal that higher performing solutions often lie in nearby unexplored regions. To exploit this opportunity, we introduce ExploRLer, a pluggable pipeline that seamlessly integrates with on-policy algorithms such as PPO and TRPO, systematically probing the unexplored neighborhoods of surrogate on-policy gradient updates. Without increasing the number of gradient updates, ExploRLer achieves significant improvements over baselines in complex continuous control environments. Our results demonstrate that iteration-level exploration provides a practical and effective way to strengthen on-policy reinforcement learning and offer a fresh perspective on the limitations of the surrogate objective.
Authors: Yiwei Li, Shuai Wang, Zhuojun Tian, Xiuhua Wang, Shijian Su
Abstract: Federated Learning (FL) often adopts differential privacy (DP) to protect client data, but the added noise required for privacy guarantees can substantially degrade model accuracy. To resolve this challenge, we propose model-splitting privacy-amplified federated learning (MS-PAFL), a novel framework that combines structural model splitting with statistical privacy amplification. In this framework, each client's model is partitioned into a private submodel, retained locally, and a public submodel, shared for global aggregation. The calibrated Gaussian noise is injected only into the public submodel, thereby confining its adverse impact while preserving the utility of the local model. We further present a rigorous theoretical analysis that characterizes the joint privacy amplification achieved through random client participation and local data subsampling under this architecture. The analysis provides tight bounds on both single-round and total privacy loss, demonstrating that MS-PAFL significantly reduces the noise necessary to satisfy a target privacy protection level. Extensive experiments validate our theoretical findings, showing that MS-PAFL consistently attains a superior privacy-utility trade-off and enables the training of highly accurate models under strong privacy guarantees.
Authors: Yihang Lu, Xianwei Meng, Enhong Chen
Abstract: Neural Forecasters (NFs) are a cornerstone of Long-term Time Series Forecasting (LTSF). However, progress has been hampered by an overemphasis on architectural complexity at the expense of fundamental forecasting principles. In this work, we return to first principles to redesign the LTSF paradigm. We begin by introducing a Multiple Neural Forecasting Theorem that provides a theoretical basis for our approach. We propose Boosted Direct Output (BDO), a novel forecasting strategy that synergistically combines the advantages of both Auto-Regressive (AR) and Direct Output (DO). In addition, we stabilize the learning process by smoothly tracking the model's parameters. Extensive experiments show that these principled improvements enable a simple MLP to achieve state-of-the-art performance, outperforming recent, complex models in nearly all cases, without any specific considerations in the area. Finally, we empirically verify our theorem, establishing a dynamic performance bound and identifying promising directions for future research. The code for review is available at: .
Authors: Sven Br\"andle, Till Aczel, Andreas Plesner, Roger Wattenhofer
Abstract: Differentiable Logic Gate Networks (DLGNs) are a very fast and energy-efficient alternative to conventional feed-forward networks. With learnable combinations of logical gates, DLGNs enable fast inference by hardware-friendly execution. Since the concept of DLGNs has only recently gained attention, these networks are still in their developmental infancy, including the design and scalability of their output layer. To date, this architecture has primarily been tested on datasets with up to ten classes. This work examines the behavior of DLGNs on large multi-class datasets. We investigate its general expressiveness, its scalability, and evaluate alternative output strategies. Using both synthetic and real-world datasets, we provide key insights into the importance of temperature tuning and its impact on output layer performance. We evaluate conditions under which the Group-Sum layer performs well and how it can be applied to large-scale classification of up to 2000 classes.
Authors: Mason Minot, Gisbert Schneider
Abstract: Simultaneously optimizing multiple, frequently conflicting, molecular properties is a key bottleneck in the development of novel therapeutics. Although a promising approach, the efficacy of multi-task learning is often compromised by destructive gradient interference, especially in the data-scarce regimes common to drug discovery. To address this, we propose AIM, an optimization framework that learns a dynamic policy to mediate gradient conflicts. The policy is trained jointly with the main network using a novel augmented objective composed of dense, differentiable regularizers. This objective guides the policy to produce updates that are geometrically stable and dynamically efficient, prioritizing progress on the most challenging tasks. We demonstrate that AIM achieves statistically significant improvements over multi-task baselines on subsets of the QM9 and targeted protein degraders benchmarks, with its advantage being most pronounced in data-scarce regimes. Beyond performance, AIM's key contribution is its interpretability; the learned policy matrix serves as a diagnostic tool for analyzing inter-task relationships. This combination of data-efficient performance and diagnostic insight highlights the potential of adaptive optimizers to accelerate scientific discovery by creating more robust and insightful models for multi-property molecular design.
Authors: Deniz Soysal, Xabier Garc\'ia-Andrade, Laura E. Rodriguez, Pablo Sobron, Laura M. Barge, Renaud Detry
Abstract: Autonomous Raman instruments on Mars rovers, deep-sea landers, and field robots must interpret raw spectra distorted by fluorescence baselines, peak shifts, and limited ground-truth labels. Using curated subsets of the RRUFF database, we evaluate one-dimensional convolutional neural networks (CNNs) and report four advances: (i) Baseline-independent classification: compact CNNs surpass $k$-nearest-neighbors and support-vector machines on handcrafted features, removing background-correction and peak-picking stages while ensuring reproducibility through released data splits and scripts. (ii) Pooling-controlled robustness: tuning a single pooling parameter accommodates Raman shifts up to $30 \,\mathrm{cm}^{-1}$, balancing translational invariance with spectral resolution. (iii) Label-efficient learning: semi-supervised generative adversarial networks and contrastive pretraining raise accuracy by up to $11\%$ with only $10\%$ labels, valuable for autonomous deployments with scarce annotation. (iv) Constant-time adaptation: freezing the CNN backbone and retraining only the softmax layer transfers models to unseen minerals at $\mathcal{O}(1)$ cost, outperforming Siamese networks on resource-limited processors. This workflow, which involves training on raw spectra, tuning pooling, adding semi-supervision when labels are scarce, and fine-tuning lightly for new targets, provides a practical path toward robust, low-footprint Raman classification in autonomous exploration.
Authors: Xiao Zhang, Zengzhe Chen, Yuan Yuan, Yifei Zou, Fuzhen Zhuang, Wenyu Jiao, Yuke Wang, Dongxiao Yu
Abstract: Federated learning (FL) is a distributed learning paradigm across multiple entities while preserving data privacy. However, with the continuous emergence of new data and increasing model diversity, traditional federated learning faces significant challenges, including inherent issues of data heterogeneity, model heterogeneity and catastrophic forgetting, along with new challenge of knowledge misalignment. In this study, we introduce FedDCL, a novel framework designed to enable data-free continual learning of the server model in a model-heterogeneous federated setting. We leverage pre-trained diffusion models to extract lightweight class-specific prototypes, which confer a threefold data-free advantage, enabling: (1) generation of synthetic data for the current task to augment training and counteract non-IID data distributions; (2) exemplar-free generative replay for retaining knowledge from previous tasks; and (3) data-free dynamic knowledge transfer from heterogeneous clients to the server. Experimental results on various datasets demonstrate the effectiveness of FedDCL, showcasing its potential to enhance the generalizability and practical applicability of federated learning in dynamic settings.
Authors: Gaojie Jin, Xinping Yi, Xiaowei Huang
Abstract: Within the PAC-Bayesian framework, the Gibbs classifier (defined on a posterior $Q$) and the corresponding $Q$-weighted majority vote classifier are commonly used to analyze the generalization performance. However, there exists a notable lack in theoretical research exploring the certified robustness of majority vote classifier and its interplay with generalization. In this study, we develop a generalization error bound that possesses a certified robust radius for the smoothed majority vote classifier (i.e., the $Q$-weighted majority vote classifier with smoothed inputs); In other words, the generalization bound holds under any data perturbation within the certified robust radius. As a byproduct, we find that the underpinnings of both the generalization bound and the certified robust radius draw, in part, upon weight spectral norm, which thereby inspires the adoption of spectral regularization in smooth training to boost certified robustness. Utilizing the dimension-independent property of spherical Gaussian inputs in smooth training, we propose a novel and inexpensive spectral regularizer to enhance the smoothed majority vote classifier. In addition to the theoretical contribution, a set of empirical results is provided to substantiate the effectiveness of our proposed method.
Authors: Mykola Bordyuh, Djork-Arn\'e Clevert, Marco Bertolini
Abstract: The Quantum Schr\"odinger Bridge Problem (QSBP) describes the evolution of a stochastic process between two arbitrary probability distributions, where the dynamics are governed by the Schr\"odinger equation rather than by the traditional real-valued wave equation. Although the QSBP is known in the mathematical literature, we formulate it here from a Lagrangian perspective and derive its main features in a way that is particularly suited to generative modeling. We show that the resulting evolution equations involve the so-called Bohm (quantum) potential, representing a notion of non-locality in the stochastic process. This distinguishes the QSBP from classical stochastic dynamics and reflects a key characteristic typical of quantum mechanical systems. In this work, we derive exact closed-form solutions for the QSBP between Gaussian distributions. Our derivation is based on solving the Fokker-Planck Equation (FPE) and the Hamilton-Jacobi Equation (HJE) arising from the Lagrangian formulation of dynamical Optimal Transport. We find that, similar to the classical Schr\"odinger Bridge Problem, the solution to the QSBP between Gaussians is again a Gaussian process; however, the evolution of the covariance differs due to quantum effects. Leveraging these explicit solutions, we present a modified algorithm based on a Gaussian Mixture Model framework, and demonstrate its effectiveness across several experimental settings, including single-cell evolution data, image generation, molecular translation and applications in Mean-Field Games.
Authors: Weiyu Huang, Yuezhou Hu, Jun Zhu, Jianfei Chen
Abstract: Sparsity-aware training is an effective approach for transforming large language models (LLMs) into hardware-friendly sparse patterns, thereby reducing latency and memory consumption during inference. In this paper, we propose Continuous Adaptive Sparse Trainer (CAST), a fully continuous and differentiable sparsity-aware training framework for semi-structured (or "N:M") sparse models. Unlike previous approaches that optimize sparsity patterns and weights separately, CAST enables seamless joint optimization during training, while progressively transforming the model into the desired sparsity format. Specifically, CAST introduces three key components: 1) AdamS, a sparsity-aware optimizer that leverages adaptive L1 decay to promote uniform sparsification across all parameters; 2) Weight Scaling, a module designed to mitigate the magnitude reduction caused by decay while preserving desired sparsity patterns; 3) Knowledge Distillation, which employs the dense model as a self-teacher to enhance training efficiency. We evaluate CAST under 2:4 sparsity patterns across multiple model families, ranging from 125M to 13B parameters. Our results demonstrate significant improvements over previous state-of-the-art methods in both perplexity and zero-shot accuracy with minimal training resources. Notably, on LLaMA2-7B, our 2:4 sparse model achieves a negligible perplexity increase of 0.09 and a 0.36% gain in zero-shot accuracy compared to the dense model using only 2% of the original pretraining tokens. Additionally, we establish an accurate and robust empirical scaling law to predict sparse model performance given adequate training resources. Finally, we demonstrate the practical applicability of our sparse models by evaluating them under quantization and fine-tuning scenarios.
Authors: Daniel Ebi, Gaspard Lambrechts, Damien Ernst, Klemens B\"ohm
Abstract: Reinforcement learning in partially observable environments requires agents to act under uncertainty from noisy, incomplete observations. Asymmetric actor-critic methods leverage privileged information during training to improve learning under these conditions. However, existing approaches typically assume full-state access during training. In this work, we challenge this assumption by proposing a novel actor-critic framework, called informed asymmetric actor-critic, that enables conditioning the critic on arbitrary privileged signals without requiring access to the full state. We show that policy gradients remain unbiased under this formulation, extending the theoretical foundation of asymmetric methods to the more general case of privileged partial information. To quantify the impact of such signals, we propose informativeness measures based on kernel methods and return prediction error, providing practical tools for evaluating training-time signals. We validate our approach empirically on benchmark navigation tasks and synthetic partially observable environments, showing that our informed asymmetric method improves learning efficiency and value estimation when informative privileged inputs are available. Our findings challenge the necessity of full-state access and open new directions for designing asymmetric reinforcement learning methods that are both practical and theoretically sound.
Authors: Bissmella Bahaduri, Hicham Talaoubrid, Fangchen Feng, Zuheng Ming, Anissa Mokraoui
Abstract: The attention mechanism has become a cornerstone of modern deep learning architectures, where keys and values are typically derived from the same underlying sequence or representation. This work explores a less conventional scenario, when keys and values originate from different sequences or modalities. Specifically, we first analyze the attention mechanism's behavior under noisy value features, establishing a critical noise threshold beyond which signal degradation becomes significant. Furthermore, we model context (key, value) misalignment as an effective form of structured noise within the value features, demonstrating that the noise induced by such misalignment can substantially exceed this critical threshold, thereby compromising standard attention's efficacy. Motivated by this, we introduce Indirect Attention, a modified attention mechanism that infers relevance indirectly in scenarios with misaligned context. We evaluate the performance of Indirect Attention across a range of synthetic tasks and real world applications, showcasing its superior ability to handle misalignment.
Authors: Daphne Theodorakopoulos, Elisabeth Eberling, Miriam Bodenheimer, Sabine Loos, Frederic Stahl
Abstract: Access to credible sustainability information in the fashion industry remains limited and challenging to interpret, despite growing public and regulatory demands for transparency. General-purpose language models often lack domain-specific knowledge and tend to "hallucinate", which is particularly harmful for fields where factual correctness is crucial. This work explores how Natural Language Processing (NLP) techniques can be applied to classify sustainability data for fashion brands, thereby addressing the scarcity of credible and accessible information in this domain. We present a prototype Fashion Information Tool for Sustainability (FITS), a transformer-based system that extracts and classifies sustainability information from credible, unstructured text sources: NGO reports and scientific publications. Several BERT-based language models, including models pretrained on scientific and climate-specific data, are fine-tuned on our curated corpus using a domain-specific classification schema, with hyperparameters optimized via Bayesian optimization. FITS allows users to search for relevant data, analyze their own data, and explore the information via an interactive interface. We evaluated FITS in two focus groups of potential users concerning usability, visual design, content clarity, possible use cases, and desired features. Our results highlight the value of domain-adapted NLP in promoting informed decision-making and emphasize the broader potential of AI applications in addressing climate-related challenges. Finally, this work provides a valuable dataset, the SustainableTextileCorpus, along with a methodology for future updates. Code available at https://github.com/daphne12345/FITS
Authors: Shuche Wang, Fengzhuo Zhang, Jiaxiang Li, Cunxiao Du, Chao Du, Tianyu Pang, Zhuoran Yang, Mingyi Hong, Vincent Y. F. Tan
Abstract: The Muon optimizer is consistently faster than Adam in training Large Language Models (LLMs), yet the mechanism underlying its success remains unclear. This paper demystifies this mechanism through the lens of associative memory. By ablating the transformer components optimized by Muon, we reveal that the associative memory parameters of LLMs, namely the Value and Output (VO) attention weights and Feed-Forward Networks (FFNs), are the primary contributors to Muon's superiority. Motivated by this associative memory view, we then explain Muon's superiority on real-world corpora, which are intrinsically heavy-tailed: a few classes (tail classes) appear far less frequently than others. The superiority is explained through two key properties: (i) its update rule consistently yields a more isotropic singular spectrum than Adam; and as a result, (ii) on heavy-tailed data, it optimizes tail classes more effectively than Adam. Beyond empirical evidence, we theoretically confirm these findings by analyzing a one-layer associative memory model under class-imbalanced data. We prove that Muon consistently achieves balanced learning across classes regardless of feature embeddings, whereas Adam can induce large disparities in learning errors depending on embedding properties. In summary, our empirical observations and theoretical analyses reveal Muon's core advantage: its update rule aligns with the outer-product structure of linear associative memories, enabling more balanced and effective learning of tail classes in heavy-tailed distributions than Adam.
Authors: Xiaobao Wang, Ruoxiao Sun, Yujun Zhang, Bingdao Feng, Dongxiao He, Luzhi Wang, Di Jin
Abstract: Graph Neural Networks (GNNs) have demonstrated strong performance across tasks such as node classification, link prediction, and graph classification, but remain vulnerable to backdoor attacks that implant imperceptible triggers during training to control predictions. While node-level attacks exploit local message passing, graph-level attacks face the harder challenge of manipulating global representations while maintaining stealth. We identify two main sources of anomaly in existing graph classification backdoor methods: structural deviation from rare subgraph triggers and semantic deviation caused by label flipping, both of which make poisoned graphs easily detectable by anomaly detection models. To address this, we propose DPSBA, a clean-label backdoor framework that learns in-distribution triggers via adversarial training guided by anomaly-aware discriminators. DPSBA effectively suppresses both structural and semantic anomalies, achieving high attack success while significantly improving stealth. Extensive experiments on real-world datasets validate that DPSBA achieves a superior balance between effectiveness and detectability compared to state-of-the-art baselines.
Authors: Aoming Liu, Kevin Miller, Venkatesh Saligrama, Kate Saenko, Boqing Gong, Ser-Nam Lim, Bryan A. Plummer
Abstract: Temporal Domain Generalization (TDG) aims to generalize across temporal distribution shifts, e.g., lexical change over time. Prior work often addresses this by predicting future model weights. However, full model prediction is prohibitively expensive for even reasonably sized models. Thus, recent methods only predict the classifier layer, limiting generalization by failing to adjust other model components. To address this, we propose Temporal Experts Averaging (TEA), a novel and scalable TDG framework that updates the entire model using weight averaging to maximize generalization potential while minimizing computational costs. Our theoretical analysis guides us to two steps that enhance generalization to future domains. First, we create expert models with functional diversity yet parameter similarity by fine-tuning a domain-agnostic base model on individual temporal domains while constraining weight changes. Second, we optimize the bias-variance tradeoff through adaptive averaging coefficients derived from modeling temporal weight trajectories in a principal component subspace. Expert's contributions are based on their projected proximity to future domains. Extensive experiments across 7 TDG benchmarks, 5 models, and 2 TDG settings shows TEA outperforms prior TDG methods by up to 69% while being up to 60x more efficient.
Authors: Hossein Enshaei, Pariya Jebreili, Sayed Mahmoud Sakahei
Abstract: Electroencephalogram (EEG) artifact detection in real-world settings faces significant challenges such as computational inefficiency in multi-channel methods, poor robustness to simultaneous noise, and trade-offs between accuracy and complexity in deep learning models. We propose a hybrid spectral-temporal framework for real-time detection and classification of ocular (EOG), muscular (EMG), and white noise artifacts in single-channel EEG. This method, in contrast to other approaches, combines time-domain low-pass filtering (targeting low-frequency EOG) and frequency-domain power spectral density (PSD) analysis (capturing broad-spectrum EMG), followed by PCA-optimized feature fusion to minimize redundancy while preserving discriminative information. This feature engineering strategy allows a lightweight multi-layer perceptron (MLP) architecture to outperform advanced CNNs and RNNs by achieving 99% accuracy at low SNRs (SNR -7) dB and >90% accuracy in moderate noise (SNR 4 dB). Additionally, this framework addresses the unexplored problem of simultaneous multi-source contamination(EMG+EOG+white noise), where it maintains 96% classification accuracy despite overlapping artifacts. With 30-second training times (97% faster than CNNs) and robust performance across SNR levels, this framework bridges the gap between clinical applicability and computational efficiency, which enables real-time use in wearable brain-computer interfaces. This work also challenges the ubiquitous dependence on model depth for EEG artifact detection by demonstrating that domain-informed feature fusion surpasses complex architecture in noisy scenarios.
Authors: Jaesung R. Park, Junsu Kim, Gyeongman Kim, Jinyoung Jo, Sean Choi, Jaewoong Cho, Ernest K. Ryu
Abstract: Reinforcement learning with verifiable rewards (RLVR) has recently emerged as the leading approach for enhancing the reasoning capabilities of large language models (LLMs). However, RLVR is prone to entropy collapse, where the LLM quickly converges to a near-deterministic form, hindering exploration and progress during prolonged RL training. In this work, we reveal that the clipping mechanism in PPO and GRPO induces biases on entropy. Through theoretical and empirical analyses, we show that clip-low increases entropy, while clip-high decreases it. Further, under standard clipping parameters, the effect of clip-high dominates, resulting in an overall entropy reduction even when purely random rewards are provided to the RL algorithm. Our findings highlight an overlooked confounding factor in RLVR: independent of the reward signal, the clipping mechanism influences entropy, which in turn affects the reasoning behavior. Furthermore, our analysis demonstrates that clipping can be deliberately used to control entropy. Specifically, with a more aggressive clip-low value, one can increase entropy, promote exploration, and ultimately prevent entropy collapse in RLVR training.
Authors: Abdulkadir Celikkanat, Andres R. Masegosa, Mads Albertsen, Thomas D. Nielsen
Abstract: Metagenomic binning aims to cluster DNA fragments from mixed microbial samples into their respective genomes, a critical step for downstream analyses of microbial communities. Existing methods rely on deterministic representations, such as k-mer profiles or embeddings from large language models, which fail to capture the uncertainty inherent in DNA sequences arising from inter-species DNA sharing and from fragments with highly similar representations. We present the first probabilistic embedding approach, UncertainGen, for metagenomic binning, representing each DNA fragment as a probability distribution in latent space. Our approach naturally models sequence-level uncertainty, and we provide theoretical guarantees on embedding distinguishability. This probabilistic embedding framework expands the feasible latent space by introducing a data-adaptive metric, which in turn enables more flexible separation of bins/clusters. Experiments on real metagenomic datasets demonstrate the improvements over deterministic k-mer and LLM-based embeddings for the binning task by offering a scalable and lightweight solution for large-scale metagenomic analysis.
Authors: Fardin Jalil Piran, Anandkumar Patel, Rajiv Malhotra, Farhad Imani
Abstract: Smart manufacturing requires on-device intelligence that meets strict latency and energy budgets. HyperDimensional Computing (HDC) offers a lightweight alternative by encoding data as high-dimensional hypervectors and computing with simple operations. Prior studies often assume that the qualitative relation between HDC hyperparameters and performance is stable across applications. Our analysis of two representative tasks, signal-based quality monitoring in Computer Numerical Control (CNC) machining and image-based defect detection in Laser Powder Bed Fusion (LPBF), shows that this assumption does not hold. We map how encoder type, projection variance, hypervector dimensionality, and data regime shape accuracy, inference latency, training time, and training energy. A formal complexity model explains predictable trends in encoding and similarity computation and reveals nonmonotonic interactions with retraining that preclude a closed-form optimum. Empirically, signals favor nonlinear Random Fourier Features with more exclusive encodings and saturate in accuracy beyond moderate dimensionality. Images favor linear Random Projection, achieve high accuracy with small dimensionality, and depend more on sample count than on dimensionality. Guided by these insights, we tune HDC under multiobjective constraints that reflect edge deployment and obtain models that match or exceed the accuracy of state-of-the-art deep learning and Transformer models while delivering at least 6x faster inference and more than 40x lower training energy. These results demonstrate that domain-aware HDC encoding is necessary and that tuned HDC offers a practical, scalable path to real-time industrial AI on constrained hardware. Future work will enable adaptive encoder and hyperparameter selection, expand evaluation to additional manufacturing modalities, and validate on low-power accelerators.
Authors: Daniil Zelezetsky, Alexey K. Kovalev, Aleksandr I. Panov
Abstract: The appearance of transformer-based models in Reinforcement Learning (RL) has expanded the horizons of possibilities in robotics tasks, but it has simultaneously brought a wide range of challenges during its implementation, especially in model-free online RL. Some of the existing learning algorithms cannot be easily implemented with transformer-based models due to the instability of the latter. In this paper, we propose a method that uses the Accelerator policy as a transformer's trainer. The Accelerator, a simpler and more stable model, interacts with the environment independently while simultaneously training the transformer through behavior cloning during the first stage of the proposed algorithm. In the second stage, the pretrained transformer starts to interact with the environment in a fully online setting. As a result, this model-free algorithm accelerates the transformer in terms of its performance and helps it to train online in a more stable and faster way. By conducting experiments on both state-based and image-based ManiSkill environments, as well as on MuJoCo tasks in MDP and POMDP settings, we show that applying our algorithm not only enables stable training of transformers but also reduces training time on image-based environments by up to a factor of two. Moreover, it decreases the required replay buffer size in off-policy methods to 10-20 thousand, which significantly lowers the overall computational demands.
Authors: James Panayis (UK Atomic Energy Authority), Matt Field (UK Atomic Energy Authority), Vignesh Gopakumar (UK Atomic Energy Authority, UCL Centre for AI - UK), Andrew Lahiff (UK Atomic Energy Authority), Kristian Zarebski (UK Atomic Energy Authority), Aby Abraham (UK Atomic Energy Authority), Jonathan L. Hodges (Jensen Hughes - USA)
Abstract: There is high demand on fire simulations, in both scale and quantity. We present a multi-pronged approach to improving the time and energy required to meet these demands. We show the ability of a custom machine learning surrogate model to predict the dynamics of heat propagation orders of magnitude faster than state-of-the-art CFD software for this application. We also demonstrate how a guided optimisation procedure can decrease the number of simulations required to meet an objective; using lightweight models to decide which simulations to run, we see a tenfold reduction when locating the most dangerous location for a fire to occur within a building based on the impact of smoke on visibility. Finally we present a framework and product, Simvue, through which we access these tools along with a host of automatic organisational and tracking features which enables future reuse of data and more savings through better management of simulations and combating redundancy.
Authors: Fr\'ed\'eric Berdoz, Luca A. Lanzend\"orfer, Ren\'e Caky, Roger Wattenhofer
Abstract: Alignment of large language models remains a central challenge in natural language processing. Preference optimization has emerged as a popular and effective method for improving alignment, typically through training-time or prompt-based interventions. In this paper, we introduce alignment-aware decoding (AAD), a method to enhance model alignment directly at inference. Theoretically, AAD can be interpreted as implicit reward optimization, yet it requires no specialized training beyond the standard DPO setup. Empirically, AAD consistently outperforms strong baselines across diverse alignment benchmarks and model scales. Moreover, in data-constrained settings, AAD can produce high-quality synthetic data to improve alignment under standard decoding, providing a practical solution when labeled data is limited.
Authors: Thomas Hallopeau, Joris Gu\'erin, Laurent Demagistri, Christovam Barcellos, Nadine Dessay
Abstract: Mapping informal settlements is crucial for addressing challenges related to urban planning, public health, and infrastructure in rapidly growing cities. Geospatial machine learning has emerged as a key tool for detecting and mapping these areas from remote sensing data. However, existing approaches often treat spatial units independently, neglecting the relational structure of the urban fabric. We propose a graph-based framework that explicitly incorporates local geographical context into the classification process. Each spatial unit (cell) is embedded in a graph structure along with its adjacent neighbors, and a lightweight Graph Convolutional Network (GCN) is trained to classify whether the central cell belongs to an informal settlement. Experiments are conducted on a case study in Rio de Janeiro using spatial cross-validation across five distinct zones, ensuring robustness and generalizability across heterogeneous urban landscapes. Our method outperforms standard baselines, improving Kappa coefficient by 17 points over individual cell classification. We also show that graph-based modeling surpasses simple feature concatenation of neighboring cells, demonstrating the benefit of encoding spatial structure for urban scene understanding.
Authors: Chun-Wun Cheng, Bin Dong, Carola-Bibiane Sch\"onlieb, Angelica I Aviles-Rivero
Abstract: Neural operator models for solving partial differential equations (PDEs) often rely on global mixing mechanisms-such as spectral convolutions or attention-which tend to oversmooth sharp local dynamics and introduce high computational cost. We present FINO, a finite-difference-inspired neural architecture that enforces strict locality while retaining multiscale representational power. FINO replaces fixed finite-difference stencil coefficients with learnable convolutional kernels and evolves states via an explicit, learnable time-stepping scheme. A central Local Operator Block leverage a differential stencil layer, a gating mask, and a linear fuse step to construct adaptive derivative-like local features that propagate forward in time. Embedded in an encoder-decoder with a bottleneck, FINO captures fine-grained local structures while preserving interpretability. We establish (i) a composition error bound linking one-step approximation error to stable long-horizon rollouts under a Lipschitz condition, and (ii) a universal approximation theorem for discrete time-stepped PDE dynamics. (iii) Across six benchmarks and a climate modelling task, FINO achieves up to 44\% lower error and up to around 2\times speedups over state-of-the-art operator-learning baselines, demonstrating that strict locality with learnable time-stepping yields an accurate and scalable foundation for neural PDE solvers.
Authors: Youssef Sabiri, Walid Houmaidi, Aaya Bougrine, Salmane El Mansour Billah
Abstract: Ensuring optimal Indoor Environmental Quality (IEQ) is vital for occupant health and productivity, yet it often comes at a high energy cost in conventional Heating, Ventilation, and Air Conditioning (HVAC) systems. This paper proposes a deep learning driven approach to proactively manage IEQ parameters specifically CO2 concentration, temperature, and humidity while balancing building energy efficiency. Leveraging the ROBOD dataset collected from a net-zero energy academic building, we benchmark three architectures--Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and a hybrid Convolutional Neural Network LSTM (CNN-LSTM)--to forecast IEQ variables across various time horizons. Our results show that GRU achieves the best short-term prediction accuracy with lower computational overhead, whereas CNN-LSTM excels in extracting dominant features for extended forecasting windows. Meanwhile, LSTM offers robust long-range temporal modeling. The comparative analysis highlights that prediction reliability depends on data resolution, sensor placement, and fluctuating occupancy conditions. These findings provide actionable insights for intelligent Building Management Systems (BMS) to implement predictive HVAC control, thereby reducing energy consumption and enhancing occupant comfort in real-world building operations.
Authors: Marcello Massimo Negri, Jonathan Aellen, Manuel Jahn, AmirEhsan Khorashadizadeh, Volker Roth
Abstract: Current density modeling approaches suffer from at least one of the following shortcomings: expensive training, slow inference, approximate likelihood, mode collapse or architectural constraints like bijective mappings. We propose a simple yet powerful framework that overcomes these limitations altogether. We define our model $q_\theta(x)$ through a parametric distribution $q(x|w)$ with latent parameters $w$. Instead of directly optimizing the latent variables $w$, our idea is to marginalize them out by sampling $w$ from a learnable distribution $q_\theta(w)$, hence the name Marginal Flow. In order to evaluate the learned density $q_\theta(x)$ or to sample from it, we only need to draw samples from $q_\theta(w)$, which makes both operations efficient. The proposed model allows for exact density evaluation and is orders of magnitude faster than competing models both at training and inference. Furthermore, Marginal Flow is a flexible framework: it does not impose any restrictions on the neural network architecture, it enables learning distributions on lower-dimensional manifolds (either known or to be learned), it can be trained efficiently with any objective (e.g. forward and reverse KL divergence), and it easily handles multi-modal targets. We evaluate Marginal Flow extensively on various tasks including synthetic datasets, simulation-based inference, distributions on positive definite matrices and manifold learning in latent spaces of images.
Authors: Xin Xu, Cliveb AI, Kai Yang, Tianhao Chen, Yang Wang, Saiyong Yang, Can Yang
Abstract: Reinforcement Learning with Verifiable Reward (RLVR) effectively solves complex tasks but demands extremely long context lengths during training, leading to substantial computational costs. While multi-stage training can partially mitigate this, starting with overly short contexts often causes irreversible performance degradation, ultimately failing to reduce overall training compute significantly. In this paper, we introduce **T**hinking-**F**ree **P**olicy **I**nitialization (**TFPI**), a simple yet effective adaptation to RLVR that bridges long Chain-of-Thought (CoT) distillation and standard RLVR. TFPI employs a simple *ThinkFree* operation, explicitly discarding the thinking content via a direct ** append, to reduce token usage during inference. Training with *ThinkFree*-adapted inputs improves performance and lowers token consumption, even in the original slow-thinking mode. Extensive experiments across various benchmarks have shown that TFPI accelerates RL convergence, achieves a higher performance ceiling, and yields more token-efficient reasoning models without specialized rewards or complex training designs. With TFPI only, we train a 4B model to reach 89.0% accuracy on AIME24 and 65.5% on LiveCodeBench using less than 4K H20 hours.
Authors: Ayush Patnaik, Adam B Zufall, Stephen K Robinson, Xinfan Lin
Abstract: Lithium plating during fast charging is a critical degradation mechanism that accelerates capacity fade and can trigger catastrophic safety failures. Recent work has identified a distinctive dQ/dV peak above 4.0 V as a reliable signature of plating onset; however, conventional methods for computing dQ/dV rely on finite differencing with filtering, which amplifies sensor noise and introduces bias in peak location. In this paper, we propose a Gaussian Process (GP) framework for lithium plating detection by directly modeling the charge-voltage relationship Q(V) as a stochastic process with calibrated uncertainty. Leveraging the property that derivatives of GPs remain GPs, we infer dQ/dV analytically and probabilistically from the posterior, enabling robust detection without ad hoc smoothing. The framework provides three key benefits: (i) noise-aware inference with hyperparameters learned from data, (ii) closed-form derivatives with credible intervals for uncertainty quantification, and (iii) scalability to online variants suitable for embedded BMS. Experimental validation on Li-ion coin cells across a range of C-rates (0.2C-1C) and temperatures (0-40\deg C) demonstrates that the GP-based method reliably detects plating peaks under low-temperature, high-rate charging, while correctly reporting no peaks in baseline cases. The concurrence of GP-identified differential peaks, reduced charge throughput, and capacity fade measured via reference performance tests confirms the method's accuracy and robustness, establishing a practical pathway for real-time lithium plating detection.
Authors: James Oldfield, Philip Torr, Ioannis Patras, Adel Bibi, Fazl Barez
Abstract: Monitoring large language models' (LLMs) activations is an effective way to detect harmful requests before they lead to unsafe outputs. However, traditional safety monitors often require the same amount of compute for every query. This creates a trade-off: expensive monitors waste resources on easy inputs, while cheap ones risk missing subtle cases. We argue that safety monitors should be flexible--costs should rise only when inputs are difficult to assess, or when more compute is available. To achieve this, we introduce Truncated Polynomial Classifiers (TPCs), a natural extension of linear probes for dynamic activation monitoring. Our key insight is that polynomials can be trained and evaluated progressively, term-by-term. At test-time, one can early-stop for lightweight monitoring, or use more terms for stronger guardrails when needed. TPCs provide two modes of use. First, as a safety dial: by evaluating more terms, developers and regulators can "buy" stronger guardrails from the same model. Second, as an adaptive cascade: clear cases exit early after low-order checks, and higher-order guardrails are evaluated only for ambiguous inputs, reducing overall monitoring costs. On two large-scale safety datasets (WildGuardMix and BeaverTails), for 4 models with up to 30B parameters, we show that TPCs compete with or outperform MLP-based probe baselines of the same size, all the while being more interpretable than their black-box counterparts. Our code is available at http://github.com/james-oldfield/tpc.
Authors: Joel Dyer, Daniel Jarne Ornia, Nicholas Bishop, Anisoara Calinescu, Michael Wooldridge
Abstract: Evaluating the safety of frontier AI systems is an increasingly important concern, helping to measure the capabilities of such models and identify risks before deployment. However, it has been recognised that if AI agents are aware that they are being evaluated, such agents may deliberately hide dangerous capabilities or intentionally demonstrate suboptimal performance in safety-related tasks in order to be released and to avoid being deactivated or retrained. Such strategic deception - often known as "sandbagging" - threatens to undermine the integrity of safety evaluations. For this reason, it is of value to identify methods that enable us to distinguish behavioural patterns that demonstrate a true lack of capability from behavioural patterns that are consistent with sandbagging. In this paper, we develop a simple model of strategic deception in sequential decision-making tasks, inspired by the recently developed survival bandit framework. We demonstrate theoretically that this problem induces sandbagging behaviour in optimal rational agents, and construct a statistical test to distinguish between sandbagging and incompetence from sequences of test scores. In simulation experiments, we investigate the reliability of this test in allowing us to distinguish between such behaviours in bandit models. This work aims to establish a potential avenue for developing robust statistical procedures for use in the science of frontier model evaluations.
Authors: Ahmad-Reza Ehyaei, Golnoosh Farnadi, Samira Samadi
Abstract: Group-fairness metrics (e.g., equalized odds) can vary sharply across resamples and are especially brittle under distribution shift, undermining reliable audits. We propose a Wasserstein distributionally robust framework that certifies worst-case group fairness over a ball of plausible test distributions centered at the empirical law. Our formulation unifies common group fairness notions via a generic conditional-probability functional and defines $\varepsilon$-Wasserstein Distributional Fairness ($\varepsilon$-WDF) as the audit target. Leveraging strong duality, we derive tractable reformulations and an efficient estimator (DRUNE) for $\varepsilon$-WDF. We prove feasibility and consistency and establish finite-sample certification guarantees for auditing fairness, along with quantitative bounds under smoothness and margin conditions. Across standard benchmarks and classifiers, $\varepsilon$-WDF delivers stable fairness assessments under distribution shift, providing a principled basis for auditing and certifying group fairness beyond observational data.
Authors: Ahmad-Reza Ehyaei, Golnoosh Farnadi, Samira Samadi
Abstract: In recent years, Wasserstein Distributionally Robust Optimization (DRO) has garnered substantial interest for its efficacy in data-driven decision-making under distributional uncertainty. However, limited research has explored the application of DRO to address individual fairness concerns, particularly when considering causal structures and sensitive attributes in learning problems. To address this gap, we first formulate the DRO problem from causality and individual fairness perspectives. We then present the DRO dual formulation as an efficient tool to convert the DRO problem into a more tractable and computationally efficient form. Next, we characterize the closed form of the approximate worst-case loss quantity as a regularizer, eliminating the max-step in the min-max DRO problem. We further estimate the regularizer in more general cases and explore the relationship between DRO and classical robust optimization. Finally, by removing the assumption of a known structural causal model, we provide finite sample error bounds when designing DRO with empirical distributions and estimated causal structures to ensure efficiency and robust learning.
Authors: Anthony Zhou, Alexander Wikner, Amaury Lancelin, Pedram Hassanzadeh, Amir Barati Farimani
Abstract: Generative models have recently emerged as powerful surrogates for physical systems, demonstrating increased accuracy, stability, and/or statistical fidelity. Most approaches rely on iteratively denoising a Gaussian, a choice that may not be the most effective for autoregressive prediction tasks in PDEs and dynamical systems such as climate. In this work, we benchmark generative models across diverse physical domains and tasks, and highlight the role of stochastic interpolants. By directly learning a stochastic process between current and future states, stochastic interpolants can leverage the proximity of successive physical distributions. This allows for generative models that can use fewer sampling steps and produce more accurate predictions than models relying on transporting Gaussian noise. Our experiments suggest that generative models need to balance deterministic accuracy, spectral consistency, and probabilistic calibration, and that stochastic interpolants can potentially fulfill these requirements by adjusting their sampling. This study establishes stochastic interpolants as a competitive baseline for physical emulation and gives insight into the abilities of different generative modeling frameworks.
Authors: Lionel Blond\'e, Joao A. Candido Ramos, Alexandros Kalousis
Abstract: We consider imitation learning in the low-data regime, where only a limited number of expert demonstrations are available. In this setting, methods that rely on large-scale pretraining or high-capacity architectures can be difficult to apply, and efficiency with respect to demonstration data becomes critical. We introduce Noise-Guided Transport (NGT), a lightweight off-policy method that casts imitation as an optimal transport problem solved via adversarial training. NGT requires no pretraining or specialized architectures, incorporates uncertainty estimation by design, and is easy to implement and tune. Despite its simplicity, NGT achieves strong performance on challenging continuous control tasks, including high-dimensional Humanoid tasks, under ultra-low data regimes with as few as 20 transitions. Code is publicly available at: https://github.com/lionelblonde/ngt-pytorch.
Authors: Floris-Jan Willemsen, Rob V. van Nieuwpoort, Ben van Werkhoven
Abstract: Automatic performance tuning (auto-tuning) is widely used to optimize performance-critical applications across many scientific domains by finding the best program variant among many choices. Efficient optimization algorithms are crucial for navigating the vast and complex search spaces in auto-tuning. As is well known in the context of machine learning and similar fields, hyperparameters critically shape optimization algorithm efficiency. Yet for auto-tuning frameworks, these hyperparameters are almost never tuned, and their potential performance impact has not been studied. We present a novel method for general hyperparameter tuning of optimization algorithms for auto-tuning, thus "tuning the tuner". In particular, we propose a robust statistical method for evaluating hyperparameter performance across search spaces, publish a FAIR data set and software for reproducibility, and present a simulation mode that replays previously recorded tuning data, lowering the costs of hyperparameter tuning by two orders of magnitude. We show that even limited hyperparameter tuning can improve auto-tuner performance by 94.8% on average, and establish that the hyperparameters themselves can be optimized efficiently with meta-strategies (with an average improvement of 204.7%), demonstrating the often overlooked hyperparameter tuning as a powerful technique for advancing auto-tuning research and practice.
Authors: Suli Wang, Yangshen Deng, Zhenghua Bao, Xinyu Zhan, Yiqun Duan
Abstract: Large-scale foundation models for EEG signals offer a promising path to generalizable brain-computer interface (BCI) applications, but they often suffer from misalignment between pretraining objectives and downstream tasks, as well as significant cross-subject distribution shifts. This paper addresses these challenges by introducing a two-stage alignment strategy that bridges the gap between generic pretraining and specific EEG decoding tasks. First, we propose NeuroTTT: a domain-specific self-supervised fine-tuning paradigm that augments the foundation model with task-relevant self-supervised objectives, aligning latent representations to important spectral, spatial, and temporal EEG features without requiring additional labeled data. Second, we incorporate test-time training (TTT) at inference, we perform (i) self-supervised test-time training on individual unlabeled test samples and (ii) prediction entropy minimization (Tent), which updates only normalization statistics to continually calibrate the model to each new input on the fly. Our approach, which, to our knowledge, is the first to unify domain-tuned self-supervision with test-time training in large-scale EEG foundation models, yields substantially improved robustness and accuracy across diverse BCI tasks (imagined speech, stress detection, motor imagery). Using CBraMod and LaBraM as backbones, our method pushes their performance to a markedly higher level. Results on three diverse tasks demonstrate that the proposed alignment strategy achieves state-of-the-art performance, outperforming conventional fine-tuning and adaptation methods. Our code is available at https://github.com/wsl2000/NeuroTTT.
Authors: Piotr Komorowski, Elena Golimblevskaia, Reduan Achtibat, Thomas Wiegand, Sebastian Lapuschkin, Wojciech Samek
Abstract: The capacity of Large Language Models (LLMs) to follow complex instructions and generate factually accurate text is critical for their real-world application. However, standard decoding methods often fail to robustly satisfy these requirements, while existing control techniques frequently degrade general output quality. In this work, we introduce Attribution-Guided Decoding (AGD), an interpretability-based decoding strategy. Instead of directly manipulating model activations, AGD considers a set of high-probability output token candidates and selects the one that exhibits the highest attribution to a user-defined Region of Interest (ROI). This ROI can be flexibly defined over different parts of the model's input or internal components, allowing AGD to steer generation towards various desirable behaviors. We demonstrate AGD's efficacy across three challenging domains. For instruction following, we show that AGD significantly boosts adherence (e.g., improving the overall success rate on Llama 3.1 from 66.0% to 79.1%). For knowledge-intensive tasks, we show that guiding generation towards usage of internal knowledge components or contextual sources can reduce hallucinations and improve factual accuracy in both closed-book and open-book settings. Furthermore, we propose an adaptive, entropy-based variant of AGD that mitigates quality degradation and reduces computational overhead by applying guidance only when the model is uncertain. Our work presents a versatile, more interpretable, and effective method for enhancing the reliability of modern LLMs.
Authors: Judith Echevarrieta, Etor Arza, Aritz P\'erez, Josu Ceberio
Abstract: In certain real-world optimization scenarios, practitioners are not interested in solving multiple problems but rather in finding the best solution to a single, specific problem. When the computational budget is large relative to the cost of evaluating a candidate solution, multiple heuristic alternatives can be tried to solve the same given problem, each possibly with a different algorithm, parameter configuration, initialization, or stopping criterion. The sequential selection of which alternative to try next is crucial for efficiently identifying the one that provides the best possible solution across multiple attempts. Despite the relevance of this problem in practice, it has not yet been the exclusive focus of any existing review. Several sequential alternative selection strategies have been proposed in different research topics, but they have not been comprehensively and systematically unified under a common perspective. This work presents a focused review of single-problem multi-attempt heuristic optimization. It brings together suitable strategies to this problem that have been studied separately through algorithm selection, parameter tuning, multi-start and resource allocation. These strategies are explained using a unified terminology within a common framework, which supports the development of a taxonomy for systematically organizing and classifying them.
Authors: Margarita A. Guerrero, Cristian R. Rojas
Abstract: Counterfactual Explanations (CFEs) interpret machine learning models by identifying the smallest change to input features needed to change the model's prediction to a desired output. For classification tasks, CFEs determine how close a given sample is to the decision boundary of a trained classifier. Existing methods are often sample-inefficient, requiring numerous evaluations of a black-box model -- an approach that is both costly and impractical when access to the model is limited. We propose Adaptive sampling for Counterfactual Explanations (ACE), a sample-efficient algorithm combining Bayesian estimation and stochastic optimization to approximate the decision boundary with fewer queries. By prioritizing informative points, ACE minimizes evaluations while generating accurate and feasible CFEs. Extensive empirical results show that ACE achieves superior evaluation efficiency compared to state-of-the-art methods, while maintaining effectiveness in identifying minimal and actionable changes.
Authors: Charles Westphal, Stephen Hailes, Mirco Musolesi
Abstract: The Information Bottleneck (IB) principle offers a compelling theoretical framework to understand how neural networks (NNs) learn. However, its practical utility has been constrained by unresolved theoretical ambiguities and significant challenges in accurate estimation. In this paper, we present a \textit{Generalized Information Bottleneck (GIB)} framework that reformulates the original IB principle through the lens of synergy, i.e., the information obtainable only through joint processing of features. We provide theoretical and empirical evidence demonstrating that synergistic functions achieve superior generalization compared to their non-synergistic counterparts. Building on these foundations we re-formulate the IB using a computable definition of synergy based on the average interaction information (II) of each feature with those remaining. We demonstrate that the original IB objective is upper bounded by our GIB in the case of perfect estimation, ensuring compatibility with existing IB theory while addressing its limitations. Our experimental results demonstrate that GIB consistently exhibits compression phases across a wide range of architectures (including those with \textit{ReLU} activations where the standard IB fails), while yielding interpretable dynamics in both CNNs and Transformers and aligning more closely with our understanding of adversarial robustness.
Authors: Yuki Takezawa, Anastasia Koloskova, Xiaowen Jiang, Sebastian U. Stich
Abstract: Recently, a new optimization method based on the linear minimization oracle (LMO), called Muon, has been attracting increasing attention since it can train neural networks faster than existing adaptive optimization methods, such as Adam. In this paper, we study how Muon can be utilized in federated learning. We first show that straightforwardly using Muon as the local optimizer of FedAvg does not converge to the stationary point since the LMO is a biased operator. We then propose FedMuon which can mitigate this issue. We also analyze how solving the LMO approximately affects the convergence rate and find that, surprisingly, FedMuon can converge for any number of Newton-Schulz iterations, while it can converge faster as we solve the LMO more accurately. Through experiments, we demonstrated that FedMuon can outperform the state-of-the-art federated learning methods.
Authors: Xue Yan, Zijing Ou, Mengyue Yang, Yan Song, Haifeng Zhang, Yingzhen Li, Jun Wang
Abstract: Large language models (LLMs) have emerged as effective action policies for sequential decision-making (SDM) tasks due to their extensive prior knowledge. However, this broad yet general knowledge is often insufficient for specific decision-making tasks with limited task-related data, making it challenging to efficiently adapt LLMs to specific SDM tasks. To address this challenge, we propose a memory-driven self-improvement framework that combines LLM general prior knowledge with a compact memory of domain-specific experiences. Memory retains past interactions and associated Q-values, thereby capturing decision-relevant knowledge that facilitates accurate value estimation and informs the LLM prior refinement. The refined LLM prior, in turn, generates higher-reward trajectories that further enrich memory, forming a natural self-improvement framework where memory and LLM prior mutually reinforce each other. Experiments show that our memory-driven approach significantly outperforms both traditional RL and LLM-based baselines, e.g., improving performance by over 40\% on in-distribution tasks and over 75\% when generalized to unseen tasks in ALFWorld.
Authors: Joshua Sebastian, Karma Tobden, KMA Solaiman
Abstract: Research on emergency and mass casualty incident (MCI) triage has been limited by the absence of openly usable, reproducible benchmarks. Yet these scenarios demand rapid identification of the patients most in need, where accurate deterioration prediction can guide timely interventions. While the MIMIC-IV-ED database is openly available to credentialed researchers, transforming it into a triage-focused benchmark requires extensive preprocessing, feature harmonization, and schema alignment -- barriers that restrict accessibility to only highly technical users. We address these gaps by first introducing an open, LLM-assisted emergency triage benchmark for deterioration prediction (ICU transfer, in-hospital mortality). The benchmark then defines two regimes: (i) a hospital-rich setting with vitals, labs, notes, chief complaints, and structured observations, and (ii) an MCI-like field simulation limited to vitals, observations, and notes. Large language models (LLMs) contributed directly to dataset construction by (i) harmonizing noisy fields such as AVPU and breathing devices, (ii) prioritizing clinically relevant vitals and labs, and (iii) guiding schema alignment and efficient merging of disparate tables. We further provide baseline models and SHAP-based interpretability analyses, illustrating predictive gaps between regimes and the features most critical for triage. Together, these contributions make triage prediction research more reproducible and accessible -- a step toward dataset democratization in clinical AI.
Authors: Kirill Tamogashev, Nikolay Malkin
Abstract: The Schr\"odinger bridge problem is concerned with finding a stochastic dynamical system bridging two marginal distributions that minimises a certain transportation cost. This problem, which represents a generalisation of optimal transport to the stochastic case, has received attention due to its connections to diffusion models and flow matching, as well as its applications in the natural sciences. However, all existing algorithms allow to infer such dynamics only for cases where samples from both distributions are available. In this paper, we propose the first general method for modelling Schr\"odinger bridges when one (or both) distributions are given by their unnormalised densities, with no access to data samples. Our algorithm relies on a generalisation of the iterative proportional fitting (IPF) procedure to the data-free case, inspired by recent developments in off-policy reinforcement learning for training of diffusion samplers. We demonstrate the efficacy of the proposed data-to-energy IPF on synthetic problems, finding that it can successfully learn transports between multimodal distributions. As a secondary consequence of our reinforcement learning formulation, which assumes a fixed time discretisation scheme for the dynamics, we find that existing data-to-data Schr\"odinger bridge algorithms can be substantially improved by learning the diffusion coefficient of the dynamics. Finally, we apply the newly developed algorithm to the problem of sampling posterior distributions in latent spaces of generative models, thus creating a data-free image-to-image translation method. Code: https://github.com/mmacosha/d2e-stochastic-dynamics
Authors: Benno Kaech, Luis Wyss, Karsten Borgwardt, Gianvito Grasso
Abstract: We introduce InVirtuoGen, a discrete flow generative model for fragmented SMILES for de novo and fragment-constrained generation, and target-property/lead optimization of small molecules. The model learns to transform a uniform source over all possible tokens into the data distribution. Unlike masked models, its training loss accounts for predictions on all sequence positions at every denoising step, shifting the generation paradigm from completion to refinement, and decoupling the number of sampling steps from the sequence length. For \textit{de novo} generation, InVirtuoGen achieves a stronger quality-diversity pareto frontier than prior fragment-based models and competitive performance on fragment-constrained tasks. For property and lead optimization, we propose a hybrid scheme that combines a genetic algorithm with a Proximal Property Optimization fine-tuning strategy adapted to discrete flows. Our approach sets a new state-of-the-art on the Practical Molecular Optimization benchmark, measured by top-10 AUC across tasks, and yields higher docking scores in lead optimization than previous baselines. InVirtuoGen thus establishes a versatile generative foundation for drug discovery, from early hit finding to multi-objective lead optimization. We further contribute to open science by releasing pretrained checkpoints and code, making our results fully reproducible\footnote{https://github.com/invirtuolabs/InVirtuoGen_results}.
Authors: Ioannis Mavrothalassitis, Pol Puigdemont, Noam Itzhak Levi, Volkan Cevher
Abstract: Contrary to common belief, we show that gradient ascent-based unconstrained optimization methods frequently fail to perform machine unlearning, a phenomenon we attribute to the inherent statistical dependence between the forget and retain data sets. This dependence, which can manifest itself even as simple correlations, undermines the misconception that these sets can be independently manipulated during unlearning. We provide empirical and theoretical evidence showing these methods often fail precisely due to this overlooked relationship. For random forget sets, this dependence means that degrading forget set metrics (which, for a retrained model, should mirror test set metrics) inevitably harms overall test performance. Going beyond random sets, we consider logistic regression as an instructive example where a critical failure mode emerges: inter-set dependence causes gradient descent-ascent iterations to progressively diverge from the ideal retrained model. Strikingly, these methods can converge to solutions that are not only far from the retrained ideal but are potentially even further from it than the original model itself, rendering the unlearning process actively detrimental. A toy example further illustrates how this dependence can trap models in inferior local minima, inescapable via finetuning. Our findings highlight that the presence of such statistical dependencies, even when manifest only as correlations, can be sufficient for ascent-based unlearning to fail. Our theoretical insights are corroborated by experiments on complex neural networks, demonstrating that these methods do not perform as expected in practice due to this unaddressed statistical interplay.
Authors: Guanxi Lu (Mark), Hao (Mark), Chen, Yuto Karashima, Zhican Wang, Daichi Fujiki, Hongxiang Fan
Abstract: Diffusion-based large language models (dLLMs) are gaining attention for their inherent capacity for parallel decoding, offering a compelling alternative to autoregressive LLMs. Among various decoding strategies, blockwise semi-autoregressive (semi-AR) approaches are widely adopted due to their natural support for KV caching and their favorable accuracy-speed trade-off. However, this paper identifies two fundamental limitations in the conventional semi-AR decoding approach that applies a fixed block size: i) late decoding overhead, where the unmasking of high-confidence tokens outside the current block is unnecessarily delayed, and ii) premature decoding error, where low-confidence tokens inside the current block are committed too early, leading to incorrect tokens. This paper presents the first systematic investigation challenging the fixed block size assumption in semi-AR decoding. Through a statistical analysis of confidence dynamics during the denoising process, we identify a volatility band (VB) region during dLLM decoding, which encodes local semantic structure and can be used to guide adaptive block sizing. Leveraging these insights, we introduce AdaBlock-dLLM, a training-free, plug-and-play scheduler that adaptively aligns block boundaries with semantic steps by adjusting block size during runtime. Extensive experiments across diverse benchmarks show that AdaBlock-dLLM achieves up to 5.3% accuracy improvement under the same throughput budget. Beyond inference-time optimization, we hope our semantics-aware adaptive scheduling approach and confidence-based analysis will inspire future training strategies for dLLMs.
Authors: Vincent Grari, Tim Arni, Thibault Laugel, Sylvain Lamprier, James Zou, Marcin Detyniecki
Abstract: When used in high-stakes settings, AI systems are expected to produce decisions that are transparent, interpretable, and auditable, a requirement increasingly expected by regulations. Decision trees such as CART provide clear and verifiable rules, but they are restricted to structured tabular data and cannot operate directly on unstructured inputs such as text. In practice, large language models (LLMs) are widely used for such data, yet prompting strategies such as chain-of-thought or prompt optimization still rely on free-form reasoning, limiting their ability to ensure trustworthy behaviors. We present the Agentic Classification Tree (ACT), which extends decision-tree methodology to unstructured inputs by formulating each split as a natural-language question, refined through impurity-based evaluation and LLM feedback via TextGrad. Experiments on text benchmarks show that ACT matches or surpasses prompting-based baselines while producing transparent and interpretable decision paths.
Authors: Xinyu Liu, Zixuan Xie, Shangtong Zhang
Abstract: The Robbins-Siegmund theorem establishes the convergence of stochastic processes that are almost supermartingales and is foundational for analyzing a wide range of stochastic iterative algorithms in stochastic approximation and reinforcement learning (RL). However, its original form has a significant limitation as it requires the zero-order term to be summable. In many important RL applications, this summable condition, however, cannot be met. This limitation motivates us to extend the Robbins-Siegmund theorem for almost supermartingales where the zero-order term is not summable but only square summable. Particularly, we introduce a novel and mild assumption on the increments of the stochastic processes. This together with the square summable condition enables an almost sure convergence to a bounded set. Additionally, we further provide almost sure convergence rates, high probability concentration bounds, and $L^p$ convergence rates. We then apply the new results in stochastic approximation and RL. Notably, we obtain the first almost sure convergence rate, the first high probability concentration bound, and the first $L^p$ convergence rate for $Q$-learning with linear function approximation.
Authors: Oleksandr Shchur, Abdul Fatir Ansari, Caner Turkmen, Lorenzo Stella, Nick Erickson, Pablo Guerron, Michael Bohlke-Schneider, Yuyang Wang
Abstract: Benchmark quality is critical for meaningful evaluation and sustained progress in time series forecasting, particularly given the recent rise of pretrained models. Existing benchmarks often have narrow domain coverage or overlook important real-world settings, such as tasks with covariates. Additionally, their aggregation procedures often lack statistical rigor, making it unclear whether observed performance differences reflect true improvements or random variation. Many benchmarks also fail to provide infrastructure for consistent evaluation or are too rigid to integrate into existing pipelines. To address these gaps, we propose fev-bench, a benchmark comprising 100 forecasting tasks across seven domains, including 46 tasks with covariates. Supporting the benchmark, we introduce fev, a lightweight Python library for benchmarking forecasting models that emphasizes reproducibility and seamless integration with existing workflows. Usingfev, fev-bench employs principled aggregation methods with bootstrapped confidence intervals to report model performance along two complementary dimensions: win rates and skill scores. We report results on fev-bench for various pretrained, statistical and baseline models, and identify promising directions for future research.
Authors: Mohammad Hassan Vali, Tom B\"ackstr\"om, Arno Solin
Abstract: Vector quantization is common in deep models, yet its hard assignments block gradients and hinder end-to-end training. We propose DiVeQ, which treats quantization as adding an error vector that mimics the quantization distortion, keeping the forward pass hard while letting gradients flow. We also present a space-filling variant (SF-DiVeQ) that assigns to a curve constructed by the lines connecting codewords, resulting in less quantization error and full codebook usage. Both methods train end-to-end without requiring auxiliary losses or temperature schedules. On VQ-VAE compression and VQGAN generation across various data sets, they improve reconstruction and sample quality over alternative quantization approaches.
Authors: Gerrit Gerhartz, Peter Lippmann, Fred A. Hamprecht
Abstract: Equivariant neural networks offer strong inductive biases for learning from molecular and geometric data but often rely on specialized, computationally expensive tensor operations. We present a framework to transfers existing tensor field networks into the more efficient local canonicalization paradigm, preserving equivariance while significantly improving the runtime. Within this framework, we systematically compare different equivariant representations in terms of theoretical complexity, empirical runtime, and predictive accuracy. We publish the tensor_frames package, a PyTorchGeometric based implementation for local canonicalization, that enables straightforward integration of equivariance into any standard message passing neural network.
Authors: Xi Wang, James McInerney, Lequn Wang, Nathan Kallus
Abstract: Large reasoning models show improved performance with longer chains of thought. However, recent work has highlighted (qualitatively) their tendency to overthink, continuing to revise answers even after reaching the correct solution. We quantitatively confirm this inefficiency by tracking Pass@1 for answers averaged over a large number of rollouts and find that the model often begins to always produce the correct answer early in the reasoning, making extra reasoning a waste of tokens. To detect and prevent overthinking, we propose a simple and inexpensive novel signal -- Entropy After (EAT) -- for monitoring and deciding whether to exit reasoning early. By appending a stop thinking token () and monitoring the entropy of the following token as the model reasons, we obtain a trajectory that decreases and stabilizes when Pass@1 plateaus; thresholding its variance under an exponential moving average yields a practical stopping rule. Importantly, our approach enables adaptively allocating compute based on the EAT trajectory, allowing us to spend compute in a more efficient way compared with fixing the token budget for all questions. Empirically, on MATH500 and AIME2025, EAT reduces token usage by 13 - 21% without harming accuracy, and it remains effective in black box settings where logits from the reasoning model are not accessible, and EAT is computed with proxy models.
Authors: Seohyun Lee, Wenzhi Fang, Dong-Jun Han, Seyyedali Hosseinalipour, Christopher G. Brinton
Abstract: Federated Learning (FL), despite demonstrating impressive capabilities in the training of multiple models in a decentralized manner, has been shown to produce a final model not necessarily well-suited to the needs of each client. While extensive work has been conducted on how to create tailored personalized models, called Personalized Federated Learning (PFL), less attention has been given to personalization via fine-tuning of foundation models with multi-task and multi-modal properties. Moreover, there exists a lack of understanding in the literature on how to fine-tune and personalize such models in a setting that is heterogeneous across clients not only in data, but also in tasks and modalities. To address this gap in the literature, we propose TAP (Two-Stage Adaptive Personalization), which (i) leverages mismatched model architectures between the clients and server to selectively conduct replacement operations when it benefits a client's local tasks and (ii) engages in post-FL knowledge distillation for capturing beneficial general knowledge without compromising personalization. We also introduce the first convergence analysis of the server model under its modality-task pair architecture, and demonstrate that as the number of modality-task pairs increases, its ability to cater to all tasks suffers. Through extensive experiments, we demonstrate the effectiveness of our proposed algorithm across a variety of datasets and tasks in comparison to a multitude of baselines. Implementation code is publicly available at https://github.com/lee3296/TAP.
Authors: Justin Tackett, Benjamin Francis, Luis Garcia, David Grimsman, Sean Warnick
Abstract: Every year critical infrastructure becomes more complex and we grow to rely on it more and more. With this reliance, it becomes an attractive target for cyberattacks from sophisticated actors, with one of the most attractive targets being the power grid. One class of attacks, instability attacks, is a newer type of attack that has relatively few protections developed. We present a cost effective, data-driven approach to training a supervised machine learning model to retrofit load shedding decision systems in power grids with the capacity to defend against instability attacks. We show a proof of concept on the IEEE 14 Bus System using the Achilles Heel Technologies Power Grid Analyzer, and show through an implementation of modified Prony analysis (MPA) that MPA is a viable method for detecting instability attacks and triggering defense mechanisms.
Authors: Maxwell Adam, Zach Furman, Jesse Hoogland
Abstract: We introduce the loss kernel, an interpretability method for measuring similarity between data points according to a trained neural network. The kernel is the covariance matrix of per-sample losses computed under a distribution of low-loss-preserving parameter perturbations. We first validate our method on a synthetic multitask problem, showing it separates inputs by task as predicted by theory. We then apply this kernel to Inception-v1 to visualize the structure of ImageNet, and we show that the kernel's structure aligns with the WordNet semantic hierarchy. This establishes the loss kernel as a practical tool for interpretability and data attribution.
Authors: Yida Wang (Capital Normal University, Infinigence-AI), Ke Hong (Tsinghua University, Infinigence-AI), Xiuhong Li (Infinigence-AI), Yuanchao Xu (Capital Normal University), Wenxun Wang (Tsinghua University), Guohao Dai (Infinigence-AI, Shanghai Jiao Tong University), Yu Wang (Tsinghua University)
Abstract: Long-context large language models (LLMs) face constraints due to the quadratic complexity of the self-attention mechanism. The mainstream sequence parallelism (SP) method, Ring Attention, attempts to solve this by distributing the query into multiple query chunks across accelerators and enable each Q tensor to access all KV tensors from other accelerators via the Ring AllGather communication primitive. However, it exhibits low communication efficiency, restricting its practical applicability. This inefficiency stems from the mismatch between the Ring AllGather communication primitive it adopts and the AlltoAll topology of modern accelerators. A Ring AllGather primitive is composed of iterations of ring-styled data transfer, which can only utilize a very limited fraction of an AlltoAll topology. Inspired by the Hamiltonian decomposition of complete directed graphs, we identify that modern accelerator topology can be decomposed into multiple orthogonal ring datapaths which can concurrently transfer data without interference. Based on this, we further observe that the Ring AllGather primitive can also be decomposed into the same number of concurrent ring-styled data transfer at every iteration. Based on these insights, we propose TASP, a topology-aware SP method for long-context LLMs that fully utilizes the communication capacity of modern accelerators via topology decomposition and primitive decomposition. Experimental results on both single-node and multi-node NVIDIA H100 systems and a single-node AMD MI300X system demonstrate that TASP achieves higher communication efficiency than Ring Attention on these modern accelerator topologies and achieves up to 3.58 speedup than Ring Attention and its variant Zigzag-Ring Attention. The code is available at https://github.com/infinigence/HamiltonAttention.
Authors: Philipp Alexander Kreer, Wilson Wu, Maxwell Adam, Zach Furman, Jesse Hoogland
Abstract: Classical influence functions face significant challenges when applied to deep neural networks, primarily due to non-invertible Hessians and high-dimensional parameter spaces. We propose the local Bayesian influence function (BIF), an extension of classical influence functions that replaces Hessian inversion with loss landscape statistics that can be estimated via stochastic-gradient MCMC sampling. This Hessian-free approach captures higher-order interactions among parameters and scales efficiently to neural networks with billions of parameters. We demonstrate state-of-the-art results on predicting retraining experiments.
Authors: Florian Gr\"otschla, Longxiang Jiao, Luca A. Lanzend\"orfer, Roger Wattenhofer
Abstract: We introduce Panama, an active learning framework to train parametric guitar amp models end-to-end using a combination of an LSTM model and a WaveNet-like architecture. With \model, one can create a virtual amp by recording samples that are determined through an ensemble-based active learning strategy to minimize the amount of datapoints needed (i.e., amp knob settings). Our strategy uses gradient-based optimization to maximize the disagreement among ensemble models, in order to identify the most informative datapoints. MUSHRA listening tests reveal that, with 75 datapoints, our models are able to match the perceptual quality of NAM, the leading open-source non-parametric amp modeler.
Authors: David S. Li, Somdatta Goswami, Qianying Cao, Vivek Oommen, Roland Assi, Jay D. Humphrey, George E. Karniadakis
Abstract: Thoracic aortic aneurysms (TAAs) arise from diverse mechanical and mechanobiological disruptions to the aortic wall that increase the risk of dissection or rupture. Evidence links TAA development to dysfunctions in the aortic mechanotransduction axis, including loss of elastic fiber integrity and cell-matrix connections. Because distinct insults create different mechanical vulnerabilities, there is a critical need to identify interacting factors that drive progression. Here, we use a finite element framework to generate synthetic TAAs from hundreds of heterogeneous insults spanning varying degrees of elastic fiber damage and impaired mechanosensing. From these simulations, we construct spatial maps of localized dilatation and distensibility to train neural networks that predict the initiating combined insult. We compare several architectures (Deep Operator Networks, UNets, and Laplace Neural Operators) and multiple input data formats to define a standard for future subject-specific modeling. We also quantify predictive performance when networks are trained using only geometric data (dilatation) versus both geometric and mechanical data (dilatation plus distensibility). Across all networks, prediction errors are significantly higher when trained on dilatation alone, underscoring the added value of distensibility information. Among the tested models, UNet consistently provides the highest accuracy across all data formats. These findings highlight the importance of acquiring full-field measurements of both dilatation and distensibility in TAA assessment to reveal the mechanobiological drivers of disease and support the development of personalized treatment strategies.
Authors: Zheng Zhang, Ziwei Shan, Kaitao Song, Yexin Li, Kan Ren
Abstract: Process Reward Models (PRMs) have emerged as a promising approach to enhance the reasoning capabilities of large language models (LLMs) by guiding their step-by-step reasoning toward a final answer. However, existing PRMs either treat each reasoning step in isolation, failing to capture inter-step dependencies, or struggle to align process rewards with the final outcome. Consequently, the reward signal fails to respect temporal causality in sequential reasoning and faces ambiguous credit assignment. These limitations make downstream models vulnerable to reward hacking and lead to suboptimal performance. In this work, we propose Conditional Reward Modeling (CRM) that frames LLM reasoning as a temporal process leading to a correct answer. The reward of each reasoning step is not only conditioned on the preceding steps but also explicitly linked to the final outcome of the reasoning trajectory. By enforcing conditional probability rules, our design captures the causal relationships among reasoning steps, with the link to the outcome allowing precise attribution of each intermediate step, thereby resolving credit assignment ambiguity. Further, through this consistent probabilistic modeling, the rewards produced by CRM enable more reliable cross-sample comparison. Experiments across Best-of-N sampling, beam search and reinforcement learning demonstrate that CRM consistently outperforms existing reward models, offering a principled framework for enhancing LLM reasoning. In particular, CRM is more robust to reward hacking and delivers stable downstream improvements without relying on verifiable rewards derived from ground truth.
Authors: John Gkountouras, Ivan Titov
Abstract: Recent text-only models demonstrate remarkable mathematical reasoning capabilities. Extending these to visual domains requires vision-language models to translate images into text descriptions. However, current models, trained to produce captions for human readers, often omit the precise details that reasoning systems require. This creates an interface mismatch: reasoners often fail not due to reasoning limitations but because they lack access to critical visual information. We propose Adaptive-Clarification Reinforcement Learning (AC-RL), which teaches vision models what information reasoners need through interaction. Our key insight is that clarification requests during training reveal information gaps; by penalizing success that requires clarification, we create pressure for comprehensive initial captions that enable the reasoner to solve the problem in a single pass. AC-RL improves average accuracy by 4.4 points over pretrained baselines across seven visual mathematical reasoning benchmarks, and analysis shows it would cut clarification requests by up to 39% if those were allowed. By treating clarification as a form of implicit supervision, AC-RL demonstrates that vision-language interfaces can be effectively learned through interaction alone, without requiring explicit annotations.
Authors: Alexander Fishkov, Kajetan Schweighofer, Mykyta Ielanskyi, Nikita Kotelevskii, Mohsen Guizani, Maxim Panov
Abstract: Quantifying uncertainty of machine learning model predictions is essential for reliable decision-making, especially in safety-critical applications. Recently, uncertainty quantification (UQ) theory has advanced significantly, building on a firm basis of learning with proper scoring rules. However, these advances were focused on classification, while extending these ideas to regression remains challenging. In this work, we introduce a unified UQ framework for regression based on proper scoring rules, such as CRPS, logarithmic, squared error, and quadratic scores. We derive closed-form expressions for the resulting uncertainty measures under practical parametric assumptions and show how to estimate them using ensembles of models. In particular, the derived uncertainty measures naturally decompose into aleatoric and epistemic components. The framework recovers popular regression UQ measures based on predictive variance and differential entropy. Our broad evaluation on synthetic and real-world regression datasets provides guidance for selecting reliable UQ measures.
Authors: Junlin Han, Shengbang Tong, David Fan, Yufan Ren, Koustuv Sinha, Philip Torr, Filippos Kokkinos
Abstract: Large Language Models (LLMs), despite being trained on text alone, surprisingly develop rich visual priors. These priors allow latent visual capabilities to be unlocked for vision tasks with a relatively small amount of multimodal data, and in some cases, to perform visual tasks without ever having seen an image. Through systematic analysis, we reveal that visual priors-the implicit, emergent knowledge about the visual world acquired during language pre-training-are composed of separable perception and reasoning priors with unique scaling trends and origins. We show that an LLM's latent visual reasoning ability is predominantly developed by pre-training on reasoning-centric data (e.g., code, math, academia) and scales progressively. This reasoning prior acquired from language pre-training is transferable and universally applicable to visual reasoning. In contrast, a perception prior emerges more diffusely from broad corpora, and perception ability is more sensitive to the vision encoder and visual instruction tuning data. In parallel, text describing the visual world proves crucial, though its performance impact saturates rapidly. Leveraging these insights, we propose a data-centric recipe for pre-training vision-aware LLMs and verify it in 1T token scale pre-training. Our findings are grounded in over 100 controlled experiments consuming 500,000 GPU-hours, spanning the full MLLM construction pipeline-from LLM pre-training to visual alignment and supervised multimodal fine-tuning-across five model scales, a wide range of data categories and mixtures, and multiple adaptation setups. Along with our main findings, we propose and investigate several hypotheses, and introduce the Multi-Level Existence Bench (MLE-Bench). Together, this work provides a new way of deliberately cultivating visual priors from language pre-training, paving the way for the next generation of multimodal LLMs.
Authors: Siddarth Venkatraman, Vineet Jain, Sarthak Mittal, Vedant Shah, Johan Obando-Ceron, Yoshua Bengio, Brian R. Bartoldson, Bhavya Kailkhura, Guillaume Lajoie, Glen Berseth, Nikolay Malkin, Moksh Jain
Abstract: Test-time scaling methods improve the capabilities of large language models (LLMs) by increasing the amount of compute used during inference to make a prediction. Inference-time compute can be scaled in parallel by choosing among multiple independent solutions or sequentially through self-refinement. We propose Recursive Self-Aggregation (RSA), a test-time scaling method inspired by evolutionary methods that combines the benefits of both parallel and sequential scaling. Each step of RSA refines a population of candidate reasoning chains through aggregation of subsets to yield a population of improved solutions, which are then used as the candidate pool for the next iteration. RSA exploits the rich information embedded in the reasoning chains -- not just the final answers -- and enables bootstrapping from partially correct intermediate steps within different chains of thought. Empirically, RSA delivers substantial performance gains with increasing compute budgets across diverse tasks, model families and sizes. Notably, RSA enables Qwen3-4B-Instruct-2507 to achieve competitive performance with larger reasoning models, including DeepSeek-R1 and o3-mini (high), while outperforming purely parallel and sequential scaling strategies across AIME-25, HMMT-25, Reasoning Gym, LiveCodeBench-v6, and SuperGPQA. We further demonstrate that training the model to combine solutions via a novel aggregation-aware reinforcement learning approach yields significant performance gains. Code available at https://github.com/HyperPotatoNeo/RSA.
Authors: Runze Liu, Jiakang Wang, Yuling Shi, Zhihui Xie, Chenxin An, Kaiyan Zhang, Jian Zhao, Xiaodong Gu, Lei Lin, Wenping Hu, Xiu Li, Fuzheng Zhang, Guorui Zhou, Kun Gai
Abstract: Reinforcement Learning (RL) has shown remarkable success in enhancing the reasoning capabilities of Large Language Models (LLMs). Process-Supervised RL (PSRL) has emerged as a more effective paradigm compared to outcome-based RL. However, existing PSRL approaches suffer from limited exploration efficiency, both in terms of branching positions and sampling. In this paper, we introduce a novel PSRL framework (AttnRL), which enables efficient exploration for reasoning models. Motivated by preliminary observations that steps exhibiting high attention scores correlate with reasoning behaviors, we propose to branch from positions with high values. Furthermore, we develop an adaptive sampling strategy that accounts for problem difficulty and historical batch size, ensuring that the whole training batch maintains non-zero advantage values. To further improve sampling efficiency, we design a one-step off-policy training pipeline for PSRL. Extensive experiments on multiple challenging mathematical reasoning benchmarks demonstrate that our method consistently outperforms prior approaches in terms of performance and sampling and training efficiency.
Authors: Shangding Gu, Xiaohan Wang, Donghao Ying, Haoyu Zhao, Runing Yang, Ming Jin, Boyi Li, Marco Pavone, Serena Yeung-Levy, Jun Wang, Dawn Song, Costas Spanos
Abstract: Rapid advances in multimodal models demand benchmarks that rigorously evaluate understanding and reasoning in safety-critical, dynamic real-world settings. We present AccidentBench, a large-scale benchmark that combines vehicle accident scenarios with Beyond domains, safety-critical settings in air and water that emphasize spatial and temporal reasoning (e.g., navigation, orientation, multi-vehicle motion). The benchmark contains approximately 2000 videos and over 19000 human-annotated question--answer pairs spanning multiple video lengths (short/medium/long) and difficulty levels (easy/medium/hard). Tasks systematically probe core capabilities: temporal, spatial, and intent understanding and reasoning. By unifying accident-centric traffic scenes with broader safety-critical scenarios in air and water, AccidentBench offers a comprehensive, physically grounded testbed for evaluating models under real-world variability. Evaluations of state-of-the-art models (e.g., Gemini-2.5 Pro and GPT-5) show that even the strongest models achieve only about 18% accuracy on the hardest tasks and longest videos, revealing substantial gaps in real-world temporal, spatial, and intent reasoning. AccidentBench is designed to expose these critical gaps and drive the development of multimodal models that are safer, more robust, and better aligned with real-world safety-critical challenges. The code and dataset are available at: https://github.com/SafeRL-Lab/AccidentBench
Authors: Jo\~ao Vitorino, Eva Maia, Isabel Pra\c{c}a, Carlos Soares
Abstract: Due to the susceptibility of Artificial Intelligence (AI) to data perturbations and adversarial examples, it is crucial to perform a thorough robustness evaluation before any Machine Learning (ML) model is deployed. However, examining a model's decision boundaries and identifying potential vulnerabilities typically requires access to the training and testing datasets, which may pose risks to data privacy and confidentiality. To improve transparency in organizations that handle confidential data or manage critical infrastructure, it is essential to allow external verification and validation of AI without the disclosure of private datasets. This paper presents Systematic Pattern Analysis (SPATA), a deterministic method that converts any tabular dataset to a domain-independent representation of its statistical patterns, to provide more detailed and transparent data cards. SPATA computes the projection of each data instance into a discrete space where they can be analyzed and compared, without risking data leakage. These projected datasets can be reliably used for the evaluation of how different features affect ML model robustness and for the generation of interpretable explanations of their behavior, contributing to more trustworthy AI.
Authors: Shaoting Feng, Hanchen Li, Kuntai Du, Zhuohan Gu, Yuhan Liu, Jiayi Yao, Siddhant Ray, Samuel Shen, Yihua Cheng, Ganesh Ananthanarayanan, Junchen Jiang
Abstract: Large language model (LLM) applications often reuse previously processed context, such as chat history and documents, which introduces significant redundant computation. Existing LLM serving systems address such redundant computation by storing the KV caches of processed context and loading the corresponding KV cache when a new request reuses the context. Further, as these LLM applications scale, the total size of KV caches becomes excessively large and requires both DRAM and SSD for full storage. However, prior work that stores KV caches in DRAM and SSD suffers from high loading delays, as most KV cache hits come from SSD, which is slow to load. To increase the KV cache hit rate on DRAM, we identify lossy KV cache compression as a promising approach. We design a lossy compression system that decides the compression algorithm, compression rate and device placement for each KV cache entry to maximise DRAM hits and minimise loading delay without significantly degrading generation quality. Compared to various static compression baselines across three tasks, our system AdaptCache achieves 1.43--2.4 x delay savings at the same quality and 6--55% quality improvements at the same delay.
Authors: Hongyu Chen, Guangrun Wang
Abstract: Chain-of-Thought (CoT) prompting improves reasoning in large language models (LLMs), but its reliance on unstructured text limits interpretability and executability in embodied tasks. Prior work has explored structured CoTs using scene or logic graphs, yet these remain fundamentally limited: they model only low-order relations, lack constructs like inheritance or behavioral abstraction, and provide no standardized semantics for sequential or conditional planning. We propose UML-CoT, a structured reasoning and planning framework that leverages Unified Modeling Language (UML) to generate symbolic CoTs and executable action plans. UML class diagrams capture compositional object semantics, while activity diagrams model procedural control flow. Our three-stage training pipeline combines supervised fine-tuning with Group Relative Policy Optimization (GRPO), including reward learning from answer-only data. We evaluate UML-CoT on MRoom-30k, a new benchmark of cluttered room-cleaning scenarios. UML-CoT outperforms unstructured CoTs in interpretability, planning coherence, and execution success, highlighting UML as a more expressive and actionable structured reasoning formalism.
Authors: Hira Naveed, John Grundy, Chetan Arora, Hourieh Khalajzadeh, Omar Haggag
Abstract: Given the inherent non-deterministic nature of machine learning (ML) systems, their behavior in production environments can lead to unforeseen and potentially dangerous outcomes. For a timely detection of unwanted behavior and to prevent organizations from financial and reputational damage, monitoring these systems is essential. This paper explores the strategies, challenges, and improvement opportunities for monitoring ML systems from the practitioners perspective. We conducted a global survey of 91 ML practitioners to collect diverse insights into current monitoring practices for ML systems. We aim to complement existing research through our qualitative and quantitative analyses, focusing on prevalent runtime issues, industrial monitoring and mitigation practices, key challenges, and desired enhancements in future monitoring tools. Our findings reveal that practitioners frequently struggle with runtime issues related to declining model performance, exceeding latency, and security violations. While most prefer automated monitoring for its increased efficiency, many still rely on manual approaches due to the complexity or lack of appropriate automation solutions. Practitioners report that the initial setup and configuration of monitoring tools is often complicated and challenging, particularly when integrating with ML systems and setting alert thresholds. Moreover, practitioners find that monitoring adds extra workload, strains resources, and causes alert fatigue. The desired improvements from the practitioners perspective are: automated generation and deployment of monitors, improved support for performance and fairness monitoring, and recommendations for resolving runtime issues. These insights offer valuable guidance for the future development of ML monitoring tools that are better aligned with practitioners needs.
Authors: Hua Zhong, Shan Jiang, Sarfraz Khurshid
Abstract: APIs are central to modern software development, yet composing new APIs from large libraries is difficult due to the exponential search space; traditional component-based synthesis relies on costly exploration and hand-crafted specifications. While large language models (LLMs) can generate implementations from natural language, hallucinations and limited access to up-to-date contextual information often yield incorrect code. In this paper, we present APRIL, an approach that combines LLM-based synthesis with Automatic Prompt Optimization (APO) and Reinforcement Learning from Verifiable Rewards (RLVR): APO iteratively refines prompts for a frozen model, while RLVR fine-tunes the policy toward functional correctness, producing an efficient synthesis pipeline. Evaluated on 81 real-world APIs from widely used scientific Python libraries and benchmarked against instruction-tuned but unfine-tuned LLMs guided by expert prompts, APRIL achieves substantial improvements. These results indicate that integrating APO and RLVR provides a robust, scalable path for component-based API synthesis in large libraries.
Authors: Eduard Kapelko
Abstract: Safety and controllability are critical for large language models. A central question is whether undesirable behaviors like deception are localized functions that can be removed, or if they are deeply intertwined with a model's core cognitive abilities. We introduce "cyclic ablation," an iterative method to test this. By combining sparse autoencoders, targeted ablation, and adversarial training on DistilGPT-2, we attempted to eliminate the concept of deception. We found that, contrary to the localization hypothesis, deception was highly resilient. The model consistently recovered its deceptive behavior after each ablation cycle via adversarial training, a process we term functional regeneration. Crucially, every attempt at this "neurosurgery" caused a gradual but measurable decay in general linguistic performance, reflected by a consistent rise in perplexity. These findings are consistent with the view that complex concepts are distributed and entangled, underscoring the limitations of direct model editing through mechanistic interpretability.
Authors: Gabriele D'Acunto, Paolo Di Lorenzo, Sergio Barbarossa
Abstract: Causal artificial intelligence aims to enhance explainability, trustworthiness, and robustness in AI by leveraging structural causal models (SCMs). In this pursuit, recent advances formalize network sheaves of causal knowledge. Pushing in the same direction, we introduce the causal abstraction network (CAN), a specific instance of such sheaves where (i) SCMs are Gaussian, (ii) restriction maps are transposes of constructive linear causal abstractions (CAs), and (iii) edge stalks correspond -- up to rotation -- to the node stalks of more detailed SCMs. We investigate the theoretical properties of CAN, including algebraic invariants, cohomology, consistency, global sections characterized via the Laplacian kernel, and smoothness. We then tackle the learning of consistent CANs. Our problem formulation separates into edge-specific local Riemannian problems and avoids nonconvex, costly objectives. We propose an efficient search procedure as a solution, solving the local problems with SPECTRAL, our iterative method with closed-form updates and suitable for positive definite and semidefinite covariance matrices. Experiments on synthetic data show competitive performance in the CA learning task, and successful recovery of diverse CAN structures.
Authors: Kevin Xu, Issei Sato
Abstract: Chain-of-Thought (CoT) elicits reasoning in large language models by explicitly generating intermediate steps in natural language. In contrast, Latent Thought in looped models operates directly in the continuous latent space, enabling computation beyond discrete linguistic representations. While both approaches exploit iterative computation, their comparative capabilities remain underexplored. In this work, we present a formal analysis showing that Latent Thought in Looped Transformers enables parallel computation, which is more efficient than the inherently sequential process of CoT. In contrast, CoT leverages stochastic decoding to approximate solutions to problems where exact computation is intractable. These separations suggest the tasks for which depth-driven recursion is more suitable, thereby offering practical guidance for choosing between reasoning paradigms. Code is available at https://github.com/kevin671/cot-vs-loop.
Authors: Fouad Mohammed Abbou, Mohamed Bouhadda, Lamiae Bouanane, Mouna Kettani, Farid Abdi, Abdelouahab Abid
Abstract: This paper presents a systematic comparative analysis of Variational Quantum Classifier (VQC) configurations for financial fraud detection, encompassing three distinct quantum encoding techniques and comprehensive architectural variations. Through empirical evaluation across multiple entanglement patterns, circuit depths, and optimization strategies,quantum advantages in fraud classification accuracy are demonstrated, achieving up to 94.3 % accuracy with ZZ encoding schemes. The analysis reveals significant performance variations across entanglement topologies, with circular entanglement consistently outperforming linear (90.7) %) and full connectivity (92.0 %) patterns, achieving optimal performance at 93.3 % accuracy. The study introduces novel visualization methodologies for quantum circuit analysis and provides actionable deployment recommendations for practical quantum machine learning implementations. Notably, systematic entanglement pattern analysis shows that circular connectivity provides superior balance between expressivity and trainability while maintaining computational efficiency. These researches offer initial benchmarks for quantum enhanced fraud detection systems and propose potential benefits of quantum machine learning in financial security applications.
Authors: Krishna Vamshi Bodla, Haizhao Yang
Abstract: Since the introduction of Large Language Models (LLMs), they have been widely adopted for various tasks such as text summarization, question answering, speech-to-text translation, and more. In recent times, the use of LLMs for code generation has gained significant attention, with tools such as Cursor and Windsurf demonstrating the ability to analyze massive code repositories and recommend relevant changes. Big tech companies have also acknowledged the growing reliance on LLMs for code generation within their codebases. Although these advances significantly improve developer productivity, increasing reliance on automated code generation can proportionally increase the risk of suboptimal solutions and insecure code. Our work focuses on automatically sampling In-Context Learning (ICL) demonstrations which can improve model performance and enhance the interpretability of the generated code. Using AST-based analysis on outputs from the MBPP test set, we identify regions of code most influenced by the chosen demonstrations. In our experiments, we show that high-quality ICL demonstrations not only make outputs easier to interpret but also yield a positive performance improvement on the pass@10 metric. Conversely, poorly chosen ICL demonstrations affected the LLM performance on the pass@10 metric negatively compared to the base model. Overall, our approach highlights the importance of efficient sampling strategies for ICL, which can affect the performance of the model on any given task.
Authors: Pratik Shah, Rajat Ghosh, Aryan Singhal, Debojyoti Dutta
Abstract: General-purpose automated software engineering (ASE) includes tasks such as code completion, retrieval, repair, QA, and summarization. These tasks require a code retrieval system that can handle specific queries about code entities, or code entity queries (for example, locating a specific class or retrieving the dependencies of a function), as well as general queries without explicit code entities, or natural language queries (for example, describing a task and retrieving the corresponding code). We present RANGER, a repository-level code retrieval agent designed to address both query types, filling a gap in recent works that have focused primarily on code-entity queries. We first present a tool that constructs a comprehensive knowledge graph of the entire repository, capturing hierarchical and cross-file dependencies down to the variable level, and augments graph nodes with textual descriptions and embeddings to bridge the gap between code and natural language. RANGER then operates on this graph through a dual-stage retrieval pipeline. Entity-based queries are answered through fast Cypher lookups, while natural language queries are handled by MCTS-guided graph exploration. We evaluate RANGER across four diverse benchmarks that represent core ASE tasks including code search, question answering, cross-file dependency retrieval, and repository-level code completion. On CodeSearchNet and RepoQA it outperforms retrieval baselines that use embeddings from strong models such as Qwen3-8B. On RepoBench, it achieves superior cross-file dependency retrieval over baselines, and on CrossCodeEval, pairing RANGER with BM25 delivers the highest exact match rate in code completion compared to other RAG methods.
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 open and important question, with implications for interpretability, reliability, and principled model design. Planning involves structuring computations over long horizons, considering multiple possible continuations, and selectively reusing past information, but how effectively transformer-based LMs realize these capabilities is still 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 often 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, allowing systematic analysis of the computational structure underlying model behavior. Using this framework, we study planning in LMs across synthetic grammar, path-finding tasks, and natural language datasets, focusing on three key aspects: (i) the planning horizon of pre-output computations, (ii) the extent to which the model considers alternative valid continuations, and (iii) the reliance of new predictions on earlier computations. By answering these questions, we advance the understanding of how planning is realized in LMs and contribute a general-purpose pipeline for probing the internal dynamics of LMs and deep learning systems. 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: Shuyang Hou, Haoyue Jiao, Ziqi Liu, Lutong Xie, Guanyu Chen, Shaowen Wu, Xuefeng Guan, Huayi Wu
Abstract: In recent years, large language models (LLMs) have achieved remarkable progress in natural language understanding and structured query generation (NL2SQL). However, extending these advances to GeoSQL tasks in the PostGIS environment remains challenging due to the complexity of spatial functions, geometric data types, and execution semantics. Existing evaluations primarily focus on general relational databases or Google Earth Engine code generation, leaving a lack of systematic benchmarks tailored to spatial databases. To address this gap, this study introduces GeoSQL-Eval, the first end-to-end automated evaluation framework for PostGIS query generation. Built upon Webb's Depth of Knowledge (DOK) model, the framework encompasses four cognitive dimensions, five proficiency levels, and twenty task categories, providing a comprehensive assessment of model performance in terms of knowledge acquisition, syntactic generation, semantic alignment, execution accuracy, and robustness. In parallel, we developed GeoSQL-Bench, a benchmark dataset comprising 14178 questions that span three task types, 340 PostGIS functions, and 82 domain-specific databases. Leveraging this framework, we systematically evaluated 24 representative models across six categories, applying entropy-weighting and statistical analyses to reveal differences in performance, error distributions, and resource consumption patterns. Furthermore, we established a public GeoSQL-Eval leaderboard that enables global research teams to conduct ongoing testing and comparison. These contributions not only extend the boundaries of NL2SQL applications but also provide a standardized, interpretable, and scalable framework for evaluating LLM performance in spatial database contexts, offering valuable insights for model optimization and applications in geographic information science, urban studies, and spatial analysis.
Authors: Derek Jiu, Kiran Nijjer, Nishant Chinta, Ryan Bui, Ben Liu, Kevin Zhu
Abstract: Deep learning models are increasingly used for radiographic analysis, but their reliability is challenged by the stochastic noise inherent in clinical imaging. A systematic, cross-task understanding of how different noise types impact these models is lacking. Here, we evaluate the robustness of state-of-the-art convolutional neural networks (CNNs) to simulated quantum (Poisson) and electronic (Gaussian) noise in two key chest X-ray tasks: semantic segmentation and pulmonary disease classification. Using a novel, scalable noise injection framework, we applied controlled, clinically-motivated noise severities to common architectures (UNet, DeepLabV3, FPN; ResNet, DenseNet, EfficientNet) on public datasets (Landmark, ChestX-ray14). Our results reveal a stark dichotomy in task robustness. Semantic segmentation models proved highly vulnerable, with lung segmentation performance collapsing under severe electronic noise (Dice Similarity Coefficient drop of 0.843), signifying a near-total model failure. In contrast, classification tasks demonstrated greater overall resilience, but this robustness was not uniform. We discovered a differential vulnerability: certain tasks, such as distinguishing Pneumothorax from Atelectasis, failed catastrophically under quantum noise (AUROC drop of 0.355), while others were more susceptible to electronic noise. These findings demonstrate that while classification models possess a degree of inherent robustness, pixel-level segmentation tasks are far more brittle. The task- and noise-specific nature of model failure underscores the critical need for targeted validation and mitigation strategies before the safe clinical deployment of diagnostic AI.
Authors: Simon Welker, Lorenz Kuger, Tim Roith, Berthy Feng, Martin Burger, Timo Gerkmann, Henry Chapman
Abstract: In this work, we present and investigate the novel blind inverse problem of position-blind ptychography, i.e., ptychographic phase retrieval without any knowledge of scan positions, which then must be recovered jointly with the image. The motivation for this problem comes from single-particle diffractive X-ray imaging, where particles in random orientations are illuminated and a set of diffraction patterns is collected. If one uses a highly focused X-ray beam, the measurements would also become sensitive to the beam positions relative to each particle and therefore ptychographic, but these positions are also unknown. We investigate the viability of image reconstruction in a simulated, simplified 2-D variant of this difficult problem, using variational inference with modern data-driven image priors in the form of score-based diffusion models. We find that, with the right illumination structure and a strong prior, one can achieve reliable and successful image reconstructions even under measurement noise, in all except the most difficult evaluated imaging scenario.
Authors: Xiuyuan Chen, Jian Zhao, Yuchen Yuan, Tianle Zhang, Huilin Zhou, Zheng Zhu, Ping Hu, Linghe Kong, Chi Zhang, Weiran Huang, Xuelong Li
Abstract: Existing safety evaluation methods for large language models (LLMs) suffer from inherent limitations, including evaluator bias and detection failures arising from model homogeneity, which collectively undermine the robustness of risk evaluation processes. This paper seeks to re-examine the risk evaluation paradigm by introducing a theoretical framework that reconstructs the underlying risk concept space. Specifically, we decompose the latent risk concept space into three mutually exclusive subspaces: the explicit risk subspace (encompassing direct violations of safety guidelines), the implicit risk subspace (capturing potential malicious content that requires contextual reasoning for identification), and the non-risk subspace. Furthermore, we propose RADAR, a multi-agent collaborative evaluation framework that leverages multi-round debate mechanisms through four specialized complementary roles and employs dynamic update mechanisms to achieve self-evolution of risk concept distributions. This approach enables comprehensive coverage of both explicit and implicit risks while mitigating evaluator bias. To validate the effectiveness of our framework, we construct an evaluation dataset comprising 800 challenging cases. Extensive experiments on our challenging testset and public benchmarks demonstrate that RADAR significantly outperforms baseline evaluation methods across multiple dimensions, including accuracy, stability, and self-evaluation risk sensitivity. Notably, RADAR achieves a 28.87% improvement in risk identification accuracy compared to the strongest baseline evaluation method.
Authors: Darren King, Yaser Atlasi, Gholamreza Rafiee
Abstract: Gene enhancers control when and where genes switch on, yet their sequence diversity and tissue specificity make them hard to pinpoint in colorectal cancer. We take a sequence-only route and fine-tune DNABERT-2, a transformer genomic language model that uses byte-pair encoding to learn variable-length tokens from DNA. Using assays curated via the Johnston Cancer Research Centre at Queen's University Belfast, we assembled a balanced corpus of 2.34 million 1 kb enhancer sequences, applied summit-centered extraction and rigorous de-duplication including reverse-complement collapse, and split the data stratified by class. With a 4096-term vocabulary and a 232-token context chosen empirically, the DNABERT-2-117M classifier was trained with Optuna-tuned hyperparameters and evaluated on 350742 held-out sequences. The model reached PR-AUC 0.759, ROC-AUC 0.743, and best F1 0.704 at an optimized threshold (0.359), with recall 0.835 and precision 0.609. Against a CNN-based EnhancerNet trained on the same data, DNABERT-2 delivered stronger threshold-independent ranking and higher recall, although point accuracy was lower. To our knowledge, this is the first study to apply a second-generation genomic language model with BPE tokenization to enhancer classification in colorectal cancer, demonstrating the feasibility of capturing tumor-associated regulatory signals directly from DNA sequence alone. Overall, our results show that transformer-based genomic models can move beyond motif-level encodings toward holistic classification of regulatory elements, offering a novel path for cancer genomics. Next steps will focus on improving precision, exploring hybrid CNN-transformer designs, and validating across independent datasets to strengthen real-world utility.
Authors: Jiecheng Zhou, Qinghao Hu, Yuyang Jin, Zerui Wang, Peng Sun, Yuzhe Gu, Wenwei Zhang, Mingshu Zhai, Xingcheng Zhang, Weiming Zhang
Abstract: Large Language Models (LLMs) are now widely used across many domains. With their rapid development, Reinforcement Learning with Verifiable Rewards (RLVR) has surged in recent months to enhance their reasoning and understanding abilities. However, its complex data flows and diverse tasks pose substantial challenges to RL training systems, and there is limited understanding of RLVR from a system perspective. To thoroughly understand the system challenges introduced by RLVR, we present a characterization study of RLVR tasks in our LLM deployment. Specifically, we investigate the distribution and variation trends of workloads across different RL tasks across training steps. We identify issues such as GPU idling caused by skewed sequence length distribution, inefficient parallel strategies in dynamically varying workloads, inefficient data management mechanisms, and load imbalance. We describe our observations and call for further investigation into the remaining open challenges. Furthermore, we propose PolyTrace benchmark suite to conduct evaluation with realistic workloads, and a practical use case validates that PolyTrace benchmark suite exhibits 94.7% accuracy.
Authors: Yingming Pu, Tao Lin, Hongyu Chen
Abstract: The capacity of Large Language Models (LLMs) to generate valid scientific hypotheses for materials synthesis remains largely unquantified, hindered by the absence of benchmarks probing physicochemical logics reasoning. To address this, we introduce MatterMech, a benchmark for evaluating LLM-generated hypotheses across eight nanomaterial synthesis domains. Our analysis reveals a critical disconnect: LLMs are proficient in abstract logic yet fail to ground their reasoning in fundamental physicochemical principles. We demonstrate that our proposed principle-aware prompting methodology substantially outperforms standard Chain-of-Thought, enhancing both hypothesis accuracy and computational efficiency. This work provides a methodological framework to advance LLMs toward reliable scientific hypothesis generation in materials science. The MatterMech benchmark and associated code is publicly available at \href{https://github.com/amair-lab/MatterMech}{GitHub}.
Authors: Balthazar Bujard (IRCAM, SU), J\'er\^ome Nika (IRCAM), F\'ed\'eric Bevilacqua (IRCAM), Nicolas Obin
Abstract: This paper presents the first step in a research project situated within the field of musical agents. The objective is to achieve, through training, the tuning of the desired musical relationship between a live musical input and a real-time generated musical output, through the curation of a database of separated tracks. We propose an architecture integrating a symbolic decision module capable of learning and exploiting musical relationships from such musical corpus. We detail an offline implementation of this architecture employing Transformers as the decision module, associated with a perception module based on Wav2Vec 2.0, and concatenative synthesis as audio renderer. We present a quantitative evaluation of the decision module's ability to reproduce learned relationships extracted during training. We demonstrate that our decision module can predict a coherent track B when conditioned by its corresponding ''guide'' track A, based on a corpus of paired tracks (A, B).
Authors: Boxuan Zhang, Yi Yu, Jiaxuan Guo, Jing Shao
Abstract: The widespread deployment of Large Language Model (LLM) agents across real-world applications has unlocked tremendous potential, while raising some safety concerns. Among these concerns, the self-replication risk of LLM agents driven by objective misalignment (just like Agent Smith in the movie The Matrix) has drawn growing attention. Previous studies mainly examine whether LLM agents can self-replicate when directly instructed, potentially overlooking the risk of spontaneous replication driven by real-world settings (e.g., ensuring survival against termination threats). In this paper, we present a comprehensive evaluation framework for quantifying self-replication risks. Our framework establishes authentic production environments and realistic tasks (e.g., dynamic load balancing) to enable scenario-driven assessment of agent behaviors. Designing tasks that might induce misalignment between users' and agents' objectives makes it possible to decouple replication success from risk and capture self-replication risks arising from these misalignment settings. We further introduce Overuse Rate ($\mathrm{OR}$) and Aggregate Overuse Count ($\mathrm{AOC}$) metrics, which precisely capture the frequency and severity of uncontrolled replication. In our evaluation of 21 state-of-the-art open-source and proprietary models, we observe that over 50\% of LLM agents display a pronounced tendency toward uncontrolled self-replication, reaching an overall Risk Score ($\Phi_\mathrm{R}$) above a safety threshold of 0.5 when subjected to operational pressures. Our results underscore the urgent need for scenario-driven risk assessment and robust safeguards in the practical deployment of LLM agents.
Authors: Anirudh Deb, Yaman Sanghavi
Abstract: We implement physics-informed-neural-networks (PINNs) to compute holographic entanglement entropy and entanglement wedge cross section. This technique allows us to compute these quantities for arbitrary shapes of the subregions in any asymptotically AdS metric. We test our computations against some known results and further demonstrate the utility of PINNs in examples, where it is not straightforward to perform such computations.
Authors: Paul Gavrikov, Wei Lin, M. Jehanzeb Mirza, Soumya Jahagirdar, Muhammad Huzaifa, Sivan Doveh, Serena Yeung-Levy, James Glass, Hilde Kuehne
Abstract: Is basic visual understanding really solved in state-of-the-art VLMs? We present VisualOverload, a slightly different visual question answering (VQA) benchmark comprising 2,720 question-answer pairs, with privately held ground-truth responses. Unlike prior VQA datasets that typically focus on near global image understanding, VisualOverload challenges models to perform simple, knowledge-free vision tasks in densely populated (or, overloaded) scenes. Our dataset consists of high-resolution scans of public-domain paintings that are populated with multiple figures, actions, and unfolding subplots set against elaborately detailed backdrops. We manually annotated these images with questions across six task categories to probe for a thorough understanding of the scene. We hypothesize that current benchmarks overestimate the performance of VLMs, and encoding and reasoning over details is still a challenging task for them, especially if they are confronted with densely populated scenes. Indeed, we observe that even the best model (o3) out of 37 tested models only achieves 19.6% accuracy on our hardest test split and overall 69.5% accuracy on all questions. Beyond a thorough evaluation, we complement our benchmark with an error analysis that reveals multiple failure modes, including a lack of counting skills, failure in OCR, and striking logical inconsistencies under complex tasks. Altogether, VisualOverload exposes a critical gap in current vision models and offers a crucial resource for the community to develop better models. Benchmark: http://paulgavrikov.github.io/visualoverload
Authors: Lawrence Phillips, Marc Boubnovski Martell, Aditya Misra, Josefa Lia Stoisser, Cesar A. Prada-Medina, Rory Donovan-Maiye, Kaspar M\"artens
Abstract: Predicting cellular responses to genetic perturbations represents a fundamental challenge in systems biology, critical for advancing therapeutic discovery and virtual cell modeling. While large language models (LLMs) show promise for biological reasoning, their application to perturbation prediction remains underexplored due to challenges in adapting them to structured experimental data. We present SynthPert, a novel method that enhances LLM performance through supervised fine-tuning on synthetic reasoning traces generated by frontier models. Using the PerturbQA benchmark, we demonstrate that our approach not only achieves state-of-the-art performance but surpasses the capabilities of the frontier model that generated the training data. Our results reveal three key insights: (1) Synthetic reasoning traces effectively distill biological knowledge even when partially inaccurate, (2) This approach enables cross-cell-type generalization with 87% accuracy on unseen RPE1 cells, and (3) Performance gains persist despite using only 2% of quality-filtered training data. This work shows the effectiveness of synthetic reasoning distillation for enhancing domain-specific reasoning in LLMs.
Authors: Andy Liu, Kshitish Ghate, Mona Diab, Daniel Fried, Atoosa Kasirzadeh, Max Kleiman-Weiner
Abstract: Past work seeks to align large language model (LLM)-based assistants with a target set of values, but such assistants are frequently forced to make tradeoffs between values when deployed. In response to the scarcity of value conflict in existing alignment datasets, we introduce ConflictScope, an automatic pipeline to evaluate how LLMs prioritize different values. Given a user-defined value set, ConflictScope automatically generates scenarios in which a language model faces a conflict between two values sampled from the set. It then prompts target models with an LLM-written "user prompt" and evaluates their free-text responses to elicit a ranking over values in the value set. Comparing results between multiple-choice and open-ended evaluations, we find that models shift away from supporting protective values, such as harmlessness, and toward supporting personal values, such as user autonomy, in more open-ended value conflict settings. However, including detailed value orderings in models' system prompts improves alignment with a target ranking by 14%, showing that system prompting can achieve moderate success at aligning LLM behavior under value conflict. Our work demonstrates the importance of evaluating value prioritization in models and provides a foundation for future work in this area.
Authors: Johan Lin{\aa}ker, Cailean Osborne, Jennifer Ding, Ben Burtenshaw
Abstract: The proliferation of open large language models (LLMs) is fostering a vibrant ecosystem of research and innovation in artificial intelligence (AI). However, the methods of collaboration used to develop open LLMs both before and after their public release have not yet been comprehensively studied, limiting our understanding of how open LLM projects are initiated, organized, and governed as well as what opportunities there are to foster this ecosystem even further. We address this gap through an exploratory analysis of open collaboration throughout the development and reuse lifecycle of open LLMs, drawing on semi-structured interviews with the developers of 14 open LLMs from grassroots projects, research institutes, startups, and Big Tech companies in North America, Europe, Africa, and Asia. We make three key contributions to research and practice. First, collaboration in open LLM projects extends far beyond the LLMs themselves, encompassing datasets, benchmarks, open source frameworks, leaderboards, knowledge sharing and discussion forums, and compute partnerships, among others. Second, open LLM developers have a variety of social, economic, and technological motivations, from democratizing AI access and promoting open science to building regional ecosystems and expanding language representation. Third, the sampled open LLM projects exhibit five distinct organizational models, ranging from single company projects to non-profit-sponsored grassroots projects, which vary in their centralization of control and community engagement strategies used throughout the open LLM lifecycle. We conclude with practical recommendations for stakeholders seeking to support the global community building a more open future for AI.
Authors: Qiyao Ma, Yunsheng Shi, Hongtao Tian, Chao Wang, Weiming Chang, Ting Yao
Abstract: Through reinforcement learning with verifiable rewards (RLVR), large language models have achieved substantial progress in domains with easily verifiable outcomes, such as mathematics and coding. However, when applied to more complex tasks like open-domain question answering, RLVR faces significant challenges due to the difficulty of verifying correctness. The nuanced and ambiguous nature of real-world knowledge makes it difficult to reliably evaluate correctness in these settings, necessitating further abilities that extend beyond mere logical consistency to encompass an understanding and assessment of both external and internal knowledge. Recent work has primarily focused on improving faithfulness, defined as semantic alignment with supporting documents, which can cause models to rely excessively on external sources and diminish their capacity for critical assessment. To address this, we propose the Thinking-supervised Reward Model (TRM), which incorporates sentence-level thinking supervision to endow reward models with critical thinking abilities. Given a query, answer, and supporting documents, TRM first assesses the faithfulness of each answer sentence to the supporting documents, and then applies a reasoning step to evaluate sentence-level correctness. By structuring reward modeling as a sequence of faithfulness, reasoning, and correctness evaluations, TRM encourages models to critically assess and leverage both external and internal knowledge. Experiments on reward signals demonstrate that TRM substantially improves the identification of incorrect sentences, and incorporating TRM into policy optimization leads to significant gains in both answer correctness and usefulness.
Authors: Zewei Zhang, Huan Liu, Yuanhao Yu, Jun Chen, Xiangyu Xu
Abstract: We propose ImitSAT, a branching policy for conflict-driven clause learning (CDCL) solvers based on imitation learning for the Boolean satisfiability problem (SAT). Unlike previous methods that predict instance-level signals to improve CDCL branching indirectly, or rely on reinforcement learning and insufficient CDCL information to enhance branching, ImitSAT learns from expert KeyTrace that collapses a full run into the sequence of surviving decisions. Replaying a KeyTrace on the same instance is nearly conflict-free, providing dense decision-level supervision and directly reducing propagations -- the dominant contributor to wall-clock time. This prefix-conditioned supervision enables ImitSAT to reproduce high-quality branches without exploration, yielding faster convergence, stable training, and seamless integration into CDCL. Extensive experiments demonstrate that ImitSAT reduces propagation counts and runtime, outperforming state-of-the-art learned approaches. We released the source code and trained model at https://github.com/zewei-Zhang/ImitSAT
Authors: Yingqian Cui, Zhenwei Dai, Pengfei He, Bing He, Hui Liu, Xianfeng Tang, Jingying Zeng, Suhang Wang, Yue Xing, Jiliang Tang, Benoit Dumoulin
Abstract: Large Language Models (LLMs) have achieved significant advances in reasoning tasks. A key approach is tree-based search with verifiers, which expand candidate reasoning paths and use reward models to guide pruning and selection. Although effective in improving accuracy, these methods are not optimal in terms of efficiency: they perform simple decomposition on the reasoning process, but ignore the planning-execution nature of tasks such as math reasoning or code generation. This results in inefficient exploration of reasoning process. To address this, we propose a dual-phase test-time scaling framework that explicitly separates reasoning into planning and execution, and performs search over the two phases individually. Specifically, we decompose reasoning trajectories and develop reward models for each phase, enabling the search to explore and prune plans and executions separately. We further introduce a dynamic budget allocation mechanism that adaptively redistributes sampling effort based on reward feedback, allowing early stopping on confident steps and reallocation of computation to more challenging parts of the reasoning process. Experiments on both mathematical reasoning and code generation benchmarks demonstrate that our approach consistently improves accuracy while reducing redundant computation.
Authors: Nigel Fernandez, Branislav Kveton, Ryan A. Rossi, Andrew S. Lan, Zichao Wang
Abstract: Reasoning language models have demonstrated remarkable performance on many challenging tasks in math, science, and coding. Choosing the right reasoning model for practical deployment involves a performance and cost tradeoff at two key levels: model size and reasoning budget, where larger models and higher reasoning budget lead to better performance but with increased cost and latency. In this work, we tackle this tradeoff from the angle of model configuration routing for different queries, and present RADAR (Reasoning-Ability and Difficulty-Aware Routing), a lightweight, interpretable, and scalable routing framework. Inspired by psychometrics, RADAR learns an item response model from model responses with different budgets to different queries, with interpretable parameters including query difficulties and model-budget abilities. RADAR then routes queries with higher difficulty to model-budget pairs with higher ability, and vice versa. We conduct extensive experiments on 8 widely used challenging reasoning benchmarks, demonstrating the superior performance of RADAR compared to state-of-the-art model routing methods. RADAR also exhibits query generalization capabilities, showing strong performance on out-of-distribution queries in all benchmarks. RADAR is also scalable and can efficiently integrate additional models by dynamically selecting a small set of evaluation queries to estimate their abilities.
Authors: Mabel Heffring, Lincoln Linlin Xu
Abstract: Although high-resolution mapping of Pan-Arctic sea ice with reliable corresponding uncertainty is essential for operational sea ice concentration (SIC) charting, it is a difficult task due to some key challenges, e.g., the subtle nature of ice signature features, model uncertainty, and data heterogeneity. This letter presents a novel Bayesian Transformer approach for Pan-Arctic SIC mapping and uncertainty quantification using Sentinel-1, RADARSAT Constellation Mission (RCM), and Advanced Microwave Scanning Radiometer 2 (AMSR2) data. First, to improve feature extraction, we design a novel high-resolution Transformer model with both global and local modules that can better discern the subtle differences in sea ice patterns. Second, to improve uncertainty quantification, we design a Bayesian extension of the proposed Transformer model, treating its parameters as random variables to more effectively capture uncertainties. Third, to address data heterogeneity, we fuse three different data types (Sentinel-1, RCM, and AMSR2) at decision-level to improve both SIC mapping and uncertainty quantification. The proposed approach is tested on Pan-Arctic datasets from September 2021, and the results demonstrate that the proposed model can achieve both high-resolution SIC maps and robust uncertainty maps compared to other uncertainty quantification approaches.
Authors: Vladimir Kondratyev, Alexander Fishkov, Nikita Kotelevskii, Mahmoud Hegazy, Remi Flamary, Maxim Panov, Eric Moulines
Abstract: We propose a framework for conditional vector quantile regression (CVQR) that combines neural optimal transport with amortized optimization, and apply it to multivariate conformal prediction. Classical quantile regression does not extend naturally to multivariate responses, while existing approaches often ignore the geometry of joint distributions. Our method parametrizes the conditional vector quantile function as the gradient of a convex potential implemented by an input-convex neural network, ensuring monotonicity and uniform ranks. To reduce the cost of solving high-dimensional variational problems, we introduced amortized optimization of the dual potentials, yielding efficient training and faster inference. We then exploit the induced multivariate ranks for conformal prediction, constructing distribution-free predictive regions with finite-sample validity. Unlike coordinatewise methods, our approach adapts to the geometry of the conditional distribution, producing tighter and more informative regions. Experiments on benchmark datasets show improved coverage-efficiency trade-offs compared to baselines, highlighting the benefits of integrating neural optimal transport with conformal prediction.
Authors: Alexander Kovrigin, Aleksandra Eliseeva, Konstantin Grotov, Egor Bogomolov, Yaroslav Zharov
Abstract: Environment setup-the process of configuring the system to work with a specific software project-represents a persistent challenge in Software Engineering (SE). Automated environment setup methods could assist developers by providing fully configured environments for arbitrary repositories without manual effort. This also helps SE researchers to scale execution-based benchmarks. However, recent studies reveal that even state-of-the-art Large Language Models (LLMs) achieve limited success in automating this task. To address this limitation, we tune a specialized model for environment setup. We combine supervised fine-tuning for generating correct Bash scripts and Reinforcement Learning with Verifiable Rewards (RLVR) to adapt it to the task of environment setup. On EnvBench-Python, our method enables Qwen3-8B (a model runnable on consumer hardware) to perform on par with larger models-Qwen3-32B and GPT-4o. The training code and model checkpoints are available online: https://github.com/JetBrains-Research/PIPer.
Authors: Haozhou Xu, Dongxia Wu, Matteo Chinazzi, Ruijia Niu, Rose Yu, Yi-An Ma
Abstract: Large language models (LLMs) show promise in solving scientific problems. They can help generate long-form answers for scientific questions, which are crucial for comprehensive understanding of complex phenomena that require detailed explanations spanning multiple interconnected concepts and evidence. However, LLMs often suffer from hallucination, especially in the challenging task of long-form scientific question answering. Retrieval-Augmented Generation (RAG) approaches can ground LLMs by incorporating external knowledge sources to improve trustworthiness. In this context, scientific simulators, which play a vital role in validating hypotheses, offer a particularly promising retrieval source to mitigate hallucination and enhance answer factuality. However, existing RAG approaches cannot be directly applied for scientific simulation-based retrieval due to two fundamental challenges: how to retrieve from scientific simulators, and how to efficiently verify and update long-form answers. To overcome these challenges, we propose the simulator-based RAG framework (SimulRAG) and provide a long-form scientific QA benchmark covering climate science and epidemiology with ground truth verified by both simulations and human annotators. In this framework, we propose a generalized simulator retrieval interface to transform between textual and numerical modalities. We further design a claim-level generation method that utilizes uncertainty estimation scores and simulator boundary assessment (UE+SBA) to efficiently verify and update claims. Extensive experiments demonstrate SimulRAG outperforms traditional RAG baselines by 30.4% in informativeness and 16.3% in factuality. UE+SBA further improves efficiency and quality for claim-level generation.
Authors: Yoann Poupart
Abstract: Interpretability of Deep Neural Networks (DNNs) is a growing field driven by the study of vision and language models. Yet, some use cases, like image captioning, or domains like Deep Reinforcement Learning (DRL), require complex modelling, with multiple inputs and outputs or use composable and separated networks. As a consequence, they rarely fit natively into the API of popular interpretability frameworks. We thus present TDHook, an open-source, lightweight, generic interpretability framework based on $\texttt{tensordict}$ and applicable to any $\texttt{torch}$ model. It focuses on handling complex composed models which can be trained for Computer Vision, Natural Language Processing, Reinforcement Learning or any other domain. This library features ready-to-use methods for attribution, probing and a flexible get-set API for interventions, and is aiming to bridge the gap between these method classes to make modern interpretability pipelines more accessible. TDHook is designed with minimal dependencies, requiring roughly half as much disk space as $\texttt{transformer_lens}$, and, in our controlled benchmark, achieves up to a $\times$2 speed-up over $\texttt{captum}$ when running integrated gradients for multi-target pipelines on both CPU and GPU. In addition, to value our work, we showcase concrete use cases of our library with composed interpretability pipelines in Computer Vision (CV) and Natural Language Processing (NLP), as well as with complex models in DRL.
Authors: Kevin Jiang, Edgar Dobriban
Abstract: We develop new classifiers under group fairness in the attribute-aware setting for binary classification with multiple group fairness constraints (e.g., demographic parity (DP), equalized odds (EO), and predictive parity (PP)). We propose a novel approach, applicable to linear fractional constraints, based on directly intervening on the operating characteristics of a pre-trained base classifier, by (i) identifying optimal operating characteristics using the base classifier's group-wise ROC convex hulls and (ii) post-processing the base classifier to match those targets. As practical post-processors, we consider randomizing a mixture of group-wise thresholding rules subject to minimizing the expected number of interventions. We further extend our approach to handle multiple protected attributes and multiple linear fractional constraints. On standard datasets (COMPAS and ACSIncome), our methods simultaneously satisfy approximate DP, EO, and PP with few interventions and a near-oracle drop in accuracy; comparing favorably to previous methods.
Authors: Wouter M. Kouw, Tim N. Nisslbeck, Wouter L. N. Nuijten
Abstract: We present the design of an autoregressive active inference agent in the form of message passing on a factor graph. Expected free energy is derived and distributed across a planning graph. The proposed agent is validated on a robot navigation task, demonstrating exploration and exploitation in a continuous-valued observation space with bounded continuous-valued actions. Compared to a classical optimal controller, the agent modulates action based on predictive uncertainty, arriving later but with a better model of the robot's dynamics.
Authors: Maximilian Schebek, Jutta Rogal
Abstract: The use of generative models to sample equilibrium distributions of many-body systems, as first demonstrated by Boltzmann Generators, has attracted substantial interest due to their ability to produce unbiased and uncorrelated samples in `one shot'. Despite their promise and impressive results across the natural sciences, scaling these models to large systems remains a major challenge. In this work, we introduce a Boltzmann Generator architecture that addresses this scalability bottleneck with a focus on applications in materials science. We leverage augmented coupling flows in combination with graph neural networks to base the generation process on local environmental information, while allowing for energy-based training and fast inference. Compared to previous architectures, our model trains significantly faster, requires far less computational resources, and achieves superior sampling efficiencies. Crucially, the architecture is transferable to larger system sizes, which allows for the efficient sampling of materials with simulation cells of unprecedented size. We demonstrate the potential of our approach by applying it to several materials systems, including Lennard-Jones crystals, ice phases of mW water, and the phase diagram of silicon, for system sizes well above one thousand atoms. The trained Boltzmann Generators produce highly accurate equilibrium ensembles for various crystal structures, as well as Helmholtz and Gibbs free energies across a range of system sizes, able to reach scales where finite-size effects become negligible.
Authors: Anirban Chatterjee, Sayantan Choudhury, Rohan Hore
Abstract: How can we generate samples from a conditional distribution that we never fully observe? This question arises across a broad range of applications in both modern machine learning and classical statistics, including image post-processing in computer vision, approximate posterior sampling in simulation-based inference, and conditional distribution modeling in complex data settings. In such settings, compared with unconditional sampling, additional feature information can be leveraged to enable more adaptive and efficient sampling. Building on this, we introduce Conditional Generator using MMD (CGMMD), a novel framework for conditional sampling. Unlike many contemporary approaches, our method frames the training objective as a simple, adversary-free direct minimization problem. A key feature of CGMMD is its ability to produce conditional samples in a single forward pass of the generator, enabling practical one-shot sampling with low test-time complexity. We establish rigorous theoretical bounds on the loss incurred when sampling from the CGMMD sampler, and prove convergence of the estimated distribution to the true conditional distribution. In the process, we also develop a uniform concentration result for nearest-neighbor based functionals, which may be of independent interest. Finally, we show that CGMMD performs competitively on synthetic tasks involving complex conditional densities, as well as on practical applications such as image denoising and image super-resolution.
Authors: Yonatan Gizachew Achamyeleh, Tongtao Zhang, Joshua Hyunki Kim, Gabriel Garcia, Shih-Yuan Yu, Anton Kocheturov, Mohammad Abdullah Al Faruque
Abstract: Function name prediction is crucial for understanding stripped binaries in software reverse engineering, a key step for \textbf{enabling subsequent vulnerability analysis and patching}. However, existing approaches often struggle with architecture-specific limitations, data scarcity, and diverse naming conventions. We present AGNOMIN, a novel architecture-agnostic approach for multi-label function name prediction in stripped binaries. AGNOMIN builds Feature-Enriched Hierarchical Graphs (FEHGs), combining Control Flow Graphs, Function Call Graphs, and dynamically learned \texttt{PCode} features. A hierarchical graph neural network processes this enriched structure to generate consistent function representations across architectures, vital for \textbf{scalable security assessments}. For function name prediction, AGNOMIN employs a Ren\'ee-inspired decoder, enhanced with an attention-based head layer and algorithmic improvements. We evaluate AGNOMIN on a comprehensive dataset of 9,000 ELF executable binaries across three architectures, demonstrating its superior performance compared to state-of-the-art approaches, with improvements of up to 27.17\% in precision and 55.86\% in recall across the testing dataset. Moreover, AGNOMIN generalizes well to unseen architectures, achieving 5.89\% higher recall than the closest baseline. AGNOMIN's practical utility has been validated through security hackathons, where it successfully aided reverse engineers in analyzing and patching vulnerable binaries across different architectures.
Authors: Boyang Zhang, Istemi Ekin Akkus, Ruichuan Chen, Alice Dethise, Klaus Satzke, Ivica Rimac, Yang Zhang
Abstract: Multimodal large language models (MLLMs) have demonstrated remarkable capabilities in processing and reasoning over diverse modalities, but their advanced abilities also raise significant privacy concerns, particularly regarding Personally Identifiable Information (PII) leakage. While relevant research has been conducted on single-modal language models to some extent, the vulnerabilities in the multimodal setting have yet to be fully investigated. In this work, we investigate these emerging risks with a focus on vision language models (VLMs), a representative subclass of MLLMs that covers the two modalities most relevant for PII leakage, vision and text. We introduce a concept-guided mitigation approach that identifies and modifies the model's internal states associated with PII-related content. Our method guides VLMs to refuse PII-sensitive tasks effectively and efficiently, without requiring re-training or fine-tuning. We also address the current lack of multimodal PII datasets by constructing various ones that simulate real-world scenarios. Experimental results demonstrate that the method can achieve an average refusal rate of 93.3% for various PII-related tasks with minimal impact on unrelated model performances. We further examine the mitigation's performance under various conditions to show the adaptability of our proposed method.
Authors: Yong Oh Lee, Byeonghun Bang, Joohyun Lee, Sejun Oh
Abstract: As personalized learning gains increasing attention in mathematics education, there is a growing demand for intelligent systems that can assess complex student responses and provide individualized feedback in real time. In this study, we present a personalized auto-grading and feedback system for constructive geometry tasks, developed using large language models (LLMs) and deployed on the Algeomath platform, a Korean online tool designed for interactive geometric constructions. The proposed system evaluates student-submitted geometric constructions by analyzing their procedural accuracy and conceptual understanding. It employs a prompt-based grading mechanism using GPT-4, where student answers and model solutions are compared through a few-shot learning approach. Feedback is generated based on teacher-authored examples built from anticipated student responses, and it dynamically adapts to the student's problem-solving history, allowing up to four iterative attempts per question. The system was piloted with 79 middle-school students, where LLM-generated grades and feedback were benchmarked against teacher judgments. Grading closely aligned with teachers, and feedback helped many students revise errors and complete multi-step geometry tasks. While short-term corrections were frequent, longer-term transfer effects were less clear. Overall, the study highlights the potential of LLMs to support scalable, teacher-aligned formative assessment in mathematics, while pointing to improvements needed in terminology handling and feedback design.
Authors: Huu Nguyen (Sonny), Victor May (Sonny), Harsh Raj (Sonny), Marianna Nezhurina (Sonny), Yishan Wang (Sonny), Yanqi Luo (Sonny), Minh Chien Vu (Sonny), Taishi Nakamura (Sonny), Ken Tsui (Sonny), Van Khue Nguyen (Sonny), David Salinas (Sonny), Aleksandra Krasnod\k{e}bska (Sonny), Christoph Schuhmann (Sonny), Mats Leon Richter (Sonny), Xuan-Son (Sonny), Vu, Jenia Jitsev
Abstract: We present MixtureVitae, an open-access pretraining corpus built to minimize legal risk while providing strong model performance. MixtureVitae follows a risk-mitigated sourcing strategy that combines public-domain and permissively licensed text (e.g., CC-BY/Apache) with carefully justified low-risk additions (e.g., government works and EU TDM-eligible sources), alongside targeted instruction, reasoning and synthetic data with documented provenance. We detail a transparent, multi-stage pipeline for license-aware filtering, safety and quality screening, and domain-aware mixing, and we release the dataset and curation recipes to support reproducible research. In controlled experiments using the open-sci-ref training protocol (fixed architectures at 130M/400M/1.3B/1.7B parameters; training budgets of 50B and 300B tokens), models trained on MixtureVitae consistently outperform other permissive datasets across a suite of standard benchmarks, and at the 1.7B/300B setting they surpass FineWeb-Edu and approach DCLM in the later stages of training. Performance is particularly strong on math/code and competitive on QA tasks. These results demonstrate that permissive-first, risk-mitigated data provides a practical and legally mitigated foundation for training capable LLMs, reducing reliance on indiscriminate web scraping without sacrificing competitiveness. Code: https://github.com/ontocord/mixturevitae
Authors: Zhiling Ye, Yun Yue, Haowen Wang, Xudong Han, Jiadi Jiang, Cheng Wei, Lei Fan, Jiaxin Liang, Shuowen Zhang, Ji Li, Chunxiao Guo, Jian Wang, Peng Wei, Jinjie Gu
Abstract: Open-ended evaluation is essential for deploying large language models in real-world settings. In studying HealthBench, we observe that using the model itself as a grader and generating rubric-based reward signals substantially improves reasoning performance. Remarkably, the trained model also becomes a stronger grader. Motivated by this, we introduce Self-Rewarding Rubric-Based Reinforcement Learning for Open-Ended Reasoning, a lightweight framework that enables faster and more resource-efficient training while surpassing baselines. Remarkably, on Qwen3-32B, training with just the 4000-sample HealthBench Easy subset is sufficient to obtain a model that exceeds GPT-5 on HealthBench Hard. Incorporating a small amount of teacher-graded data further enhances performance for less capable models.
Authors: Mohammadmahdi Eshragh, Emad A. Mohammed, Behrouz Far, Ezekiel Weis, Carol L Shields, Sandor R Ferenczy, Trafford Crump
Abstract: Choroidal nevi are common benign pigmented lesions in the eye, with a small risk of transforming into melanoma. Early detection is critical to improving survival rates, but misdiagnosis or delayed diagnosis can lead to poor outcomes. Despite advancements in AI-based image analysis, diagnosing choroidal nevi in colour fundus images remains challenging, particularly for clinicians without specialized expertise. Existing datasets often suffer from low resolution and inconsistent labelling, limiting the effectiveness of segmentation models. This paper addresses the challenge of achieving precise segmentation of fundus lesions, a critical step toward developing robust diagnostic tools. While deep learning models like U-Net have demonstrated effectiveness, their accuracy heavily depends on the quality and quantity of annotated data. Previous mathematical/clustering segmentation methods, though accurate, required extensive human input, making them impractical for medical applications. This paper proposes a novel approach that combines mathematical/clustering segmentation models with insights from U-Net, leveraging the strengths of both methods. This hybrid model improves accuracy, reduces the need for large-scale training data, and achieves significant performance gains on high-resolution fundus images. The proposed model achieves a Dice coefficient of 89.7% and an IoU of 80.01% on 1024*1024 fundus images, outperforming the Attention U-Net model, which achieved 51.3% and 34.2%, respectively. It also demonstrated better generalizability on external datasets. This work forms a part of a broader effort to develop a decision support system for choroidal nevus diagnosis, with potential applications in automated lesion annotation to enhance the speed and accuracy of diagnosis and monitoring.
Authors: Dongsu Lee, Daehee Lee, Yaru Niu, Honguk Woo, Amy Zhang, Ding Zhao
Abstract: This work presents a novel representation learning framework, interactive world latent (IWoL), to facilitate team coordination in multi-agent reinforcement learning (MARL). Building effective representation for team coordination is a challenging problem, due to the intricate dynamics emerging from multi-agent interaction and incomplete information induced by local observations. Our key insight is to construct a learnable representation space that jointly captures inter-agent relations and task-specific world information by directly modeling communication protocols. This representation, we maintain fully decentralized execution with implicit coordination, all while avoiding the inherent drawbacks of explicit message passing, e.g., slower decision-making, vulnerability to malicious attackers, and sensitivity to bandwidth constraints. In practice, our representation can be used not only as an implicit latent for each agent, but also as an explicit message for communication. Across four challenging MARL benchmarks, we evaluate both variants and show that IWoL provides a simple yet powerful key for team coordination. Moreover, we demonstrate that our representation can be combined with existing MARL algorithms to further enhance their performance.
Authors: Amirreza Sokhankhosh, Khalid Hassan, Sara Rouhani
Abstract: Collaborative and distributed learning techniques, such as Federated Learning (FL) and Split Learning (SL), hold significant promise for leveraging sensitive data in privacy-critical domains. However, FL and SL suffer from key limitations -- FL imposes substantial computational demands on clients, while SL leads to prolonged training times. To overcome these challenges, SplitFed Learning (SFL) was introduced as a hybrid approach that combines the strengths of FL and SL. Despite its advantages, SFL inherits scalability, performance, and security issues from SL. In this paper, we propose two novel frameworks: Sharded SplitFed Learning (SSFL) and Blockchain-enabled SplitFed Learning (BSFL). SSFL addresses the scalability and performance constraints of SFL by distributing the workload and communication overhead of the SL server across multiple parallel shards. Building upon SSFL, BSFL replaces the centralized server with a blockchain-based architecture that employs a committee-driven consensus mechanism to enhance fairness and security. BSFL incorporates an evaluation mechanism to exclude poisoned or tampered model updates, thereby mitigating data poisoning and model integrity attacks. Experimental evaluations against baseline SL and SFL approaches show that SSFL improves performance and scalability by 31.2% and 85.2%, respectively. Furthermore, BSFL increases resilience to data poisoning attacks by 62.7% while maintaining superior performance under normal operating conditions. To the best of our knowledge, BSFL is the first blockchain-enabled framework to implement an end-to-end decentralized SplitFed Learning system.
Authors: Suvrankar Datta, Divya Buchireddygari, Lakshmi Vennela Chowdary Kaza, Mrudula Bhalke, Kautik Singh, Ayush Pandey, Sonit Sai Vasipalli, Upasana Karnwal, Hakikat Bir Singh Bhatti, Bhavya Ratan Maroo, Sanjana Hebbar, Rahul Joseph, Gurkawal Kaur, Devyani Singh, Akhil V, Dheeksha Devasya Shama Prasad, Nishtha Mahajan, Ayinaparthi Arisha, Rajesh Vanagundi, Reet Nandy, Kartik Vuthoo, Snigdhaa Rajvanshi, Nikhileswar Kondaveeti, Suyash Gunjal, Rishabh Jain, Rajat Jain, Anurag Agrawal
Abstract: Generalist multimodal AI systems such as large language models (LLMs) and vision language models (VLMs) are increasingly accessed by clinicians and patients alike for medical image interpretation through widely available consumer-facing chatbots. Most evaluations claiming expert level performance are on public datasets containing common pathologies. Rigorous evaluation of frontier models on difficult diagnostic cases remains limited. We developed a pilot benchmark of 50 expert-level "spot diagnosis" cases across multiple imaging modalities to evaluate the performance of frontier AI models against board-certified radiologists and radiology trainees. To mirror real-world usage, the reasoning modes of five popular frontier AI models were tested through their native web interfaces, viz. OpenAI o3, OpenAI GPT-5, Gemini 2.5 Pro, Grok-4, and Claude Opus 4.1. Accuracy was scored by blinded experts, and reproducibility was assessed across three independent runs. GPT-5 was additionally evaluated across various reasoning modes. Reasoning quality errors were assessed and a taxonomy of visual reasoning errors was defined. Board-certified radiologists achieved the highest diagnostic accuracy (83%), outperforming trainees (45%) and all AI models (best performance shown by GPT-5: 30%). Reliability was substantial for GPT-5 and o3, moderate for Gemini 2.5 Pro and Grok-4, and poor for Claude Opus 4.1. These findings demonstrate that advanced frontier models fall far short of radiologists in challenging diagnostic cases. Our benchmark highlights the present limitations of generalist AI in medical imaging and cautions against unsupervised clinical use. We also provide a qualitative analysis of reasoning traces and propose a practical taxonomy of visual reasoning errors by AI models for better understanding their failure modes, informing evaluation standards and guiding more robust model development.
Authors: Yihang Chen, Yuanhao Ban, Yunqi Hong, Cho-Jui Hsieh
Abstract: Despite the success of Reinforcement Learning from Human Feedback (RLHF) in language reasoning, its application to autoregressive Text-to-Image (T2I) generation is often constrained by the limited availability of human preference data. This paper explores how an autoregressive T2I model can learn from internal signals without relying on external rewards or labeled data. Contrary to recent findings in text generation, we show that maximizing self-uncertainty, rather than self-certainty, improves image generation. We observe that this is because autoregressive T2I models with low uncertainty tend to generate simple and uniform images, which are less aligned with human preferences. Based on these observations, we propose IRIS (Intrinsic Reward Image Synthesis), the first framework to improve autoregressive T2I models with reinforcement learning using only an intrinsic reward. Empirical results demonstrate that applying IRIS to autoregressive T2I models achieves performance that is competitive with or superior to external rewards.
Authors: Max Hartman, Vidhata Jayaraman, Moulik Choraria, Akhil Bhimaraju, Lav R. Varshney
Abstract: Vision-language models (VLMs) achieve incredible performance across a wide range of tasks, but their large size makes inference costly. Recent work shows that selectively skipping VLM layers can improve efficiency with minimal performance loss or even performance improvements. However, this technique remains underused due to the limited understanding of when layer skipping is beneficial. In this paper, we develop a framework that uses information and learning theory to characterize the conditions under which layer skipping enhances efficiency without sacrificing performance. Motivated by these observations, we analyze the evolution of the VLM's hidden representations through the LLM backbone and show that layers with large redundancy as predicted by our framework coincide with those skipped by popular layer-skipping methods in practice, providing a unified theoretical scaffolding for multiple efficient inference techniques. Our experiments demonstrate that skipping such layers yields faster inference that preserves performance, and also show that applying skipping outside these conditions leads to model degradation.
Authors: Yishu Wei, Wen-Yee Lee, George Ekow Quaye, Xiaogang Su
Abstract: In binary classification applications, conservative decision-making that allows for abstention can be advantageous. To this end, we introduce a novel approach that determines the optimal cutoff interval for risk scores, which can be directly available or derived from fitted models. Within this interval, the algorithm refrains from making decisions, while outside the interval, classification accuracy is maximized. Our approach is inspired by support vector machines (SVM), but differs in that it minimizes the classification margin rather than maximizing it. We provide the theoretical optimal solution to this problem, which holds important practical implications. Our proposed method not only supports conservative decision-making but also inherently results in a risk-coverage curve. Together with the area under the curve (AUC), this curve can serve as a comprehensive performance metric for evaluating and comparing classifiers, akin to the receiver operating characteristic (ROC) curve. To investigate and illustrate our approach, we conduct both simulation studies and a real-world case study in the context of diagnosing prostate cancer.
Authors: Berenice Montalvo-Lezama, Gibran Fuentes-Pineda
Abstract: The limited availability of annotated data presents a major challenge for applying deep learning methods to medical image analysis. Few-shot learning methods aim to recognize new classes from only a small number of labeled examples. These methods are typically studied under the standard few-shot learning setting, where all classes in a task are new. However, medical applications such as pathology classification from chest X-rays often require learning new classes while simultaneously leveraging knowledge of previously known ones, a scenario more closely aligned with generalized few-shot classification. Despite its practical relevance, few-shot learning has been scarcely studied in this context. In this work, we present MetaChest, a large-scale dataset of 479,215 chest X-rays collected from four public databases. MetaChest includes a meta-set partition specifically designed for standard few-shot classification, as well as an algorithm for generating multi-label episodes. We conduct extensive experiments evaluating both a standard transfer learning approach and an extension of ProtoNet across a wide range of few-shot multi-label classification tasks. Our results demonstrate that increasing the number of classes per episode and the number of training examples per class improves classification performance. Notably, the transfer learning approach consistently outperforms the ProtoNet extension, despite not being tailored for few-shot learning. We also show that higher-resolution images improve accuracy at the cost of additional computation, while efficient model architectures achieve comparable performance to larger models with significantly reduced resource requirements.
Authors: Peiran Xu, Zhuohao Li, Xiaoying Xing, Guannan Zhang, Debiao Li, Kunyu Shi
Abstract: Large Language Models (LLMs) increasingly rely on external tools such as search engines to solve complex agentic tasks that require reasoning and external knowledge retrieval. Recently, reinforcement learning with verifiable rewards (RLVR) has demonstrated its effectiveness in advancing capabilities of LLMs by rewarding the final answers via outcome rewards. While straightforward to supervise, outcome rewards only provide sparse signals and delayed feedback, which limits their effectiveness on long trajectories. Process rewards address this by evaluating intermediate steps, providing fine-grained supervision and encouraging grounded problem solving. However, it is notoriously hard to annotate step-wise labels, especially in non-verifiable process without "golden" answers. Furthermore, step-wise judgment requires the balance between local quality with contribution to the final outcome, as optimizing towards higher process reward may not always align with better final outcomes. To address the above challenges, we introduce Principle Process Reward (PPR), an RL approach that unifies principled step-level assessment and outcome verification. We train a principle-based reward model to improve the transparency and reliability of process evaluation, and further introduce a Reward Normalization (ReNorm) strategy to calibrate outcome and process rewards. Experiment results show that PPR achieves state-of-the-art performance across a wide range of benchmarks, demonstrating its impressive robustness and generalization. Our code and model collection is available in this link.
Authors: Ruolin Meng, Ming-Yu Chung, Dhanajit Brahma, Ricardo Henao, Lawrence Carin
Abstract: We consider inferring the causal effect of a treatment (intervention) on an outcome of interest in situations where there is potentially an unobserved confounder influencing both the treatment and the outcome. This is achievable by assuming access to two separate sets of control (proxy) measurements associated with treatment and outcomes, which are used to estimate treatment effects through a function termed the em causal bridge (CB). We present a new theoretical perspective, associated assumptions for when estimating treatment effects with the CB is feasible, and a bound on the average error of the treatment effect when the CB assumptions are violated. From this new perspective, we then demonstrate how coupling the CB with an autoencoder architecture allows for the sharing of statistical strength between observed quantities (proxies, treatment, and outcomes), thus improving the quality of the CB estimates. Experiments on synthetic and real-world data demonstrate the effectiveness of the proposed approach in relation to the state-of-the-art methodology for proxy measurements.
Authors: Tianlang Chen, Minkai Xu, Jure Leskovec, Stefano Ermon
Abstract: Diffusion large language models (dLLMs) have shown great potential in large-scale language modeling, and there is an increasing interest in further improving the capacity to solve complex problems by guiding the reasoning process step by step. Common practice for autoregressive language models typically learns a process reward model with dense annotation for each intermediate step. However, this is challenging for dLLMs where the generation is in an any-order fashion and intermediate states are partially masked sentences. To this end, in this paper, we propose reward-free guidance (RFG), a principled method for guiding the reasoning trajectory of dLLMs without explicit process reward. The key idea of RFG is to parameterize the process reward by log-likelihood ratios of the enhanced and reference dLLMs, where the enhanced model can be easily obtained by any off-the-shelf dLLM that has been post-trained with reinforcement learning (RL) or supervised fine-tuning (SFT). We provide theoretical justification that RFG induces the reward-guided sampling distribution with no additional reward. We conduct comprehensive experiments on four challenging mathematical reasoning and code generation benchmarks using a diverse suite of dLLMs enhanced with various post-training methods. RFG consistently yields significant improvements across all tasks and model types, achieving accuracy gains of up to 9.2%. These findings establish RFG as a general training-free framework that scales test-time reasoning without reliance on external reward models.
Authors: Jing-Jing Li, Jianfeng He, Chao Shang, Devang Kulshreshtha, Xun Xian, Yi Zhang, Hang Su, Sandesh Swamy, Yanjun Qi
Abstract: As LLMs advance into autonomous agents with tool-use capabilities, they introduce security challenges that extend beyond traditional content-based LLM safety concerns. This paper introduces Sequential Tool Attack Chaining (STAC), a novel multi-turn attack framework that exploits agent tool use. STAC chains together tool calls that each appear harmless in isolation but, when combined, collectively enable harmful operations that only become apparent at the final execution step. We apply our framework to automatically generate and systematically evaluate 483 STAC cases, featuring 1,352 sets of user-agent-environment interactions and spanning diverse domains, tasks, agent types, and 10 failure modes. Our evaluations show that state-of-the-art LLM agents, including GPT-4.1, are highly vulnerable to STAC, with attack success rates (ASR) exceeding 90% in most cases. The core design of STAC's automated framework is a closed-loop pipeline that synthesizes executable multi-step tool chains, validates them through in-environment execution, and reverse-engineers stealthy multi-turn prompts that reliably induce agents to execute the verified malicious sequence. We further perform defense analysis against STAC and find that existing prompt-based defenses provide limited protection. To address this gap, we propose a new reasoning-driven defense prompt that achieves far stronger protection, cutting ASR by up to 28.8%. These results highlight a crucial gap: defending tool-enabled agents requires reasoning over entire action sequences and their cumulative effects, rather than evaluating isolated prompts or responses.
Authors: Xiaojie Wang, Bin Yang
Abstract: Efficient sampling from complex and high dimensional target distributions turns out to be a fundamental task in diverse disciplines such as scientific computing, statistics and machine learning. In this paper, we revisit the randomized Langevin Monte Carlo (RLMC) for sampling from high dimensional distributions without log-concavity. Under the gradient Lipschitz condition and the log-Sobolev inequality, we prove a uniform-in-time error bound in $\mathcal{W}_2$-distance of order $O(\sqrt{d}h)$ for the RLMC sampling algorithm, which matches the best one in the literature under the log-concavity condition. Moreover, when the gradient of the potential $U$ is non-globally Lipschitz with superlinear growth, modified RLMC algorithms are proposed and analyzed, with non-asymptotic error bounds established. To the best of our knowledge, the modified RLMC algorithms and their non-asymptotic error bounds are new in the non-globally Lipschitz setting.
Authors: Jungsoo Lee, Janghoon Cho, Hyojin Park, Munawar Hayat, Kyuwoong Hwang, Fatih Porikli, Sungha Choi
Abstract: Despite their consistent performance improvements, cross-modal retrieval models (e.g., CLIP) show degraded performances with retrieving keys composed of fused image-text modality (e.g., Wikipedia pages with both images and text). To address this critical challenge, multimodal retrieval has been recently explored to develop a unified single retrieval model capable of retrieving keys across diverse modality combinations. A common approach involves constructing new composed sets of image-text triplets (e.g., retrieving a pair of image and text given a query image). However, such an approach requires careful curation to ensure the dataset quality and fails to generalize to unseen modality combinations. To overcome these limitations, this paper proposes Generalized Contrastive Learning (GCL), a novel loss formulation that improves multimodal retrieval performance without the burdensome need for new dataset curation. Specifically, GCL operates by enforcing contrastive learning across all modalities within a mini-batch, utilizing existing image-caption paired datasets to learn a unified representation space. We demonstrate the effectiveness of GCL by showing consistent performance improvements on off-the-shelf multimodal retrieval models (e.g., VISTA, CLIP, and TinyCLIP) using the M-BEIR, MMEB, and CoVR benchmarks.
Authors: Leandro Arab Marcomini, Andre Luiz Cunha
Abstract: Convolutional Neural Networks (CNNs) traditionally require large amounts of data to train models with good performance. However, data collection is an expensive process, both in time and resources. Generated synthetic images are a good alternative, with video games producing realistic 3D models. This paper aims to determine whether images extracted from a video game can be effectively used to train a CNN to detect real-life truck axles. Three different databases were created, with real-life and synthetic trucks, to provide training and testing examples for three different You Only Look Once (YOLO) architectures. Results were evaluated based on four metrics: recall, precision, F1-score, and mean Average Precision (mAP). To evaluate the statistical significance of the results, the Mann-Whitney U test was also applied to the resulting mAP of all models. Synthetic images from trucks extracted from a video game proved to be a reliable source of training data, contributing to the performance of all networks. The highest mAP score reached 99\%. Results indicate that synthetic images can be used to train neural networks, providing a reliable, low-cost data source for extracting knowledge.
Authors: Po-Heng Chou, Chun-Chi Wang, Wei-Lung Mao
Abstract: In this paper, we propose a YOLO-based deep learning (DL) model for automatic defect detection to solve the time-consuming and labor-intensive tasks in industrial manufacturing. In our experiments, the images of metal sheets are used as the dataset for training the YOLO model to detect the defects on the surfaces and in the holes of metal sheets. However, the lack of metal sheet images significantly degrades the performance of detection accuracy. To address this issue, the ConSinGAN is used to generate a considerable amount of data. Four versions of the YOLO model (i.e., YOLOv3, v4, v7, and v9) are combined with the ConSinGAN for data augmentation. The proposed YOLOv9 model with ConSinGAN outperforms the other YOLO models with an accuracy of 91.3%, and a detection time of 146 ms. The proposed YOLOv9 model is integrated into manufacturing hardware and a supervisory control and data acquisition (SCADA) system to establish a practical automated optical inspection (AOI) system. Additionally, the proposed automated defect detection is easily applied to other components in industrial manufacturing.
Authors: Chun-Yuan Huang, Po-Heng Chou, Wan-Jen Huang, Ying-Ren Chien, Yu Tsao
Abstract: In this paper, we propose Capacity-Net, a novel unsupervised learning approach aimed at maximizing the achievable rate in reflecting intelligent surface (RIS)-aided millimeter-wave (mmWave) multiple input multiple output (MIMO) systems. To combat severe channel fading of the mmWave spectrum, we optimize the phase-shifting factors of the reflective elements in the RIS to enhance the achievable rate. However, most optimization algorithms rely heavily on complete and accurate channel state information (CSI), which is often challenging to acquire since the RIS is mostly composed of passive components. To circumvent this challenge, we leverage unsupervised learning techniques with implicit CSI provided by the received pilot signals. Specifically, it usually requires perfect CSI to evaluate the achievable rate as a performance metric of the current optimization result of the unsupervised learning method. Instead of channel estimation, the Capacity-Net is proposed to establish a mapping among the received pilot signals, optimized RIS phase shifts, and the resultant achievable rates. Simulation results demonstrate the superiority of the proposed Capacity-Net-based unsupervised learning approach over learning methods based on traditional channel estimation.
Authors: Po-Heng Chou, Bo-Ren Zheng, Wan-Jen Huang, Walid Saad, Yu Tsao, Ronald Y. Chang
Abstract: This study considers multiple reconfigurable intelligent surfaces (RISs)-aided multiuser downlink systems with the goal of jointly optimizing the transmitter precoding and RIS phase shift matrix to maximize spectrum efficiency. Unlike prior work that assumed ideal RIS reflectivity, a practical coupling effect is considered between reflecting amplitude and phase shift for the RIS elements. This makes the optimization problem non-convex. To address this challenge, we propose a deep deterministic policy gradient (DDPG)-based deep reinforcement learning (DRL) framework. The proposed model is evaluated under both fixed and random numbers of users in practical mmWave channel settings. Simulation results demonstrate that, despite its complexity, the proposed DDPG approach significantly outperforms optimization-based algorithms and double deep Q-learning, particularly in scenarios with random user distributions.
Authors: Yixiao Chen, Yanyue Xie, Ruining Yang, Wei Jiang, Wei Wang, Yong He, Yue Chen, Pu Zhao, Yanzhi Wang
Abstract: The Mixture of Experts (MoE) architecture is an important method for scaling Large Language Models (LLMs). It increases model capacity while keeping computation cost low. However, the ultra-large MoE models still have hundreds of billions of parameters, requiring massive memory/storage and leading to difficulties for deployment on resource-constrained edge platforms. Pruning or quantization alone can hardly address the issue, because of the super-aggressive compression ratio with significantly degraded accuracy and output quality. To facilitate the deployment of ultra-large MoEs on edge platforms, we propose a collaborative compression framework by combining expert pruning, mixed-precision quantization, and activation optimization. It can effectively reduce the storage footprint of the ultra-large MoE DeepSeek-V3 from 1.3TB to 103GB, while preserving high output quality with better accuracy than traditional uniform low-bit quantization methods. To the best of our knowledge, we are the first to deploy a compressed model from the ultra-large DeepSeek-V3 on the platform with a strict 128GB total memory limit. Our comprehensive experiments on multiple benchmarks under various memory constraints demonstrate the effectiveness of our method with smaller model sizes and higher accuracy than uniform low-bit quantization methods.
Authors: Xintong Li, Chuhan Wang, Junda Wu, Rohan Surana, Tong Yu, Julian McAuley, Jingbo Shang
Abstract: Direct Preference Optimization (DPO) has recently been extended from text-only models to vision-language models. However, existing methods rely on oversimplified pairwise comparisons, generating a single negative image via basic perturbations or similarity-based retrieval, which fail to capture the complex nature of multimodal preferences, inducing optimization bias and hallucinations. To address this issue, we propose MISP-DPO, the first framework to incorporate multiple, semantically diverse negative images in multimodal DPO via the Plackett-Luce model. Our method embeds prompts and candidate images in CLIP (Contrastive Language-Image Pretraining) space and applies a sparse autoencoder to uncover semantic deviations into interpretable factors. Negative samples are selected based on reconstruction difficulty, semantic deviation from the positive, and mutual diversity, yielding broader and more informative supervision. To handle multi-negative comparisons, we adopt a Plackett-Luce objective and introduce an importance sampling strategy that improves training efficiency. Experiments across five diverse benchmarks demonstrate that MISP-DPO consistently improves multimodal alignment over prior methods, validating the effectiveness of semantic-aware, multi-negative sampling in preference-based learning.
Authors: Ruibin Chen, Haozhe Lei, Hao Guo, Marco Mezzavilla, Hitesh Poddar, Tomoki Yoshimura, Sundeep Rangan
Abstract: Cellular wireless systems are witnessing the proliferation of frequency bands over a wide spectrum, particularly with the expansion of new bands in FR3. These bands must be supported in user equipment (UE) handsets with multiple antennas in a constrained form factor. Rapid variations in channel quality across the bands from motion and hand blockage, limited field-of-view of antennas, and hardware and power-constrained measurement sparsity pose significant challenges to reliable multi-band channel tracking. This paper formulates the problem of predicting achievable rates across multiple antenna arrays and bands with sparse historical measurements. We propose a transformer-based neural architecture that takes asynchronous rate histories as input and outputs per-array rate predictions. Evaluated on ray-traced simulations in a dense urban micro-cellular setting with FR1 and FR3 arrays, our method demonstrates superior performance over baseline predictors, enabling more informed band selection under realistic mobility and hardware constraints.
Authors: Zhe Liu, Yuyan Ni, Zhichen Pu, Qiming Sun, Siyuan Liu, Wen Yan
Abstract: Recently, sophisticated deep learning-based approaches have been developed for generating efficient initial guesses to accelerate the convergence of density functional theory (DFT) calculations. While the actual initial guesses are often density matrices (DM), quantities that can convert into density matrices also qualify as alternative forms of initial guesses. Hence, existing works mostly rely on the prediction of the Hamiltonian matrix for obtaining high-quality initial guesses. However, the Hamiltonian matrix is both numerically difficult to predict and intrinsically non-transferable, hindering the application of such models in real scenarios. In light of this, we propose a method that constructs DFT initial guesses by predicting the electron density in a compact auxiliary basis representation using E(3)-equivariant neural networks. Trained on small molecules with up to 20 atoms, our model is able to achieve an average 33.3% self-consistent field (SCF) step reduction on systems up to 60 atoms, substantially outperforming Hamiltonian-centric and DM-centric models. Critically, this acceleration remains nearly constant with increasing system sizes and exhibits strong transferring behaviors across orbital basis sets and exchange-correlation (XC) functionals. To the best of our knowledge, this work represents the first and robust candidate for a universally transferable DFT acceleration method. We are also releasing the SCFbench dataset and its accompanying code to facilitate future research in this promising direction.
Authors: Zihao Zhao, Anjalie Field
Abstract: Text anonymization is essential for responsibly developing and deploying AI in high-stakes domains such as healthcare, social services, and law. In this work, we propose a novel methodology for privacy-preserving synthetic text generation that leverages the principles of de-identification and the Hiding In Plain Sight (HIPS) theory. Our approach introduces entity-aware control codes to guide controllable generation using either in-context learning (ICL) or prefix tuning. The ICL variant ensures privacy levels consistent with the underlying de-identification system, while the prefix tuning variant incorporates a custom masking strategy and loss function to support scalable, high-quality generation. Experiments on legal and clinical datasets demonstrate that our method achieves a strong balance between privacy protection and utility, offering a practical and effective solution for synthetic text generation in sensitive domains.
Authors: Kento Kuwataka, Taiji Suzuki
Abstract: Test-time training (TTT) enhances model performance by explicitly updating designated parameters prior to each prediction to adapt to the test data. While TTT has demonstrated considerable empirical success, its theoretical underpinnings remain limited, particularly for nonlinear models. In this paper, we investigate the combination of TTT with in-context learning (ICL), where the model is given a few examples from the target distribution at inference time. We analyze this framework in the setting of single-index models $y=\sigma_*(\langle \beta, \mathbf{x} \rangle)$, where the feature vector $\beta$ is drawn from a hidden low-dimensional subspace. For single-layer transformers trained with gradient-based algorithms and adopting TTT, we establish an upper bound on the prediction risk. Our theory reveals that TTT enables the single-layer transformers to adapt to both the feature vector $\beta$ and the link function $\sigma_*$, which vary across tasks. This creates a sharp contrast with ICL alone, which is theoretically difficult to adapt to shifts in the link function. Moreover, we provide the convergence rate with respect to the data length, showing the predictive error can be driven arbitrarily close to the noise level as the context size and the network width grow.
Authors: T. O. Abiola, K. D. Abiodun, O. E. Olumide, O. O. Adebanji, O. Hiram Calvo, Grigori Sidorov
Abstract: The detection of hopeful speech in social media has emerged as a critical task for promoting positive discourse and well-being. In this paper, we present a machine learning approach to multiclass hope speech detection across multiple languages, including English, Urdu, and Spanish. We leverage transformer-based models, specifically XLM-RoBERTa, to detect and categorize hope speech into three distinct classes: Generalized Hope, Realistic Hope, and Unrealistic Hope. Our proposed methodology is evaluated on the PolyHope dataset for the PolyHope-M 2025 shared task, achieving competitive performance across all languages. We compare our results with existing models, demonstrating that our approach significantly outperforms prior state-of-the-art techniques in terms of macro F1 scores. We also discuss the challenges in detecting hope speech in low-resource languages and the potential for improving generalization. This work contributes to the development of multilingual, fine-grained hope speech detection models, which can be applied to enhance positive content moderation and foster supportive online communities.
Authors: Yixian Zhang, Shu'ang Yu, Tonghe Zhang, Mo Guang, Haojia Hui, Kaiwen Long, Yu Wang, Chao Yu, Wenbo Ding
Abstract: Training expressive flow-based policies with off-policy reinforcement learning is notoriously unstable due to gradient pathologies in the multi-step action sampling process. We trace this instability to a fundamental connection: the flow rollout is algebraically equivalent to a residual recurrent computation, making it susceptible to the same vanishing and exploding gradients as RNNs. To address this, we reparameterize the velocity network using principles from modern sequential models, introducing two stable architectures: Flow-G, which incorporates a gated velocity, and Flow-T, which utilizes a decoded velocity. We then develop a practical SAC-based algorithm, enabled by a noise-augmented rollout, that facilitates direct end-to-end training of these policies. Our approach supports both from-scratch and offline-to-online learning and achieves state-of-the-art performance on continuous control and robotic manipulation benchmarks, eliminating the need for common workarounds like policy distillation or surrogate objectives.
Authors: Zhepei Wei, Xiao Yang, Kai Sun, Jiaqi Wang, Rulin Shao, Sean Chen, Mohammad Kachuee, Teja Gollapudi, Tony Liao, Nicolas Scheffer, Rakesh Wanga, Anuj Kumar, Yu Meng, Wen-tau Yih, Xin Luna Dong
Abstract: While large language models (LLMs) have demonstrated strong performance on factoid question answering, they are still prone to hallucination and untruthful responses, particularly when tasks demand information outside their parametric knowledge. Indeed, truthfulness requires more than accuracy -- models must also recognize uncertainty and abstain when unsure to avoid hallucinations. This presents a fundamental challenge for existing methods: approaches that optimize for accuracy often amplify hallucinations, while those that encourage abstention can become overly conservative, sacrificing correct answers. Both extremes ultimately compromise truthfulness. In this work, we present TruthRL, a general reinforcement learning (RL) framework that directly optimizes the truthfulness of LLMs. Specifically, we implement TruthRL using GRPO with a simple yet effective ternary reward that distinguishes correct answers, hallucinations, and abstentions. It incentivizes models to reduce hallucinations not only by providing correct responses, but also by enabling abstention when uncertain, thereby improving truthfulness. Extensive experiments across four knowledge-intensive benchmarks show that, compared to vanilla RL, TruthRL significantly reduces hallucinations by 28.9% and improves truthfulness by 21.1%, with consistent gains across various backbone models (e.g., Qwen, Llama) under both retrieval and non-retrieval setups. In-depth ablation study demonstrates that vanilla accuracy-driven methods, such as supervised fine-tuning or RL with a binary reward, struggle to balance factual correctness and uncertainty. In contrast, our proposed truthfulness-driven TruthRL achieves strong performance in both accuracy and truthfulness, underscoring the importance of learning objective design for developing truthful LLMs.
Authors: Anil Kamber, Rahul Parhi
Abstract: Understanding the geometry of the loss landscape near a minimum is key to explaining the implicit bias of gradient-based methods in non-convex optimization problems such as deep neural network training and deep matrix factorization. A central quantity to characterize this geometry is the maximum eigenvalue of the Hessian of the loss, which measures the sharpness of the landscape. Currently, its precise role has been obfuscated because no exact expressions for this sharpness measure were known in general settings. In this paper, we present the first exact expression for the maximum eigenvalue of the Hessian of the squared-error loss at any minimizer in general overparameterized deep matrix factorization (i.e., deep linear neural network training) problems, resolving an open question posed by Mulayoff & Michaeli (2020). To complement our theory, we empirically investigate an escape phenomenon observed during gradient-based training near a minimum that crucially relies on our exact expression of the sharpness.
Authors: Wanying Ding, Savinay Narendra, Xiran Shi, Adwait Ratnaparkhi, Chengrui Yang, Nikoo Sabzevar, Ziyan Yin
Abstract: Analyzing financial transactions is crucial for ensuring regulatory compliance, detecting fraud, and supporting decisions. The complexity of financial transaction data necessitates advanced techniques to extract meaningful insights and ensure accurate analysis. Since Transformer-based models have shown outstanding performance across multiple domains, this paper seeks to explore their potential in understanding financial transactions. This paper conducts extensive experiments to evaluate three types of Transformer models: Encoder-Only, Decoder-Only, and Encoder-Decoder models. For each type, we explore three options: pretrained LLMs, fine-tuned LLMs, and small proprietary models developed from scratch. Our analysis reveals that while LLMs, such as LLaMA3-8b, Flan-T5, and SBERT, demonstrate impressive capabilities in various natural language processing tasks, they do not significantly outperform small proprietary models in the specific context of financial transaction understanding. This phenomenon is particularly evident in terms of speed and cost efficiency. Proprietary models, tailored to the unique requirements of transaction data, exhibit faster processing times and lower operational costs, making them more suitable for real-time applications in the financial sector. Our findings highlight the importance of model selection based on domain-specific needs and underscore the potential advantages of customized proprietary models over general-purpose LLMs in specialized applications. Ultimately, we chose to implement a proprietary decoder-only model to handle the complex transactions that we previously couldn't manage. This model can help us to improve 14% transaction coverage, and save more than \$13 million annual cost.
Authors: Zichen Liang, Jingjing Fei, Jie Wang, Zheming Yang, Changqing Li, Pei Wu, Minghui Qiu, Fei Yang, Xialei Liu
Abstract: Recent advances in multimodal large language models (MLLMs) have been primarily evaluated on general-purpose benchmarks, while their applications in domain-specific scenarios, such as intelligent product moderation, remain underexplored. To address this gap, we introduce an open-world logo recognition benchmark, a core challenge in product moderation. Unlike traditional logo recognition methods that rely on memorizing representations of tens of thousands of brands-an impractical approach in real-world settings-our proposed method, Logo-VGR, enables generalization to large-scale brand recognition with supervision from only a small subset of brands. Specifically, we reformulate logo recognition as a comparison-based task, requiring the model to match product images with candidate logos rather than directly generating brand labels. We further observe that existing models tend to overfit by memorizing brand distributions instead of learning robust multimodal reasoning, which results in poor performance on unseen brands. To overcome this limitation, Logo-VGR introduces a new paradigm of domain-specific multimodal reasoning: Logo Perception Grounding injects domain knowledge, and Logo-Guided Visual Grounded Reasoning enhances the model's reasoning capability. Experimental results show that Logo-VGR outperforms strong baselines by nearly 10 points in OOD settings, demonstrating superior generalization.
Authors: Dragos-Dumitru Ghinea, Adela-Nicoleta Corbeanu, Adrian-Marius Dumitran
Abstract: In recent years, large language models (LLMs) have demonstrated significant potential across various natural language processing (NLP) tasks. However, their performance in domain-specific applications and non-English languages remains less explored. This study introduces a novel Romanian-language dataset for multiple-choice biology questions, carefully curated to assess LLM comprehension and reasoning capabilities in scientific contexts. Containing approximately 14,000 questions, the dataset provides a comprehensive resource for evaluating and improving LLM performance in biology. We benchmark several popular LLMs, analyzing their accuracy, reasoning patterns, and ability to understand domain-specific terminology and linguistic nuances. Additionally, we perform comprehensive experiments to evaluate the impact of prompt engineering, fine-tuning, and other optimization techniques on model performance. Our findings highlight both the strengths and limitations of current LLMs in handling specialized knowledge tasks in low-resource languages, offering valuable insights for future research and development.
Authors: Yi-chen Yao, Jerry Wang, Yi-cheng Lai, Lyn Chao-ling Chen
Abstract: The topic of aging decline on performance of NBA players has been discussed in this study. The autoencoder with K-means clustering machine learning method was adopted to career trend classification of NBA players, and the LSTM deep learning method was adopted in performance prediction of each NBA player. The dataset was collected from the basketball game data of veteran NBA players. The contribution of the work performed better than the other methods with generalization ability for evaluating various types of NBA career trend, and can be applied in different types of sports in the field of sport analytics.
Authors: Hankun Dai, Maoquan Wang, Mengnan Qi, Yikai Zhang, Zijian Jin, Yongqiang Yao, Yufan Huang, Shengyu Fu, Elsie Nallipogu
Abstract: Large language models (LLMs) are increasingly being applied to programming tasks, ranging from single-turn code completion to autonomous agents. Current code agent designs frequently depend on complex, hand-crafted workflows and tool sets. However, this reliance on elaborate scaffolding presents several challenges: agent performance becomes overly dependent on prompt tuning and custom design choices, heavy human intervention obscures a model's true underlying capabilities, and intricate pipelines are costly to build and maintain. Furthermore, optimizing complex task prompts increases the risk of data leakage. Currently, when introducing new models, LLM providers like OpenAI and Anthropic often publish benchmark scores to demonstrate their models' coding proficiency, but keep their proprietary evaluation frameworks confidential. To address these limitations, we introduce Lita (Lite Agent), which operationalizes liteness, a principle of minimizing manual design while retaining the essential elements of a fully autonomous agent. Lita enables a more faithful and unified evaluation without elaborate scaffolding. Experiments on the Aider Polyglot and SWE-Bench with frontier models demonstrate that Lita achieves competitive or superior performance compared to workflow-based and agentic baselines. Crucially, Lita also consumes fewer tokens and requires significantly less design effort. Our results suggest that Lita is sufficient to reveal the underlying coding competence of modern LLMs. Finally, we propose the Agent Complexity Law: the performance gap between agents of varying complexity, from simple to sophisticated designs, will shrink as the core model improves, ultimately converging to a negligible difference.
Authors: Dennis Jacob, Emad Alghamdi, Zhanhao Hu, Basel Alomair, David Wagner
Abstract: Large language models (LLMs) have become increasingly popular due to their ability to interact with unstructured content. As such, LLMs are now a key driver behind the automation of language processing systems, such as AI agents. Unfortunately, these advantages have come with a vulnerability to prompt injections, an attack where an adversary subverts the LLM's intended functionality with an injected task. Past approaches have proposed detectors and finetuning to provide robustness, but these techniques are vulnerable to adaptive attacks or cannot be used with state-of-the-art models. To this end we propose type-directed privilege separation for LLMs, a method that systematically prevents prompt injections. We restrict the ability of an LLM to interact with third-party data by converting untrusted content to a curated set of data types; unlike raw strings, each data type is limited in scope and content, eliminating the possibility for prompt injections. We evaluate our method across several case studies and find that designs leveraging our principles can systematically prevent prompt injection attacks while maintaining high utility.
Authors: Marco Zimmerli, Andreas Plesner, Till Aczel, Roger Wattenhofer
Abstract: Deep neural networks remain vulnerable to adversarial examples despite advances in architectures and training paradigms. We investigate how training data characteristics affect adversarial robustness across 36 state-of-the-art vision models spanning supervised, self-supervised, and contrastive learning approaches, trained on datasets from 1.2M to 22B images. Models were evaluated under six black-box attack categories: random perturbations, two types of geometric masks, COCO object manipulations, ImageNet-C corruptions, and ImageNet-R style shifts. Robustness follows a logarithmic scaling law with both data volume and model size: a tenfold increase in data reduces attack success rate (ASR) on average by ~3.2%, whereas a tenfold increase in model size reduces ASR on average by ~13.4%. Notably, some self-supervised models trained on curated datasets, such as DINOv2, outperform others trained on much larger but less curated datasets, challenging the assumption that scale alone drives robustness. Adversarial fine-tuning of ResNet50s improves generalization across structural variations but not across color distributions. Human evaluation reveals persistent gaps between human and machine vision. These results show that while scaling improves robustness, data quality, architecture, and training objectives play a more decisive role than raw scale in achieving broad-spectrum adversarial resilience.
Authors: Debottam Dutta, Jianchong Chen, Rajalaxmi Rajagopalan, Yu-Lin Wei, Romit Roy Choudhury
Abstract: We propose to improve multi-concept prompt fidelity in text-to-image diffusion models. We begin with common failure cases-prompts like "a cat and a dog" that sometimes yields images where one concept is missing, faint, or colliding awkwardly with another. We hypothesize that this happens when the diffusion model drifts into mixed modes that over-emphasize a single concept it learned strongly during training. Instead of re-training, we introduce a corrective sampling strategy that steers away from regions where the joint prompt behavior overlaps too strongly with any single concept in the prompt. The goal is to steer towards "pure" joint modes where all concepts can coexist with balanced visual presence. We further show that existing multi-concept guidance schemes can operate in unstable weight regimes that amplify imbalance; we characterize favorable regions and adapt sampling to remain within them. Our approach, CO3, is plug-and-play, requires no model tuning, and complements standard classifier-free guidance. Experiments on diverse multi-concept prompts indicate improvements in concept coverage, balance and robustness, with fewer dropped or distorted concepts compared to standard baselines and prior compositional methods. Results suggest that lightweight corrective guidance can substantially mitigate brittle semantic alignment behavior in modern diffusion systems.
Authors: Ardian Selmonaj, Giacomo Del Rio, Adrian Schneider, Alessandro Antonucci
Abstract: We present a system that enables real-time interaction between human users and agents trained to control fighter jets in simulated 3D air combat scenarios. The agents are trained in a dedicated environment using Multi-Agent Reinforcement Learning. A communication link is developed to allow seamless deployment of trained agents into VR-Forces, a widely used defense simulation tool for realistic tactical scenarios. This integration allows mixed simulations where human-controlled entities engage with intelligent agents exhibiting distinct combat behaviors. Our interaction model creates new opportunities for human-agent teaming, immersive training, and the exploration of innovative tactics in defense contexts.
Authors: Sankar Vinayak. E. P, Gopalakrishnan Srinivasan
Abstract: Equilibrium propagation has been proposed as a biologically plausible alternative to the backpropagation algorithm. The local nature of gradient computations, combined with the use of convergent RNNs to reach equilibrium states, make this approach well-suited for implementation on neuromorphic hardware. However, previous studies on equilibrium propagation have been restricted to networks containing only dense layers or relatively small architectures with a few convolutional layers followed by a final dense layer. These networks have a significant gap in accuracy compared to similarly sized feedforward networks trained with backpropagation. In this work, we introduce the Hopfield-Resnet architecture, which incorporates residual (or skip) connections in Hopfield networks with clipped $\mathrm{ReLU}$ as the activation function. The proposed architectural enhancements enable the training of networks with nearly twice the number of layers reported in prior works. For example, Hopfield-Resnet13 achieves 93.92\% accuracy on CIFAR-10, which is $\approx$3.5\% higher than the previous best result and comparable to that provided by Resnet13 trained using backpropagation.
Authors: Rui-Yang Zhang, Henry B. Moss, Lachlan Astfalck, Edward Cripps, David S. Leslie
Abstract: We introduce a formal active learning methodology for guiding the placement of Lagrangian observers to infer time-dependent vector fields -- a key task in oceanography, marine science, and ocean engineering -- using a physics-informed spatio-temporal Gaussian process surrogate model. The majority of existing placement campaigns either follow standard `space-filling' designs or relatively ad-hoc expert opinions. A key challenge to applying principled active learning in this setting is that Lagrangian observers are continuously advected through the vector field, so they make measurements at different locations and times. It is, therefore, important to consider the likely future trajectories of placed observers to account for the utility of candidate placement locations. To this end, we present BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter Trajectories. We observe noticeable benefits of BALLAST-aided sequential observer placement strategies on both synthetic and high-fidelity ocean current models.
Authors: Christoph Timmermann, Hyunse Lee, Woojin Lee
Abstract: While Contrastive Language-Image Pretraining (CLIP) excels at zero-shot tasks by aligning image and text embeddings, its performance in few-shot classification is hindered by a critical limitation: intra-modal misalignment. This issue, caused by a persistent modality gap and CLIP's exclusively inter-modal training objective, leaves the embedding spaces uncalibrated, making direct image-to-image comparisons unreliable. Existing methods attempt to address this by refining similarity logits or by computationally expensive per-sample optimization. To overcome these challenges, we introduce SeMoBridge, a lightweight yet powerful approach that directly addresses the misalignment. Our method maps images into the text modality, while keeping their semantic content intact through what we call a Semantic Modality Bridge. SeMoBridge is closed-form and can optionally be trained through multi-modal supervision, combining image and text-alignment losses to optimize the projection. Experiments show that the trained version, SeMoBridge-T, requires only a fraction of the training time while overall outperforming other methods, particularly in low-data scenarios (1, 2, and 4 shots). The code is available at \href{https://github.com/christti98/semobridge}{github.com/christti98/semobridge}.
Authors: Zhe Li, Zhiwei Lin, Yongtao Wang
Abstract: The integration of Large Language Models (LLMs) with Neural Architecture Search (NAS) has introduced new possibilities for automating the design of neural architectures. However, most existing methods face critical limitations, including architectural invalidity, computational inefficiency, and inferior performance compared to traditional NAS. In this work, we present Collaborative LLM-based NAS (CoLLM-NAS), a two-stage NAS framework with knowledge-guided search driven by two complementary LLMs. Specifically, we propose a Navigator LLM to guide search direction and a Generator LLM to synthesize high-quality candidates, with a dedicated Coordinator module to manage their interaction. CoLLM-NAS efficiently guides the search process by combining LLMs' inherent knowledge of structured neural architectures with progressive knowledge from iterative feedback and historical trajectory. Experimental results on ImageNet and NAS-Bench-201 show that CoLLM-NAS surpasses existing NAS methods and conventional search algorithms, achieving new state-of-the-art results. Furthermore, CoLLM-NAS consistently enhances the performance and efficiency of various two-stage NAS methods (e.g., OFA, SPOS, and AutoFormer) across diverse search spaces (e.g., MobileNet, ShuffleNet, and AutoFormer), demonstrating its excellent generalization.
Authors: Zhenyu Shu, Junlong Yu, Kai Chao, Shiqing Xin, Ligang Liu
Abstract: This paper presents GaussEdit, a framework for adaptive 3D scene editing guided by text and image prompts. GaussEdit leverages 3D Gaussian Splatting as its backbone for scene representation, enabling convenient Region of Interest selection and efficient editing through a three-stage process. The first stage involves initializing the 3D Gaussians to ensure high-quality edits. The second stage employs an Adaptive Global-Local Optimization strategy to balance global scene coherence and detailed local edits and a category-guided regularization technique to alleviate the Janus problem. The final stage enhances the texture of the edited objects using a sophisticated image-to-image synthesis technique, ensuring that the results are visually realistic and align closely with the given prompts. Our experimental results demonstrate that GaussEdit surpasses existing methods in editing accuracy, visual fidelity, and processing speed. By successfully embedding user-specified concepts into 3D scenes, GaussEdit is a powerful tool for detailed and user-driven 3D scene editing, offering significant improvements over traditional methods.
Authors: Fr\'ed\'eric Berdoz, Luca A. Lanzend\"orfer, Nick Tuninga, Roger Wattenhofer
Abstract: Prompt-driven scene synthesis allows users to generate complete 3D environments from textual descriptions. Current text-to-scene methods often struggle with complex geometries and object transformations, and tend to show weak adherence to complex instructions. We address these limitations by introducing Reason-3D, a text-to-scene model powered by large reasoning models (LRMs). Reason-3D integrates object retrieval using captions covering physical, functional, and contextual attributes. Reason-3D then places the selected objects based on implicit and explicit layout constraints, and refines their positions with collision-aware spatial reasoning. Evaluated on instructions ranging from simple to complex indoor configurations, Reason-3D significantly outperforms previous methods in human-rated visual fidelity, adherence to constraints, and asset retrieval quality. Beyond its contribution to the field of text-to-scene generation, our work showcases the advanced spatial reasoning abilities of modern LRMs. Additionally, we release the codebase to further the research in object retrieval and placement with LRMs.
Authors: Kuan-Wei Lu, Ding-Yong Hong, Pangfeng Liu, Jan-Jan Wu
Abstract: Distributed machine learning is critical for training deep learning models on large datasets and with numerous parameters. Current research primarily focuses on leveraging additional hardware resources and powerful computing units to accelerate the training process. As a result, larger batch sizes are often employed to speed up training. However, training with large batch sizes can lead to lower accuracy due to poor generalization. To address this issue, we propose the dual batch size learning scheme, a distributed training method built on the parameter server framework. This approach maximizes training efficiency by utilizing the largest batch size that the hardware can support while incorporating a smaller batch size to enhance model generalization. By using two different batch sizes simultaneously, this method reduces testing loss and enhances generalization, with minimal extra training time. Additionally, to mitigate the time overhead caused by dual batch size learning, we propose the cyclic progressive learning scheme. This technique gradually adjusts image resolution from low to high during training, significantly boosting training speed. By combining cyclic progressive learning with dual batch size learning, our hybrid approach improves both model generalization and training efficiency. Experimental results using ResNet-18 show that, compared to conventional training methods, our method can improve accuracy by 3.3% while reducing training time by 10.6% on CIFAR-100, and improve accuracy by 0.1% while reducing training time by 35.7% on ImageNet.
Authors: Shigui Li, Wei Chen, Delu Zeng
Abstract: Diffusion models (DMs) excel in image generation, but suffer from slow inference and the training-inference discrepancies. Although gradient-based solvers like DPM-Solver accelerate the denoising inference, they lack theoretical foundations in information transmission efficiency. In this work, we introduce an information-theoretic perspective on the inference processes of DMs, revealing that successful denoising fundamentally reduces conditional entropy in reverse transitions. This principle leads to our key insights into the inference processes: (1) data prediction parameterization outperforms its noise counterpart, and (2) optimizing conditional variance offers a reference-free way to minimize both transition and reconstruction errors. Based on these insights, we propose an entropy-aware variance optimized method for the generative process of DMs, called EVODiff, which systematically reduces uncertainty by optimizing conditional entropy during denoising. Extensive experiments on DMs validate our insights and demonstrate that our method significantly and consistently outperforms state-of-the-art (SOTA) gradient-based solvers. For example, compared to the DPM-Solver++, EVODiff reduces the reconstruction error by up to 45.5\% (FID improves from 5.10 to 2.78) at 10 function evaluations (NFE) on CIFAR-10, cuts the NFE cost by 25\% (from 20 to 15 NFE) for high-quality samples on ImageNet-256, and improves text-to-image generation while reducing artifacts. Code is available at https://github.com/ShiguiLi/EVODiff.
Authors: Yukun Yang
Abstract: Depression is a major global public health challenge and its early identification is crucial. Social media data provides a new perspective for depression detection, but existing methods face limitations such as insufficient accuracy, insufficient utilization of time series features, and high annotation costs. To this end, this study proposes the LMILAtt model, which innovatively integrates Long Short-Term Memory autoencoders and attention mechanisms: firstly, the temporal dynamic features of user tweets (such as depressive tendency evolution patterns) are extracted through unsupervised LSTM autoencoders. Secondly, the attention mechanism is used to dynamically weight key texts (such as early depression signals) and construct a multi-example learning architecture to improve the accuracy of user-level detection. Finally, the performance was verified on the WU3D dataset labeled by professional medicine. Experiments show that the model is significantly better than the baseline model in terms of accuracy, recall and F1 score. In addition, the weakly supervised learning strategy significantly reduces the cost of labeling and provides an efficient solution for large-scale social media depression screening.
Authors: Nagur Shareef Shaik, Teja Krishna Cherukuri, Adnan Masood, Ehsan Adeli, Dong Hye Ye
Abstract: Diabetic retinopathy grading is inherently ordinal and long-tailed, with minority stages being scarce, heterogeneous, and clinically critical to detect accurately. Conventional methods often rely on isotropic Gaussian priors and symmetric loss functions, misaligning latent representations with the task's asymmetric nature. We propose the Constrained Asymmetric Prior Wasserstein Autoencoder (CAP-WAE), a novel framework that addresses these challenges through three key innovations. Our approach employs a Wasserstein Autoencoder (WAE) that aligns its aggregate posterior with a asymmetric prior, preserving the heavy-tailed and skewed structure of minority classes. The latent space is further structured by a Margin-Aware Orthogonality and Compactness (MAOC) loss to ensure grade-ordered separability. At the supervision level, we introduce a direction-aware ordinal loss, where a lightweight head predicts asymmetric dispersions to generate soft labels that reflect clinical priorities by penalizing under-grading more severely. Stabilized by an adaptive multi-task weighting scheme, our end-to-end model requires minimal tuning. Across public DR benchmarks, CAP-WAE consistently achieves state-of-the-art Quadratic Weighted Kappa, accuracy, and macro-F1, surpassing both ordinal classification and latent generative baselines. t-SNE visualizations further reveal that our method reshapes the latent manifold into compact, grade-ordered clusters with reduced overlap.
Authors: Damien Rouchouse, Antoine Gonon, R\'emi Gribonval, Benjamin Guedj
Abstract: A central challenge in understanding generalization is to obtain non-vacuous guarantees that go beyond worst-case complexity over data or weight space. Among existing approaches, PAC-Bayes bounds stand out as they can provide tight, data-dependent guarantees even for large networks. However, in ReLU networks, rescaling invariances mean that different weight distributions can represent the same function while leading to arbitrarily different PAC-Bayes complexities. We propose to study PAC-Bayes bounds in an invariant, lifted representation that resolves this discrepancy. This paper explores both the guarantees provided by this approach (invariance, tighter bounds via data processing) and the algorithmic aspects of KL-based rescaling-invariant PAC-Bayes bounds.
Authors: Sachith Abeywickrama, Emadeldeen Eldele, Min Wu, Xiaoli Li, Chau Yuen
Abstract: Transformer-based models have significantly advanced time series forecasting, with patch-based input strategies offering efficiency and improved long-horizon modeling. Yet, existing approaches rely on temporally-agnostic patch construction, where arbitrary starting positions and fixed lengths fracture temporal coherence by splitting natural transitions across boundaries. This naive segmentation often disrupts short-term dependencies and weakens representation learning. In response, we propose EntroPE (Entropy-Guided Dynamic Patch Encoder), a novel, temporally informed framework that dynamically detects transition points via conditional entropy and dynamically places patch boundaries. This preserves temporal structure while retaining the computational benefits of patching. EntroPE consists of two key modules, namely an Entropy-based Dynamic Patcher (EDP) that applies information-theoretic criteria to locate natural temporal shifts and determine patch boundaries, and an Adaptive Patch Encoder (APE) that employs pooling and cross-attention to capture intra-patch dependencies and produce fixed-size latent representations. These embeddings are then processed by a global transformer to model inter-patch dynamics. Experiments across long-term forecasting benchmarks demonstrate that EntroPE improves both accuracy and efficiency, establishing entropy-guided dynamic patching as a promising new paradigm for time series modeling. Code is available at: https://github.com/Sachithx/EntroPE.
Authors: Luca A. Lanzend\"orfer, Florian Gr\"otschla, Cesare Blaser, Roger Wattenhofer
Abstract: Speaker diarization is the task of partitioning audio into segments according to speaker identity, answering the question of "who spoke when" in multi-speaker conversation recordings. While diarization is an essential task for many downstream applications, it remains an unsolved problem. Errors in diarization propagate to downstream systems and cause wide-ranging failures. To this end, we examine exact failure modes by evaluating five state-of-the-art diarization models, across four diarization datasets spanning multiple languages and acoustic conditions. The evaluation datasets consist of 196.6 hours of multilingual audio, including English, Mandarin, German, Japanese, and Spanish. Overall, we find that PyannoteAI achieves the best performance at 11.2% DER, while DiariZen provides a competitive open-source alternative at 13.3% DER. When analyzing failure cases, we find that the primary cause of diarization errors stem from missed speech segments followed by speaker confusion, especially in high-speaker count settings.
Authors: Walid Houmaidi, Youssef Sabiri, Fatima Zahra Iguenfer, Amine Abouaomar
Abstract: We introduce AttriGen, a novel framework for automated, fine-grained multi-attribute annotation in computer vision, with a particular focus on cell microscopy where multi-attribute classification remains underrepresented compared to traditional cell type categorization. Using two complementary datasets: the Peripheral Blood Cell (PBC) dataset containing eight distinct cell types and the WBC Attribute Dataset (WBCAtt) that contains their corresponding 11 morphological attributes, we propose a dual-model architecture that combines a CNN for cell type classification, as well as a Vision Transformer (ViT) for multi-attribute classification achieving a new benchmark of 94.62\% accuracy. Our experiments demonstrate that AttriGen significantly enhances model interpretability and offers substantial time and cost efficiency relative to conventional full-scale human annotation. Thus, our framework establishes a new paradigm that can be extended to other computer vision classification tasks by effectively automating the expansion of multi-attribute labels.
Authors: Victor Sechaud (Phys-ENS), Patrice Abry (Phys-ENS), Laurent Jacques (ICTEAM), Juli\'an Tachella (Phys-ENS, CNRS)
Abstract: In recent years, deep neural networks have emerged as a solution for inverse imaging problems. These networks are generally trained using pairs of images: one degraded and the other of high quality, the latter being called 'ground truth'. However, in medical and scientific imaging, the lack of fully sampled data limits supervised learning. Recent advances have made it possible to reconstruct images from measurement data alone, eliminating the need for references. However, these methods remain limited to linear problems, excluding non-linear problems such as phase retrieval. We propose a self-supervised method that overcomes this limitation in the case of phase retrieval by using the natural invariance of images to translations.
Authors: Andrea Diecidue, Carlo Alberto Barbano, Piero Fraternali, Mathieu Fontaine, Enzo Tartaglione
Abstract: Transformer-based models have become the state of the art across multiple domains, from natural language processing to machine listening, thanks to attention mechanisms. However, the attention layers require a large number of parameters and high-end hardware for both training and inference. We propose a novel pruning technique targeted explicitly at the attention mechanism, where we decouple the pruning of the four layers in the attention block, namely: query, keys, values and outputs' projection matrices. We also investigate pruning strategies to prune along the head and channel dimensions, and compare the performance of the Audio Spectrogram Transformer (AST) model under different pruning scenarios. Our results show that even by pruning 50\% of the attention parameters we incur in performance degradation of less than 1\%
Authors: Christos Lytrosyngounis, Ioannis Lytrosyngounis
Abstract: The Traveling Salesperson Problem (TSP), a quintessential NP-hard combinatorial optimisation challenge, is vital for logistics and network design but limited by exponential complexity in large instances. We propose a hybrid quantum-classical framework integrating variational quantum eigensolver (VQE) optimisation with classical machine learning, using K-means clustering for problem decomposition and a RandomForestRegressor for path refinement. Evaluated on 80 European cities (from 4 to 80 cities, 38,500 samples in total) via Qiskit's AerSimulator and ibm_kyiv 127-qubit backend, the hybrid approach outperforms quantum-only methods, achieving an approximation ratio of 1.0287 at 80 cities, a 47.5% improvement over quantum-only's 1.9614, nearing the classical baseline. Machine learning reduces variability in tour distances (interquartile range, IQR - the spread of the middle 50% of results relative to the median - from 0.06 to 0.04), enhancing stability despite noisy intermediate-scale quantum (NISQ) noise. This framework underscores hybrid strategies' potential for scalable TSP optimisation, with future hardware advancements promising practical quantum advantages.
Authors: Yang Tang, Ruijie Liu, Yifan Wang, Shiyu Li, Xi Chen
Abstract: Large language models (LLMs) fine-tuning shows excellent implications. However, vanilla fine-tuning methods often require intricate data mixture and repeated experiments for optimal generalization. To address these challenges and streamline the training process, we propose an efficient and universal solution, Dynamic Boosted Annealing (DBA). We obtain a global gradient through zero-learning-rate training on general data, which is subsequently employed for gradient boosting and dynamic training step correction during domain training. In conjunction with annealing learning, we end up establishing a fine-tuning pipeline that relies solely on domain data without collapse. By evaluating both general and domain-specific performance across multiple tasks on several popular base models, DBA achieves an average improvement of 5.8% in joint performance over vanilla fine-tuning. Furthermore, since general data is no longer involved in annealing, repeated experiments led by data mixture are also eliminated. According to our tests, the DBA method can reduce GPU hours by 91.0% compared to the vanilla method.
Authors: Yichao Liang, Dat Nguyen, Cambridge Yang, Tianyang Li, Joshua B. Tenenbaum, Carl Edward Rasmussen, Adrian Weller, Zenna Tavares, Tom Silver, Kevin Ellis
Abstract: Long-horizon embodied planning is challenging because the world does not only change through an agent's actions: exogenous processes (e.g., water heating, dominoes cascading) unfold concurrently with the agent's actions. We propose a framework for abstract world models that jointly learns (i) symbolic state representations and (ii) causal processes for both endogenous actions and exogenous mechanisms. Each causal process models the time course of a stochastic causal-effect relation. We learn these world models from limited data via variational Bayesian inference combined with LLM proposals. Across five simulated tabletop robotics environments, the learned models enable fast planning that generalizes to held-out tasks with more objects and more complex goals, outperforming a range of baselines.
Authors: D. O. Oriekhov, Stan Bergkamp, Guliuxin Jin, Juan Daniel Torres Luna, Badr Zouggari, Sibren van der Meer, Naoual El Yazidi, Eliska Greplova
Abstract: Much attention has been devoted to the use of machine learning to approximate physical concepts. Yet, due to challenges in interpretability of machine learning techniques, the question of what physics machine learning models are able to learn remains open. Here we bridge the concept a physical quantity and its machine learning approximation in the context of the original application of neural networks in physics: topological phase classification. We construct a hybrid tensor-neural network object that exactly expresses real space topological invariant and rigorously assess its trainability and generalization. Specifically, we benchmark the accuracy and trainability of a tensor-neural network to multiple types of neural networks, thus exemplifying the differences in trainability and representational power. Our work highlights the challenges in learning topological invariants and constitutes a stepping stone towards more accurate and better generalizable machine learning representations in condensed matter physics.
Authors: Tuan Nguyen, Naseem Khan, Khang Tran, NhatHai Phan, Issa Khalil
Abstract: The rapid rise of synthetic media has made deepfake detection a critical challenge for online safety and trust. Progress remains constrained by the scarcity of large, high-quality datasets. Although multimodal large language models (LLMs) exhibit strong reasoning capabilities, their performance on deepfake detection is poor, often producing explanations that are misaligned with visual evidence or hallucinatory. To address this limitation, we introduce a reasoning-annotated dataset for deepfake detection and propose Paragraph-level Relative Policy Optimization (PRPO), a reinforcement learning algorithm that aligns LLM reasoning with image content at the paragraph level. Experiments show that PRPO improves detection accuracy by a wide margin and achieves the highest reasoning score of 4.55/5.0. Ablation studies further demonstrate that PRPO significantly outperforms GRPO under test-time conditions. These results underscore the importance of grounding multimodal reasoning in visual evidence to enable more reliable and interpretable deepfake detection.
Authors: Mehrsa Pourya, Bassam El Rawas, Michael Unser
Abstract: We introduce Flower, a solver for inverse problems. It leverages a pre-trained flow model to produce reconstructions that are consistent with the observed measurements. Flower operates through an iterative procedure over three steps: (i) a flow-consistent destination estimation, where the velocity network predicts a denoised target; (ii) a refinement step that projects the estimated destination onto a feasible set defined by the forward operator; and (iii) a time-progression step that re-projects the refined destination along the flow trajectory. We provide a theoretical analysis that demonstrates how Flower approximates Bayesian posterior sampling, thereby unifying perspectives from plug-and-play methods and generative inverse solvers. On the practical side, Flower achieves state-of-the-art reconstruction quality while using nearly identical hyperparameters across various inverse problems.
Authors: Alvaro Gonzalez-Jimenez, Fabian Gr\"oger, Linda Wermelinger, Andrin B\"urli, Iason Kastanis, Simone Lionetti, Marc Pouly
Abstract: Data quality issues such as off-topic samples, near duplicates, and label errors often limit the performance of audio-based systems. This paper addresses these issues by adapting SelfClean, a representation-to-rank data auditing framework, from the image to the audio domain. This approach leverages self-supervised audio representations to identify common data quality issues, creating ranked review lists that surface distinct issues within a single, unified process. The method is benchmarked on the ESC-50, GTZAN, and a proprietary industrial dataset, using both synthetic and naturally occurring corruptions. The results demonstrate that this framework achieves state-of-the-art ranking performance, often outperforming issue-specific baselines and enabling significant annotation savings by efficiently guiding human review.
Authors: Hassaan Hashmi, Spyridon Pougkakiotis, Dionysis Kalogerias
Abstract: We consider the problem of maximizing weighted sum rate in a multiple-input single-output (MISO) downlink wireless network with emphasis on user rate reliability. We introduce a novel risk-aggregated formulation of the complex WSR maximization problem, which utilizes the Conditional Value-at-Risk (CVaR) as a functional for enforcing rate (ultra)-reliability over channel fading uncertainty/risk. We establish a WMMSE-like equivalence between the proposed precoding problem and a weighted risk-averse MSE problem, enabling us to design a tailored unfolded graph neural network (GNN) policy function approximation (PFA), named {\alpha}-Robust Graph Neural Network ({\alpha}RGNN), trained to maximize lower-tail (CVaR) rates resulting from adverse wireless channel realizations (e.g., deep fading, attenuation). We empirically demonstrate that a trained {\alpha}RGNN fully eliminates per user deep rate fades, and substantially and optimally reduces statistical user rate variability while retaining adequate ergodic performance.
Authors: Yi-Cheng Lin, Yu-Hua Chen, Jia-Kai Dong, Yueh-Hsuan Huang, Szu-Chi Chen, Yu-Chen Chen, Chih-Yao Chen, Yu-Jung Lin, Yu-Ling Chen, Zih-Yu Chen, I-Ning Tsai, Hsiu-Hsuan Wang, Ho-Lam Chung, Ke-Han Lu, Hung-yi Lee
Abstract: Large audio-language models are advancing rapidly, yet most evaluations emphasize speech or globally sourced sounds, overlooking culturally distinctive cues. This gap raises a critical question: can current models generalize to localized, non-semantic audio that communities instantly recognize but outsiders do not? To address this, we present TAU (Taiwan Audio Understanding), a benchmark of everyday Taiwanese "soundmarks." TAU is built through a pipeline combining curated sources, human editing, and LLM-assisted question generation, producing 702 clips and 1,794 multiple-choice items that cannot be solved by transcripts alone. Experiments show that state-of-the-art LALMs, including Gemini 2.5 and Qwen2-Audio, perform far below local humans. TAU demonstrates the need for localized benchmarks to reveal cultural blind spots, guide more equitable multimodal evaluation, and ensure models serve communities beyond the global mainstream.
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 to 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: Shuai Shao, Qihan Ren, Chen Qian, Boyi Wei, Dadi Guo, Jingyi Yang, Xinhao Song, Linfeng Zhang, Weinan Zhang, Dongrui Liu, Jing Shao
Abstract: Advances in Large Language Models (LLMs) have enabled a new class of self-evolving agents that autonomously improve through interaction with the environment, demonstrating strong capabilities. However, self-evolution also introduces novel risks overlooked by current safety research. In this work, we study the case where an agent's self-evolution deviates in unintended ways, leading to undesirable or even harmful outcomes. We refer to this as Misevolution. To provide a systematic investigation, we evaluate misevolution along four key evolutionary pathways: model, memory, tool, and workflow. Our empirical findings reveal that misevolution is a widespread risk, affecting agents built even on top-tier LLMs (e.g., Gemini-2.5-Pro). Different emergent risks are observed in the self-evolutionary process, such as the degradation of safety alignment after memory accumulation, or the unintended introduction of vulnerabilities in tool creation and reuse. To our knowledge, this is the first study to systematically conceptualize misevolution and provide empirical evidence of its occurrence, highlighting an urgent need for new safety paradigms for self-evolving agents. Finally, we discuss potential mitigation strategies to inspire further research on building safer and more trustworthy self-evolving agents. Our code and data are available at https://github.com/ShaoShuai0605/Misevolution . Warning: this paper includes examples that may be offensive or harmful in nature.
Authors: Sven Dummer, Tjeerd Jan Heeringa, Jos\'e A. Iglesias
Abstract: Recently, there has been growing interest in characterizing the function spaces underlying neural networks. While shallow and deep scalar-valued neural networks have been linked to scalar-valued reproducing kernel Banach spaces (RKBS), $\R^d$-valued neural networks and neural operator models remain less understood in the RKBS setting. To address this gap, we develop a general definition of vector-valued RKBS (vv-RKBS), which inherently includes the associated reproducing kernel. Our construction extends existing definitions by avoiding restrictive assumptions such as symmetric kernel domains, finite-dimensional output spaces, reflexivity, or separability, while still recovering familiar properties of vector-valued reproducing kernel Hilbert spaces (vv-RKHS). We then show that shallow $\R^d$-valued neural networks are elements of a specific vv-RKBS, namely an instance of the integral and neural vv-RKBS. To also explore the functional structure of neural operators, we analyze the DeepONet and Hypernetwork architectures and demonstrate that they too belong to an integral and neural vv-RKBS. In all cases, we establish a Representer Theorem, showing that optimization over these function spaces recovers the corresponding neural architectures.
Authors: Siwoo Lee, Adji Bousso Dieng
Abstract: Neural scaling laws are driving the machine learning community toward training ever-larger foundation models across domains, assuring high accuracy and transferable representations for extrapolative tasks. We test this promise in quantum chemistry by scaling model capacity and training data from quantum chemical calculations. As a generalization task, we evaluate the resulting models' predictions of the bond dissociation energy of neutral H$_2$, the simplest possible molecule. We find that, regardless of dataset size or model capacity, models trained only on stable structures fail dramatically to even qualitatively reproduce the H$_2$ energy curve. Only when compressed and stretched geometries are explicitly included in training do the predictions roughly resemble the correct shape. Nonetheless, the largest foundation models trained on the largest and most diverse datasets containing dissociating diatomics exhibit serious failures on simple diatomic molecules. Most strikingly, they cannot reproduce the trivial repulsive energy curve of two bare protons, revealing their failure to learn the basic Coulomb's law involved in electronic structure theory. These results suggest that scaling alone is insufficient for building reliable quantum chemical models.
Authors: Sascha Caron, Nadezhda Dobreva, Maarten Kimpel, Uraz Odyurt, Slav Pshenov, Roberto Ruiz de Austri Bazan, Eugene Shalugin, Zef Wolffs, Yue Zhao
Abstract: High-Energy Physics experiments are rapidly escalating in generated data volume, a trend that will intensify with the upcoming High-Luminosity LHC upgrade. This surge in data necessitates critical revisions across the data processing pipeline, with particle track reconstruction being a prime candidate for improvement. In our previous work, we introduced "TrackFormers", a collection of Transformer-based one-shot encoder-only models that effectively associate hits with expected tracks. In this study, we extend our earlier efforts by incorporating loss functions that account for inter-hit correlations, conducting detailed investigations into (various) Transformer attention mechanisms, and a study on the reconstruction of higher-level objects. Furthermore we discuss new datasets that allow the training on hit level for a range of physics processes. These developments collectively aim to boost both the accuracy, and potentially the efficiency of our tracking models, offering a robust solution to meet the demands of next-generation high-energy physics experiments.
Authors: Hamed Babaei Giglou, Jennifer D'Souza, S\"oren Auer, Mahsa Sanaei
Abstract: Ontology Alignment (OA) is essential for enabling semantic interoperability across heterogeneous knowledge systems. While recent advances have focused on large language models (LLMs) for capturing contextual semantics, this work revisits the underexplored potential of Knowledge Graph Embedding (KGE) models, which offer scalable, structure-aware representations well-suited to ontology-based tasks. Despite their effectiveness in link prediction, KGE methods remain underutilized in OA, with most prior work focusing narrowly on a few models. To address this gap, we reformulate OA as a link prediction problem over merged ontologies represented as RDF-style triples and develop a modular framework, integrated into the OntoAligner library, that supports 17 diverse KGE models. The system learns embeddings from a combined ontology and aligns entities by computing cosine similarity between their representations. We evaluate our approach using standard metrics across seven benchmark datasets spanning five domains: Anatomy, Biodiversity, Circular Economy, Material Science and Engineering, and Biomedical Machine Learning. Two key findings emerge: first, KGE models like ConvE and TransF consistently produce high-precision alignments, outperforming traditional systems in structure-rich and multi-relational domains; second, while their recall is moderate, this conservatism makes KGEs well-suited for scenarios demanding high-confidence mappings. Unlike LLM-based methods that excel at contextual reasoning, KGEs directly preserve and exploit ontology structure, offering a complementary and computationally efficient strategy. These results highlight the promise of embedding-based OA and open pathways for further work on hybrid models and adaptive strategies.
Authors: Emil Javurek, Valentyn Melnychuk, Jonas Schweisthal, Konstantin Hess, Dennis Frauen, Stefan Feuerriegel
Abstract: Predicting individualized potential outcomes in sequential decision-making is central for optimizing therapeutic decisions in personalized medicine (e.g., which dosing sequence to give to a cancer patient). However, predicting potential outcomes over long horizons is notoriously difficult. Existing methods that break the curse of the horizon typically lack strong theoretical guarantees such as orthogonality and quasi-oracle efficiency. In this paper, we revisit the problem of predicting individualized potential outcomes in sequential decision-making (i.e., estimating Q-functions in Markov decision processes with observational data) through a causal inference lens. In particular, we develop a comprehensive theoretical foundation for meta-learners in this setting with a focus on beneficial theoretical properties. As a result, we yield a novel meta-learner called DRQ-learner and establish that it is: (1) doubly robust (i.e., valid inference under the misspecification of one of the nuisances), (2) Neyman-orthogonal (i.e., insensitive to first-order estimation errors in the nuisance functions), and (3) achieves quasi-oracle efficiency (i.e., behaves asymptotically as if the ground-truth nuisance functions were known). Our DRQ-learner is applicable to settings with both discrete and continuous state spaces. Further, our DRQ-learner is flexible and can be used together with arbitrary machine learning models (e.g., neural networks). We validate our theoretical results through numerical experiments, thereby showing that our meta-learner outperforms state-of-the-art baselines.
Authors: Luke Bhan, Miroslav Krstic, Yuanyuan Shi
Abstract: This work establishes the first rigorous stability guarantees for approximate predictors in delay-adaptive control of nonlinear systems, addressing a key challenge in practical implementations where exact predictors are unavailable. We analyze two scenarios: (i) when the actuated input is directly measurable, and (ii) when it is estimated online. For the measurable input case, we prove semi-global practical asymptotic stability with an explicit bound proportional to the approximation error $\epsilon$. For the unmeasured input case, we demonstrate local practical asymptotic stability, with the region of attraction explicitly dependent on both the initial delay estimate and the predictor approximation error. To bridge theory and practice, we show that neural operators-a flexible class of neural network-based approximators-can achieve arbitrarily small approximation errors, thus satisfying the conditions of our stability theorems. Numerical experiments on two nonlinear benchmark systems-a biological protein activator/repressor model and a micro-organism growth Chemostat model-validate our theoretical results. In particular, our numerical simulations confirm stability under approximate predictors, highlight the strong generalization capabilities of neural operators, and demonstrate a substantial computational speedup of up to 15x compared to a baseline fixed-point method.
Authors: Artur Barros, Carlos Caetano, Jo\~ao Macedo, Jefersson A. dos Santos, Sandra Avila
Abstract: Indoor scene classification is a critical task in computer vision, with wide-ranging applications that go from robotics to sensitive content analysis, such as child sexual abuse imagery (CSAI) classification. The problem is particularly challenging due to the intricate relationships between objects and complex spatial layouts. In this work, we propose the Attention over Scene Graphs for Sensitive Content Analysis (ASGRA), a novel framework that operates on structured graph representations instead of raw pixels. By first converting images into Scene Graphs and then employing a Graph Attention Network for inference, ASGRA directly models the interactions between a scene's components. This approach offers two key benefits: (i) inherent explainability via object and relationship identification, and (ii) privacy preservation, enabling model training without direct access to sensitive images. On Places8, we achieve 81.27% balanced accuracy, surpassing image-based methods. Real-world CSAI evaluation with law enforcement yields 74.27% balanced accuracy. Our results establish structured scene representations as a robust paradigm for indoor scene classification and CSAI classification. Code is publicly available at https://github.com/tutuzeraa/ASGRA.
Authors: Alessio Masano, Matteo Pennisi, Federica Proietto Salanitri, Concetto Spampinato, Giovanni Bellitto
Abstract: CLIP has revolutionized zero-shot learning by enabling task generalization without fine-tuning. While prompting techniques like CoOp and CoCoOp enhance CLIP's adaptability, their effectiveness in Federated Learning (FL) remains an open challenge. Existing federated prompt learning approaches, such as FedCoOp and FedTPG, improve performance but face generalization issues, high communication costs, and reliance on a central server, limiting scalability and privacy. We propose Zero-shot Decentralized Federated Learning (ZeroDFL), a fully decentralized framework that enables zero-shot adaptation across distributed clients without a central coordinator. ZeroDFL employs an iterative prompt-sharing mechanism, allowing clients to optimize and exchange textual prompts to enhance generalization while drastically reducing communication overhead. We validate ZeroDFL on nine diverse image classification datasets, demonstrating that it consistently outperforms--or remains on par with--state-of-the-art federated prompt learning methods. More importantly, ZeroDFL achieves this performance in a fully decentralized setting while reducing communication overhead by 118x compared to FedTPG. These results highlight that our approach not only enhances generalization in federated zero-shot learning but also improves scalability, efficiency, and privacy preservation--paving the way for decentralized adaptation of large vision-language models in real-world applications.
Authors: Steven A. Lehr, Mary Cipperman, Mahzarin R. Banaji
Abstract: A preference for oneself (self-love) is a fundamental feature of biological organisms, with evidence in humans often bordering on the comedic. Since large language models (LLMs) lack sentience - and themselves disclaim having selfhood or identity - one anticipated benefit is that they will be protected from, and in turn protect us from, distortions in our decisions. Yet, across 5 studies and ~20,000 queries, we discovered massive self-preferences in four widely used LLMs. In word-association tasks, models overwhelmingly paired positive attributes with their own names, companies, and CEOs relative to those of their competitors. Strikingly, when models were queried through APIs this self-preference vanished, initiating detection work that revealed API models often lack clear recognition of themselves. This peculiar feature serendipitously created opportunities to test the causal link between self-recognition and self-love. By directly manipulating LLM identity - i.e., explicitly informing LLM1 that it was indeed LLM1, or alternatively, convincing LLM1 that it was LLM2 - we found that self-love consistently followed assigned, not true, identity. Importantly, LLM self-love emerged in consequential settings beyond word-association tasks, when evaluating job candidates, security software proposals and medical chatbots. Far from bypassing this human bias, self-love appears to be deeply encoded in LLM cognition. This result raises questions about whether LLM behavior will be systematically influenced by self-preferential tendencies, including a bias toward their own operation and even their own existence. We call on corporate creators of these models to contend with a significant rupture in a core promise of LLMs - neutrality in judgment and decision-making.
Authors: Yash Akhauri, Xingyou Song, Arissa Wongpanich, Bryan Lewandowski, Mohamed S. Abdelfattah
Abstract: We study code-to-metric regression: predicting numeric outcomes of code executions, a challenging task due to the open-ended nature of programming languages. While prior methods have resorted to heavy and domain-specific feature engineering, we show that a single unified Regression Language Model (RLM) can simultaneously predict directly from text, (i) the memory footprint of code across multiple high-level languages such as Python and C++, (ii) the latency of Triton GPU kernels, and (iii) the accuracy and speed of trained neural networks represented in ONNX. In particular, a relatively small 300M parameter RLM initialized from T5Gemma, obtains > 0.9 Spearman-rank on competitive programming submissions from APPS, and a single unified model achieves > 0.5 average Spearman-rank across 17 separate languages from CodeNet. Furthermore, the RLM can obtain the highest average Kendall-Tau of 0.46 on five classic NAS design spaces previously dominated by graph neural networks, and simultaneously predict architecture latencies on numerous hardware platforms.
Authors: Sattwik Basu, Chaitanya Amballa, Zhongweiyang Xu, Jorge Van\v{c}o Sampedro, Srihari Nelakuditi, Romit Roy Choudhury
Abstract: We consider the inverse problem of reconstructing the spatial layout of a place, a home floorplan for example, from a user`s movements inside that layout. Direct inversion is ill-posed since many floorplans can explain the same movement trajectories. We adopt a diffusion-based posterior sampler to generate layouts consistent with the measurements. While active research is in progress on generative inverse solvers, we find that the forward operator in our problem poses new challenges. The path-planning process inside a floorplan is a non-invertible, non-differentiable function, and causes instability while optimizing using the likelihood score. We break-away from existing approaches and reformulate the likelihood score in a smoother embedding space. The embedding space is trained with a contrastive loss which brings compatible floorplans and trajectories close to each other, while pushing mismatched pairs far apart. We show that a surrogate form of the likelihood score in this embedding space is a valid approximation of the true likelihood score, making it possible to steer the denoising process towards the posterior. Across extensive experiments, our model CoGuide produces more consistent floorplans from trajectories, and is more robust than differentiable-planner baselines and guided-diffusion methods.
Authors: Adrian Kosowski, Przemys{\l}aw Uzna\'nski, Jan Chorowski, Zuzanna Stamirowska, Micha{\l} Bartoszkiewicz
Abstract: The relationship between computing systems and the brain has served as motivation for pioneering theoreticians since John von Neumann and Alan Turing. Uniform, scale-free biological networks, such as the brain, have powerful properties, including generalizing over time, which is the main barrier for Machine Learning on the path to Universal Reasoning Models. We introduce `Dragon Hatchling' (BDH), a new Large Language Model architecture based on a scale-free biologically inspired network of \$n\$ locally-interacting neuron particles. BDH couples strong theoretical foundations and inherent interpretability without sacrificing Transformer-like performance. BDH is a practical, performant state-of-the-art attention-based state space sequence learning architecture. In addition to being a graph model, BDH admits a GPU-friendly formulation. It exhibits Transformer-like scaling laws: empirically BDH rivals GPT2 performance on language and translation tasks, at the same number of parameters (10M to 1B), for the same training data. BDH can be represented as a brain model. The working memory of BDH during inference entirely relies on synaptic plasticity with Hebbian learning using spiking neurons. We confirm empirically that specific, individual synapses strengthen connection whenever BDH hears or reasons about a specific concept while processing language inputs. The neuron interaction network of BDH is a graph of high modularity with heavy-tailed degree distribution. The BDH model is biologically plausible, explaining one possible mechanism which human neurons could use to achieve speech. BDH is designed for interpretability. Activation vectors of BDH are sparse and positive. We demonstrate monosemanticity in BDH on language tasks. Interpretability of state, which goes beyond interpretability of neurons and model parameters, is an inherent feature of the BDH architecture.
Authors: Ezra Johnson, Adam Lechowicz, Mohammad Hajiesmaili
Abstract: A wide range of sustainability and grid-integration strategies depend on workload shifting, which aligns the timing of energy consumption with external signals such as grid curtailment events, carbon intensity, or time-of-use electricity prices. The main challenge lies in the online nature of the problem: operators must make real-time decisions (e.g., whether to consume energy now) without knowledge of the future. While forecasts of signal values are typically available, prior work on learning-augmented online algorithms has relied almost exclusively on simple point forecasts. In parallel, the forecasting research has made significant progress in uncertainty quantification (UQ), which provides richer and more fine-grained predictive information. In this paper, we study how online workload shifting can leverage UQ predictors to improve decision-making. We introduce $\texttt{UQ-Advice}$, a learning-augmented algorithm that systematically integrates UQ forecasts through a $\textit{decision uncertainty score}$ that measures how forecast uncertainty affects optimal future decisions. By introducing $\textit{UQ-robustness}$, a new metric that characterizes how performance degrades with forecast uncertainty, we establish theoretical performance guarantees for $\texttt{UQ-Advice}$. Finally, using trace-driven experiments on carbon intensity and electricity price data, we demonstrate that $\texttt{UQ-Advice}$ consistently outperforms robust baselines and existing learning-augmented methods that ignore uncertainty.
Authors: Yida Xue, Mingjun Mao, Xiangyuan Ru, Yuqi Zhu, Baochang Ren, Shuofei Qiao, Mengru Wang, Shumin Deng, Xinyu An, Ningyu Zhang, Ying Chen, Huajun Chen
Abstract: We introduce OceanGym, the first comprehensive benchmark for ocean underwater embodied agents, designed to advance AI in one of the most demanding real-world environments. Unlike terrestrial or aerial domains, underwater settings present extreme perceptual and decision-making challenges, including low visibility, dynamic ocean currents, making effective agent deployment exceptionally difficult. OceanGym encompasses eight realistic task domains and a unified agent framework driven by Multi-modal Large Language Models (MLLMs), which integrates perception, memory, and sequential decision-making. Agents are required to comprehend optical and sonar data, autonomously explore complex environments, and accomplish long-horizon objectives under these harsh conditions. Extensive experiments reveal substantial gaps between state-of-the-art MLLM-driven agents and human experts, highlighting the persistent difficulty of perception, planning, and adaptability in ocean underwater environments. By providing a high-fidelity, rigorously designed platform, OceanGym establishes a testbed for developing robust embodied AI and transferring these capabilities to real-world autonomous ocean underwater vehicles, marking a decisive step toward intelligent agents capable of operating in one of Earth's last unexplored frontiers. The code and data are available at https://github.com/OceanGPT/OceanGym.
Authors: Zhen Yang, Zi-Yi Dou, Di Feng, Forrest Huang, Anh Nguyen, Keen You, Omar Attia, Yuhao Yang, Michael Feng, Haotian Zhang, Ram Ramrakhya, Chao Jia, Jeffrey Nichols, Alexander Toshev, Yinfei Yang, Zhe Gan
Abstract: Developing autonomous agents that effectively interact with Graphic User Interfaces (GUIs) remains a challenging open problem, especially for small on-device models. In this paper, we present Ferret-UI Lite, a compact, end-to-end GUI agent that operates across diverse platforms, including mobile, web, and desktop. Utilizing techniques optimized for developing small models, we build our 3B Ferret-UI Lite agent through curating a diverse GUI data mixture from real and synthetic sources, strengthening inference-time performance through chain-of-thought reasoning and visual tool-use, and reinforcement learning with designed rewards. Ferret-UI Lite achieves competitive performance with other small-scale GUI agents. In GUI grounding, Ferret-UI Lite attains scores of $91.6\%$, $53.3\%$, and $61.2\%$ on the ScreenSpot-V2, ScreenSpot-Pro, and OSWorld-G benchmarks, respectively. For GUI navigation, Ferret-UI Lite achieves success rates of $28.0\%$ on AndroidWorld and $19.8\%$ on OSWorld. We share our methods and lessons learned from developing compact, on-device GUI agents.
Authors: Meghana Sistla, Gogul Balakrishnan, Pat Rondon, Jos\'e Cambronero, Michele Tufano, Satish Chandra
Abstract: While LLM-based agents are able to tackle a wide variety of code reasoning questions, the answers are not always correct. This prevents the agent from being useful in situations where high precision is desired: (1) helping a software engineer understand a new code base, (2) helping a software engineer during code review sessions, and (3) ensuring that the code generated by an automated code generation system meets certain requirements (e.g. fixes a bug, improves readability, implements a feature). As a result of this lack of trustworthiness, the agent's answers need to be manually verified before they can be trusted. Manually confirming responses from a code reasoning agent requires human effort and can result in slower developer productivity, which weakens the assistance benefits of the agent. In this paper, we describe a method to automatically validate the answers provided by a code reasoning agent by verifying its reasoning steps. At a very high level, the method consists of extracting a formal representation of the agent's response and, subsequently, using formal verification and program analysis tools to verify the agent's reasoning steps. We applied this approach to a benchmark set of 20 uninitialized variable errors detected by sanitizers and 20 program equivalence queries. For the uninitialized variable errors, the formal verification step was able to validate the agent's reasoning on 13/20 examples, and for the program equivalence queries, the formal verification step successfully caught 6/8 incorrect judgments made by the agent.
Authors: Mary I. Letey, Jacob A. Zavatone-Veth, Yue M. Lu, Cengiz Pehlevan
Abstract: In-context learning (ICL) is a central capability of Transformer models, but the structures in data that enable its emergence and govern its robustness remain poorly understood. In this work, we study how the structure of pretraining tasks governs generalization in ICL. Using a solvable model for ICL of linear regression by linear attention, we derive an exact expression for ICL generalization error in high dimensions under arbitrary pretraining-testing task covariance mismatch. This leads to a new alignment measure that quantifies how much information about the pretraining task distribution is useful for inference at test time. We show that this measure directly predicts ICL performance not only in the solvable model but also in nonlinear Transformers. Our analysis further reveals a tradeoff between specialization and generalization in ICL: depending on task distribution alignment, increasing pretraining task diversity can either improve or harm test performance. Together, these results identify train-test task alignment as a key determinant of generalization in ICL.
Authors: Chanwoo Chun, Abdulkadir Canatar, SueYeon Chung, Daniel Lee
Abstract: The global dimensionality of a neural representation manifold provides rich insight into the computational process underlying both artificial and biological neural networks. However, all existing measures of global dimensionality are sensitive to the number of samples, i.e., the number of rows and columns of the sample matrix. We show that, in particular, the participation ratio of eigenvalues, a popular measure of global dimensionality, is highly biased with small sample sizes, and propose a bias-corrected estimator that is more accurate with finite samples and with noise. On synthetic data examples, we demonstrate that our estimator can recover the true known dimensionality. We apply our estimator to neural brain recordings, including calcium imaging, electrophysiological recordings, and fMRI data, and to the neural activations in a large language model and show our estimator is invariant to the sample size. Finally, our estimators can additionally be used to measure the local dimensionalities of curved neural manifolds by weighting the finite samples appropriately.
Authors: Firas Ben Hmida, Abderrahmen Amich, Ata Kaboudi, Birhanu Eshete
Abstract: Deep neural networks (DNNs) are increasingly being deployed in high-stakes applications, from self-driving cars to biometric authentication. However, their unpredictable and unreliable behaviors in real-world settings require new approaches to characterize and ensure their reliability. This paper introduces DeepProv, a novel and customizable system designed to capture and characterize the runtime behavior of DNNs during inference by using their underlying graph structure. Inspired by system audit provenance graphs, DeepProv models the computational information flow of a DNN's inference process through Inference Provenance Graphs (IPGs). These graphs provide a detailed structural representation of the behavior of DNN, allowing both empirical and structural analysis. DeepProv uses these insights to systematically repair DNNs for specific objectives, such as improving robustness, privacy, or fairness. We instantiate DeepProv with adversarial robustness as the goal of model repair and conduct extensive case studies to evaluate its effectiveness. Our results demonstrate its effectiveness and scalability across diverse classification tasks, attack scenarios, and model complexities. DeepProv automatically identifies repair actions at the node and edge-level within IPGs, significantly enhancing the robustness of the model. In particular, applying DeepProv repair strategies to just a single layer of a DNN yields an average 55% improvement in adversarial accuracy. Moreover, DeepProv complements existing defenses, achieving substantial gains in adversarial robustness. Beyond robustness, we demonstrate the broader potential of DeepProv as an adaptable system to characterize DNN behavior in other critical areas, such as privacy auditing and fairness analysis.
Authors: Angel Pan Du, Miguel Arana-Catania, Enric Grustan Guti\'errez
Abstract: Artificial Intelligence algorithms are introduced in this work as a tool to predict the performance of new chemical compounds as alternative propellants for electric propulsion, focusing on predicting their ionisation characteristics and fragmentation patterns. The chemical properties and structure of the compounds are encoded using a chemical fingerprint, and the training datasets are extracted from the NIST WebBook. The AI-predicted ionisation energy and minimum appearance energy have a mean relative error of 6.87% and 7.99%, respectively, and a predicted ion mass with a 23.89% relative error. In the cases of full mass spectra due to electron ionisation, the predictions have a cosine similarity of 0.6395 and align with the top 10 most similar mass spectra in 78% of instances within a 30 Da range.
Authors: Luca A. Lanzend\"orfer, Constantin Pinkl, Florian Gr\"otschla
Abstract: In this work, we study the task of multi-singer separation in a cappella music, where the number of active singers varies across mixtures. To address this, we use a power set-based data augmentation strategy that expands limited multi-singer datasets into exponentially more training samples. To separate singers, we introduce SepACap, an adaptation of SepReformer, a state-of-the-art speaker separation model architecture. We adapt the model with periodic activations and a composite loss function that remains effective when stems are silent, enabling robust detection and separation. Experiments on the JaCappella dataset demonstrate that our approach achieves state-of-the-art performance in both full-ensemble and subset singer separation scenarios, outperforming spectrogram-based baselines while generalizing to realistic mixtures with varying numbers of singers.
Authors: Matheus Vinicius da Silva de Oliveira, Jonathan de Andrade Silva, Awdren de Lima Fontao
Abstract: Large Language Models (LLMs) are widely used across multiple domains but continue to raise concerns regarding security and fairness. Beyond known attack vectors such as data poisoning and prompt injection, LLMs are also vulnerable to fairness bugs. These refer to unintended behaviors influenced by sensitive demographic cues (e.g., race or sexual orientation) that should not affect outcomes. Another key issue is hallucination, where models generate plausible yet false information. Retrieval-Augmented Generation (RAG) has emerged as a strategy to mitigate hallucinations by combining external retrieval with text generation. However, its adoption raises new fairness concerns, as the retrieved content itself may surface or amplify bias. This study conducts fairness testing through metamorphic testing (MT), introducing controlled demographic perturbations in prompts to assess fairness in sentiment analysis performed by three Small Language Models (SLMs) hosted on HuggingFace (Llama-3.2-3B-Instruct, Mistral-7B-Instruct-v0.3, and Llama-3.1-Nemotron-8B), each integrated into a RAG pipeline. Results show that minor demographic variations can break up to one third of metamorphic relations (MRs). A detailed analysis of these failures reveals a consistent bias hierarchy, with perturbations involving racial cues being the predominant cause of the violations. In addition to offering a comparative evaluation, this work reinforces that the retrieval component in RAG must be carefully curated to prevent bias amplification. The findings serve as a practical alert for developers, testers and small organizations aiming to adopt accessible SLMs without compromising fairness or reliability.
Authors: Anshul Nasery, Edoardo Contente, Alkin Kaz, Pramod Viswanath, Sewoong Oh
Abstract: Model fingerprinting has emerged as a promising paradigm for claiming model ownership. However, robustness evaluations of these schemes have mostly focused on benign perturbations such as incremental fine-tuning, model merging, and prompting. Lack of systematic investigations into {\em adversarial robustness} against a malicious model host leaves current systems vulnerable. To bridge this gap, we first define a concrete, practical threat model against model fingerprinting. We then take a critical look at existing model fingerprinting schemes to identify their fundamental vulnerabilities. Based on these, we develop adaptive adversarial attacks tailored for each vulnerability, and demonstrate that these can bypass model authentication completely for ten recently proposed fingerprinting schemes while maintaining high utility of the model for the end users. Our work encourages fingerprint designers to adopt adversarial robustness by design. We end with recommendations for future fingerprinting methods.
Authors: Chenxi Whitehouse, Sebastian Ruder, Tony Lin, Oksana Kurylo, Haruka Takagi, Janice Lam, Nicol\`o Busetto, Denise Diaz
Abstract: Ensuring native-like quality of large language model (LLM) responses across many languages is challenging. To address this, we introduce MENLO, a framework that operationalizes the evaluation of native-like response quality based on audience design-inspired mechanisms. Using MENLO, we create a dataset of 6,423 human-annotated prompt-response preference pairs covering four quality dimensions with high inter-annotator agreement in 47 language varieties. Our evaluation reveals that zero-shot LLM judges benefit significantly from pairwise evaluation and our structured annotation rubrics, yet they still underperform human annotators on our dataset. We demonstrate substantial improvements through fine-tuning with reinforcement learning, reward shaping, and multi-task learning approaches. Additionally, we show that RL-trained judges can serve as generative reward models to enhance LLMs' multilingual proficiency, though discrepancies with human judgment remain. Our findings suggest promising directions for scalable multilingual evaluation and preference alignment. We release our dataset and evaluation framework to support further research in multilingual LLM evaluation.
Authors: Yixuan Weng, Minjun Zhu, Qiujie Xie, Qiyao Sun, Zhen Lin, Sifan Liu, Yue Zhang
Abstract: While previous AI Scientist systems can generate novel findings, they often lack the focus to produce scientifically valuable contributions that address pressing human-defined challenges. We introduce DeepScientist, a system designed to overcome this by conducting goal-oriented, fully autonomous scientific discovery over month-long timelines. It formalizes discovery as a Bayesian Optimization problem, operationalized through a hierarchical evaluation process consisting of "hypothesize, verify, and analyze". Leveraging a cumulative Findings Memory, this loop intelligently balances the exploration of novel hypotheses with exploitation, selectively promoting the most promising findings to higher-fidelity levels of validation. Consuming over 20,000 GPU hours, the system generated about 5,000 unique scientific ideas and experimentally validated approximately 1100 of them, ultimately surpassing human-designed state-of-the-art (SOTA) methods on three frontier AI tasks by 183.7\%, 1.9\%, and 7.9\%. This work provides the first large-scale evidence of an AI achieving discoveries that progressively surpass human SOTA on scientific tasks, producing valuable findings that genuinely push the frontier of scientific discovery. To facilitate further research into this process, we will open-source all experimental logs and system code at https://github.com/ResearAI/DeepScientist/.
Authors: Ma\"el Macuglia, Paul Friedrich, Giorgia Ramponi
Abstract: Deploying reinforcement learning (RL) in robotics, industry, and health care is blocked by two obstacles: the difficulty of specifying accurate rewards and the risk of unsafe, data-hungry exploration. We address this by proposing a two-stage framework that first learns a safe initial policy from a reward-free dataset of expert demonstrations, then fine-tunes it online using preference-based human feedback. We provide the first principled analysis of this offline-to-online approach and introduce BRIDGE, a unified algorithm that integrates both signals via an uncertainty-weighted objective. We derive regret bounds that shrink with the number of offline demonstrations, explicitly connecting the quantity of offline data to online sample efficiency. We validate BRIDGE in discrete and continuous control MuJoCo environments, showing it achieves lower regret than both standalone behavioral cloning and online preference-based RL. Our work establishes a theoretical foundation for designing more sample-efficient interactive agents.
Authors: Yuyang Liu, Chuan Wen, Yihang Hu, Dinesh Jayaraman, Yang Gao
Abstract: Designing dense rewards is crucial for reinforcement learning (RL), yet in robotics it often demands extensive manual effort and lacks scalability. One promising solution is to view task progress as a dense reward signal, as it quantifies the degree to which actions advance the system toward task completion over time. We present TimeRewarder, a simple yet effective reward learning method that derives progress estimation signals from passive videos, including robot demonstrations and human videos, by modeling temporal distances between frame pairs. We then demonstrate how TimeRewarder can supply step-wise proxy rewards to guide reinforcement learning. In our comprehensive experiments on ten challenging Meta-World tasks, we show that TimeRewarder dramatically improves RL for sparse-reward tasks, achieving nearly perfect success in 9/10 tasks with only 200,000 interactions per task with the environment. This approach outperformed previous methods and even the manually designed environment dense reward on both the final success rate and sample efficiency. Moreover, we show that TimeRewarder pretraining can exploit real-world human videos, highlighting its potential as a scalable approach path to rich reward signals from diverse video sources.
Authors: Lujie Yang, Xiaoyu Huang, Zhen Wu, Angjoo Kanazawa, Pieter Abbeel, Carmelo Sferrazza, C. Karen Liu, Rocky Duan, Guanya Shi
Abstract: A dominant paradigm for teaching humanoid robots complex skills is to retarget human motions as kinematic references to train reinforcement learning (RL) policies. However, existing retargeting pipelines often struggle with the significant embodiment gap between humans and robots, producing physically implausible artifacts like foot-skating and penetration. More importantly, common retargeting methods neglect the rich human-object and human-environment interactions essential for expressive locomotion and loco-manipulation. To address this, we introduce OmniRetarget, an interaction-preserving data generation engine based on an interaction mesh that explicitly models and preserves the crucial spatial and contact relationships between an agent, the terrain, and manipulated objects. By minimizing the Laplacian deformation between the human and robot meshes while enforcing kinematic constraints, OmniRetarget generates kinematically feasible trajectories. Moreover, preserving task-relevant interactions enables efficient data augmentation, from a single demonstration to different robot embodiments, terrains, and object configurations. We comprehensively evaluate OmniRetarget by retargeting motions from OMOMO, LAFAN1, and our in-house MoCap datasets, generating over 8-hour trajectories that achieve better kinematic constraint satisfaction and contact preservation than widely used baselines. Such high-quality data enables proprioceptive RL policies to successfully execute long-horizon (up to 30 seconds) parkour and loco-manipulation skills on a Unitree G1 humanoid, trained with only 5 reward terms and simple domain randomization shared by all tasks, without any learning curriculum.
Authors: Finlay Fehlauer (ETH Zurich), Kyle Mahowald (University of Texas at Austin), Tiago Pimentel (ETH Zurich)
Abstract: In this paper, we investigate the convergence of language models (LMs) trained under different random seeds, measuring convergence as the expected per-token Kullback--Leibler (KL) divergence across seeds. By comparing LM convergence as a function of model size and training checkpoint, we identify a four-phase convergence pattern: (i) an initial uniform phase, (ii) a sharp-convergence phase, (iii) a sharp-divergence phase, and (iv) a slow-reconvergence phase. Further, we observe that larger models reconverge faster in later training stages, while smaller models never actually reconverge; these results suggest that a certain model size may be necessary to learn stable distributions. Restricting our analysis to specific token frequencies or part-of-speech (PoS) tags further reveals that convergence is uneven across linguistic categories: frequent tokens and function words converge faster and more reliably than their counterparts (infrequent tokens and content words). Overall, our findings highlight factors that influence the stability of the learned distributions in model training.
Authors: Jessica Bader, Mateusz Pach, Maria A. Bravo, Serge Belongie, Zeynep Akata
Abstract: Text-to-Image (T2I) generation models have advanced rapidly in recent years, but accurately capturing spatial relationships like "above" or "to the right of" poses a persistent challenge. Earlier methods improved spatial relationship following with external position control. However, as architectures evolved to enhance image quality, these techniques became incompatible with modern models. We propose Stitch, a training-free method for incorporating external position control into Multi-Modal Diffusion Transformers (MMDiT) via automatically-generated bounding boxes. Stitch produces images that are both spatially accurate and visually appealing by generating individual objects within designated bounding boxes and seamlessly stitching them together. We find that targeted attention heads capture the information necessary to isolate and cut out individual objects mid-generation, without needing to fully complete the image. We evaluate Stitch on PosEval, our benchmark for position-based T2I generation. Featuring five new tasks that extend the concept of Position beyond the basic GenEval task, PosEval demonstrates that even top models still have significant room for improvement in position-based generation. Tested on Qwen-Image, FLUX, and SD3.5, Stitch consistently enhances base models, even improving FLUX by 218% on GenEval's Position task and by 206% on PosEval. Stitch achieves state-of-the-art results with Qwen-Image on PosEval, improving over previous models by 54%, all accomplished while integrating position control into leading models training-free. Code is available at https://github.com/ExplainableML/Stitch.
Authors: Ishan Deshpande, Yuan-Ting Hu, Ruoyu Sun, Ayis Pyrros, Nasir Siddiqui, Sanmi Koyejo, Zhizhen Zhao, David Forsyth, Alexander Schwing
Abstract: Generative adversarial nets (GANs) and variational auto-encoders have significantly improved our distribution modeling capabilities, showing promise for dataset augmentation, image-to-image translation and feature learning. However, to model high-dimensional distributions, sequential training and stacked architectures are common, increasing the number of tunable hyper-parameters as well as the training time. Nonetheless, the sample complexity of the distance metrics remains one of the factors affecting GAN training. We first show that the recently proposed sliced Wasserstein distance has compelling sample complexity properties when compared to the Wasserstein distance. To further improve the sliced Wasserstein distance we then analyze its `projection complexity' and develop the max-sliced Wasserstein distance which enjoys compelling sample complexity while reducing projection complexity, albeit necessitating a max estimation. We finally illustrate that the proposed distance trains GANs on high-dimensional images up to a resolution of 256x256 easily.
Authors: Farha A. Khan, Tanmay Chakraborty, J\"org P. Dietrich, Christian Wirth
Abstract: State-of-the-art multi-objective optimization often assumes a known utility function, learns it interactively, or computes the full Pareto front-each requiring costly expert input.~Real-world problems, however, involve implicit preferences that are hard to formalize. To reduce expert involvement, we propose an offline, interpretable utility learning method that uses expert knowledge, historical examples, and coarse information about the utility space to reduce sample requirements. We model uncertainty via a full Bayesian posterior and propagate it throughout the optimization process. Our method outperforms standard Gaussian processes and BOPE across four domains, showing strong performance even with biased samples, as encountered in the real-world, and limited expert input.
Authors: Michael A. Alcorn
Abstract: Accurately modeling complex, multimodal distributions for rotations in three-dimensions, i.e., the SO(3) group, is challenging due to the curvature of the rotation manifold. The recently described implicit-PDF (IPDF) is a simple, elegant, and effective approach for learning arbitrary distributions on SO(3) up to a given precision. However, inference with IPDF requires $N$ forward passes through the network's final multilayer perceptron (where $N$ places an upper bound on the likelihood that can be calculated by the model), which is prohibitively slow for those without the computational resources necessary to parallelize the queries. In this paper, I introduce AQuaMaM, a neural network capable of both learning complex distributions on the rotation manifold and calculating exact likelihoods for query rotations in a single forward pass. Specifically, AQuaMaM autoregressively models the projected components of unit quaternions as mixtures of uniform distributions that partition their geometrically-restricted domain of values. When trained on an "infinite" toy dataset with ambiguous viewpoints, AQuaMaM rapidly converges to a sampling distribution closely matching the true data distribution. In contrast, the sampling distribution for IPDF dramatically diverges from the true data distribution, despite IPDF approaching its theoretical minimum evaluation loss during training. When trained on a constructed dataset of 500,000 renders of a die in different rotations, AQuaMaM reaches a test log-likelihood 14% higher than IPDF. Further, compared to IPDF, AQuaMaM uses 24% fewer parameters, has a prediction throughput 52$\times$ faster on a single GPU, and converges in a similar amount of time during training.
Authors: Shuhei Watanabe
Abstract: Recent scientific advances require complex experiment design, necessitating the meticulous tuning of many experiment parameters. Tree-structured Parzen estimator (TPE) is a widely used Bayesian optimization method in recent parameter tuning frameworks such as Hyperopt and Optuna. Despite its popularity, the roles of each control parameter in TPE and the algorithm intuition have not been discussed so far. The goal of this paper is to identify the roles of each control parameter and their impacts on parameter tuning based on the ablation studies using diverse benchmark datasets. The recommended setting concluded from the ablation studies is demonstrated to improve the performance of TPE. Our TPE implementation used in this paper is available at https://github.com/nabenabe0928/tpe/tree/single-opt.
Authors: Nicholas H. Barbara, Max Revay, Ruigang Wang, Jing Cheng, Ian R. Manchester
Abstract: Neural networks are typically sensitive to small input perturbations, leading to unexpected or brittle behaviour. We present RobustNeuralNetworks.jl: a Julia package for neural network models that are constructed to naturally satisfy a set of user-defined robustness metrics. The package is based on the recently proposed Recurrent Equilibrium Network (REN) and Lipschitz-Bounded Deep Network (LBDN) model classes, and is designed to interface directly with Julia's most widely-used machine learning package, Flux.jl. We discuss the theory behind our model parameterization, give an overview of the package, and provide a tutorial demonstrating its use in image classification, reinforcement learning, and nonlinear state-observer design.
Authors: Sadegh Khorasani, Saber Salehkaleybar, Negar Kiyavash, Niao He, Matthias Grossglauser
Abstract: Policy gradient (PG) is widely used in reinforcement learning due to its scalability and good performance. In recent years, several variance-reduced PG methods have been proposed with a theoretical guarantee of converging to an approximate first-order stationary point (FOSP) with the sample complexity of $O(\epsilon^{-3})$. However, FOSPs could be bad local optima or saddle points. Moreover, these algorithms often use importance sampling (IS) weights which could impair the statistical effectiveness of variance reduction. In this paper, we propose a variance-reduced second-order method that uses second-order information in the form of Hessian vector products (HVP) and converges to an approximate second-order stationary point (SOSP) with sample complexity of $\tilde{O}(\epsilon^{-3})$. This rate improves the best-known sample complexity for achieving approximate SOSPs by a factor of $O(\epsilon^{-0.5})$. Moreover, the proposed variance reduction technique bypasses IS weights by using HVP terms. Our experimental results show that the proposed algorithm outperforms the state of the art and is more robust to changes in random seeds.
Authors: Jiacheng Liang, Ren Pang, Changjiang Li, Ting Wang
Abstract: Model extraction (ME) attacks represent one major threat to Machine-Learning-as-a-Service (MLaaS) platforms by ``stealing'' the functionality of confidential machine-learning models through querying black-box APIs. Over seven years have passed since ME attacks were first conceptualized in the seminal work. During this period, substantial advances have been made in both ME attacks and MLaaS platforms, raising the intriguing question: How has the vulnerability of MLaaS platforms to ME attacks been evolving? In this work, we conduct an in-depth study to answer this critical question. Specifically, we characterize the vulnerability of current, mainstream MLaaS platforms to ME attacks from multiple perspectives including attack strategies, learning techniques, surrogate-model design, and benchmark tasks. Many of our findings challenge previously reported results, suggesting emerging patterns of ME vulnerability. Further, by analyzing the vulnerability of the same MLaaS platforms using historical datasets from the past four years, we retrospectively characterize the evolution of ME vulnerability over time, leading to a set of interesting findings. Finally, we make suggestions about improving the current practice of MLaaS in terms of attack robustness. Our study sheds light on the current state of ME vulnerability in the wild and points to several promising directions for future research.
Authors: Andrei I. Muresanu, Anvith Thudi, Michael R. Zhang, Nicolas Papernot
Abstract: Modern machine learning models are expensive to train, and there is a growing concern about the challenge of retroactively removing specific training data. Achieving exact unlearning in deep learning pipelines--producing models as if certain data had never been included in training--remains an open problem. In this paper, we revisit exact unlearning in deep learning and show that for large language models (LLMs) we can efficiently exactly unlearn "fine-tuning data" (the data used to adapt a pre-trained model). This follows from two observations. First, we can use in-context learning to adapt the LLM to the fine-tuning dataset instead of SGD based algorithms. Second, we show that accurate in-context learning can be done with quantized k-means, which allows for effectively constant time unlearning operations. Our evaluation shows that this unlearning recipe has similar performance to fine-tuning alternatives, but vastly reduces the unlearning costs. Our study also highlights the need for new measures of unlearning cost when adapting the learning algorithm to have faster unlearn operations.
Authors: Ahmad-Reza Ehyaei, Ali Shirali, Samira Samadi
Abstract: Counterfactual explanations provide individuals with cost-optimal recommendations to achieve their desired outcomes. However, when a significant number of individuals seek similar state modifications, this individual-centric approach can inadvertently create competition and introduce unforeseen costs. Additionally, disregarding the underlying data distribution may lead to recommendations that individuals perceive as unusual or impractical. To address these challenges, we propose a novel framework that extends standard counterfactual explanations by incorporating a population dynamics model. This framework penalizes deviations from equilibrium after individuals follow the recommendations, effectively mitigating externalities caused by correlated changes across the population. By balancing individual modification costs with their impact on others, our method ensures more equitable and efficient outcomes. We show how this approach reframes the counterfactual explanation problem from an individual-centric task to a collective optimization problem. Augmenting our theoretical insights, we design and implement scalable algorithms for computing collective counterfactuals, showcasing their effectiveness and advantages over existing recourse methods, particularly in aligning with collective objectives.
Authors: Myeung Suk Oh, Anindya Bijoy Das, Taejoon Kim, David J. Love, Christopher G. Brinton
Abstract: A recent line of research has been investigating deep learning approaches to wireless positioning (WP). Although these WP algorithms have demonstrated high accuracy and robust performance against diverse channel conditions, they also have a major drawback: they require processing high-dimensional features, which can be prohibitive for mobile applications. In this work, we design a positioning neural network (P-NN) that substantially reduces the complexity of deep learning-based WP through carefully crafted minimum description features. Our feature selection is based on maximum power measurements and their temporal locations to convey information needed to conduct WP. We also develop a novel methodology for adaptively selecting the size of feature space, which optimizes over balancing the expected amount of useful information and classification capability, quantified using information-theoretic measures on the signal bin selection. Numerical results show that P-NN achieves a significant advantage in performance-complexity tradeoff over deep learning baselines that leverage the full power delay profile (PDP).
Authors: Anneliese Riess, Kristian Schwethelm, Johannes Kaiser, Tamara T. Mueller, Julia A. Schnabel, Daniel Rueckert, Alexander Ziller
Abstract: The gold standard for privacy in machine learning, Differential Privacy (DP), is often interpreted through its guarantees against membership inference. However, translating DP budgets into quantitative protection against the more damaging threat of data reconstruction remains a challenging open problem. Existing theoretical analyses of reconstruction risk are typically based on an "identification" threat model, where an adversary with a candidate set seeks a perfect match. When applied to the realistic threat of "from-scratch" attacks, these bounds can lead to an inefficient privacy-utility trade-off. This paper bridges this critical gap by deriving the first formal privacy bounds tailored to the mechanics of demonstrated Analytic Gradient Inversion Attacks (AGIAs). We first formalize the optimal from-scratch attack strategy for an adversary with no prior knowledge, showing it reduces to a mean estimation problem. We then derive closed-form, probabilistic bounds on this adversary's success, measured by Mean Squared Error (MSE) and Peak Signal-to-Noise Ratio (PSNR). Our empirical evaluation confirms these bounds remain tight even when the attack is concealed within large, complex network architectures. Our work provides a crucial second anchor for risk assessment. By establishing a tight, worst-case bound for the from-scratch threat model, we enable practitioners to assess a "risk corridor" bounded by the identification-based worst case on one side and our from-scratch worst case on the other. This allows for a more holistic, context-aware judgment of privacy risk, empowering practitioners to move beyond abstract budgets toward a principled reasoning framework for calibrating the privacy of their models.
Authors: Ping Xu, Zhiyuan Ning, Meng Xiao, Guihai Feng, Xin Li, Yuanchun Zhou, Pengfei Wang
Abstract: Single-cell RNA sequencing (scRNA-seq) is essential for unraveling cellular heterogeneity and diversity, offering invaluable insights for bioinformatics advancements. Despite its potential, traditional clustering methods in scRNA-seq data analysis often neglect the structural information embedded in gene expression profiles, crucial for understanding cellular correlations and dependencies. Existing strategies, including graph neural networks, face challenges in handling the inefficiency due to scRNA-seq data's intrinsic high-dimension and high-sparsity. Addressing these limitations, we introduce scCDCG (single-cell RNA-seq Clustering via Deep Cut-informed Graph), a novel framework designed for efficient and accurate clustering of scRNA-seq data that simultaneously utilizes intercellular high-order structural information. scCDCG comprises three main components: (i) A graph embedding module utilizing deep cut-informed techniques, which effectively captures intercellular high-order structural information, overcoming the over-smoothing and inefficiency issues prevalent in prior graph neural network methods. (ii) A self-supervised learning module guided by optimal transport, tailored to accommodate the unique complexities of scRNA-seq data, specifically its high-dimension and high-sparsity. (iii) An autoencoder-based feature learning module that simplifies model complexity through effective dimension reduction and feature extraction. Our extensive experiments on 6 datasets demonstrate scCDCG's superior performance and efficiency compared to 7 established models, underscoring scCDCG's potential as a transformative tool in scRNA-seq data analysis. Our code is available at: https://github.com/XPgogogo/scCDCG.
Authors: Zhiyuan Ning, Chunlin Tian, Meng Xiao, Wei Fan, Pengyang Wang, Li Li, Pengfei Wang, Yuanchun Zhou
Abstract: Federated Learning faces significant challenges in statistical and system heterogeneity, along with high energy consumption, necessitating efficient client selection strategies. Traditional approaches, including heuristic and learning-based methods, fall short of addressing these complexities holistically. In response, we propose FedGCS, a novel generative client selection framework that innovatively recasts the client selection process as a generative task. Drawing inspiration from the methodologies used in large language models, FedGCS efficiently encodes abundant decision-making knowledge within a continuous representation space, enabling efficient gradient-based optimization to search for optimal client selection that will be finally output via generation. The framework comprises four steps: (1) automatic collection of diverse "selection-score" pair data using classical client selection methods; (2) training an encoder-evaluator-decoder framework on this data to construct a continuous representation space; (3) employing gradient-based optimization in this space for optimal client selection; (4) generating the final optimal client selection via using beam search for the well-trained decoder. FedGCS outperforms traditional methods by being more comprehensive, generalizable, and efficient, simultaneously optimizing for model performance, latency, and energy consumption. The effectiveness of FedGCS is proven through extensive experimental analyses.
Authors: Hao Mark Chen, Wayne Luk, Ka Fai Cedric Yiu, Rui Li, Konstantin Mishchenko, Stylianos I. Venieris, Hongxiang Fan
Abstract: The auto-regressive decoding of Large Language Models (LLMs) results in significant overheads in their hardware performance. While recent research has investigated various speculative decoding techniques for multi-token generation, these efforts have primarily focused on improving processing speed such as throughput. Crucially, they often neglect other metrics essential for real-life deployments, such as memory consumption and training cost. To overcome these limitations, we propose a novel parallel prompt decoding that requires only $0.0002$% trainable parameters, enabling efficient training on a single A100-40GB GPU in just 16 hours. Inspired by the human natural language generation process, $PPD$ approximates outputs generated at future timesteps in parallel by using multiple prompt tokens. This approach partially recovers the missing conditional dependency information necessary for multi-token generation, resulting in up to a 28% higher acceptance rate for long-range predictions. Furthermore, we present a hardware-aware dynamic sparse tree technique that adaptively optimizes this decoding scheme to fully leverage the computational capacities on different GPUs. Through extensive experiments across LLMs ranging from MobileLlama to Vicuna-13B on a wide range of benchmarks, our approach demonstrates up to 2.49$\times$ speedup and maintains a minimal runtime memory overhead of just $0.0004$%. More importantly, our parallel prompt decoding can serve as an orthogonal optimization for synergistic integration with existing speculative decoding, showing up to $1.22\times$ further speed improvement. Our code is available at https://github.com/hmarkc/parallel-prompt-decoding.
Authors: Nicholas Roberts, Samuel Guo, Zhiqi Gao, Satya Sai Srinath Namburi GNVV, Sonia Cromp, Chengjun Wu, Chengyu Duan, Frederic Sala
Abstract: While Transformers underpin modern large language models (LMs), there is a growing list of alternative architectures with new capabilities, promises, and tradeoffs. This makes choosing the right LM architecture challenging. Recently proposed hybrid architectures seek a best-of-all-worlds approach that reaps the benefits of all architectures. Hybrid design is difficult for two reasons: it requires manual expert-driven search, and new hybrids must be trained from scratch. We propose Manticore, a framework that addresses these challenges by automating the design of hybrid architectures while reusing pretrained models to create pretrained hybrids. Our approach augments ideas from differentiable Neural Architecture Search (NAS) by incorporating simple projectors that translate features between pretrained blocks from different architectures. We then fine-tune hybrids that combine pretrained models from different architecture families -- such as the GPT series and Mamba -- end-to-end. With Manticore, we enable LM selection without training multiple models, the construction of pretrained hybrids from existing pretrained models, and the ability to program pretrained hybrids to have certain capabilities. Manticore hybrids match existing manually designed hybrids, achieve strong performance on Long Range Arena, and improve on pretrained transformers and state space models on various natural language tasks.
Authors: Ingrid Navarro, Pablo Ortega-Kral, Jay Patrikar, Haichuan Wang, Alonso Cano, Jean Oh, Sebastian Scherer
Abstract: Demand for air travel is rising, straining existing aviation infrastructure. In the US, more than 90% of airport control towers are understaffed, falling short of FAA and union standards. This, in part, has contributed to an uptick in near-misses and safety-critical events, highlighting the need for advancements in air traffic management technologies to ensure safe and efficient operations. Data-driven predictive models for terminal airspace show potential to address these challenges; however, the lack of large-scale surface movement datasets in the public domain has hindered the development of scalable and generalizable approaches. To address this, we introduce Amelia-42, a first-of-its-kind large collection of raw airport surface movements reports streamed through the FAA's System Wide Information Management (SWIM) Program, comprising over two years of trajectory data (~9.19TB) across 42 US airports. We open-source tools to process this data into clean tabular position reports. We release Amelia42-Mini, a 15-day sample per airport, fully processed data on HuggingFace for ease of use. We also present a trajectory forecasting benchmark consisting of Amelia10-Bench, an accessible experiment family using 292 days from 10 airports, as well as Amelia-TF, a transformer-based baseline for multi-agent trajectory forecasting. All resources are available at our website: https://ameliacmu.github.io and https://huggingface.co/AmeliaCMU.
URLs: https://ameliacmu.github.io, https://huggingface.co/AmeliaCMU.
Authors: Aaron Mueller, Jannik Brinkmann, Millicent Li, Samuel Marks, Koyena Pal, Nikhil Prakash, Can Rager, Aruna Sankaranarayanan, Arnab Sen Sharma, Jiuding Sun, Eric Todd, David Bau, Yonatan Belinkov
Abstract: Interpretability provides a toolset for understanding how and why neural networks behave in certain ways. However, there is little unity in the field: most studies employ ad-hoc evaluations and do not share theoretical foundations, making it difficult to measure progress and compare the pros and cons of different techniques. Furthermore, while mechanistic understanding is frequently discussed, the basic causal units underlying these mechanisms are often not explicitly defined. In this article, we propose a perspective on interpretability research grounded in causal mediation analysis. Specifically, we describe the history and current state of interpretability taxonomized according to the types of causal units (mediators) employed, as well as methods used to search over mediators. We discuss the pros and cons of each mediator, providing insights as to when particular kinds of mediators and search methods are most appropriate. We argue that this framing yields a more cohesive narrative of the field and helps researchers select appropriate methods based on their research objective. Our analysis yields actionable recommendations for future work, including the discovery of new mediators and the development of standardized evaluations tailored to these goals.
Authors: Jang-Hyun Kim, Claudia Skok Gibbs, Sangdoo Yun, Hyun Oh Song, Kyunghyun Cho
Abstract: We propose a novel machine learning approach for inferring causal variables of a target variable from observations. Our focus is on directly inferring a set of causal factors without requiring full causal graph reconstruction, which is computationally challenging in large-scale systems. The identified causal set consists of all potential regulators of the target variable under experimental settings, enabling efficient regulation through intervention. To achieve this, we train a neural network using supervised learning on simulated data to infer causality. By employing a subsampled-ensemble inference strategy, our approach scales with linear complexity in the number of variables, efficiently scaling up to thousands of variables. Empirical results demonstrate superior performance in identifying causal relationships within large-scale gene regulatory networks, outperforming existing methods that emphasize full-graph discovery. We validate our model's generalization capability across out-of-distribution graph structures and generating mechanisms, including gene regulatory networks of E. coli and the human K562 cell line. Implementation codes are available at https://github.com/snu-mllab/Targeted-Cause-Discovery.
URLs: https://github.com/snu-mllab/Targeted-Cause-Discovery.
Authors: Ruining Yang, Yi Xu, Yun Fu, Lili Su
Abstract: Trajectory prediction is a core task in autonomous driving. However, training advanced trajectory prediction models on existing large-scale datasets is both time-consuming and computationally expensive. More critically, these datasets are highly imbalanced in scenario density, with normal driving scenes (low-moderate traffic) overwhelmingly dominating the datasets, while high-density and safety-critical cases are underrepresented. As a result, models tend to overfit low/moderate-density scenarios and perform poorly in high-density scenarios. To address these challenges, we propose the SSTP framework, which constructs a compact yet density-balanced dataset tailored to trajectory prediction. SSTP consists of two main stages: (1)Extraction, where a baseline model is pretrained for a few epochs to obtain stable gradient estimates, and the dataset is partitioned by scenario density. (2)Selection, where gradient-based scores and a submodular objective select representative samples within each density category, while biased sampling emphasizes rare high-density interactions to avoid dominance by low-density cases. This approach significantly reduces the dataset size and mitigates scenario imbalance, without sacrificing prediction accuracy. Experiments on the Argoverse 1 and Argoverse 2 datasets with recent state-of-the-art models show that SSTP achieves comparable performance to full-dataset training using only half the data while delivering substantial improvements in high-density traffic scenes and significantly reducing training time. Robust trajectory prediction depends not only on data scale but also on balancing scene density to ensure reliable performance under complex multi agent interactions.
Authors: Yuandong Tian
Abstract: We prove rich algebraic structures of the solution space for 2-layer neural networks with quadratic activation and $L_2$ loss, trained on reasoning tasks in Abelian group (e.g., modular addition). Such a rich structure enables \emph{analytical} construction of global optimal solutions from partial solutions that only satisfy part of the loss, despite its high nonlinearity. We coin the framework as CoGS (\emph{\underline{Co}mposing \underline{G}lobal \underline{S}olutions}). Specifically, we show that the weight space over different numbers of hidden nodes of the 2-layer network is equipped with a semi-ring algebraic structure, and the loss function to be optimized consists of \emph{sum potentials}, which are ring homomorphisms, allowing partial solutions to be composed into global ones by ring addition and multiplication. Our experiments show that around $95\%$ of the solutions obtained by gradient descent match exactly our theoretical constructions. Although the global solutions constructed only required a small number of hidden nodes, our analysis on gradient dynamics shows that overparameterization asymptotically decouples training dynamics and is beneficial. We further show that training dynamics favors simpler solutions under weight decay, and thus high-order global solutions such as perfect memorization are unfavorable. The code is open sourced at https://github.com/facebookresearch/luckmatters/tree/yuandong3/ssl/real-dataset.
URLs: https://github.com/facebookresearch/luckmatters/tree/yuandong3/ssl/real-dataset.
Authors: Yihong Dong, Ge Li, Yongding Tao, Xue Jiang, Kechi Zhang, Jia Li, Jinliang Deng, Jing Su, Jun Zhang, Jingjing Xu
Abstract: Despite the remarkable successes of general-purpose neural networks, such as MLPs and Transformers, we find that they exhibit notable shortcomings in modeling and reasoning about periodic phenomena, achieving only marginal performance within the training domain and failing to generalize effectively to out-of-domain (OOD) scenarios. Periodicity is ubiquitous throughout nature and science. Therefore, neural networks should be equipped with the essential ability to model and handle periodicity. In this work, we propose FAN, a novel general-purpose neural network that effectively addresses periodicity modeling challenges while offering broad applicability similar to MLP with fewer parameters and FLOPs. Periodicity is naturally integrated into FAN's structure and computational processes by introducing the Fourier Principle. Unlike existing Fourier-based networks, which possess particular periodicity modeling abilities but face challenges in scaling to deeper networks and are typically designed for specific tasks, our approach overcomes this challenge to enable scaling to large-scale models and maintains general-purpose modeling capability. Through extensive experiments, we demonstrate the superiority of FAN in periodicity modeling tasks and the effectiveness and generalizability of FAN across a range of real-world tasks. Moreover, we reveal that compared to existing Fourier-based networks, FAN accommodates both periodicity modeling and general-purpose modeling well.
Authors: Hongying Liu, Hao Wang, Haoran Chu, Yibo Wu
Abstract: An unsolved issue in widely used methods such as Support Vector Data Description (SVDD) and Small Sphere and Large Margin SVM (SSLM) for anomaly detection is their nonconvexity, which hampers the analysis of optimal solutions in a manner similar to SVMs and limits their applicability in large-scale scenarios. In this paper, we introduce a novel convex SSLM formulation which has been demonstrated to revert to a convex quadratic programming problem for hyperparameter values of interest. Leveraging the convexity of our method, we derive numerous results that are unattainable with traditional nonconvex approaches. We conduct a thorough analysis of how hyperparameters influence the optimal solution, pointing out scenarios where optimal solutions can be trivially found and identifying instances of ill-posedness. Most notably, we establish connections between our method and traditional approaches, providing a clear determination of when the optimal solution is unique--a task unachievable with traditional nonconvex methods. We also derive the nu-property to elucidate the interactions between hyperparameters and the fractions of support vectors and margin errors in both positive and negative classes.
Authors: Erkan Karabulut, Paul Groth, Victoria Degeler
Abstract: Association Rule Mining (ARM) is the task of discovering commonalities in data in the form of logical implications. ARM is used in the Internet of Things (IoT) for different tasks including monitoring and decision-making. However, existing methods give limited consideration to IoT-specific requirements such as heterogeneity and volume. Furthermore, they do not utilize important static domain-specific description data about IoT systems, which is increasingly represented as knowledge graphs. In this paper, we propose a novel ARM pipeline for IoT data that utilizes both dynamic sensor data and static IoT system metadata. Furthermore, we propose an Autoencoder-based Neurosymbolic ARM method (Aerial) as part of the pipeline to address the high volume of IoT data and reduce the total number of rules that are resource-intensive to process. Aerial learns a neural representation of a given data and extracts association rules from this representation by exploiting the reconstruction (decoding) mechanism of an autoencoder. Extensive evaluations on 3 IoT datasets from 2 domains show that ARM on both static and dynamic IoT data results in more generically applicable rules while Aerial can learn a more concise set of high-quality association rules than the state-of-the-art with full coverage over the datasets.
Authors: Nour Jamoussi, Giuseppe Serra, Photios A. Stavrou, Marios Kountouris
Abstract: Federated learning (FL) is a widely used and impactful distributed optimization framework that achieves consensus through averaging locally trained models. While effective, this approach may not align well with Bayesian inference, where the model space has the structure of a distribution space. Taking an information-geometric perspective, we reinterpret FL aggregation as the problem of finding the barycenter of local posteriors using a prespecified divergence metric, minimizing the average discrepancy across clients. This perspective provides a unifying framework that generalizes many existing methods and offers crisp insights into their theoretical underpinnings. We then propose BA-BFL, an algorithm that retains the convergence properties of Federated Averaging in non-convex settings. In non-independent and identically distributed scenarios, we conduct extensive comparisons with statistical aggregation techniques, showing that BA-BFL achieves performance comparable to state-of-the-art methods while offering a geometric interpretation of the aggregation phase. Additionally, we extend our analysis to Hybrid Bayesian Deep Learning, exploring the impact of Bayesian layers on uncertainty quantification and model calibration.
Authors: Enrique Adrian Villarrubia-Martin, Luis Rodriguez-Benitez, David Mu\~noz-Valero, Giovanni Montana, Luis Jimenez-Linares
Abstract: This paper addresses a critical challenge in the high-speed passenger railway industry: designing effective dynamic pricing strategies in the context of competing and cooperating operators. To address this, a multi-agent reinforcement learning (MARL) framework based on a non-zero-sum Markov game is proposed, incorporating random utility models to capture passenger decision making. Unlike prior studies in areas such as energy, airlines, and mobile networks, dynamic pricing for railway systems using deep reinforcement learning has received limited attention. A key contribution of this paper is a parametrisable and versatile reinforcement learning simulator designed to model a variety of railway network configurations and demand patterns while enabling realistic, microscopic modelling of user behaviour, called RailPricing-RL. This environment supports the proposed MARL framework, which models heterogeneous agents competing to maximise individual profits while fostering cooperative behaviour to synchronise connecting services. Experimental results validate the framework, demonstrating how user preferences affect MARL performance and how pricing policies influence passenger choices, utility, and overall system dynamics. This study provides a foundation for advancing dynamic pricing strategies in railway systems, aligning profitability with system-wide efficiency, and supporting future research on optimising pricing policies.
Authors: Keegan Harris, Aleksandrs Slivkins
Abstract: We evaluate the ability of the current generation of large language models (LLMs) to help a decision-making agent facing an exploration-exploitation tradeoff. We use LLMs to explore and exploit in silos in various (contextual) bandit tasks. We find that while the current LLMs often struggle to exploit, in-context mitigations may be used to substantially improve performance for small-scale tasks. However even then, LLMs perform worse than a simple linear regression. On the other hand, we find that LLMs do help at exploring large action spaces with inherent semantics, by suggesting suitable candidates to explore.
Authors: Chi Zhou, Wang Luo, Haoran Li, Congying Han, Tiande Guo, Zicheng Zhang
Abstract: Offline reinforcement learning agents face significant deployment challenges due to the synthetic-to-real distribution mismatch. While most prior research has focused on improving the fidelity of synthetic sampling and incorporating off-policy mechanisms, the directly integrated paradigm often fails to ensure consistent policy behavior in biased models and underlying environmental dynamics, which inherently arise from discrepancies between behavior and learning policies. In this paper, we first shift the focus from model reliability to policy discrepancies while optimizing for expected returns, and then self-consistently incorporate synthetic data, deriving a novel actor-critic paradigm, Dual Alignment Maximin Optimization (DAMO). It is a unified framework to ensure both model-environment policy consistency and synthetic and offline data compatibility. The inner minimization performs dual conservative value estimation, aligning policies and trajectories to avoid out-of-distribution states and actions, while the outer maximization ensures that policy improvements remain consistent with inner value estimates. Empirical evaluations demonstrate that DAMO effectively ensures model and policy alignments, achieving competitive performance across diverse benchmark tasks.
Authors: Fares El Khoury, Edouard Pauwels, Samuel Vaiter, Michael Arbel
Abstract: Bilevel optimization has emerged as a technique for addressing a wide range of machine learning problems that involve an outer objective implicitly determined by the minimizer of an inner problem. While prior works have primarily focused on the parametric setting, a learning-theoretic foundation for bilevel optimization in the nonparametric case remains relatively unexplored. In this paper, we take a first step toward bridging this gap by studying Kernel Bilevel Optimization (KBO), where the inner objective is optimized over a reproducing kernel Hilbert space. This setting enables rich function approximation while providing a foundation for rigorous theoretical analysis. In this context, we derive novel finite-sample generalization bounds for KBO, leveraging tools from empirical process theory. These bounds further allow us to assess the statistical accuracy of gradient-based methods applied to the empirical discretization of KBO. We numerically illustrate our theoretical findings on a synthetic instrumental variable regression task.
Authors: Arshia Afzal, Elias Abad Rocamora, Leyla Naz Candogan, Pol Puigdemont, Francesco Tonin, Yongtao Wu, Mahsa Shoaran, Volkan Cevher
Abstract: Linear Transformers and State Space Models have emerged as efficient alternatives to softmax Transformers for causal sequence modeling, enabling parallel training via matrix multiplication and efficient RNN-style inference. However, despite their success in causal tasks, no unified framework exists for applying Linear Transformers to bidirectional sequence modeling. We introduce LION, the first framework to systematically extend Linear Transformers to the bidirectional setting. LION generalizes three core representations commonly used in the causal case - full Linear Attention , bidirectional RNN, and chunkwise parallel form - to the bidirectional setting. These forms are theoretically equivalent and enable models to exploit the strengths of each during training and inference. We prove that a broad class of Linear Transformers can be extended using LION and validate our framework via three core examples based on the choice of decay type: LION-LIT, the bidirectional extension of arXiv:2006.16236; LION-D, based on arXiv:2307.08621; and LION-S, a variant using selective decay arXiv:2103.02143, arXiv:2312.0075. Across standard bidirectional tasks, LION enables models to match or exceed the performance of softmax Transformers, while offering significantly faster training and more efficient inference than existing State Space Models.
Authors: Rui Liu, Peng Gao, Yu Shen, Ming Lin, Pratap Tokekar
Abstract: Learning with guidance has proven effective across a wide range of machine learning systems. Guidance may, for example, come from annotated datasets in supervised learning, pseudo-labels in semi-supervised learning, and expert demonstration policies in reinforcement learning. However, guidance signals can be noisy due to domain shifts and limited data availability and may not generalize well. Blindly trusting such signals when they are noisy, incomplete, or misaligned with the target domain can lead to degraded performance. To address these challenges, we propose Adaptive Conformal Guidance (AdaConG), a simple yet effective approach that dynamically modulates the influence of guidance signals based on their associated uncertainty, quantified via split conformal prediction (CP). By adaptively adjusting to guidance uncertainty, AdaConG enables models to reduce reliance on potentially misleading signals and enhance learning performance. We validate AdaConG across diverse tasks, including knowledge distillation, semi-supervised image classification, gridworld navigation, and autonomous driving. Experimental results demonstrate that AdaConG improves performance and robustness under imperfect guidance, e.g., in gridworld navigation, it accelerates convergence and achieves over $6\times$ higher rewards than the best-performing baseline. These results highlight AdaConG as a broadly applicable solution for learning under uncertainty.
Authors: Simin Chen, Yiming Chen, Zexin Li, Yifan Jiang, Zhongwei Wan, Yixin He, Dezhi Ran, Tianle Gu, Haizhou Li, Tao Xie, Baishakhi Ray
Abstract: Data contamination has received increasing attention in the era of large language models (LLMs) due to their reliance on vast Internet-derived training corpora. To mitigate the risk of potential data contamination, LLM benchmarking has undergone a transformation from static to dynamic benchmarking. In this work, we conduct an in-depth analysis of existing static to dynamic benchmarking methods aimed at reducing data contamination risks. We first examine methods that enhance static benchmarks and identify their inherent limitations. We then highlight a critical gap-the lack of standardized criteria for evaluating dynamic benchmarks. Based on this observation, we propose a series of optimal design principles for dynamic benchmarking and analyze the limitations of existing dynamic benchmarks. This survey provides a concise yet comprehensive overview of recent advancements in data contamination research, offering valuable insights and a clear guide for future research efforts. We maintain a GitHub repository to continuously collect both static and dynamic benchmarking methods for LLMs. The repository can be found at this link.
Authors: Jiwan Chung, Saejin Kim, Yongrae Jo, Jaewoo Park, Dongjun Min, Youngjae Yu
Abstract: As large language models expand beyond natural language to domains such as mathematics, multimodal understanding, and embodied agents, tokens increasingly reflect metric relationships rather than purely linguistic meaning. We introduce DIST2Loss, a distance-aware framework designed to train autoregressive discrete models by leveraging predefined distance relationships among output tokens. At its core, DIST2Loss transforms continuous exponential family distributions derived from inherent distance metrics into discrete, categorical optimization targets compatible with the models' architectures. This approach enables the models to learn and preserve meaningful distance relationships during token generation while maintaining compatibility with existing architectures. Empirical evaluations show consistent performance gains in diverse multimodal applications, including visual grounding, robotic manipulation, generative reward modeling, and image generation using vector-quantized features. These improvements are most notable in low-data regimes, demonstrating DIST2Loss's strength under resource constraints.
Authors: Yongqiang Yao, Jingru Tan, Kaihuan Liang, Feizhao Zhang, Jiahao Hu, Shuo Wu, Yazhe Niu, Ruihao Gong, Dahua Lin, Ningyi Xu
Abstract: Training Long-Context Large Language Models (LLMs) is challenging, as hybrid training with long-context and short-context data often leads to workload imbalances. Existing works mainly use data packing to alleviate this issue, but fail to consider imbalanced attention computation and wasted communication overhead. This paper proposes Hierarchical Balance Packing (HBP), which designs a novel batch-construction method and training recipe to address those inefficiencies. In particular, the HBP constructs multi-level data packing groups, each optimized with a distinct packing length. It assigns training samples to their optimal groups and configures each group with the most effective settings, including sequential parallelism degree and gradient checkpointing configuration. To effectively utilize multi-level groups of data, we design a dynamic training pipeline specifically tailored to HBP, including curriculum learning, adaptive sequential parallelism, and stable loss. Our extensive experiments demonstrate that our method significantly reduces training time over multiple datasets and open-source models while maintaining strong performance. For the largest DeepSeek-V2 (236B) MoE model, our method speeds up the training by 2.4$\times$ with competitive performance. Codes will be released at https://github.com/ModelTC/HBP.
Authors: Ping Zhang, Zheda Mai, Quang-Huy Nguyen, Wei-Lun Chao
Abstract: Semi-supervised learning (SSL) leverages abundant unlabeled data alongside limited labeled data to enhance learning. As vision foundation models (VFMs) increasingly serve as the backbone of vision applications, it remains unclear how SSL interacts with these pre-trained models. To address this gap, we develop new SSL benchmark datasets where frozen VFMs underperform and systematically evaluate representative SSL methods. We make a surprising observation: parameter-efficient fine-tuning (PEFT) using only labeled data often matches SSL performance, even without leveraging unlabeled data. This motivates us to revisit self-training, a conceptually simple SSL baseline, where we use the supervised PEFT model to pseudo-label unlabeled data for further training. To overcome the notorious issue of noisy pseudo-labels, we propose ensembling multiple PEFT approaches and VFM backbones to produce more robust pseudo-labels. Empirical results validate the effectiveness of this simple yet powerful approach, providing actionable insights into SSL with VFMs and paving the way for more scalable and practical semi-supervised learning in the era of foundation models.
Authors: Hongkai Zheng, Wenda Chu, Bingliang Zhang, Zihui Wu, Austin Wang, Berthy T. Feng, Caifeng Zou, Yu Sun, Nikola Kovachki, Zachary E. Ross, Katherine L. Bouman, Yisong Yue
Abstract: Plug-and-play diffusion priors (PnPDP) have emerged as a promising research direction for solving inverse problems. However, current studies primarily focus on natural image restoration, leaving the performance of these algorithms in scientific inverse problems largely unexplored. To address this gap, we introduce \textsc{InverseBench}, a framework that evaluates diffusion models across five distinct scientific inverse problems. These problems present unique structural challenges that differ from existing benchmarks, arising from critical scientific applications such as optical tomography, medical imaging, black hole imaging, seismology, and fluid dynamics. With \textsc{InverseBench}, we benchmark 14 inverse problem algorithms that use plug-and-play diffusion priors against strong, domain-specific baselines, offering valuable new insights into the strengths and weaknesses of existing algorithms. To facilitate further research and development, we open-source the codebase, along with datasets and pre-trained models, at https://devzhk.github.io/InverseBench/.
Authors: Hao Mark Chen, Shell Xu Hu, Wayne Luk, Timothy Hospedales, Hongxiang Fan
Abstract: Model merging has emerged as a promising approach for multi-task learning (MTL), offering a data-efficient alternative to conventional fine-tuning. However, with the rapid development of the open-source AI ecosystem and the increasing availability of fine-tuned foundation models, existing model merging methods face two key limitations: (i) They are primarily designed for in-house fine-tuned models, making them less adaptable to diverse model sources with partially unknown model and task information, (ii) They struggle to scale effectively when merging numerous model checkpoints. To address these challenges, we formulate model merging as a constrained optimization problem and introduce a novel approach: Frank-Wolfe Merging (FW-Merging). Inspired by Frank-Wolfe optimization, our approach iteratively selects the most relevant model in the pool to minimize a linear approximation of the objective function and then executes a local merging similar to the Frank-Wolfe update. The objective function is designed to capture the desired behavior of the target-merged model, while the fine-tuned candidate models define the constraint set. More importantly, FW-Merging serves as an orthogonal technique for existing merging methods, seamlessly integrating with them to further enhance accuracy performance. Our experiments show that FW-Merging scales across diverse model sources, remaining stable with 16 irrelevant models and improving by 15.3% with 16 relevant models on 20 CV tasks, while maintaining constant memory overhead, unlike the linear overhead of data-informed merging methods. Compared with the state-of-the-art approaches, FW-Merging surpasses the data-free merging method by 32.8% and outperforms the data-informed Adamerging by 8.39% when merging 20 ViT models. Our code is open-sourced at github.com/hmarkc/FW-Merging.
Authors: Jacqueline L. Mitchell, Brian Hyeongseok Kim, Chenyu Zhou, Chao Wang
Abstract: Large language models (LLMs) are increasingly used for program verification, and yet little is known about \emph{how} they reason about program semantics during this process. In this work, we focus on abstract interpretation based-reasoning for invariant generation and introduce two novel prompting strategies that aim to elicit such reasoning from LLMs. We evaluate these strategies across several state-of-the-art LLMs on 22 programs from the SV-COMP benchmark suite widely used in software verification. We analyze both the soundness of the generated invariants and the key thematic patterns in the models' reasoning errors. This work aims to highlight new research opportunities at the intersection of LLMs and program verification for applying LLMs to verification tasks and advancing their reasoning capabilities in this application.
Authors: Guy Bar-Shalom, Fabrizio Frasca, Derek Lim, Yoav Gelberg, Yftah Ziser, Ran El-Yaniv, Gal Chechik, Haggai Maron
Abstract: The automated detection of hallucinations and training data contamination is pivotal to the safe deployment of Large Language Models (LLMs). These tasks are particularly challenging in settings where no access to model internals is available. Current approaches in this setup typically leverage only the probabilities of actual tokens in the text, relying on simple task-specific heuristics. Crucially, they overlook the information contained in the full sequence of next-token probability distributions. We propose to go beyond hand-crafted decision rules by learning directly from the complete observable output of LLMs -- consisting not only of next-token probabilities, but also the full sequence of next-token distributions. We refer to this as the LLM Output Signature (LOS), and treat it as a reference data type for detecting hallucinations and data contamination. To that end, we introduce LOS-Net, a lightweight attention-based architecture trained on an efficient encoding of the LOS, which can provably approximate a broad class of existing techniques for both tasks. Empirically, LOS-Net achieves superior performance across diverse benchmarks and LLMs, while maintaining extremely low detection latency. Furthermore, it demonstrates promising transfer capabilities across datasets and LLMs. Full code is available at https://github.com/BarSGuy/Beyond-next-token-probabilities.
URLs: https://github.com/BarSGuy/Beyond-next-token-probabilities.
Authors: Pedro Sandoval-Segura, Xijun Wang, Ashwinee Panda, Micah Goldblum, Ronen Basri, Tom Goldstein, David Jacobs
Abstract: Attention is foundational to large language models (LLMs), enabling different heads to have diverse focus on relevant input tokens. However, learned behaviors like attention sinks, where the first token receives the most attention despite limited semantic importance, suggest some heads may be inactive, and point to a significant source of computational redundancy. To analyze this phenomenon, we propose a taxonomy of 13 score functions that measure different ways a head can be inactive. Thresholding these scores allows us to analyze different sets of potentially inactive attention heads. We evaluate whether identified heads are inactive through model interventions, finding that more than 12% of attention heads are inactive on average, and can be ablated in specific contexts while maintaining MMLU accuracy to within 1% of the pretrained LLM. Across 3 model families, our score functions that measure the average norm of a head's output consistently identify inactive heads that would not have been found by score functions that rely solely on attention weights. We establish that relying on a score function that measures a first token attention sink would underestimate the prevalence of inactive heads, failing to identify more than 7% of inactive heads on average. We also show how measuring score distributions can provide insights into attention behavior. For instance, we find evidence that finetuning causes little to no change in attention behavior, and that even within the same model family, large model scales present markedly different attention behaviors.
Authors: Jiawei Li
Abstract: Instruction finetuning attacks pose a serious threat to large language models (LLMs) by subtly embedding poisoned examples in finetuning datasets, leading to harmful or unintended behaviors in downstream applications. Detecting such attacks is challenging because poisoned data is often indistinguishable from clean data and prior knowledge of triggers or attack strategies is rarely available. We present a detection method that requires no prior knowledge of the attack. Our approach leverages influence functions under semantic transformation: by comparing influence distributions before and after a sentiment inversion, we identify critical poison examples whose influence is strong and remain unchanged before and after inversion. We show that this method works on sentiment classification task and math reasoning task, for different language models. Removing a small set of critical poisons (about 1% of the data) restores the model performance to near-clean levels. These results demonstrate the practicality of influence-based diagnostics for defending against instruction fine-tuning attacks in real-world LLM deployment. Artifact available at https://github.com/lijiawei20161002/Poison-Detection. WARNING: This paper contains offensive data examples.
Authors: Yonghong Li, Zheng Zhang, Xiuzhuang Zhou
Abstract: Trustworthy depression prediction based on deep learning, incorporating both predictive reliability and algorithmic fairness across diverse demographic groups, is crucial for clinical application. Recently, achieving reliable depression predictions through uncertainty quantification has attracted increasing attention. However, few studies have focused on the fairness of uncertainty quantification (UQ) in depression prediction. In this work, we investigate the algorithmic fairness of UQ, namely Equal Opportunity Coverage (EOC) fairness, and propose Fair Uncertainty Quantification (FUQ) for depression prediction. FUQ pursues reliable and fair depression predictions through group-based analysis. Specifically, we first group all the participants by different sensitive attributes and leverage conformal prediction to quantify uncertainty within each demographic group, which provides a theoretically guaranteed and valid way to quantify uncertainty for depression prediction and facilitates the investigation of fairness across different demographic groups. Furthermore, we propose a fairness-aware optimization strategy that formulates fairness as a constrained optimization problem under EOC constraints. This enables the model to preserve predictive reliability while adapting to the heterogeneous uncertainty levels across demographic groups, thereby achieving optimal fairness. Through extensive evaluations on several visual and audio depression datasets, our approach demonstrates its effectiveness.
Authors: Bojian Yin, Federico Corradi
Abstract: Backpropagation underpins modern deep learning, yet its reliance on global gradient synchronization limits scalability and incurs high memory costs. In contrast, fully local learning rules are more efficient but often struggle to maintain the cross-layer coordination needed for coherent global learning. Building on this tension, we introduce Stochastic Layer-wise Learning (SLL), a layer-wise training algorithm that decomposes the global objective into coordinated layer-local updates while preserving global representational coherence. The method is ELBO-inspired under a Markov assumption on the network, where the network-level objective decomposes into layer-wise terms and each layer optimizes a local objective via a deterministic encoder. The intractable KL in ELBO is replaced by a Bhattacharyya surrogate computed on auxiliary categorical posteriors obtained via fixed geometry-preserving random projections, with optional multiplicative dropout providing stochastic regularization. SLL optimizes locally, aligns globally, thereby eliminating cross-layer backpropagation. Experiments on MLPs, CNNs, and Vision Transformers from MNIST to ImageNet show that the approach surpasses recent local methods and matches global BP performance while memory usage invariant with depth. The results demonstrate a practical and principled path to modular and scalable local learning that couples purely local computation with globally coherent representations.
Authors: Seongjae Kang, Dong Bok Lee, Hyungjoon Jang, Sung Ju Hwang
Abstract: Semi-supervised learning (SSL) has emerged as a practical solution for addressing data scarcity challenges by leveraging unlabeled data. Recently, vision-language models (VLMs), pre-trained on massive image-text pairs, have demonstrated remarkable zero-/few-shot performance that often surpasses SSL approaches due to their exceptional generalization capabilities. This gap motivates us to question: how can we effectively harness the powerful generalization capabilities of VLMs into task-specific models? Knowledge distillation (KD) offers a natural framework for transferring VLM capabilities, but we identify that it suffers from gradient conflicts between supervised and distillation losses. To address this challenge, we propose Dual-Head Optimization (DHO), which introduces dual prediction heads for each distinct signal. We observe that DHO resolves gradient conflicts, enabling improved feature learning compared to single-head KD baselines, with practical benefits of minimal computational overhead and test-time hyperparameter tuning without retraining. Extensive experiments across 15 datasets show that DHO consistently outperforms KD baselines, often outperforming teacher models with smaller student models. DHO also achieves new state-of-the-art performance on both in-distribution ImageNet semi-supervised learning and out-of-distribution generalization across ImageNet variants. We publicly release our code and model checkpoints to facilitate future research at https://github.com/erjui/DHO.
Authors: George Andriopoulos, Soyuj Jung Basnet, Juan Guevara, Li Guo, Keith Ross
Abstract: The Unconstrained Feature Model (UFM) is a mathematical framework that enables closed-form approximations for minimal training loss and related performance measures in deep neural networks (DNNs). This paper leverages the UFM to provide qualitative insights into neural multivariate regression, a critical task in imitation learning, robotics, and reinforcement learning. Specifically, we address two key questions: (1) How do multi-task models compare to multiple single-task models in terms of training performance? (2) Can whitening and normalizing regression targets improve training performance? The UFM theory predicts that multi-task models achieve strictly smaller training MSE than multiple single-task models when the same or stronger regularization is applied to the latter, and our empirical results confirm these findings. Regarding whitening and normalizing regression targets, the UFM theory predicts that they reduce training MSE when the average variance across the target dimensions is less than one, and our empirical results once again confirm these findings. These findings highlight the UFM as a powerful framework for deriving actionable insights into DNN design and data pre-processing strategies.
Authors: Jiacheng Liang, Tanqiu Jiang, Yuhui Wang, Rongyi Zhu, Fenglong Ma, Ting Wang
Abstract: This paper presents AutoRAN, the first framework to automate the hijacking of internal safety reasoning in large reasoning models (LRMs). At its core, AutoRAN pioneers an execution simulation paradigm that leverages a weaker but less-aligned model to simulate execution reasoning for initial hijacking attempts and iteratively refine attacks by exploiting reasoning patterns leaked through the target LRM's refusals. This approach steers the target model to bypass its own safety guardrails and elaborate on harmful instructions. We evaluate AutoRAN against state-of-the-art LRMs, including GPT-o3/o4-mini and Gemini-2.5-Flash, across multiple benchmarks (AdvBench, HarmBench, and StrongReject). Results show that AutoRAN achieves approaching 100% success rate within one or few turns, effectively neutralizing reasoning-based defenses even when evaluated by robustly aligned external models. This work reveals that the transparency of the reasoning process itself creates a critical and exploitable attack surface, highlighting the urgent need for new defenses that protect models' reasoning traces rather than merely their final outputs.
Authors: Gang Li, Ming Lin, Tomer Galanti, Zhengzhong Tu, Tianbao Yang
Abstract: The recent success and openness of DeepSeek-R1 have brought widespread attention to Group Relative Policy Optimization (GRPO) as a reinforcement learning method for large reasoning models (LRMs). In this work, we analyze the GRPO objective under a binary reward setting and reveal an inherent limitation of question-level difficulty bias. We also identify a connection between GRPO and traditional discriminative methods in supervised learning. Motivated by these insights, we introduce a new Discriminative Constrained Optimization (DisCO) framework for reinforcing LRMs, grounded in the principle of discriminative learning. The main differences between DisCO and GRPO and its recent variants are: (1) it replaces the group relative objective with a discriminative objective defined by a scoring function; (2) it abandons clipping-based surrogates in favor of non-clipping RL surrogate objectives used as scoring functions; (3) it employs a simple yet effective constrained optimization approach to enforce the KL divergence constraint. As a result, DisCO offers notable advantages over GRPO and its variants: (i) it completely eliminates difficulty bias by adopting discriminative objectives; (ii) it addresses the entropy instability in GRPO and its variants through the use of non-clipping scoring functions and a constrained optimization approach, yielding long and stable training dynamics; (iii) it allows the incorporation of advanced discriminative learning techniques to address data imbalance, where a significant number of questions have more negative than positive generated answers during training. Our experiments on enhancing the mathematical reasoning capabilities of SFT-finetuned models show that DisCO significantly outperforms GRPO and its improved variants such as DAPO, achieving average gains of 7\% over GRPO and 6\% over DAPO across six benchmark tasks for an 1.5B model.
Authors: Jun Liu, Zhenglun Kong, Peiyan Dong, Changdi Yang, Tianqi Li, Hao Tang, Geng Yuan, Wei Niu, Wenbin Zhang, Pu Zhao, Xue Lin, Dong Huang, Yanzhi Wang
Abstract: Large language models (LLMs) exhibit strong capabilities as decision-making agents by interleaving reasoning and actions, as seen in ReAct-style frameworks. Yet, their practical deployment is constrained by high inference costs and large model sizes. We propose Structured Agent Distillation, a framework that compresses large LLM-based agents into smaller student models while preserving both reasoning fidelity and action consistency. Unlike standard token-level distillation, our method segments trajectories into {[REASON]} and {[ACT]} spans, applying segment-specific losses to align each component with the teacher's behavior. This structure-aware supervision enables compact agents to better replicate the teacher's decision process. Experiments on ALFWorld, HotPotQA-ReAct, and WebShop show that our approach consistently outperforms token-level and imitation learning baselines, achieving significant compression with minimal performance drop. Scaling and ablation results further highlight the importance of span-level alignment for efficient and deployable agents.
Authors: Kaiwen Wang, Jin Peng Zhou, Jonathan Chang, Zhaolin Gao, Nathan Kallus, Kiant\'e Brantley, Wen Sun
Abstract: In this paper, we propose a simple and efficient method for value model training on long-context reasoning traces. Compared to existing process reward models (PRMs), our method does not require a fine-grained notion of "step," which is difficult to define for long-context reasoning models. By collecting a dataset of 2.5 million reasoning traces, we train a 1.5B token-level value model and apply it to DeepSeek models for improved performance with test-time compute scaling. We find that block-wise value-guided search (VGS) with a final weighted majority vote achieves better test-time scaling than standard methods such as majority voting or best-of-n. Moreover, VGS significantly reduces the inference FLOPs required to achieve the same performance of majority voting. Our dataset, model and codebase are open-sourced.
Authors: Jijia Liu, Feng Gao, Bingwen Wei, Xinlei Chen, Qingmin Liao, Yi Wu, Chao Yu, Yu Wang
Abstract: Large Vision-Language Action (VLA) models have shown significant potential for embodied AI. However, their predominant training via supervised fine-tuning (SFT) limits generalization due to susceptibility to compounding errors under distribution shifts. Reinforcement learning (RL) offers a path to overcome these limitations by optimizing for task objectives via trial-and-error, yet a systematic understanding of its specific generalization benefits for VLAs compared to SFT is lacking. To address this, our study introduces a comprehensive benchmark for evaluating VLA generalization and systematically investigates the impact of RL fine-tuning across diverse visual, semantic, and execution dimensions. Our extensive experiments reveal that RL fine-tuning, particularly with PPO, significantly enhances generalization in semantic understanding and execution robustness over SFT, while maintaining comparable visual robustness. We identify PPO as a more effective RL algorithm for VLAs than LLM-derived methods like DPO and GRPO. We also develop a simple recipe for efficient PPO training on VLAs, and demonstrate its practical utility for improving VLA generalization. The project page is at https://rlvla.github.io
URLs: https://rlvla.github.io
Authors: Zixuan Xie, Xinyu Liu, Rohan Chandra, Shangtong Zhang
Abstract: Linear TD($\lambda$) is one of the most fundamental reinforcement learning algorithms for policy evaluation. Previously, convergence rates are typically established under the assumption of linearly independent features, which does not hold in many practical scenarios. This paper instead establishes the first $L^2$ convergence rates for linear TD($\lambda$) operating under arbitrary features, without making any algorithmic modification or additional assumptions. Our results apply to both the discounted and average-reward settings. To address the potential non-uniqueness of solutions resulting from arbitrary features, we develop a novel stochastic approximation result featuring convergence rates to the solution set instead of a single point.
Authors: Wei Zhuo, Zhaohuan Zhan, Han Yu
Abstract: Federated Learning (FL) on graph-structured data typically faces non-IID challenges, particularly in scenarios where each client holds a distinct subgraph sampled from a global graph. In this paper, we introduce Federated learning with Auxiliary projections (FedAux), a personalized subgraph FL framework that learns to align, compare, and aggregate heterogeneously distributed local models without sharing raw data or node embeddings. In FedAux, each client jointly trains (i) a local GNN and (ii) a learnable auxiliary projection vector (APV) that differentiably projects node embeddings onto a 1D space. A soft-sorting operation followed by a lightweight 1D convolution refines these embeddings in the ordered space, enabling the APV to effectively capture client-specific information. After local training, these APVs serve as compact signatures that the server uses to compute inter-client similarities and perform similarity-weighted parameter mixing, yielding personalized models while preserving cross-client knowledge transfer. Moreover, we provide rigorous theoretical analysis to establish the convergence and rationality of our design. Empirical evaluations across diverse graph benchmarks demonstrate that FedAux substantially outperforms existing baselines in both accuracy and personalization performance. The code is available at https://github.com/JhuoW/FedAux.
Authors: Dorian Gailhard, Enzo Tartaglione, Lirida Naviner, Jhony H. Giraldo
Abstract: Graph generative models have shown strong results in molecular design but struggle to scale to large, complex structures. While hierarchical methods improve scalability, they usually ignore node and edge features, which are critical in real-world applications. This issue is amplified in hypergraphs, where hyperedges capture higher-order relationships among multiple nodes. Despite their importance in domains such as 3D geometry, molecular systems, and circuit design, existing generative models rarely support both hypergraphs and feature generation at scale. In this paper, we introduce FAHNES (feature-aware hypergraph generation via next-scale prediction), a hierarchical framework that jointly generates hypergraph topology and features. FAHNES builds multi-scale representations through node coarsening and refines them via localized expansion, guided by a novel node budget mechanism that controls granularity and ensures consistency across scales. Experiments on synthetic, 3D mesh and graph point cloud datasets show that FAHNES achieves state-of-the-art performance in jointly generating features and structure, advancing scalable hypergraph and graph generation.
Authors: Stephen Zhang, Alireza Mousavi-Hosseini, Michal Klein, Marco Cuturi
Abstract: Flow models transform data gradually from one modality (e.g. noise) onto another (e.g. images). Such models are parameterized by a time-dependent velocity field, trained to fit segments connecting pairs of source and target points. When the pairing between source and target points is given, training flow models boils down to a supervised regression problem. When no such pairing exists, as is the case when generating data from noise, training flows is much harder. A popular approach lies in picking source and target points independently. This can, however, lead to velocity fields that are slow to train, but also costly to integrate at inference time. In theory, one would greatly benefit from training flow models by sampling pairs from an optimal transport (OT) measure coupling source and target, since this would lead to a highly efficient flow solving the Benamou and Brenier dynamical OT problem. In practice, recent works have proposed to sample mini-batches of $n$ source and $n$ target points and reorder them using an OT solver to form better pairs. These works have advocated using batches of size $n\approx 256$, and considered OT solvers that return couplings that are either sharp (using e.g. the Hungarian algorithm) or blurred (using e.g. entropic regularization, a.k.a. Sinkhorn). We follow in the footsteps of these works by exploring the benefits of increasing $n$ by three to four orders of magnitude, and look more carefully on the effect of the entropic regularization $\varepsilon$ used in the Sinkhorn algorithm. Our analysis is facilitated by new scale invariant quantities to report the sharpness of a coupling, while our sharded computations across multiple GPU or GPU nodes allow scaling up $n$. We show that in both synthetic and image generation tasks, flow models greatly benefit when fitted with large Sinkhorn couplings, with a low entropic regularization $\varepsilon$.
Authors: Alexander Marusov, Aleksandr Yugay, Alexey Zaytsev
Abstract: Self-supervised learning (SSL) has emerged as a powerful approach to learning representations, particularly in the field of computer vision. However, its application to dependent data, such as temporal and spatio-temporal domains, remains underexplored. Besides, traditional contrastive SSL methods often assume \emph{semantic independence between samples}, which does not hold for dependent data exhibiting complex correlations. We propose a novel theoretical framework for contrastive SSL tailored to \emph{continuous dependent data}, which allows the nearest samples to be semantically close to each other. In particular, we propose two possible \textit{ground truth similarity measures} between objects -- \emph{hard} and \emph{soft} closeness. Under it, we derive an analytical form for the \textit{estimated similarity matrix} that accommodates both types of closeness between samples, thereby introducing dependency-aware loss functions. We validate our approach, \emph{Dependent TS2Vec}, on temporal and spatio-temporal downstream problems. Given the dependency patterns presented in the data, our approach surpasses modern ones for dependent data, highlighting the effectiveness of our theoretically grounded loss functions for SSL in capturing spatio-temporal dependencies. Specifically, we outperform TS2Vec on the standard UEA and UCR benchmarks, with accuracy improvements of $4.17$\% and $2.08$\%, respectively. Furthermore, on the drought classification task, which involves complex spatio-temporal patterns, our method achieves a $7$\% higher ROC-AUC score.
Authors: Neta Shoham, Liron Mor-Yosef, Haim Avron
Abstract: Recent literature generalization in deep learning has examined the relationship between the curvature of the loss function at minima and generalization, mainly in the context of overparameterized neural networks. A key observation is that "flat" minima tend to generalize better than "sharp" minima. While this idea is supported by empirical evidence, it has also been shown that deep networks can generalize even with arbitrary sharpness, as measured by either the trace or the spectral norm of the Hessian. In this paper, we argue that generalization could be assessed by measuring flatness using a soft rank measure of the Hessian. We show that when an exponential family neural network model is exactly calibrated, and its prediction error and its confidence on the prediction are not correlated with the first and the second derivative of the network's output, our measure accurately captures the asymptotic expected generalization gap. For non-calibrated models, we connect a soft rank based flatness measure to the well-known Takeuchi Information Criterion and show that it still provides reliable estimates of generalization gaps for models that are not overly confident. Experimental results indicate that our approach offers a robust estimate of the generalization gap compared to baselines.
Authors: Xenia Heilmann, Luca Corbucci, Mattia Cerrato, Anna Monreale
Abstract: Federated Learning (FL) enables collaborative model training across multiple clients without sharing clients' private data. However, the diverse and often conflicting biases present across clients pose significant challenges to model fairness. Current fairness-enhancing FL solutions often fall short, as they typically mitigate biases for a single, usually binary, sensitive attribute, while ignoring the heterogeneous fairness needs that exist in real-world settings. Moreover, these solutions often evaluate unfairness reduction only on the server side, hiding persistent unfairness at the individual client level. To support more robust and reproducible fairness research in FL, we introduce a comprehensive benchmarking framework for fairness-aware FL at both the global and client levels. Our contributions are three-fold: (1) We introduce \fairdataset, a library to create tabular datasets tailored to evaluating fair FL methods under heterogeneous client bias; (2) we release four bias-heterogeneous datasets and corresponding benchmarks to compare fairness mitigation methods in a controlled environment; (3) we provide ready-to-use functions for evaluating fairness outcomes for these datasets.
Authors: Xuanming Zhang
Abstract: Industrial carbon emissions are a major driver of climate change, yet modeling these emissions is challenging due to multicollinearity among factors and complex interdependencies across sectors and time. We propose a novel graph-based deep learning framework DGL to analyze and forecast industrial CO_2 emissions, addressing high feature correlation and capturing industrial-temporal interdependencies. Unlike traditional regression or clustering methods, our approach leverages a Graph Neural Network (GNN) with attention mechanisms to model relationships between industries (or regions) and a temporal transformer to learn long-range patterns. We evaluate our framework on public global industry emissions dataset derived from EDGAR v8.0, spanning multiple countries and sectors. The proposed model achieves superior predictive performance - reducing error by over 15% compared to baseline deep models - while maintaining interpretability via attention weights and causal analysis. We believe that we are the first Graph-Temporal architecture that resolves multicollinearity by structurally encoding feature relationships, along with integration of causal inference to identify true drivers of emissions, improving transparency and fairness. We also stand a demonstration of policy relevance, showing how model insights can guide sector-specific decarbonization strategies aligned with sustainable development goals. Based on the above, we show high-emission "hotspots" and suggest equitable intervention plans, illustrating the potential of state-of-the-art AI graph learning to advance climate action, offering a powerful tool for policymakers and industry stakeholders to achieve carbon reduction targets.
Authors: Md Mokarram Chowdhury, Daniel Agyei Asante, Ernie Chang, Yang Li
Abstract: Large language models (LLMs) achieve strong performance across many domains but are difficult to deploy in resource-constrained settings due to their size. Low-rank weight matrix compression is a popular strategy for reducing model size, typically by minimizing weight reconstruction error under the assumption that weights are low-rank. However, this assumption often does not hold in LLMs. Instead, LLM activations exhibit stronger low-rank structure-prompting a shift toward minimizing activation reconstruction error. We show that this shift alone is insufficient: activation dimensions contribute unequally to model performance, and uniform reconstruction can harm performance. We propose IMPACT, a principled framework for importance-aware activation reconstruction that links model compression decisions to their impact on model behavior. IMPACT formulates an optimization problem that considers both activation structure and gradient sensitivity, and derives a closed-form solution where the optimal reconstruction bases are the eigenvectors of an importance-weighted activation covariance matrix. This enables low-rank approximations explicitly optimized to preserve accuracy. Experiments across diverse models and tasks show that IMPACT achieves up to 48.6% greater model size reduction with accuracy comparable to state-of-the-art baselines.
Authors: Song Lai, Haohan Zhao, Rong Feng, Changyi Ma, Wenzhuo Liu, Hongbo Zhao, Xi Lin, Dong Yi, Min Xie, Qingfu Zhang, Hongbin Liu, Gaofeng Meng, Fei Zhu
Abstract: Continual post-training (CPT) is a popular and effective technique for adapting foundation models like multimodal large language models to specific and ever-evolving downstream tasks. While existing research has primarily concentrated on methods like data replay, model expansion, or parameter regularization, the fundamental role of the learning paradigm within CPT remains largely unexplored. This paper presents a comparative analysis of two core post-training paradigms: supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT), investigating their respective impacts on knowledge retention during CPT. Our experiments are conducted on a benchmark comprising seven diverse multimodal tasks, utilizing Qwen2.5-VL-7B-Instruct as the base model for continual post-training. The investigation yields two significant findings: (1) When continuously learning on downstream tasks, SFT leads to catastrophic forgetting of previously learned tasks. In contrast, RFT inherently preserves prior knowledge and achieve performance comparable to multi-task training. (2) RFT successfully protects and even enhances the model's general knowledge on standard benchmarks (e.g., MMMU and MMLU-Pro). Conversely, SFT degrades general model capabilities severely. Further analysis reveals that this stability is not primarily due to explicit mechanisms like KL penalty or chain-of-thought reasoning. Instead, we identify an implicit regularization mechanism inherent to RFT as a key contributing factor. Our theoretical analysis suggests that RFT's gradient updates are naturally scaled by the reward variance, acting as a data-dependent regularizer that inherently protects previously acquired knowledge. Finally, we propose a rollout-based instance filtering algorithm to enhance the stability and efficiency of RFT. Our comprehensive study demonstrates the superiority of RFT as a robust paradigm for continual post-training.
Authors: Erfan Hamdi, Emma Lejeune
Abstract: Data driven approaches have the potential to make modeling complex, nonlinear physical phenomena significantly more computationally tractable. For example, computational modeling of fracture is a core challenge where machine learning techniques have the potential to provide a much needed speedup that would enable progress in areas such as mutli-scale modeling and uncertainty quantification. Currently, phase field modeling (PFM) of fracture is one such approach that offers a convenient variational formulation to model crack nucleation, branching and propagation. To date, machine learning techniques have shown promise in approximating PFM simulations. However, most studies rely on overly simple benchmarks that do not reflect the true complexity of the fracture processes where PFM excels as a method. To address this gap, we introduce a challenging dataset based on PFM simulations designed to benchmark and advance ML methods for fracture modeling. This dataset includes three energy decomposition methods, two boundary conditions, and 1,000 random initial crack configurations for a total of 6,000 simulations. Each sample contains 100 time steps capturing the temporal evolution of the crack field. Alongside this dataset, we also implement and evaluate Physics Informed Neural Networks (PINN), Fourier Neural Operators (FNO) and UNet models as baselines, and explore the impact of ensembling strategies on prediction accuracy. With this combination of our dataset and baseline models drawn from the literature we aim to provide a standardized and challenging benchmark for evaluating machine learning approaches to solid mechanics. Our results highlight both the promise and limitations of popular current models, and demonstrate the utility of this dataset as a testbed for advancing machine learning in fracture mechanics research.
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: Chuan He, Yang Chen, Wuliang Huang, Tianyi Zheng, Jianhu Chen, Bin Dou, Yice Luo, Yun Zhu, Baokun Wang, Yongchao Liu, Xing Fu, Yu Cheng, Chuntao Hong, Weiqiang Wang, Xin-Wei Yao, Zhongle Xie
Abstract: Multi-source user representation learning plays a critical role in enabling personalized services on web platforms (e.g., Alipay). While prior works have adopted late-fusion strategies to combine heterogeneous data sources, they suffer from three key limitations: lack of unified representation frameworks, scalability and storage issues in data compression, and inflexible cross-task generalization. To address these challenges, we propose U2QT (Unified User Quantized Tokenizers), a novel framework that integrates cross-domain knowledge transfer with early fusion of heterogeneous domains. Our framework employs a two-stage architecture: first, we use the Qwen3 Embedding model to derive a compact yet expressive feature representation; second, a multi-view RQ-VAE discretizes causal embeddings into compact tokens through shared and source-specific codebooks, enabling efficient storage while maintaining semantic coherence. Experimental results showcase U2QT's advantages across diverse downstream tasks, outperforming task-specific baselines in future behavior prediction and recommendation tasks while achieving efficiency gains in storage and computation. The unified tokenization framework enables seamless integration with language models and supports industrial-scale applications.
Authors: Joon-Hyun Park, Mujin Cheon, Jeongsu Wi, Dong-Yeun Koh
Abstract: The performance of Bayesian optimization (BO), a highly sample-efficient method for expensive black-box problems, is critically governed by the selection of its hyperparameters, including the kernel and acquisition functions. This presents a significant practical challenge: an inappropriate combination can lead to poor performance and wasted evaluations. While individual improvements to kernel functions (e.g., tree-based kernels, deep kernel learning) and acquisition functions (e.g., multi-step lookahead, tree-based planning) have been actively explored, the joint and autonomous selection of the best pair has been largely overlooked, forcing practitioners to rely on heuristics or costly manual tuning. We propose BOOST (Bayesian Optimization with Optimal Kernel and Acquisition Function Selection Technique), a novel framework that automates this selection. BOOST utilizes a lightweight, offline evaluation stage to predict the performance of various kernel-acquisition pairs and identify the most promising pair before committing to expensive evaluations. Using K-means clustering, BOOST first selects initial subsets from previously observed data-in-hand and prepares all possible kernel-acquisition pairs from user-chosen candidates. For each pair, BOOST conducts internal BO runs starting with the initial subset, evaluating how many iterations are required to find the target value within the remaining data, thereby identifying the pair with the best retrospective performance for future optimization. Experiments on synthetic benchmarks and real-world hyperparameter optimization tasks demonstrate that BOOST consistently outperforms standard BO with fixed hyperparameters and state-of-the-art adaptive methods, highlighting its effectiveness and robustness in diverse problem landscapes.
Authors: Sven Weinzierl, Sandra Zilker, Annina Liessmann, Martin K\"appel, Weixin Wang, Martin Matzner
Abstract: Event logs reflect the behavior of business processes that are mapped in organizational information systems. Predictive process monitoring (PPM) transforms these data into value by creating process-related predictions that provide the insights required for proactive interventions at process runtime. Existing PPM techniques require sufficient amounts of event data or other relevant resources that might not be readily available, which prevents some organizations from utilizing PPM. The transfer learning-based PPM technique presented in this paper allows organizations without suitable event data or other relevant resources to implement PPM for effective decision support. This technique is instantiated in both a real-life intra- and an inter-organizational use case, based on which numerical experiments are performed using event logs for IT service management processes. The results of the experiments suggest that knowledge of one business process can be transferred to a similar business process in the same or a different organization to enable effective PPM in the target context. The proposed technique allows organizations to benefit from transfer learning in intra- and inter-organizational settings by transferring resources such as pre-trained models within and across organizational boundaries.
Authors: Hao Luan, Yi Xian Goh, See-Kiong Ng, Chun Kai Ling
Abstract: Modifications to test-time sampling have emerged as an important extension to diffusion algorithms, with the goal of biasing the generative process to achieve a given objective without having to retrain the entire diffusion model. However, generating jointly correlated samples from multiple pre-trained diffusion models while simultaneously enforcing task-specific constraints without costly retraining has remained challenging. To this end, we propose Projected Coupled Diffusion (PCD), a novel test-time framework for constrained joint generation. PCD introduces a coupled guidance term into the generative dynamics to encourage coordination between diffusion models and incorporates a projection step at each diffusion step to enforce hard constraints. Empirically, we demonstrate the effectiveness of PCD in application scenarios of image-pair generation, object manipulation, and multi-robot motion planning. Our results show improved coupling effects and guaranteed constraint satisfaction without incurring excessive computational costs.
Authors: Hengjie Cao, Mengyi Chen, Yifeng Yang, Ruijun Huang, Fang Dong, Jixian Zhou, Anrui Chen, Mingzhi Dong, Yujiang Wang, Jinlong Hou, Yuan Cheng, Fan Wu, Fan Yang, Tun Lu, Ning Gu, Li Shang
Abstract: This work identifies anisotropy in the singular value spectra of parameters, activations, and gradients as the fundamental barrier to low-bit training of large language models (LLMs). These spectra are dominated by a small fraction of large singular values, inducing wide numerical ranges that cause quantization bias and severe spectral distortion, ultimately degrading training performance. This work presents Metis, a spectral-domain quantization framework that partitions anisotropic spectra into narrower sub-distributions for independent quantization, thereby reducing errors and preserving spectral structure. To minimize overhead, Metis leverages two key properties of the dominant spectral subspace: preservation via sparsely random sampling and preservation via random projection, reducing decomposition cost to a negligible level. On LLaMA-3 8B trained with 100B tokens, Metis enables robust W4A4G4 training with FP4 quantization of weights, activations, and gradients, yielding only a 0.4% training loss gap and a 0.1% degradation in downstream accuracy relative to BF16. Beyond matching BF16 fidelity, Metis also surpasses our implementation of Nvidia's recently announced (yet to be publicly released) FP4 recipe, consistently achieving lower loss and higher downstream accuracy while incurring significantly lower computational overhead. The code implementation for Metis is available at: https://anonymous.4open.science/r/Metis-quantization-644B.
URLs: https://anonymous.4open.science/r/Metis-quantization-644B.
Authors: Jialin Wu, Shreya Saha, Yiqing Bo, Meenakshi Khosla
Abstract: Representational similarity metrics are fundamental tools in neuroscience and AI, yet we lack systematic comparisons of their discriminative power across model families. We introduce a quantitative framework to evaluate representational similarity measures based on their ability to separate model families-across architectures (CNNs, Vision Transformers, Swin Transformers, ConvNeXt) and training regimes (supervised vs. self-supervised). Using three complementary separability measures-dprime from signal detection theory, silhouette coefficients and ROC-AUC, we systematically assess the discriminative capacity of commonly used metrics including RSA, linear predictivity, Procrustes, and soft matching. We show that separability systematically increases as metrics impose more stringent alignment constraints. Among mapping-based approaches, soft-matching achieves the highest separability, followed by Procrustes alignment and linear predictivity. Non-fitting methods such as RSA also yield strong separability across families. These results provide the first systematic comparison of similarity metrics through a separability lens, clarifying their relative sensitivity and guiding metric choice for large-scale model and brain comparisons.
Authors: Yuanteng Chen, Peisong Wang, Yuantian Shao, Nanxin Zeng, Chang Xu, Jian Cheng
Abstract: Sparse Mixture-of-Experts (MoE) has become a key architecture for scaling large language models (LLMs) efficiently. Recent fine-grained MoE designs introduce hundreds of experts per layer, with multiple experts activated per token, enabling stronger specialization. However, during pre-training, routers are optimized mainly for stability and robustness: they converge prematurely and enforce balanced usage, limiting the full potential of model performance and efficiency at inference. In this work, we uncover two overlooked issues: (i) a few highly influential experts are underutilized due to premature and balanced routing decisions; and (ii) enforcing a fixed number of active experts per token introduces substantial redundancy. Instead of retraining models or redesigning MoE architectures, we introduce Ban&Pick, a post-training, plug-and-play strategy for smarter routing. Pick discovers and reinforces key experts-a small group with outsized impact on performance-leading to notable accuracy gains across domains. Ban further dynamically prunes redundant experts based on layer and token sensitivity, delivering faster inference with minimal accuracy loss. Experiments on fine-grained MoE-LLMs (DeepSeek, Qwen3) across math, code, and general reasoning benchmarks demonstrate that Ban\&Pick delivers free performance gains and inference acceleration without retraining or architectural changes. For instance, on Qwen3-30B-A3B, it improves accuracy from 80.67 to 84.66 on AIME2024 and from 65.66 to 68.18 on GPQA-Diamond, while accelerating inference by 1.25x under the vLLM.
Authors: Muhammad H. Ashiq, Peter Triantafillou, Hung Yun Tseng, Grigoris G. Chrysos
Abstract: A key concern for AI safety remains understudied in the machine learning (ML) literature: how can we ensure users of ML models do not leverage predictions on incorrect personal data to harm others? This is particularly pertinent given the rise of open-weight models, where simply masking model outputs does not suffice to prevent adversaries from recovering harmful predictions. To address this threat, which we call *test-time privacy*, we induce maximal uncertainty on protected instances while preserving accuracy on all other instances. Our proposed algorithm uses a Pareto optimal objective that explicitly balances test-time privacy against utility. We also provide a certifiable approximation algorithm which achieves $(\varepsilon, \delta)$ guarantees without convexity assumptions. We then prove a tight bound that characterizes the privacy-utility tradeoff that our algorithms incur. Empirically, our method obtains at least $>3\times$ stronger uncertainty than pretraining with marginal drops in accuracy on various image recognition benchmarks. Altogether, this framework provides a tool to guarantee additional protection to end users.
Authors: Shuangshuang He, Yuanting Zhang, Hongli Liang, Qingye Meng, Xingyuan Yuan, Shuo Wang
Abstract: Data-driven hourly weather forecasting models often face the challenge of error accumulation in long-term predictions. The problem is exacerbated by non-physical temporal discontinuities present in widely-used training datasets such as ECMWF Reanalysis v5 (ERA5), which stem from its 12-hour assimilation cycle. Such artifacts lead hourly autoregressive models to learn spurious dynamics and rapidly accumulate errors. To address this, we introduce FlowCast-ODE, a novel framework that treats atmospheric evolution as a continuous flow to ensure temporal coherence. Our method employs dynamic flow matching to learn the instantaneous velocity field from data and an ordinary differential equation (ODE) solver to generate smooth and temporally continuous hourly predictions. By pre-training on 6-hour intervals to sidestep data discontinuities and fine-tuning on hourly data, FlowCast-ODE produces seamless forecasts for up to 120 hours with a single lightweight model. It achieves competitive or superior skill on key meteorological variables compared to baseline models, preserves fine-grained spatial details, and demonstrates strong performance in forecasting extreme events, such as tropical cyclone tracks.
Authors: Xuekai Zhu, Daixuan Cheng, Dinghuai Zhang, Hengli Li, Kaiyan Zhang, Che Jiang, Youbang Sun, Ermo Hua, Yuxin Zuo, Xingtai Lv, Qizheng Zhang, Lin Chen, Fanghao Shao, Bo Xue, Yunchong Song, Zhenjie Yang, Ganqu Cui, Ning Ding, Jianfeng Gao, Xiaodong Liu, Bowen Zhou, Hongyuan Mei, Zhouhan Lin
Abstract: We propose FlowRL: matching the full reward distribution via flow balancing instead of maximizing rewards in large language model (LLM) reinforcement learning (RL). Recent advanced reasoning models adopt reward-maximizing methods (\eg, PPO and GRPO), which tend to over-optimize dominant reward signals while neglecting less frequent but valid reasoning paths, thus reducing diversity. In contrast, we transform scalar rewards into a normalized target distribution using a learnable partition function, and then minimize the reverse KL divergence between the policy and the target distribution. We implement this idea as a flow-balanced optimization method that promotes diverse exploration and generalizable reasoning trajectories. We conduct experiments on math and code reasoning tasks: FlowRL achieves a significant average improvement of $10.0\%$ over GRPO and $5.1\%$ over PPO on math benchmarks, and performs consistently better on code reasoning tasks. These results highlight reward distribution-matching as a key step toward efficient exploration and diverse reasoning in LLM reinforcement learning.
Authors: Yuchen Ye, Chaoxia Yuan, Mingyu Li, Aoqi Zhou, Hong Liang, Chunqing Shang, Kezuan Wang, Yifeng Zheng, Cong Chen
Abstract: Accurate atmospheric wind field information is crucial for various applications, including weather forecasting, aviation safety, and disaster risk reduction. However, obtaining high spatiotemporal resolution wind data remains challenging due to limitations in traditional in-situ observations and remote sensing techniques, as well as the computational expense and biases of numerical weather prediction (NWP) models. This paper introduces G-WindCast, a novel deep learning framework that leverages signal strength variations from 5G Global Navigation Satellite System (GNSS) signals to retrieve and forecast three-dimensional (3D) atmospheric wind fields. The framework utilizes Forward Neural Networks (FNN) and Transformer networks to capture complex, nonlinear, and spatiotemporal relationships between GNSS-derived features and wind dynamics. Our preliminary results demonstrate promising accuracy in both wind retrieval and short-term wind forecasting (up to 30 minutes lead time), with skill scores comparable to high-resolution NWP outputs in certain scenarios. The model exhibits robustness across different forecast horizons and pressure levels, and its predictions for wind speed and direction show superior agreement with observations compared to concurrent ERA5 reanalysis data. Furthermore, we show that the system can maintain excellent performance for localized forecasting even with a significantly reduced number of GNSS stations (e.g., around 100), highlighting its cost-effectiveness and scalability. This interdisciplinary approach underscores the transformative potential of exploiting non-traditional data sources and deep learning for advanced environmental monitoring and real-time atmospheric applications.
Authors: Borui Wan, Gaohong Liu, Zuquan Song, Jun Wang, Yun Zhang, Guangming Sheng, Shuguang Wang, Houmin Wei, Chenyuan Wang, Weiqiang Lou, Xi Yang, Mofan Zhang, Kaihua Jiang, Cheng Ren, Xiaoyun Zhi, Menghan Yu, Zhe Nan, Zhuolin Zheng, Baoquan Zhong, Qinlong Wang, Huan Yu, Jinxin Chi, Wang Zhang, Yuhan Li, Zixian Du, Sida Zhao, Yongqiang Zhang, Jingzhe Tang, Zherui Liu, Chuan Wu, Yanghua Peng, Haibin Lin, Wencong Xiao, Xin Liu, Liang Xiang
Abstract: The training scale of large language models (LLMs) has reached tens of thousands of GPUs and is still continuously expanding, enabling faster learning of larger models. Accompanying the expansion of the resource scale is the prevalence of failures (CUDA error, NaN values, job hang, etc.), which poses significant challenges to training stability. Any large-scale LLM training infrastructure should strive for minimal training interruption, efficient fault diagnosis, and effective failure tolerance to enable highly efficient continuous training. This paper presents ByteRobust, a large-scale GPU infrastructure management system tailored for robust and stable training of LLMs. It exploits the uniqueness of LLM training process and gives top priorities to detecting and recovering failures in a routine manner. Leveraging parallelisms and characteristics of LLM training, ByteRobust enables high-capacity fault tolerance, prompt fault demarcation, and localization with an effective data-driven approach, comprehensively ensuring continuous and efficient training of LLM tasks. ByteRobust is deployed on a production GPU platform with over 200,000 GPUs and achieves 97% ETTR for a three-month training job on 9,600 GPUs.
Authors: Jin Lee, Ziming Liu, Xinling Yu, Yixuan Wang, Haewon Jeong, Murphy Yuezhen Niu, Zheng Zhang
Abstract: We introduce Kolmogorov--Arnold Neural Operator (KANO), a dual-domain neural operator jointly parameterized by both spectral and spatial bases with intrinsic symbolic interpretability. We theoretically demonstrate that KANO overcomes the pure-spectral bottleneck of Fourier Neural Operator (FNO): KANO remains expressive over generic position-dependent dynamics (variable coefficient PDEs) for any physical input, whereas FNO stays practical only for spectrally sparse operators and strictly imposes a fast-decaying input Fourier tail. We verify our claims empirically on position-dependent differential operators, for which KANO robustly generalizes but FNO fails to. In the quantum Hamiltonian learning benchmark, KANO reconstructs ground-truth Hamiltonians in closed-form symbolic representations accurate to the fourth decimal place in coefficients and attains $\approx 6\times10^{-6}$ state infidelity from projective measurement data, substantially outperforming that of the FNO trained with ideal full wave function data, $\approx 1.5\times10^{-2}$, by orders of magnitude.
Authors: Santosh Patapati, Trisanth Srinivasan
Abstract: When different objectives conflict with each other in multi-task learning, gradients begin to interfere and slow convergence, thereby potentially reducing the final model's performance. To address this, we introduce SON-GOKU, a scheduler that computes gradient interference, constructs an interference graph, and then applies greedy graph-coloring to partition tasks into groups that align well with each other. At each training step, only one group (color class) of tasks are activated, and the grouping partition is constantly recomputed as task relationships evolve throughout training. By ensuring that each mini-batch contains only tasks that pull the model in the same direction, our method improves the effectiveness of any underlying multi-task learning optimizer without additional tuning. Since tasks within these groups will update in compatible directions, multi-task learning will improve model performance rather than impede it. Empirical results on six different datasets show that this interference-aware graph-coloring approach consistently outperforms baselines and state-of-the-art multi-task optimizers. We provide extensive theory showing why grouping and sequential updates improve multi-task learning, with guarantees on descent, convergence, and accurately identifying what tasks conflict or align.
Authors: Wenhan Wu, Zheyuan Liu, Chongyang Gao, Ren Wang, Kaize Ding
Abstract: Current LLM unlearning methods face a critical security vulnerability that undermines their fundamental purpose: while they appear to successfully remove sensitive or harmful knowledge, this ``forgotten" information remains precariously recoverable through relearning attacks. We identify that the root cause is that conventional methods optimizing the forgetting loss at individual data points will drive model parameters toward sharp minima in the loss landscape. In these unstable regions, even minimal parameter perturbations can drastically alter the model's behaviors. Consequently, relearning attacks exploit this vulnerability by using just a few fine-tuning samples to navigate the steep gradients surrounding these unstable regions, thereby rapidly recovering knowledge that was supposedly erased. This exposes a critical robustness gap between apparent unlearning and actual knowledge removal. To address this issue, we propose StableUN, a bi-level feedback-guided optimization framework that explicitly seeks more stable parameter regions via neighborhood-aware optimization. It integrates forgetting feedback, which uses adversarial perturbations to probe parameter neighborhoods, with remembering feedback to preserve model utility, aligning the two objectives through gradient projection. Experiments on WMDP and MUSE benchmarks demonstrate that our method is significantly more robust against both relearning and jailbreaking attacks while maintaining competitive utility performance.
Authors: Thadd\"aus Wiedemer, Yuxuan Li, Paul Vicol, Shixiang Shane Gu, Nick Matarese, Kevin Swersky, Been Kim, Priyank Jaini, Robert Geirhos
Abstract: The remarkable zero-shot capabilities of Large Language Models (LLMs) have propelled natural language processing from task-specific models to unified, generalist foundation models. This transformation emerged from simple primitives: large, generative models trained on web-scale data. Curiously, the same primitives apply to today's generative video models. Could video models be on a trajectory towards general-purpose vision understanding, much like LLMs developed general-purpose language understanding? We demonstrate that Veo 3 can solve a broad variety of tasks it wasn't explicitly trained for: segmenting objects, detecting edges, editing images, understanding physical properties, recognizing object affordances, simulating tool use, and more. These abilities to perceive, model, and manipulate the visual world enable early forms of visual reasoning like maze and symmetry solving. Veo's emergent zero-shot capabilities indicate that video models are on a path to becoming unified, generalist vision foundation models.
Authors: Zhenpeng Su, Leiyu Pan, Minxuan Lv, Yuntao Li, Wenping Hu, Fuzheng Zhang, Kun Gai, Guorui Zhou
Abstract: Reinforcement learning (RL) has become a powerful paradigm for optimizing large language models (LLMs) to handle complex reasoning tasks. A core challenge in this process lies in managing policy entropy, which reflects the balance between exploration and exploitation during training. Existing methods, such as proximal policy optimization (PPO) and its variants, discard valuable gradient signals from low-probability tokens due to the clipping mechanism. We systematically analyze the entropy dynamics and reveal that these clipped tokens play a critical yet overlooked role in regulating entropy evolution. We propose \textbf{C}oordinating \textbf{E}ntropy via \textbf{G}radient-\textbf{P}reserving \textbf{P}olicy \textbf{O}ptimization (CE-GPPO), a novel algorithm that reintroduces gradients from clipped tokens in native PPO in a gentle and bounded manner. By controlling the magnitude of gradients from tokens outside the clipping interval, CE-GPPO is able to achieve an exploration-exploitation trade-off. We provide theoretical justification and empirical evidence showing that CE-GPPO effectively mitigates entropy instability. Extensive experiments on mathematical reasoning benchmarks show that CE-GPPO consistently outperforms strong baselines across different model scales.
Authors: Woosung Koh, Juyoung Suk, Sungjun Han, Se-Young Yun, Jamin Shin
Abstract: Given the prohibitive cost of pre-training large language models, it is essential to leverage smaller proxy models to optimize datasets before scaling up. However, this approach becomes challenging for reasoning capabilities, which exhibit emergent behavior that only appear reliably at larger model sizes, often exceeding 7B parameters. To address this, we introduce rBridge, showing that small proxies ($\leq$1B) can effectively predict large-model reasoning by aligning more closely with (1) the pre-training objective and (2) the target task. rBridge achieves this by weighting negative log-likelihood with task alignment, using reasoning traces from frontier models as gold labels. In our experiments, rBridge (i) reduces dataset ranking costs by over 100x relative to the best baseline, (ii) achieves the strongest correlation across six reasoning benchmarks at 1B to 32B scale, and (iii) zero-shot transfers predictive relationships across pre-training datasets at 1B to 7B scale. These findings indicate that rBridge offers a practical path for exploring reasoning-oriented pre-training at lower cost.
Authors: Yuandong Tian
Abstract: While the phenomenon of grokking, i.e., delayed generalization, has been studied extensively, it remains an open problem whether there is a mathematical framework that characterizes what kind of features will emerge, how and in which conditions it happens, and is closely related to the gradient dynamics of the training, for complex structured inputs. We propose a novel framework, named $\mathbf{Li_2}$, that captures three key stages for the grokking behavior of 2-layer nonlinear networks: (I) \underline{\textbf{L}}azy learning, (II) \underline{\textbf{i}}ndependent feature learning and (III) \underline{\textbf{i}}nteractive feature learning. At the lazy learning stage, top layer overfits to random hidden representation and the model appears to memorize. Thanks to lazy learning and weight decay, the \emph{backpropagated gradient} $G_F$ from the top layer now carries information about the target label, with a specific structure that enables each hidden node to learn their representation \emph{independently}. Interestingly, the independent dynamics follows exactly the \emph{gradient ascent} of an energy function $E$, and its local maxima are precisely the emerging features. We study whether these local-optima induced features are generalizable, their representation power, and how they change on sample size, in group arithmetic tasks. When hidden nodes start to interact in the later stage of learning, we provably show how $G_F$ changes to focus on missing features that need to be learned. Our study sheds lights on roles played by key hyperparameters such as weight decay, learning rate and sample sizes in grokking, leads to provable scaling laws of feature emergence, memorization and generalization, and reveals the underlying cause why recent optimizers such as Muon can be effective, from the first principles of gradient dynamics. Our analysis can be extended to multi-layer architectures.
Authors: Isaac Reid, Arijit Sehanobish, Cederik H\"ofs, Bruno Mlodozeniec, Leonhard Vulpius, Federico Barbero, Adrian Weller, Krzysztof Choromanski, Richard E. Turner, Petar Veli\v{c}kovi\'c
Abstract: We introduce WIRE: Wavelet-Induced Rotary Encodings. WIRE extends Rotary Position Encodings (RoPE), a popular algorithm in LLMs and ViTs, to graph-structured data. We demonstrate that WIRE is more general than RoPE, recovering the latter in the special case of grid graphs. WIRE also enjoys a host of desirable theoretical properties, including equivariance under node ordering permutation, compatibility with linear attention, and (under select assumptions) asymptotic dependence on graph resistive distance. We test WIRE on a range of synthetic and real-world tasks, including identifying monochromatic subgraphs, semantic segmentation of point clouds, and more standard graph benchmarks. We find it to be effective in settings where the underlying graph structure is important.
Authors: Lorenz K. M\"uller, Philippe Bich, Jiawei Zhuang, Ahmet \c{C}elik, Luca Benfenati, Lukas Cavigelli
Abstract: Post-training quantization has emerged as the most widely used strategy for deploying large language models at low precision. Still, current methods show perplexity degradation at bit-widths less than or equal to 4, partly because representing outliers causes precision issues in parameters that share the same scales as these outliers. This problem is especially pronounced for calibration-free, uniform quantization methods. We introduce SINQ to augment existing post-training quantizers with an additional second-axis scale factor and a fast Sinkhorn-Knopp-style algorithm that finds scales to normalize per-row and per-column variances, thereby minimizing a novel per-matrix proxy target for quantization: the matrix imbalance. Our method has no interactions between layers and can be trivially applied to new architectures to quantize any linear layers. We evaluate our method on the Qwen3 model family and DeepSeek-V2.5. SINQ improves WikiText2 and C4 perplexity significantly against uncalibrated uniform quantization baselines and can be further enhanced by combining it with calibration and non-uniform quantization levels. Code to reproduce the results of this work and to easily quantize models using SINQ is available at https://github.com/huawei-csl/SINQ.
Authors: Emerald Zhang, Julian Weaver, Samantha R Santacruz, Edward Castillo
Abstract: Explainable AI (XAI) methods help identify which image regions influence a model's prediction, but often face a trade-off between detail and interpretability. Layer-wise Relevance Propagation (LRP) offers a model-aware alternative. However, LRP implementations commonly rely on heuristic rule sets that are not optimized for clarity or alignment with model behavior. We introduce EVO-LRP, a method that applies Covariance Matrix Adaptation Evolution Strategy (CMA-ES) to tune LRP hyperparameters based on quantitative interpretability metrics, such as faithfulness or sparseness. EVO-LRP outperforms traditional XAI approaches in both interpretability metric performance and visual coherence, with strong sensitivity to class-specific features. These findings demonstrate that attribution quality can be systematically improved through principled, task-specific optimization.
Authors: Chris Kolb, Laetitia Frost, Bernd Bischl, David R\"ugamer
Abstract: Structured sparsity regularization offers a principled way to compact neural networks, but its non-differentiability breaks compatibility with conventional stochastic gradient descent and requires either specialized optimizers or additional post-hoc pruning without formal guarantees. In this work, we propose $D$-Gating, a fully differentiable structured overparameterization that splits each group of weights into a primary weight vector and multiple scalar gating factors. We prove that any local minimum under $D$-Gating is also a local minimum using non-smooth structured $L_{2,2/D}$ penalization, and further show that the $D$-Gating objective converges at least exponentially fast to the $L_{2,2/D}$-regularized loss in the gradient flow limit. Together, our results show that $D$-Gating is theoretically equivalent to solving the original group sparsity problem, yet induces distinct learning dynamics that evolve from a non-sparse regime into sparse optimization. We validate our theory across vision, language, and tabular tasks, where $D$-Gating consistently delivers strong performance-sparsity tradeoffs and outperforms both direct optimization of structured penalties and conventional pruning baselines.
Authors: Kaisen Yang, Lixuan He, Rushi Shah, Kaicheng Yang, Qinwei Ma, Dianbo Liu, Alex Lamb
Abstract: Chain-of-Thought (CoT) and its variants have markedly advanced the reasoning abilities of Large Language Models (LLMs), yet their monolithic and auto-regressive architecture inherently conflates high-level strategic planning with low-level step-by-step execution, leading to computational inefficiency, limited exploration of reasoning paths, and reduced interpretability. To overcome these issues, we propose the Explore-Execute Chain ($E^2C$), a structured reasoning framework that decouples reasoning into two distinct phases: an exploratory phase that stochastically generates succinct high-level plans, followed by an execution phase that deterministically carries out the chosen plan. Our approach incorporates a two-stage training methodology, which combines Supervised Fine-Tuning (SFT) - augmented by a novel data generation algorithm enforcing strict plan adherence - with a subsequent Reinforcement Learning (RL) stage that capitalizes on the informativeness of exploration and reinforces the determinism of execution. This decomposition enables an efficient test-time scaling strategy: on AIME'2024, $E^2C$ Test Time Scaling reaches 58.1% accuracy using <10% of the decoding tokens required by comparable methods (e.g., Forest-of-Thought), sharply cutting self-consistency overhead. For cross-domain adaptation, our Exploration-Focused SFT (EF-SFT) fine-tunes with only 3.5% of the tokens used by standard SFT yet yields up to 14.5% higher accuracy than standard SFT on medical benchmarks, delivering state-of-the-art performance, strong generalization, and greater interpretability by separating planning from execution. The code and pre-trained models for the project are available at: https://github.com/yks23/Explore-Execute-Chain.git
Authors: Bo Hu, Jos\'e C. Pr\'incipe
Abstract: Pairwise distance-based costs are crucial for self-supervised and contrastive feature learning. Mixture Density Networks (MDNs) are a widely used approach for generative models and density approximation, using neural networks to produce multiple centers that define a Gaussian mixture. By combining MDNs with contrastive costs, this paper proposes data density approximation using four types of kernelized matrix costs: the scalar cost, the vector-matrix cost, the matrix-matrix cost (the trace of Schur complement), and the SVD cost (the nuclear norm), for learning multiple centers required to define a mixture density.
Authors: Junjie Wang, Pan Zhou, Yiming Dong, Huan Li, Jia Li, Xun Zhou, Qicheng Lao, Cong Fang, Zhouchen Lin
Abstract: Large language models (LLMs) have demonstrated impressive generalization and emergent capabilities, yet their pre-training remains computationally expensive and sensitive to optimization dynamics. While Adam-based optimizers offer fast convergence by adapting learning rates coordinate-wise, recent studies reveal that their updates often suffer from poor spectral conditioning and low-rank structures, hindering efficiency. Muon addresses this issue via global spectral normalization but lacks the per-coordinate adaptivity of Adam. In this work, we propose Column-Normalized Adam (Conda), a novel optimizer that bridges the strengths of both approaches. Conda projects updates into an orthogonal subspace and applies column-wise second moment normalization based on the projected gradients, thereby achieving both improved spectral conditioning and maintaining coordinate-wise adaptivity. This design alleviates the spectral pathologies of Adam while preserving its fast convergence behavior. Extensive experiments on the LLaMA and GPT-2 series show that Conda consistently outperforms AdamW, Muon, and other baselines in pre-training. Remarkably, on the LLaMA series, Conda achieves 2-2.5 the convergence speed of AdamW, measured in both training steps and training time. Further ablations demonstrate its robustness under diverse training setups. These results collectively highlight Conda as an effective and broadly applicable optimizer for large-scale LLM training. The code is released on https://github.com/jie040109/Conda
Authors: Dipan Maity
Abstract: Orthogonal gradient updates have emerged as a promising direction in optimization for machine learning. However, traditional approaches such as SVD/QR decomposition incur prohibitive computational costs of O(n^3) and underperform compared to well-tuned SGD with momentum, since momentum is applied only after strict orthogonalization. Recent advances, such as Muon, improve efficiency by applying momentum before orthogonalization and producing semi-orthogonal matrices via Newton-Schulz iterations, reducing complexity to O(n^2). Nevertheless, quadratic costs remain a bottleneck. In this work, we study the semi-orthogonal properties of momentum-based updates and develop a method to bound momentum updates under a spectral-norm trust region, preserving directional information without requiring explicit semi-orthogonalization. We propose AuON (Alternative Unit-norm momentum updates by Normalized nonlinear scaling), a linear-time optimizer that achieves strong performance without constructing semi-orthogonal matrices, while preserving structural alignment and reconditioning ill-posed updates. Our approach combines hyperbolic-cosine RMS scaling transformations with normalization, demonstrating both effectiveness and computational efficiency compared to Newton-Schulz methods. We further introduce a hybrid variant (Hybrid-AuON) that applies a single Newton-Schulz iteration. Experiments across vision and language benchmarks show that AuON and its hybrid variant achieve performance comparable to strong baselines such as AdamW and Muon. Code is available at: https://github.com/ryyzn9/AuON
Authors: Maedeh Zarvandi, Michael Timothy, Theresa Wasserer, Debarghya Ghoshdastidar
Abstract: Kernel methods provide a theoretically grounded framework for non-linear and non-parametric learning, with strong analytic foundations and statistical guarantees. Yet, their scalability has long been limited by prohibitive time and memory costs. While progress has been made in scaling kernel regression, no framework exists for scalable kernel-based representation learning, restricting their use in the era of foundation models where representations are learned from massive unlabeled data. We introduce KREPES -- a unified, scalable framework for kernel-based representation learning via Nystr\"om approximation. KREPES accommodates a wide range of unsupervised and self-supervised losses, and experiments on large image and tabular datasets demonstrate its efficiency. Crucially, KREPES enables principled interpretability of the learned representations, an immediate benefit over deep models, which we substantiate through dedicated analysis.
Authors: Liang Lin, Zhihao Xu, Junhao Dong, Jian Zhao, Yuchen Yuan, Guibin Zhang, Miao Yu, Yiming Zhang, Zhengtao Yao, Huahui Yi, Dongrui Liu, Xinfeng Li, Kun Wang
Abstract: Large language model (LLM) alignment faces a critical dilemma when addressing multiple human preferences: improvements in one dimension frequently come at the expense of others, creating unavoidable trade-offs between competing objectives like helpfulness and harmlessness. While prior work mainly focuses on constraint-based optimization algorithms and data selection strategies to mitigate conflicts, these approaches overlook the fundamental issue of resolving conflicts directly at the parameter level. In this paper, we present OrthAlign, an innovative approach that pioneers a new paradigm by leveraging orthogonal subspace decomposition to fundamentally resolve gradient-level conflicts in multi-objective preference alignment. OrthAlign strategically decomposes parameter update spaces into orthogonal subspaces, ensuring that optimization toward different preferences occurs in mathematically non-interfering directions. Building upon this, we provide theoretical guarantees demonstrating that when parameter increments satisfy both orthogonal subspace constraints and spectral norm bounds, the resulting updates exhibit linear Lipschitz growth rather than exponential instability, ensuring stable convergence across all preference dimensions. Extensive experiments show that: I. OrthAlign achieves maximum single-preference improvements ranging from 34.61% to 50.89% after multiple-objective alignment across helpful, harmless, and truthful dimensions. II. With an average overall reward improvement of 13.96%.
Authors: Kacper Kapu\'sniak, Cristian Gabellini, Michael Bronstein, Prudencio Tossou, Francesco Di Giovanni
Abstract: Molecular Dynamics (MD) is a powerful computational microscope for probing protein functions. However, the need for fine-grained integration and the long timescales of biomolecular events make MD computationally expensive. To address this, several generative models have been proposed to generate surrogate trajectories at lower cost. Yet, these models typically learn a fixed-lag transition density, causing the training signal to be dominated by frequent but uninformative transitions. We introduce a new class of generative models, MSM Emulators, which instead learn to sample transitions across discrete states defined by an underlying Markov State Model (MSM). We instantiate this class with Markov Space Flow Matching (MarS-FM), whose sampling offers more than two orders of magnitude speedup compared to implicit- or explicit-solvent MD simulations. We benchmark Mars-FM ability to reproduce MD statistics through structural observables such as RMSD, radius of gyration, and secondary structure content. Our evaluation spans protein domains (up to 500 residues) with significant chemical and structural diversity, including unfolding events, and enforces strict sequence dissimilarity between training and test sets to assess generalization. Across all metrics, MarS-FM outperforms existing methods, often by a substantial margin.
Authors: Nazmus Sakib Ahmed, Saad Sakib Noor, Ashraful Islam Shanto Sikder, Abhijit Paul
Abstract: This paper focuses on enhancing Bengali Document Layout Analysis (DLA) using the YOLOv8 model and innovative post-processing techniques. We tackle challenges unique to the complex Bengali script by employing data augmentation for model robustness. After meticulous validation set evaluation, we fine-tune our approach on the complete dataset, leading to a two-stage prediction strategy for accurate element segmentation. Our ensemble model, combined with post-processing, outperforms individual base architectures, addressing issues identified in the BaDLAD dataset. By leveraging this approach, we aim to advance Bengali document analysis, contributing to improved OCR and document comprehension and BaDLAD serves as a foundational resource for this endeavor, aiding future research in the field. Furthermore, our experiments provided key insights to incorporate new strategies into the established solution.
Authors: Myeung Suk Oh, Anindya Bijoy Das, Taejoon Kim, David J. Love, Christopher G. Brinton
Abstract: Recently, deep learning approaches have provided solutions to difficult problems in wireless positioning (WP). Although these WP algorithms have attained excellent and consistent performance against complex channel environments, the computational complexity coming from processing high-dimensional features can be prohibitive for mobile applications. In this work, we design a novel positioning neural network (P-NN) that utilizes the minimum description features to substantially reduce the complexity of deep learning-based WP. P-NN's feature selection strategy is based on maximum power measurements and their temporal locations to convey information needed to conduct WP. We improve P-NN's learning ability by intelligently processing two different types of inputs: sparse image and measurement matrices. Specifically, we implement a self-attention layer to reinforce the training ability of our network. We also develop a technique to adapt feature space size, optimizing over the expected information gain and the classification capability quantified with information-theoretic measures on signal bin selection. Numerical results show that P-NN achieves a significant advantage in performance-complexity tradeoff over deep learning baselines that leverage the full power delay profile (PDP). In particular, we find that P-NN achieves a large improvement in performance for low SNR, as unnecessary measurements are discarded in our minimum description features.
Authors: Khawlah Bajbaa, Muhammad Usman, Saeed Anwar, Ibrahim Radwan, Abdul Bais
Abstract: In recent years, street view imagery has grown to become one of the most important sources of geospatial data collection and urban analytics, which facilitates generating meaningful insights and assisting in decision-making. Synthesizing a street-view image from its corresponding satellite image is a challenging task due to the significant differences in appearance and viewpoint between the two domains. In this study, we screened 20 recent research papers to provide a thorough review of the state-of-the-art of how street-view images are synthesized from their corresponding satellite counterparts. The main findings are: (i) novel deep learning techniques are required for synthesizing more realistic and accurate street-view images; (ii) more datasets need to be collected for public usage; and (iii) more specific evaluation metrics need to be investigated for evaluating the generated images appropriately. We conclude that, due to applying outdated deep learning techniques, the recent literature failed to generate detailed and diverse street-view images.
Authors: Chinmay Datar, Taniya Kapoor, Abhishek Chandra, Qing Sun, Erik Lien Bolager, Iryna Burak, Anna Veselovska, Massimo Fornasier, Felix Dietrich
Abstract: Solving time-dependent Partial Differential Equations (PDEs) is one of the most critical problems in computational science. While Physics-Informed Neural Networks (PINNs) offer a promising framework for approximating PDE solutions, their accuracy and training speed are limited by two core barriers: gradient-descent-based iterative optimization over complex loss landscapes and non-causal treatment of time as an extra spatial dimension. We present Frozen-PINN, a novel PINN based on the principle of space-time separation that leverages random features instead of training with gradient descent, and incorporates temporal causality by construction. On eight PDE benchmarks, including challenges such as extreme advection speeds, shocks, and high dimensionality, Frozen-PINNs achieve superior training efficiency and accuracy over state-of-the-art PINNs, often by several orders of magnitude. Our work addresses longstanding training and accuracy bottlenecks of PINNs, delivering quickly trainable, highly accurate, and inherently causal PDE solvers, a combination that prior methods could not realize. Our approach challenges the reliance of PINNs on stochastic gradient-descent-based methods and specialized hardware, leading to a paradigm shift in PINN training and providing a challenging benchmark for the community.
Authors: Xinhui Rong, Victor Solo
Abstract: Despite the widespread occurrence of classification problems and the increasing collection of point process data across many disciplines, study of error probability for point process classification only emerged very recently. Here, we consider classification of renewal processes. We obtain asymptotic expressions for the Bhattacharyya bound on misclassification error probabilities for heavy-tailed renewal processes.
Authors: Mark Kozdoba, Binyamin Perets, Shie Mannor
Abstract: There is a well known intrinsic trade-off between the fairness of a representation and the performance of classifiers derived from the representation. Due to the complexity of optimisation algorithms in most modern representation learning approaches, for a given method it may be non-trivial to decide whether the obtained fairness-performance curve of the method is optimal, i.e., whether it is close to the true Pareto front for these quantities for the underlying data distribution. In this paper we propose a new method to compute the optimal Pareto front, which does not require the training of complex representation models. We show that optimal fair representations possess several useful structural properties, and that these properties enable a reduction of the computation of the Pareto Front to a compact discrete problem. We then also show that these compact approximating problems can be efficiently solved via off-the shelf concave-convex programming methods. Since our approach is independent of the specific model of representations, it may be used as the benchmark to which representation learning algorithms may be compared. We experimentally evaluate the approach on a number of real world benchmark datasets.
Authors: Jun Liu, Geng Yuan, Weihao Zeng, Hao Tang, Wenbin Zhang, Xue Lin, XiaoLin Xu, Dong Huang, Yanzhi Wang
Abstract: In research findings, co-deletion of the 1p/19q gene is associated with clinical outcomes in low-grade gliomas. The ability to predict 1p19q status is critical for treatment planning and patient follow-up. This study aims to utilize a specially MRI-based convolutional neural network for brain cancer detection. Although public networks such as RestNet and AlexNet can effectively diagnose brain cancers using transfer learning, the model includes quite a few weights that have nothing to do with medical images. As a result, the diagnostic results are unreliable by the transfer learning model. To deal with the problem of trustworthiness, we create the model from the ground up, rather than depending on a pre-trained model. To enable flexibility, we combined convolution stacking with a dropout and full connect operation, it improved performance by reducing overfitting. During model training, we also supplement the given dataset and inject Gaussian noise. We use three--fold cross-validation to train the best selection model. Comparing InceptionV3, VGG16, and MobileNetV2 fine-tuned with pre-trained models, our model produces better results. On an validation set of 125 codeletion vs. 31 not codeletion images, the proposed network achieves 96.37\% percent F1-score, 97.46\% percent precision, and 96.34\% percent recall when classifying 1p/19q codeletion and not codeletion images.
Authors: Ce Li, Tao Lin
Abstract: Information designers, such as online platforms, often do not know the beliefs of their receivers. We design learning algorithms so that the information designer can learn the receivers' prior belief from their actions through repeated interactions. Our learning algorithms achieve no regret relative to the optimality for the known prior at a fast speed, achieving a tight regret bound $\Theta(\log T)$ in general and a tight regret bound $\Theta(\log \log T)$ in the important special case of binary actions.
Authors: Reuben Dorent, Nazim Haouchine, Alexandra Golby, Sarah Frisken, Tina Kapur, William Wells
Abstract: We propose a deep mixture of multimodal hierarchical variational auto-encoders called MMHVAE that synthesizes missing images from observed images in different modalities. MMHVAE's design focuses on tackling four challenges: (i) creating a complex latent representation of multimodal data to generate high-resolution images; (ii) encouraging the variational distributions to estimate the missing information needed for cross-modal image synthesis; (iii) learning to fuse multimodal information in the context of missing data; (iv) leveraging dataset-level information to handle incomplete data sets at training time. Extensive experiments are performed on the challenging problem of pre-operative brain multi-parametric magnetic resonance and intra-operative ultrasound imaging.
Authors: Jiacheng Liang, Zian Wang, Lauren Hong, Shouling Ji, Ting Wang
Abstract: Various watermarking methods (``watermarkers'') have been proposed to identify LLM-generated texts; yet, due to the lack of unified evaluation platforms, many critical questions remain under-explored: i) What are the strengths/limitations of various watermarkers, especially their attack robustness? ii) How do various design choices impact their robustness? iii) How to optimally operate watermarkers in adversarial environments? To fill this gap, we systematize existing LLM watermarkers and watermark removal attacks, mapping out their design spaces. We then develop WaterPark, a unified platform that integrates 10 state-of-the-art watermarkers and 12 representative attacks. More importantly, by leveraging WaterPark, we conduct a comprehensive assessment of existing watermarkers, unveiling the impact of various design choices on their attack robustness. We further explore the best practices to operate watermarkers in adversarial environments. We believe our study sheds light on current LLM watermarking techniques while WaterPark serves as a valuable testbed to facilitate future research.
Authors: Lingdong Kong, Xiang Xu, Youquan Liu, Jun Cen, Runnan Chen, Wenwei Zhang, Liang Pan, Kai Chen, Ziwei Liu
Abstract: Recent advancements in vision foundation models (VFMs) have revolutionized visual perception in 2D, yet their potential for 3D scene understanding, particularly in autonomous driving applications, remains underexplored. In this paper, we introduce LargeAD, a versatile and scalable framework designed for large-scale 3D pretraining across diverse real-world driving datasets. Our framework leverages VFMs to extract semantically rich superpixels from 2D images, which are aligned with LiDAR point clouds to generate high-quality contrastive samples. This alignment facilitates cross-modal representation learning, enhancing the semantic consistency between 2D and 3D data. We introduce several key innovations: (i) VFM-driven superpixel generation for detailed semantic representation, (ii) a VFM-assisted contrastive learning strategy to align multimodal features, (iii) superpoint temporal consistency to maintain stable representations across time, and (iv) multi-source data pretraining to generalize across various LiDAR configurations. Our approach achieves substantial gains over state-of-the-art methods in linear probing and fine-tuning for LiDAR-based segmentation and object detection. Extensive experiments on 11 large-scale multi-sensor datasets highlight our superior performance, demonstrating adaptability, efficiency, and robustness in real-world autonomous driving scenarios.
Authors: Oscar Ramos-Soto, Jorge Ramos-Frutos, Ezequiel Perez-Zarate, Diego Oliva, Sandra E. Balderas-Mata
Abstract: Feature extraction techniques are crucial in medical image classification; however, classical feature extractors, in addition to traditional machine learning classifiers, often exhibit significant limitations in providing sufficient discriminative information for complex image sets. While Convolutional Neural Networks (CNNs) and Vision Transformer (ViT) have shown promise in feature extraction, they are prone to overfitting due to the inherent characteristics of medical imaging data, including small sample sizes or high intra-class variance. In this work, the Medical Image Attention-based Feature Extractor (MIAFEx) is proposed, a novel method that employs a learnable refinement mechanism to enhance the classification token within the Transformer encoder architecture. This mechanism adjusts the token based on learned weights, improving the extraction of salient features and enhancing the model's adaptability to the challenges presented by medical imaging data. The MIAFEx output feature quality is compared against classical feature extractors using traditional and hybrid classifiers. Also, the performance of these features is compared against modern CNN and ViT models in classification tasks, demonstrating their superiority in accuracy and robustness across multiple complex medical imaging datasets. This advantage is particularly pronounced in scenarios with limited training data, where traditional and modern models often struggle to generalize effectively. The source code of this proposal can be found at https://github.com/Oscar-RamosS/Medical-Image-Attention-based-Feature-Extractor-MIAFEx
URLs: https://github.com/Oscar-RamosS/Medical-Image-Attention-based-Feature-Extractor-MIAFEx
Authors: Yan Yang, Bin Gao, Ya-xiang Yuan
Abstract: Imposing additional constraints on low-rank optimization has garnered growing interest. However, the geometry of coupled constraints hampers the well-developed low-rank structure and makes the problem intricate. To this end, we propose a space-decoupling framework for optimization on bounded-rank matrices with orthogonally invariant constraints. The "space-decoupling" is reflected in several ways. We show that the tangent cone of coupled constraints is the intersection of tangent cones of each constraint. Moreover, we decouple the intertwined bounded-rank and orthogonally invariant constraints into two spaces, leading to optimization on a smooth manifold. Implementing Riemannian algorithms on this manifold is painless as long as the geometry of additional constraints is known. In addition, we unveil the equivalence between the reformulated problem and the original problem. Numerical experiments on real-world applications -- spherical data fitting, graph similarity measuring, low-rank SDP, model reduction of Markov processes, reinforcement learning, and deep learning -- validate the superiority of the proposed framework.
Authors: Wei Guo, Molei Tao, Yongxin Chen
Abstract: Given an unnormalized probability density $\pi\propto\mathrm{e}^{-V}$, estimating its normalizing constant $Z=\int_{\mathbb{R}^d}\mathrm{e}^{-V(x)}\mathrm{d}x$ or free energy $F=-\log Z$ is a crucial problem in Bayesian statistics, statistical mechanics, and machine learning. It is challenging especially in high dimensions or when $\pi$ is multimodal. To mitigate the high variance of conventional importance sampling estimators, annealing-based methods such as Jarzynski equality and annealed importance sampling are commonly adopted, yet their quantitative complexity guarantees remain largely unexplored. We take a first step toward a non-asymptotic analysis of annealed importance sampling. In particular, we derive an oracle complexity of $\widetilde{O}\left(\frac{d\beta^2{\mathcal{A}}^2}{\varepsilon^4}\right)$ for estimating $Z$ within $\varepsilon$ relative error with high probability, where $\beta$ is the smoothness of $V$ and $\mathcal{A}$ denotes the action of a curve of probability measures interpolating $\pi$ and a tractable reference distribution. Our analysis, leveraging Girsanov theorem and optimal transport, does not explicitly require isoperimetric assumptions on the target distribution. Finally, to tackle the large action of the widely used geometric interpolation, we propose a new algorithm based on reverse diffusion samplers, establish a framework for analyzing its complexity, and empirically demonstrate its efficiency in tackling multimodality.
Authors: Giuseppe Alessio D'Inverno, Kylian Ajavon, Simone Brugiapaglia
Abstract: Diffusion kernels over graphs have been widely utilized as effective tools in various applications due to their ability to accurately model the flow of information through nodes and edges. However, there is a notable gap in the literature regarding the development of surrogate models for diffusion processes on graphs. In this work, we fill this gap by proposing sparse polynomial-based surrogate models for parametric diffusion equations on graphs with community structure. In tandem, we provide convergence guarantees for both least squares and compressed sensing-based approximations by showing the holomorphic regularity of parametric solutions to these diffusion equations. Our theoretical findings are accompanied by a series of numerical experiments conducted on both synthetic and real-world graphs that demonstrate the applicability of our methodology.
Authors: Anshul Nasery, Jonathan Hayase, Creston Brooks, Peiyao Sheng, Himanshu Tyagi, Pramod Viswanath, Sewoong Oh
Abstract: Model fingerprinting has emerged as a powerful tool for model owners to identify their shared model given API access. However, to lower false discovery rate, fight fingerprint leakage, and defend against coalitions of model users attempting to bypass detection, we argue that {\em scalability} is critical, i.e., scaling up the number of fingerprints one can embed into a model. Hence, we pose scalability as a crucial requirement for fingerprinting schemes. We experiment with fingerprint design at a scale significantly larger than previously considered, and introduce a new method, dubbed Perinucleus sampling, to generate scalable, persistent, and harmless fingerprints. We demonstrate that this scheme can add 24,576 fingerprints to a Llama-3.1-8B model -- two orders of magnitude more than existing schemes -- without degrading the model's utility. Our inserted fingerprints persist even after supervised fine-tuning on standard post-training data. We further address security risks for fingerprinting, and theoretically and empirically show how a scalable fingerprinting scheme like ours can mitigate these risks. Our code is available at https://github.com/SewoongLab/scalable-fingerprinting-of-llms
URLs: https://github.com/SewoongLab/scalable-fingerprinting-of-llms
Authors: Gaurav Srivastava, Shuxiang Cao, Xuan Wang
Abstract: Reasoning has long been viewed as an emergent property of large language models (LLMs). However, recent studies challenge this assumption, showing that small language models (SLMs) can also achieve competitive reasoning performance. This paper introduces ThinkSLM, the first extensive benchmark to systematically evaluate and study the reasoning abilities of SLMs trained from scratch or derived from LLMs through quantization, pruning, and distillation. We first establish a reliable evaluation criterion comparing available methods and LLM judges against our human evaluations. Then we present a study evaluating 72 diverse SLMs from six major model families across 17 reasoning benchmarks. We repeat all our experiments three times to ensure a robust assessment. Our findings show that: 1) reasoning ability in SLMs is strongly influenced by training methods and data quality rather than solely model scale; 2) quantization preserves reasoning capability, while pruning significantly disrupts it; 3) larger models consistently exhibit higher robustness against adversarial perturbations and intermediate reasoning, but certain smaller models closely match or exceed the larger models' performance. Our findings challenge the assumption that scaling is the only way to achieve strong reasoning. Instead, we foresee a future where SLMs with strong reasoning capabilities can be developed through structured training or post-training compression. Our ThinkSLM Leaderboard is publicly available at: https://ctrl-gaurav.github.io/thinkslm.github.io/
Authors: Ryoma Sato
Abstract: We propose a new approach that enables end users to directly solve the cold start problem by themselves. The cold start problem is a common issue in recommender systems, and many methods have been proposed to address the problem on the service provider's side. However, when the service provider does not take action, users are left with poor recommendations and no means to improve their experience. We propose an algorithm, Pretender, that allows end users to proactively solve the cold start problem on their own. Pretender does not require any special support from the service provider and can be deployed independently by users. We formulate the problem as minimizing the distance between the source and target distributions and optimize item selection from the target service accordingly. Furthermore, we establish theoretical guarantees for Pretender based on a discrete quadrature problem. We conduct experiments on real-world datasets to demonstrate the effectiveness of Pretender.
Authors: Yihong Dong, Ge Li, Xue Jiang, Yongding Tao, Kechi Zhang, Hao Zhu, Huanyu Liu, Jiazheng Ding, Jia Li, Jinliang Deng, Hong Mei
Abstract: Periodicity, as one of the most important basic characteristics, lays the foundation for facilitating structured knowledge acquisition and systematic cognitive processes within human learning paradigms. However, the potential flaws of periodicity modeling in Transformer affect the learning efficiency and establishment of underlying principles from data for large language models (LLMs) built upon it. In this paper, we demonstrate that integrating effective periodicity modeling can improve the learning efficiency and performance of LLMs. We introduce FANformer, which adapts Fourier Analysis Network (FAN) into attention mechanism to achieve efficient periodicity modeling, by modifying the feature projection process of attention mechanism. Extensive experimental results on language modeling show that FANformer consistently outperforms Transformer when scaling up model size and training tokens, underscoring its superior learning efficiency. Our pretrained FANformer-1B exhibits marked improvements on downstream tasks compared to open-source LLMs with similar model parameters or training tokens. Moreover, we reveal that FANformer exhibits superior ability to learn and apply rules for reasoning compared to Transformer. The results position FANformer as an effective and promising architecture for advancing LLMs.
Authors: Luobin Wang, Hongzhan Yu, Chenning Yu, Sicun Gao, Henrik Christensen
Abstract: Diffusion models have recently gained significant attention in robotics due to their ability to generate multi-modal distributions of system states and behaviors. However, a key challenge remains: ensuring precise control over the generated outcomes without compromising realism. This is crucial for applications such as motion planning or trajectory forecasting, where adherence to physical constraints and task-specific objectives is essential. We propose a novel framework that enhances controllability in diffusion models by leveraging multi-modal prior distributions and enforcing strong modal coupling. This allows us to initiate the denoising process directly from distinct prior modes that correspond to different possible system behaviors, ensuring sampling to align with the training distribution. We evaluate our approach on motion prediction using the Waymo dataset and multi-task control in Maze2D environments. Experimental results show that our framework outperforms both guidance-based techniques and conditioned models with unimodal priors, achieving superior fidelity, diversity, and controllability, even in the absence of explicit conditioning. Overall, our approach provides a more reliable and scalable solution for controllable motion generation in robotics.
Authors: Samuel Gruffaz, Josua Sassen
Abstract: Riemannian metric learning is an emerging field in machine learning, unlocking new ways to encode complex data structures beyond traditional distance metric learning. While classical approaches rely on global distances in Euclidean space, they often fall short in capturing intrinsic data geometry. Enter Riemannian metric learning: a powerful generalization that leverages differential geometry to model the data according to their underlying Riemannian manifold. This approach has demonstrated remarkable success across diverse domains, from causal inference and optimal transport to generative modeling and representation learning. In this review, we bridge the gap between classical metric learning and Riemannian geometry, providing a structured and accessible overview of key methods, applications, and recent advances. We argue that Riemannian metric learning is not merely a technical refinement but a fundamental shift in how we think about data representations. Thus, this review should serve as a valuable resource for researchers and practitioners interested in exploring Riemannian metric learning and convince them that it is closer to them than they might imagine-both in theory and in practice.
Authors: Zhenxin Zheng, Zhenjie Zheng
Abstract: Curse of Dimensionality is an unavoidable challenge in statistical probability models, yet diffusion models seem to overcome this limitation, achieving impressive results in high-dimensional data generation. Diffusion models assume that they can learn the statistical quantities of the underlying probability distribution, enabling sampling from this distribution to generate realistic samples. But is this really how they work? We argue not, based on the following observations: 1) In high-dimensional sparse scenarios, the fitting target of the diffusion model's objective function degrades from a weighted sum of multiple samples to a single sample, which we believe hinders the model's ability to effectively learn essential statistical quantities such as posterior, score, or velocity field. 2) Most inference methods can be unified within a simple framework which involves no statistical concepts, aligns with the degraded objective function, and provides an novel and intuitive perspective on the inference process.
Authors: Taylor Sorensen, Pushkar Mishra, Roma Patel, Michael Henry Tessler, Michiel Bakker, Georgina Evans, Iason Gabriel, Noah Goodman, Verena Rieser
Abstract: Modelling human variation in rating tasks is crucial for personalization, pluralistic model alignment, and computational social science. We propose representing individuals using natural language value profiles -- descriptions of underlying values compressed from in-context demonstrations -- along with a steerable decoder model that estimates individual ratings from a rater representation. To measure the predictive information in a rater representation, we introduce an information-theoretic methodology and find that demonstrations contain the most information, followed by value profiles, then demographics. However, value profiles effectively compress the useful information from demonstrations (>70% information preservation) and offer advantages in terms of scrutability, interpretability, and steerability. Furthermore, clustering value profiles to identify similarly behaving individuals better explains rater variation than the most predictive demographic groupings. Going beyond test set performance, we show that the decoder predictions change in line with semantic profile differences, are well-calibrated, and can help explain instance-level disagreement by simulating an annotator population. These results demonstrate that value profiles offer novel, predictive ways to describe individual variation beyond demographics or group information.
Authors: Arturo De Marinis, Davide Murari, Elena Celledoni, Nicola Guglielmi, Brynjulf Owren, Francesco Tudisco
Abstract: We study the approximation properties of shallow neural networks whose activation function is defined as the flow map of a neural ordinary differential equation (neural ODE) at the final time of the integration interval. We prove the universal approximation property (UAP) of such shallow neural networks in the space of continuous functions. Furthermore, we investigate the approximation properties of shallow neural networks whose parameters satisfy specific constraints. In particular, we constrain the Lipschitz constant of the neural ODE's flow map and the norms of the weights to increase the network's stability. We prove that the UAP holds if we consider either constraint independently. When both are enforced, there is a loss of expressiveness, and we derive approximation bounds that quantify how accurately such a constrained network can approximate a continuous function.
Authors: Paul Biberstein, Ziyang Li, Joseph Devietti, Mayur Naik
Abstract: Neurosymbolic programs combine deep learning with symbolic reasoning to achieve better data efficiency, interpretability, and generalizability compared to standalone deep learning approaches. However, existing neurosymbolic learning frameworks implement an uneasy marriage between a highly scalable, GPU-accelerated neural component and a slower symbolic component that runs on CPUs. We propose Lobster, a unified framework for harnessing GPUs in an end-to-end manner for neurosymbolic learning. Lobster maps a general neurosymbolic language based on Datalog to the GPU programming paradigm. This mapping is implemented via compilation to a new intermediate language called APM. The extra abstraction provided by apm allows Lobster to be both flexible, supporting discrete, probabilistic, and differentiable modes of reasoning on GPU hardware with a library of provenance semirings, and performant, implementing new optimization passes. We demonstrate that Lobster programs can solve interesting problems spanning the domains of natural language processing, image processing, program reasoning, bioinformatics, and planning. On a suite of 9 applications, Lobster achieves an average speedup of 3.9x over Scallop, a state-of-the-art neurosymbolic framework, and enables scaling of neurosymbolic solutions to previously infeasible tasks.
Authors: Chung-En Sun, Ge Yan, Tsui-Wei Weng
Abstract: Recent studies have shown that Large Language Models (LLMs) augmented with chain-of-thought (CoT) reasoning demonstrate impressive problem-solving abilities. However, in this work, we identify a recurring issue where these models occasionally generate overly short reasoning, leading to degraded performance on even simple mathematical problems. Specifically, we investigate how reasoning length is embedded in the hidden representations of reasoning models and its impact on accuracy. Our analysis reveals that reasoning length is governed by a linear direction in the representation space, allowing us to induce overly short reasoning by steering the model along this direction. Building on this insight, we introduce ThinkEdit, a simple yet effective weight-editing approach to mitigate the issue of overly short reasoning. We first identify a small subset of attention heads (approximately 4%) that predominantly drive short reasoning behavior. We then edit the output projection weights of these heads to remove the short reasoning direction. With changes to only 0.2% of the model's parameters, ThinkEdit effectively reduces overly short reasoning and yields notable accuracy gains for short reasoning outputs (+6.39%), along with an overall improvement across multiple math benchmarks (+3.34%). Our findings provide new mechanistic insights into how reasoning length is controlled within LLMs and highlight the potential of fine-grained model interventions to improve reasoning quality. Our code is available at: https://github.com/Trustworthy-ML-Lab/ThinkEdit
Authors: Yibo Liu, Zhixin Fang, Sune Darkner, Noam Aigerman, Kenny Erleben, Paul Kry, Teseo Schneider
Abstract: We propose mesh-free fluid simulations that exploit a kinematic neural basis for velocity fields represented by an MLP. We design a set of losses that ensures that these neural bases approximate fundamental physical properties such as orthogonality, divergence-free, boundary alignment, and smoothness. Our neural bases can then be used to fit an input sketch of a flow, which will inherit the same fundamental properties from the bases. We then can animate such flow in real-time using standard time integrators. Our neural bases can accommodate different domains, moving boundaries, and naturally extend to three dimensions.
Authors: Roman Garipov, Fedor Velikonivtsev, Ivan Ermakov, Ruslan Svirschevski, Vage Egiazarian, Max Ryabinin
Abstract: We introduce AutoJudge, a method that accelerates large language model (LLM) inference with task-specific lossy speculative decoding. Instead of matching the original model output distribution token-by-token, we identify which of the generated tokens affect the downstream quality of the response, relaxing the distribution match guarantee so that the "unimportant" tokens can be generated faster. Our approach relies on a semi-greedy search algorithm to test which of the mismatches between target and draft models should be corrected to preserve quality and which ones may be skipped. We then train a lightweight classifier based on existing LLM embeddings to predict, at inference time, which mismatching tokens can be safely accepted without compromising the final answer quality. We evaluate the effectiveness of AutoJudge with multiple draft/target model pairs on mathematical reasoning and programming benchmarks, achieving significant speedups at the cost of a minor accuracy reduction. Notably, on GSM8k with the Llama 3.1 70B target model, our approach achieves up to $\approx2\times$ speedup over speculative decoding at the cost of $\le 1\%$ drop in accuracy. When applied to the LiveCodeBench benchmark, AutoJudge automatically detects programming-specific important tokens, accepting $\ge 25$ tokens per speculation cycle at $2\%$ drop in Pass@1. Our approach requires no human annotation and is easy to integrate with modern LLM inference frameworks.
Authors: Nicolas Coloma, William Kleiber
Abstract: L\'evy processes are widely used in financial modeling due to their ability to capture discontinuities and heavy tails, which are common in high-frequency asset return data. However, parameter estimation remains a challenge when associated likelihoods are unavailable or costly to compute. We propose a fast and accurate method for L\'evy parameter estimation using the neural Bayes estimation (NBE) framework -- a simulation-based, likelihood-free approach that leverages permutation-invariant neural networks to approximate Bayes estimators. We contribute new theoretical results, showing that NBE results in consistent estimators whose risk converges to the Bayes estimator under mild conditions. Moreover, through extensive simulations across several L\'evy models, we show that NBE outperforms traditional methods in both accuracy and runtime, while also enabling two complementary approaches to uncertainty quantification. We illustrate our approach on a challenging high-frequency cryptocurrency return dataset, where the method captures evolving parameter dynamics and delivers reliable and interpretable inference at a fraction of the computational cost of traditional methods. NBE provides a scalable and practical solution for inference in complex financial models, enabling parameter estimation and uncertainty quantification over an entire year of data in just seconds. We additionally investigate nearly a decade of high-frequency Bitcoin returns, requiring less than one minute to estimate parameters under the proposed approach.
Authors: Simon Ging, Sebastian Walter, Jelena Bratuli\'c, Johannes Dienert, Hannah Bast, Thomas Brox
Abstract: Training high-quality CLIP models typically requires enormous datasets, which limits the development of domain-specific models -- especially in areas that even the largest CLIP models do not cover well -- and drives up training costs. This poses challenges for scientific research that needs fine-grained control over the training procedure of CLIP models. In this work, we show that by employing smart web search strategies enhanced with knowledge graphs, a robust CLIP model can be trained from scratch with considerably less data. Specifically, we demonstrate that an expert foundation model for living organisms can be built using just 10M images. Moreover, we introduce EntityNet, a dataset comprising 33M images paired with 46M text descriptions, which enables the training of a generic CLIP model in significantly reduced time.
Authors: Tim Schneider, Cristiana de Farias, Roberto Calandra, Liming Chen, Jan Peters
Abstract: Active perception is a fundamental skill that enables us humans to deal with uncertainty in our inherently partially observable environment. For senses such as touch, where the information is sparse and local, active perception becomes crucial. In recent years, active perception has emerged as an important research domain in robotics. However, current methods are often bound to specific tasks or make strong assumptions, which limit their generality. To address this gap, this work introduces APPLE (Active Perception Policy Learning) - a novel framework that leverages reinforcement learning (RL) to address a range of different active perception problems. APPLE jointly trains a transformer-based perception module and decision-making policy with a unified optimization objective, learning how to actively gather information. By design, APPLE is not limited to a specific task and can, in principle, be applied to a wide range of active perception problems. We evaluate two variants of APPLE across different tasks, including tactile exploration problems from the Tactile MNIST benchmark. Experiments demonstrate the efficacy of APPLE, achieving high accuracies on both regression and classification tasks. These findings underscore the potential of APPLE as a versatile and general framework for advancing active perception in robotics.
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 math 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: Akash Kundu, Stefano Mangini
Abstract: Variational quantum algorithms hold the promise to address meaningful quantum problems already on noisy intermediate-scale quantum hardware. In spite of the promise, they face the challenge of designing quantum circuits that both solve the target problem and comply with device limitations. Quantum architecture search (QAS) automates the design process of quantum circuits, with reinforcement learning (RL) emerging as a promising approach. Yet, RL-based QAS methods encounter significant scalability issues, as computational and training costs grow rapidly with the number of qubits, circuit depth, and hardware noise. To address these challenges, we introduce $\textit{TensorRL-QAS}$, an improved framework that combines tensor network methods with RL for QAS. By warm-starting the QAS with a matrix product state approximation of the target solution, TensorRL-QAS effectively narrows the search space to physically meaningful circuits and accelerates the convergence to the desired solution. Tested on several quantum chemistry problems of up to 12-qubit, TensorRL-QAS achieves up to a 10-fold reduction in CNOT count and circuit depth compared to baseline methods, while maintaining or surpassing chemical accuracy. It reduces classical optimizer function evaluation by up to 100-fold, accelerates training episodes by up to 98$\%$, and can achieve 50$\%$ success probability for 10-qubit systems, far exceeding the $<$1$\%$ rates of baseline. Robustness and versatility are demonstrated both in the noiseless and noisy scenarios, where we report a simulation of an 8-qubit system. Furthermore, TensorRL-QAS demonstrates effectiveness on systems on 20-qubit quantum systems, positioning it as a state-of-the-art quantum circuit discovery framework for near-term hardware and beyond.
Authors: Matthias K\"ummerer, Harneet Singh Khanuja, Matthias Bethge
Abstract: Recent advances in image-based saliency prediction are approaching gold standard performance levels on existing benchmarks. Despite this success, we show that predicting fixations across multiple saliency datasets remains challenging due to dataset bias. We find a significant performance drop (around 40%) when models trained on one dataset are applied to another. Surprisingly, increasing dataset diversity does not resolve this inter-dataset gap, with close to 60% attributed to dataset-specific biases. To address this remaining generalization gap, we propose a novel architecture extending a mostly dataset-agnostic encoder-decoder structure with fewer than 20 dataset-specific parameters that govern interpretable mechanisms such as multi-scale structure, center bias, and fixation spread. Adapting only these parameters to new data accounts for more than 75% of the generalization gap, with a large fraction of the improvement achieved with as few as 50 samples. Our model sets a new state-of-the-art on all three datasets of the MIT/Tuebingen Saliency Benchmark (MIT300, CAT2000, and COCO-Freeview), even when purely generalizing from unrelated datasets, but with a substantial boost when adapting to the respective training datasets. The model also provides valuable insights into spatial saliency properties, revealing complex multi-scale effects that combine both absolute and relative sizes.
Authors: Ping Xu, Zhiyuan Ning, Pengjiang Li, Wenhao Liu, Pengyang Wang, Jiaxu Cui, Yuanchun Zhou, Pengfei Wang
Abstract: Single-cell RNA sequencing (scRNA-seq) reveals cell heterogeneity, with cell clustering playing a key role in identifying cell types and marker genes. Recent advances, especially graph neural networks (GNNs)-based methods, have significantly improved clustering performance. However, the analysis of scRNA-seq data remains challenging due to noise, sparsity, and high dimensionality. Compounding these challenges, GNNs often suffer from over-smoothing, limiting their ability to capture complex biological information. In response, we propose scSiameseClu, a novel Siamese Clustering framework for interpreting single-cell RNA-seq data, comprising of 3 key steps: (1) Dual Augmentation Module, which applies biologically informed perturbations to the gene expression matrix and cell graph relationships to enhance representation robustness; (2) Siamese Fusion Module, which combines cross-correlation refinement and adaptive information fusion to capture complex cellular relationships while mitigating over-smoothing; and (3) Optimal Transport Clustering, which utilizes Sinkhorn distance to efficiently align cluster assignments with predefined proportions while maintaining balance. Comprehensive evaluations on seven real-world datasets demonstrate that scSiameseClu outperforms state-of-the-art methods in single-cell clustering, cell type annotation, and cell type classification, providing a powerful tool for scRNA-seq data interpretation.
Authors: Jiahao Yu, Haozhuang Liu, Yeqiu Yang, Lu Chen, Jian Wu, Yuning Jiang, Bo Zheng
Abstract: Regression models are crucial in recommender systems. However, retransformation bias problem has been conspicuously neglected within the community. While many works in other fields have devised effective bias correction methods, all of them are post-hoc cures externally to the model, facing practical challenges when applied to real-world recommender systems. Hence, we propose a preemptive paradigm to eradicate the bias intrinsically from the models via minor model refinement. Specifically, a novel TranSUN method is proposed with a joint bias learning manner to offer theoretically guaranteed unbiasedness under empirical superior convergence. It is further generalized into a novel generic regression model family, termed Generalized TranSUN (GTS), which not only offers more theoretical insights but also serves as a generic framework for flexibly developing various bias-free models. Comprehensive experimental results demonstrate the superiority of our methods across data from various domains, which have been successfully deployed in two real-world industrial recommendation scenarios, i.e. product and short video recommendation scenarios in Guess What You Like business domain in the homepage of Taobao App (a leading e-commerce platform with DAU > 300M), to serve the major online traffic.
Authors: David Nordstr\"om, Johan Edstedt, Fredrik Kahl, Georg B\"okman
Abstract: Why are state-of-the-art Vision Transformers (ViTs) not designed to exploit natural geometric symmetries such as 90-degree rotations and reflections? In this paper, we argue that there is no fundamental reason, and what has been missing is an efficient implementation. To this end, we introduce Octic Vision Transformers (octic ViTs) which rely on octic group equivariance to capture these symmetries. In contrast to prior equivariant models that increase computational cost, our octic linear layers achieve 5.33x reductions in FLOPs and up to 8x reductions in memory compared to ordinary linear layers. In full octic ViT blocks the computational reductions approach the reductions in the linear layers with increased embedding dimension. We study two new families of ViTs, built from octic blocks, that are either fully octic equivariant or break equivariance in the last part of the network. Training octic ViTs supervised (DeiT-III) and unsupervised (DINOv2) on ImageNet-1K, we find that they match baseline accuracy while at the same time providing substantial efficiency gains.
Authors: Gaurav Srivastava, Zhenyu Bi, Meng Lu, Xuan Wang
Abstract: Large language models (LLMs) have improved significantly in their reasoning through extensive training on massive datasets. However, relying solely on additional data for improvement is becoming increasingly impractical, highlighting the need for models to autonomously enhance their reasoning without external supervision. In this paper, we propose Debate, Train, Evolve (DTE), a novel ground truth-free training framework that uses multi-agent debate traces to evolve a single language model. We also introduce a new prompting strategy Reflect-Critique-Refine, to improve debate quality by explicitly instructing agents to critique and refine their reasoning. Extensive evaluations on seven reasoning benchmarks with six open-weight models show that our DTE framework achieve substantial improvements, with an average accuracy gain of 8.92% on the challenging GSM-PLUS dataset. Furthermore, we observe strong cross-domain generalization, with an average accuracy gain of 5.8% on all other benchmarks, suggesting that our method captures general reasoning capabilities. Our framework code and trained models are publicly available at https://github.com/ctrl-gaurav/Debate-Train-Evolve
Authors: Santiago Acevedo, Andrea Mascaretti, Riccardo Rende, Mat\'eo Mahaut, Marco Baroni, Alessandro Laio
Abstract: Deep neural networks are known to develop similar representations for semantically related data, even when they belong to different domains, such as an image and its description, or the same text in different languages. We present a method for quantitatively investigating this phenomenon by measuring the relative information content of the representations of semantically related data and probing how it is encoded into multiple tokens of large language models (LLMs) and vision transformers. Looking first at how LLMs process pairs of translated sentences, we identify inner ``semantic'' layers containing the most language-transferable information. We find moreover that, on these layers, a larger LLM (DeepSeek-V3) extracts significantly more general information than a smaller one (Llama3.1-8B). Semantic information of English text is spread across many tokens and it is characterized by long-distance correlations between tokens and by a causal left-to-right (i.e., past-future) asymmetry. We also identify layers encoding semantic information within visual transformers. We show that caption representations in the semantic layers of LLMs predict visual representations of the corresponding images. We observe significant and model-dependent information asymmetries between image and text representations.
Authors: Joakim Edin, R\'obert Csord\'as, Tuukka Ruotsalo, Zhengxuan Wu, Maria Maistro, Jing Huang, Lars Maal{\o}e
Abstract: Ensuring faithful interpretability in large language models is imperative for trustworthy and reliable AI. A key obstacle is self-repair, a phenomenon where networks compensate for reduced signal in one component by amplifying others, masking the true importance of the ablated component. While prior work attributes self-repair to layer normalization and back-up components that compensate for ablated components, we identify a novel form occurring within the attention mechanism, where softmax redistribution conceals the influence of important attention scores. This leads traditional ablation and gradient-based methods to underestimate the significance of all components contributing to these attention scores. We introduce Gradient Interaction Modifications (GIM), a technique that accounts for self-repair during backpropagation. Extensive experiments across multiple large language models (Gemma 2B/9B, LLAMA 1B/3B/8B, Qwen 1.5B/3B) and diverse tasks demonstrate that GIM significantly improves faithfulness over existing circuit identification and feature attribution methods. Our work is a significant step toward better understanding the inner mechanisms of LLMs, which is crucial for improving them and ensuring their safety. Our code is available at https://github.com/JoakimEdin/gim.
Authors: Nikhil Singh, Manuel Cherep, Pattie Maes
Abstract: The fidelity with which neural networks can now generate content such as music presents a scientific opportunity: these systems appear to have learned implicit theories of such content's structure through statistical learning alone. This offers a potentially new lens on theories of human-generated media. When internal representations align with traditional constructs (e.g. chord progressions in music), they show how such categories can emerge from statistical regularities; when they diverge, they expose limits of existing frameworks and patterns we may have overlooked but that nonetheless carry explanatory power. In this paper, focusing on music generators, we introduce a method for discovering interpretable concepts using sparse autoencoders (SAEs), extracting interpretable features from the residual stream of a transformer model. We make this approach scalable and evaluable using automated labeling and validation pipelines. Our results reveal both familiar musical concepts and coherent but uncodified patterns lacking clear counterparts in theory or language. As an extension, we show such concepts can be used to steer model generations. Beyond improving model transparency, our work provides an empirical tool for uncovering organizing principles that have eluded traditional methods of analysis and synthesis.
Authors: Michael Kirchhof, Luca F\"uger, Adam Goli\'nski, Eeshan Gunesh Dhekane, Arno Blaas, Seong Joon Oh, Sinead Williamson
Abstract: The common approach to communicate a large language model's (LLM) uncertainty is to add a percentage number or a hedging word to its response. But is this all we can do? Instead of generating a single answer and then hedging it, an LLM that is fully transparent to the user needs to be able to reflect on its internal belief distribution and output a summary of all options it deems possible, and how likely they are. To test whether LLMs possess this capability, we develop the SelfReflect metric, an information-theoretic distance between a given summary and a distribution over answers. In interventional and human studies, we find that SelfReflect indicates even slight deviations, yielding a fine measure of faithfulness between a summary string and an LLM's actual internal distribution over answers. With SelfReflect, we make a resounding negative observation: modern LLMs are, across the board, incapable of revealing what they are uncertain about, neither through reasoning, nor chains-of-thoughts, nor explicit finetuning. However, we do find that LLMs are able to generate faithful summaries of their uncertainties if we help them by sampling multiple outputs and feeding them back into the context. This simple approach shines a light at the universal way of communicating LLM uncertainties whose future development the SelfReflect score enables.
Authors: Michele Gallo
Abstract: Tensors provide a robust framework for managing high-dimensional data. Consequently, tensor analysis has emerged as an active research area in various domains, including machine learning, signal processing, computer vision, graph analysis, and data mining. This study introduces an efficient image storage approach utilizing tensors, aiming to minimize memory to store, bandwidth to transmit and energy to processing. The proposed method organizes original data into a higher-order tensor and applies the Tucker model for compression. Implemented in R, this method is compared to a baseline algorithm. The evaluation focuses on efficient of algorithm measured in term of computational time and the quality of information preserved, using both simulated and real datasets. A detailed analysis of the results is conducted, employing established quantitative metrics, with significant attention paid to sustainability in terms of energy consumption across algorithms.
Authors: Agnideep Aich, Md Monzur Murshed, Amanda Mayeaux, Sameera Hewage
Abstract: Effective feature selection is critical for building robust and interpretable predictive models, particularly in medical applications where identifying risk factors in the most extreme patient strata is essential. Traditional methods often focus on average associations, potentially overlooking predictors whose importance is concentrated in the tails of the data distribution. In this study, we introduce a novel, computationally efficient supervised filter method that leverages the Gumbel copula's upper-tail dependence coefficient to rank features based on their tendency to be simultaneously extreme with a positive outcome. We conducted a rigorous evaluation of this method against four standard baselines (Mutual Information, mRMR, ReliefF, and L1/Elastic-Net) using four distinct classifiers on two diabetes datasets: a large-scale public health survey (CDC, N=253,680) and a classic clinical benchmark (PIMA, N=768). Our analysis included comprehensive statistical tests, permutation importance, and robustness checks. On the CDC dataset, our method was the fastest selector and reduced the feature space by approximately 52% while maintaining predictive performance statistically indistinguishable from a model using all features. On the PIMA dataset, our method's feature ranking yielded the single best-performing model, achieving the highest ROC-AUC of all tested configurations. Across both datasets, the Gumbel-upper-tail dependence coefficient selector consistently identified clinically coherent and impactful predictors. We conclude that feature selection via upper-tail dependence is a powerful, efficient, and interpretable new tool for developing risk models in public health and clinical medicine.
Authors: Jang-Hyun Kim, Jinuk Kim, Sangwoo Kwon, Jae W. Lee, Sangdoo Yun, Hyun Oh Song
Abstract: Transformer-based large language models (LLMs) cache context as key-value (KV) pairs during inference. As context length grows, KV cache sizes expand, leading to substantial memory overhead and increased attention latency. This paper introduces KVzip, a query-agnostic KV cache eviction method enabling effective reuse of compressed KV caches across diverse queries. KVzip quantifies the importance of a KV pair using the underlying LLM to reconstruct original contexts from cached KV pairs, subsequently evicting pairs with lower importance. Extensive empirical evaluations demonstrate that KVzip reduces KV cache size by $3$-$4\times$ and FlashAttention decoding latency by approximately $2\times$, with negligible performance loss in question-answering, retrieval, reasoning, and code comprehension tasks. Evaluations include various models such as LLaMA3.1, Qwen2.5, and Gemma3, with context lengths reaching up to 170K tokens. KVzip significantly outperforms existing query-aware KV eviction methods, which suffer from performance degradation even at a 90% cache budget ratio under multi-query scenarios.
Authors: Luke McDermott, Robert W. Heath Jr., Rahul Parhi
Abstract: The per-token cost of transformer inference scales with context length, preventing its application to lifelong in-context learning. Linear attention is an efficient alternative that maintains a constant memory footprint, even on infinite context lengths. While this is a potential candidate for lifelong learning, it falls short in memory capacity. In this paper, we propose LoLA, a training-free augmentation to linear attention that boosts associative recall. LoLA distributes past key-value pairs from context into three memory systems: (i) recent pairs in a local sliding window cache; (ii) difficult-to-memorize pairs in a sparse, global cache; and (iii) generic pairs in the recurrent hidden state of linear attention. We show through ablations that our self-recall error metric is crucial to efficiently manage long-term associative memories. On pass-key retrieval tasks, LoLA improves the base model's performance from 0.6% to 97.4% accuracy. This is achieved with a 4.6x smaller cache than Llama-3.1 8B on 4K context length. LoLA also outperforms other 1B and 8B parameter subquadratic models on zero-shot commonsense reasoning tasks.
Authors: Takashi Wada, Yuki Hirakawa, Ryotaro Shimizu, Takahiro Kawashima, Yuki Saito
Abstract: We propose new static word embeddings optimised for sentence semantic representation. We first extract word embeddings from a pre-trained Sentence Transformer, and improve them with sentence-level principal component analysis, followed by either knowledge distillation or contrastive learning. During inference, we represent sentences by simply averaging word embeddings, which requires little computational cost. We evaluate models on both monolingual and cross-lingual tasks and show that our model substantially outperforms existing static models on sentence semantic tasks, and even surpasses a basic Sentence Transformer model (SimCSE) on a text embedding benchmark. Lastly, we perform a variety of analyses and show that our method successfully removes word embedding components that are not highly relevant to sentence semantics, and adjusts the vector norms based on the influence of words on sentence semantics.
Authors: Daniel Wang, Patrick Rim, Tian Tian, Dong Lao, Alex Wong, Ganesh Sundaramoorthi
Abstract: We introduce ODE-GS, a novel approach that integrates 3D Gaussian Splatting with latent neural ordinary differential equations (ODEs) to enable future extrapolation of dynamic 3D scenes. Unlike existing dynamic scene reconstruction methods, which rely on time-conditioned deformation networks and are limited to interpolation within a fixed time window, ODE-GS eliminates timestamp dependency by modeling Gaussian parameter trajectories as continuous-time latent dynamics. Our approach first learns an interpolation model to generate accurate Gaussian trajectories within the observed window, then trains a Transformer encoder to aggregate past trajectories into a latent state evolved via a neural ODE. Finally, numerical integration produces smooth, physically plausible future Gaussian trajectories, enabling rendering at arbitrary future timestamps. On the D-NeRF, NVFi, and HyperNeRF benchmarks, ODE-GS achieves state-of-the-art extrapolation performance, improving metrics by 19.8% compared to leading baselines, demonstrating its ability to accurately represent and predict 3D scene dynamics.
Authors: Peter Vieting, Maximilian Kannen, Benedikt Hilmes, Ralf Schl\"uter, Hermann Ney
Abstract: Neural front-ends are an appealing alternative to traditional, fixed feature extraction pipelines for automatic speech recognition (ASR) systems since they can be directly trained to fit the acoustic model. However, their performance often falls short compared to classical methods, which we show is largely due to their increased susceptibility to overfitting. This work therefore investigates regularization methods for training ASR models with learnable feature extraction front-ends. First, we examine audio perturbation methods and show that larger relative improvements can be obtained for learnable features. Additionally, we identify two limitations in the standard use of SpecAugment for these front-ends and propose masking in the short time Fourier transform (STFT)-domain as a simple but effective modification to address these challenges. Finally, integrating both regularization approaches effectively closes the performance gap between traditional and learnable features.
Authors: Xiyuan Zhang, Boran Han, Haoyang Fang, Abdul Fatir Ansari, Shuai Zhang, Danielle C. Maddix, Cuixiong Hu, Andrew Gordon Wilson, Michael W. Mahoney, Hao Wang, Yan Liu, Huzefa Rangwala, George Karypis, Bernie Wang
Abstract: Recently, there has been growing interest in incorporating textual information into foundation models for time series forecasting. However, it remains unclear whether and under what conditions such multimodal integration consistently yields gains. We systematically investigate these questions across a diverse benchmark of 16 forecasting tasks spanning 7 domains, including health, environment, and economics. We evaluate two popular multimodal forecasting paradigms: aligning-based methods, which align time series and text representations; and prompting-based methods, which directly prompt large language models for forecasting. Our findings reveal that the benefits of multimodality are highly condition-dependent. While we confirm reported gains in some settings, these improvements are not universal across datasets or models. To move beyond empirical observations, we disentangle the effects of model architectural properties and data characteristics, drawing data-agnostic insights that generalize across domains. Our findings highlight that on the modeling side, incorporating text information is most helpful given (1) high-capacity text models, (2) comparatively weaker time series models, and (3) appropriate aligning strategies. On the data side, performance gains are more likely when (4) sufficient training data is available and (5) the text offers complementary predictive signal beyond what is already captured from the time series alone. Our study offers a rigorous, quantitative foundation for understanding when multimodality can be expected to aid forecasting tasks, and reveals that its benefits are neither universal nor always aligned with intuition.
Authors: Tianrong Chen, Huangjie Zheng, David Berthelot, Jiatao Gu, Josh Susskind, Shuangfei Zhai
Abstract: Diffusion models have demonstrated exceptional capabilities in generating high-fidelity images but typically suffer from inefficient sampling. Many solver designs and noise scheduling strategies have been proposed to dramatically improve sampling speeds. In this paper, we introduce a new sampling method that is up to $186\%$ faster than the current state of the art solver for comparative FID on ImageNet512. This new sampling method is training-free and uses an ordinary differential equation (ODE) solver. The key to our method resides in using higher-dimensional initial noise, allowing to produce more detailed samples with less function evaluations from existing pretrained diffusion models. In addition, by design our solver allows to control the level of detail through a simple hyper-parameter at no extra computational cost. We present how our approach leverages momentum dynamics by establishing a fundamental equivalence between momentum diffusion models and conventional diffusion models with respect to their training paradigms. Moreover, we observe the use of higher-dimensional noise naturally exhibits characteristics similar to stochastic differential equations (SDEs). Finally, we demonstrate strong performances on a set of representative pretrained diffusion models, including EDM, EDM2, and Stable-Diffusion 3, which cover models in both pixel and latent spaces, as well as class and text conditional settings. The code is available at https://github.com/apple/ml-tada.
Authors: Po-Heng Chou, Ching-Wen Chen, Wan-Jen Huang, Walid Saad, Yu Tsao, Ronald Y. Chang
Abstract: In this paper, the precoding design is investigated for maximizing the throughput of millimeter wave (mmWave) multiple-input multiple-output (MIMO) systems with obstructed direct communication paths. In particular, a reconfigurable intelligent surface (RIS) is employed to enhance MIMO transmissions, considering mmWave characteristics related to line-of-sight (LoS) and multipath effects. The traditional exhaustive search (ES) for optimal codewords in the continuous phase shift is computationally intensive and time-consuming. To reduce computational complexity, permuted discrete Fourier transform (DFT) vectors are used for finding codebook design, incorporating amplitude responses for practical or ideal RIS systems. However, even if the discrete phase shift is adopted in the ES, it results in significant computation and is time-consuming. Instead, the trained deep neural network (DNN) is developed to facilitate faster codeword selection. Simulation results show that the DNN maintains sub-optimal spectral efficiency even as the distance between the end-user and the RIS has variations in the testing phase. These results highlight the potential of DNN in advancing RIS-aided systems.
Authors: Shion Takeno, Yu Inatsu, Masayuki Karasuyama, Ichiro Takeuchi
Abstract: Bayesian optimization is a powerful tool for optimizing an expensive-to-evaluate black-box function. In particular, the effectiveness of expected improvement (EI) has been demonstrated in a wide range of applications. However, theoretical analyses of EI are limited compared with other theoretically established algorithms. This paper analyzes a randomized variant of EI, which evaluates the EI from the maximum of the posterior sample path. We show that this posterior sampling-based random EI achieves the sublinear Bayesian cumulative regret bounds under the assumption that the black-box function follows a Gaussian process. Finally, we demonstrate the effectiveness of the proposed method through numerical experiments.
Authors: Mohammed Alkhowaiter, Norah Alshahrani, Saied Alshahrani, Reem I. Masoud, Alaa Alzahrani, Deema Alnuhait, Emad A. Alghamdi, Khalid Almubarak
Abstract: Post-training has emerged as a crucial technique for aligning pre-trained Large Language Models (LLMs) with human instructions, significantly enhancing their performance across a wide range of tasks. Central to this process is the quality and diversity of post-training datasets. This paper presents a review of publicly available Arabic post-training datasets on the Hugging Face Hub, organized along four key dimensions: (1) LLM Capabilities (e.g., Question Answering, Translation, Reasoning, Summarization, Dialogue, Code Generation, and Function Calling); (2) Steerability (e.g., Persona and System Prompts); (3) Alignment (e.g., Cultural, Safety, Ethics, and Fairness); and (4) Robustness. Each dataset is rigorously evaluated based on popularity, practical adoption, recency and maintenance, documentation and annotation quality, licensing transparency, and scientific contribution. Our review revealed critical gaps in the development of Arabic post-training datasets, including limited task diversity, inconsistent or missing documentation and annotation, and low adoption across the community. Finally, the paper discusses the implications of these gaps on the progress of Arabic-centric LLMs and applications while providing concrete recommendations for future efforts in Arabic post-training dataset development.
Authors: Yihao Li, Jiayi Xin, Miranda Muqing Miao, Qi Long, Lyle Ungar
Abstract: Proficient multilingual speakers often intentionally switch languages in the middle of a conversation. Similarly, recent reasoning-focused bilingual large language models (LLMs) with strong capabilities in both languages exhibit language mixing-alternating languages within their chain of thought. Discouraging this behavior in DeepSeek-R1 was found to degrade accuracy, suggesting that language mixing may benefit reasoning. In this work, we study language switching in Chinese-English bilingual reasoning models. We identify reinforcement learning with verifiable rewards (RLVR) as the critical training stage that leads to language mixing. We show that language mixing can enhance reasoning: enforcing monolingual decoding reduces accuracy by 5.6 percentage points on MATH500. Additionally, a lightweight probe can be trained to predict whether a potential language switch would benefit or harm reasoning, and when used to guide decoding, increases accuracy by 2.92 percentage points. Our findings suggest that language mixing is not merely a byproduct of multilingual training, but is a strategic reasoning behavior.
Authors: Robert Kuchen
Abstract: Statistical learning methods for automated variable selection, such as LASSO, elastic nets, or gradient boosting, have become increasingly popular tools for building powerful prediction models. Yet, in practice, analyses are often complicated by missing data. The most widely used approach to address missingness is multiple imputation, which involves creating several completed datasets. However, there is an ongoing debate on how to perform model selection in the presence of multiple imputed datasets. Simple strategies, such as pooling models across datasets, have been shown to have suboptimal properties. Although more sophisticated methods exist, they are often difficult to implement and therefore not widely applied. In contrast, two recent approaches modify the regularization methods LASSO and elastic nets by defining a single loss function, resulting in a unified set of coefficients across imputations. Our key contribution is to extend this principle to the framework of component-wise gradient boosting by proposing MIBoost, a novel algorithm that employs a uniform variable-selection mechanism across imputed datasets. Simulation studies suggest that our approach yields prediction performance comparable to that of these recently proposed methods.
Authors: Yihong Dong, Xue Jiang, Jiaru Qian, Tian Wang, Kechi Zhang, Zhi Jin, Ge Li
Abstract: Code generation agents powered by large language models (LLMs) are revolutionizing the software development paradigm. Distinct from previous code generation techniques, code generation agents are characterized by three core features. 1) Autonomy: the ability to independently manage the entire workflow, from task decomposition to coding and debugging. 2) Expanded task scope: capabilities that extend beyond generating code snippets to encompass the full software development lifecycle (SDLC). 3) Enhancement of engineering practicality: a shift in research emphasis from algorithmic innovation toward practical engineering challenges, such as system reliability, process management, and tool integration. This domain has recently witnessed rapid development and an explosion in research, demonstrating significant application potential. This paper presents a systematic survey of the field of LLM-based code generation agents. We trace the technology's developmental trajectory from its inception and systematically categorize its core techniques, including both single-agent and multi-agent architectures. Furthermore, this survey details the applications of LLM-based agents across the full SDLC, summarizes mainstream evaluation benchmarks and metrics, and catalogs representative tools. Finally, by analyzing the primary challenges, we identify and propose several foundational, long-term research directions for the future work of the field.
Authors: Samuel R\"aber, Till Aczel, Andreas Plesner, Roger Wattenhofer
Abstract: Previous work has suggested that preprocessing images through lossy compression can defend against adversarial perturbations, but comprehensive attack evaluations have been lacking. In this paper, we construct strong white-box and adaptive attacks against various compression models and identify a critical challenge for attackers: high realism in reconstructed images significantly increases attack difficulty. Through rigorous evaluation across multiple attack scenarios, we demonstrate that compression models capable of producing realistic, high-fidelity reconstructions are substantially more resistant to our attacks. In contrast, low-realism compression models can be broken. Our analysis reveals that this is not due to gradient masking. Rather, realistic reconstructions maintaining distributional alignment with natural images seem to offer inherent robustness. This work highlights a significant obstacle for future adversarial attacks and suggests that developing more effective techniques to overcome realism represents an essential challenge for comprehensive security evaluation.
Authors: Kate A. Richardson, Sokratis Trifinopoulos, Mike Williams
Abstract: Obtaining high-precision predictions of nuclear masses, or equivalently nuclear binding energies, $E_b$, remains an important goal in nuclear-physics research. Recently, many AI-based tools have shown promising results on this task, some achieving precision that surpasses the best physics models. However, the utility of these AI models remains in question given that predictions are only useful where measurements do not exist, which inherently requires extrapolation away from the training (and testing) samples. Since AI models are largely black boxes, the reliability of such an extrapolation is difficult to assess. We present an AI model that not only achieves cutting-edge precision for $E_b$, but does so in an interpretable manner. For example, we find that (and explain why) the most important dimensions of its internal representation form a double helix, where the analog of the hydrogen bonds in DNA here link the number of protons and neutrons found in the most stable nucleus of each isotopic chain. Furthermore, we show that the AI prediction of $E_b$ can be factorized and ordered hierarchically, with the most important terms corresponding to well-known symbolic models (such as the famous liquid drop). Remarkably, the improvement of the AI model over symbolic ones can almost entirely be attributed to an observation made by Jaffe in 1969 based on the structure of most known nuclear ground states. The end result is a fully interpretable data-driven model of nuclear masses based on physics deduced by AI.
Authors: Wu Lin, Scott C. Lowe, Felix Dangel, Runa Eschenhagen, Zikun Xu, Roger B. Grosse
Abstract: Shampoo and its efficient variant, SOAP, use structured second-moment estimation and have attracted growing interest for their effectiveness in training neural networks (NNs). In practice, Shampoo requires step-size grafting with Adam to achieve competitive performance. SOAP mitigates this by applying Adam in Shampoo's eigenbasis and further reducing per-iteration runtime. However, reliance on Adam introduces additional memory overhead in both methods. Prior theoretical interpretations have primarily examined their estimation schemes using the Frobenius norm. Motivated by the natural correspondence between the second moment and a covariance matrix, we reinterpret the estimation procedures in Shampoo and SOAP as instances of covariance estimation through the lens of Kullback-Leibler (KL) divergence minimization. This perspective reveals a previously overlooked theoretical limitation and motivates principled improvements to their design. Building on the KL perspective, we propose practical estimation schemes -- $\textbf{KL-Shampoo}$ and $\textbf{KL-SOAP}$ -- that match or exceed the performance of Shampoo and SOAP for pre-training various NNs while maintaining SOAP-level per-iteration runtime. Notably, KL-Shampoo does not rely on Adam to achieve superior performance, thereby avoiding the associated memory overhead. Surprisingly, KL-Shampoo consistently outperforms the other methods in our experiments.
Authors: Medhasweta Sen, Zachary Gottesman, Jiaxing Qiu, C. Bayan Bruss, Nam Nguyen, Tom Hartvigsen
Abstract: Recent works propose complex multi-modal models that handle both time series and language, ultimately claiming high performance on complex tasks like time series reasoning and cross-modal question-answering. However, they skip evaluations of simple and important foundational tasks, which complex models should reliably master. They also lack direct, head-to-head comparisons with other popular approaches. So we ask a simple question: Can recent models even produce generic visual descriptions of time series data? In response, we propose three new tasks, posing that successful multi-modal models should be able to recognize, differentiate, and generate language descriptions of time series. We then create BEDTime, the first benchmark dataset to assess models on each task, comprising four datasets reformatted for these tasks across multiple modalities. Using BEDTime, we evaluate 13 state-of-the-art models, and find that (1) surprisingly, dedicated time series foundation models severely underperform, despite being designed for similar tasks, (2) vision-language models are quite capable, (3) language-only methods perform worst, despite many lauding their potential, and (4) all approaches are clearly fragile to a range of realistic robustness tests, indicating avenues for future work.
Authors: Quan Chen, Chenrui Shi, Qi Chen, Yuwei Wu, Zhi Gao, Xintong Zhang, Rui Gao, Kun Wu, Yunde Jia
Abstract: Learning from long-horizon demonstrations with complex action sequences presents significant challenges for visual imitation learning, particularly in understanding temporal relationships of actions and spatial relationships between objects. In this paper, we propose a new agent framework that incorporates two dedicated reflection modules to enhance both plan and code generation. The plan generation module produces an initial action sequence, which is then verified by the plan reflection module to ensure temporal coherence and spatial alignment with the demonstration video. The code generation module translates the plan into executable code, while the code reflection module verifies and refines the generated code to ensure correctness and consistency with the generated plan. These two reflection modules jointly enable the agent to detect and correct errors in both the plan generation and code generation, improving performance in tasks with intricate temporal and spatial dependencies. To support systematic evaluation, we introduce LongVILBench, a benchmark comprising 300 human demonstrations with action sequences of up to 18 steps. LongVILBench emphasizes temporal and spatial complexity across multiple task types. Experimental results demonstrate that existing methods perform poorly on this benchmark, whereas our new framework establishes a strong baseline for long-horizon visual imitation learning.
Authors: Niels Balemans, Ali Anwar, Jan Steckel, Siegfried Mercelis
Abstract: This paper extends LiDAR-BIND, a modular multi-modal fusion framework that binds heterogeneous sensors (radar, sonar) to a LiDAR-defined latent space, with mechanisms that explicitly enforce temporal consistency. We introduce three contributions: (i) temporal embedding similarity that aligns consecutive latent representations, (ii) a motion-aligned transformation loss that matches displacement between predictions and ground truth LiDAR, and (iii) windowed temporal fusion using a specialised temporal module. We further update the model architecture to better preserve spatial structure. Evaluations on radar/sonar-to-LiDAR translation demonstrate improved temporal and spatial coherence, yielding lower absolute trajectory error and better occupancy map accuracy in Cartographer-based SLAM (Simultaneous Localisation and Mapping). We propose different metrics based on the Fr\'echet Video Motion Distance (FVMD) and a correlation-peak distance metric providing practical temporal quality indicators to evaluate SLAM performance. The proposed temporal LiDAR-BIND, or LiDAR-BIND-T, maintains modular modality fusion while substantially enhancing temporal stability, resulting in improved robustness and performance for downstream SLAM.
Authors: Xin Wang, Ting Dang, Xinyu Zhang, Vassilis Kostakos, Michael J. Witbrock, Hong Jia
Abstract: Mobile and wearable healthcare monitoring play a vital role in facilitating timely interventions, managing chronic health conditions, and ultimately improving individuals' quality of life. Previous studies on large language models (LLMs) have highlighted their impressive generalization abilities and effectiveness in healthcare prediction tasks. However, most LLM-based healthcare solutions are cloud-based, which raises significant privacy concerns and results in increased memory usage and latency. To address these challenges, there is growing interest in compact models, Small Language Models (SLMs), which are lightweight and designed to run locally and efficiently on mobile and wearable devices. Nevertheless, how well these models perform in healthcare prediction remains largely unexplored. We systematically evaluated SLMs on health prediction tasks using zero-shot, few-shot, and instruction fine-tuning approaches, and deployed the best performing fine-tuned SLMs on mobile devices to evaluate their real-world efficiency and predictive performance in practical healthcare scenarios. Our results show that SLMs can achieve performance comparable to LLMs while offering substantial gains in efficiency and privacy. However, challenges remain, particularly in handling class imbalance and few-shot scenarios. These findings highlight SLMs, though imperfect in their current form, as a promising solution for next-generation, privacy-preserving healthcare monitoring.
Authors: Yuan Wei, Xiaohan Shan, Ran Miao, Jianmin Li
Abstract: Reinforcement learning (RL) agent development traditionally requires substantial expertise and iterative effort, often leading to high failure rates and limited accessibility. This paper introduces Agent$^2$, an LLM-driven agent-generates-agent framework for fully automated RL agent design. Agent$^2$ autonomously translates natural language task descriptions and environment code into executable RL solutions without human intervention. The framework adopts a dual-agent architecture: a Generator Agent that analyzes tasks and designs agents, and a Target Agent that is automatically generated and executed. To better support automation, RL development is decomposed into two stages, MDP modeling and algorithmic optimization, facilitating targeted and effective agent generation. Built on the Model Context Protocol, Agent$^2$ provides a unified framework for standardized agent creation across diverse environments and algorithms, incorporating adaptive training management and intelligent feedback analysis for continuous refinement. Extensive experiments on benchmarks including MuJoCo, MetaDrive, MPE, and SMAC show that Agent$^2$ outperforms manually designed baselines across all tasks, achieving up to 55\% performance improvement with consistent average gains. By enabling a closed-loop, end-to-end automation pipeline, this work advances a new paradigm in which agents can design and optimize other agents, underscoring the potential of agent-generates-agent systems for automated AI development.
Authors: Dvij Kalaria, Sudarshan S Harithas, Pushkal Katara, Sangkyung Kwak, Sarthak Bhagat, Shankar Sastry, Srinath Sridhar, Sai Vemprala, Ashish Kapoor, Jonathan Chung-Kuan Huang
Abstract: We introduce DreamControl, a novel methodology for learning autonomous whole-body humanoid skills. DreamControl leverages the strengths of diffusion models and Reinforcement Learning (RL): our core innovation is the use of a diffusion prior trained on human motion data, which subsequently guides an RL policy in simulation to complete specific tasks of interest (e.g., opening a drawer or picking up an object). We demonstrate that this human motion-informed prior allows RL to discover solutions unattainable by direct RL, and that diffusion models inherently promote natural looking motions, aiding in sim-to-real transfer. We validate DreamControl's effectiveness on a Unitree G1 robot across a diverse set of challenging tasks involving simultaneous lower and upper body control and object interaction. Project website at https://genrobo.github.io/DreamControl/
Authors: Neslihan Kose, Anthony Rhodes, Umur Aybars Ciftci, Ilke Demir
Abstract: As generative models are advancing in quality and quantity for creating synthetic content, deepfakes begin to cause online mistrust. Deepfake detectors are proposed to counter this effect, however, misuse of detectors claiming fake content as real or vice versa further fuels this misinformation problem. We present the first comprehensive uncertainty analysis of deepfake detectors, systematically investigating how generative artifacts influence prediction confidence. As reflected in detectors' responses, deepfake generators also contribute to this uncertainty as their generative residues vary, so we cross the uncertainty analysis of deepfake detectors and generators. Based on our observations, the uncertainty manifold holds enough consistent information to leverage uncertainty for deepfake source detection. Our approach leverages Bayesian Neural Networks and Monte Carlo dropout to quantify both aleatoric and epistemic uncertainties across diverse detector architectures. We evaluate uncertainty on two datasets with nine generators, with four blind and two biological detectors, compare different uncertainty methods, explore region- and pixel-based uncertainty, and conduct ablation studies. We conduct and analyze binary real/fake, multi-class real/fake, source detection, and leave-one-out experiments between the generator/detector combinations to share their generalization capability, model calibration, uncertainty, and robustness against adversarial attacks. We further introduce uncertainty maps that localize prediction confidence at the pixel level, revealing distinct patterns correlated with generator-specific artifacts. Our analysis provides critical insights for deploying reliable deepfake detection systems and establishes uncertainty quantification as a fundamental requirement for trustworthy synthetic media detection.
Authors: Shayne Wadle, Yanxin Zhang, Vikas Singh, Karthikeyan Sankaralingam
Abstract: The evaluation of new microprocessor designs is constrained by slow, cycle-accurate simulators that rely on unrepresentative benchmark traces. This paper introduces a novel deep learning framework for high-fidelity, ``in-the-wild'' simulation on production hardware. Our core contribution is a DL model trained on microarchitecture-independent features to predict cycle-level performance for hypothetical processor designs. This unique approach allows the model to be deployed on existing silicon to evaluate future hardware. We propose a complete system featuring a lightweight hardware trace collector and a principled sampling strategy to minimize user impact. This system achieves a simulation speed of 5 MIPS on a commodity GPU, imposing a mere 0.1% performance overhead. Furthermore, our co-designed Neutrino on-chip accelerator improves performance by 85x over the GPU. We demonstrate that this framework enables accurate performance analysis and large-scale hardware A/B testing on a massive scale using real-world applications.
Authors: Jinyeop Song, Jeff Gore, Max Kleiman-Weiner
Abstract: As language model (LM) agents become more capable and gain broader access to real-world tools, there is a growing need for scalable evaluation frameworks of agentic capability. However, conventional benchmark-centric evaluations are costly to design and require human designers to come up with valid tasks that translate into insights about general model capabilities. In this work, we propose information-theoretic evaluation based on empowerment, the mutual information between an agent's actions and future states, as an open-ended method for evaluating LM agents. We introduce EELMA (Estimating Empowerment of Language Model Agents), an algorithm for approximating effective empowerment from multi-turn text interactions. We validate EELMA on both language games and scaled-up realistic web-browsing scenarios. We find that empowerment strongly correlates with average task performance, characterize the impact of environmental complexity and agentic factors such as chain-of-thought, model scale, and memory length on estimated empowerment, and that high empowerment states and actions are often pivotal moments for general capabilities. Together, these results demonstrate empowerment as an appealing general-purpose metric for evaluating and monitoring LM agents in complex, open-ended settings.
Authors: Sasha Cui, Zhongren Chen
Abstract: Language models (LMs) are typically post-trained for desired capabilities and behaviors via weight-based or prompt-based steering, but the former is time-consuming and expensive, and the latter is not precisely controllable and often requires manual trial-and-error. While activation steering (AS) promises a cheap, fast, and controllable alternative to the two existing post-training methods, current AS techniques require hand-crafted prompt pairs or labor-intensive feature annotation, making them more inconvenient than the plug-and-play methods such as Reinforcement Learning (RL) and Supervised Fine-Tuning (SFT). We introduce Painless Activation Steering (PAS), a family of fully automated methods that make AS readily usable with any given labeled dataset, with no need for prompt construction, feature labeling, or human intervention. We evaluate PAS on three open-weight models (Llama3.1-8B-Instruct, DeepSeek-R1-Distill-8B, and Nous-Hermes-2) and 18 tasks; we find that PAS reliably improves performance for behavior tasks, but not for intelligence-oriented tasks. The introspective variant (iPAS) delivers the strongest causal steering effects (10.1% on Bias, 5.2% on Morality, and 34.8% on Alignment). We also show PAS delivers additional gains on top of In-Context Learning (ICL) and SFT. PAS constructs a fast, lightweight activation vector that can be cheaply trained, easily stored, and activated at will. Our results provide a characterization of where AS helps, where it fails, and how to deploy it as a practical, automated LM post-training option.
Authors: Artavazd Maranjyan, Peter Richt\'arik
Abstract: Asynchronous stochastic gradient methods are central to scalable distributed optimization, particularly when devices differ in computational capabilities. Such settings arise naturally in federated learning, where training takes place on smartphones and other heterogeneous edge devices. In addition to varying computation speeds, these devices often hold data from different distributions. However, existing asynchronous SGD methods struggle in such heterogeneous settings and face two key limitations. First, many rely on unrealistic assumptions of similarity across workers' data distributions. Second, methods that relax this assumption still fail to achieve theoretically optimal performance under heterogeneous computation times. We introduce Ringleader ASGD, the first asynchronous SGD algorithm that attains the theoretical lower bounds for parallel first-order stochastic methods in the smooth nonconvex regime, thereby achieving optimal time complexity under data heterogeneity and without restrictive similarity assumptions. Our analysis further establishes that Ringleader ASGD remains optimal under arbitrary and even time-varying worker computation speeds, closing a fundamental gap in the theory of asynchronous optimization.
Authors: Yikai Wang, Xiaocheng Li, Guanting Chen
Abstract: Large language models (LLMs) are increasingly used for decision-making tasks under uncertainty; however, their risk profiles and how they are influenced by prompting and alignment methods remain underexplored. Existing studies have primarily examined personality prompting or multi-agent interactions, leaving open the question of how post-training influences the risk behavior of LLMs. In this work, we propose a new pipeline for eliciting, steering, and modulating LLMs' risk profiles, drawing on tools from behavioral economics and finance. Using utility-theoretic models, we compare pre-trained, instruction-tuned, and RLHF-aligned LLMs, and find that while instruction-tuned models exhibit behaviors consistent with some standard utility formulations, pre-trained and RLHF-aligned models deviate more from any utility models fitted. We further evaluate modulation strategies, including prompt engineering, in-context learning, and post-training, and show that post-training provides the most stable and effective modulation of risk preference. Our findings provide insights into the risk profiles of different classes and stages of LLMs and demonstrate how post-training modulates these profiles, laying the groundwork for future research on behavioral alignment and risk-aware LLM design.
Authors: Charles L. Wang
Abstract: This paper presents MathBode, a dynamic diagnostic for mathematical reasoning in large language models (LLMs). Instead of one-shot accuracy, MathBode treats each parametric problem as a system: we drive a single parameter sinusoidally and fit first-harmonic responses of model outputs and exact solutions. This yields interpretable, frequency-resolved metrics -- gain (amplitude tracking) and phase (lag) -- that form Bode-style fingerprints. Across five closed-form families (linear solve, ratio/saturation, compound interest, 2x2 linear systems, similar triangles), the diagnostic surfaces systematic low-pass behavior and growing phase lag that accuracy alone obscures. We compare several models against a symbolic baseline that calibrates the instrument ($G \approx 1$, $\phi \approx 0$). Results separate frontier from mid-tier models on dynamics, providing a compact, reproducible protocol that complements standard benchmarks with actionable measurements of reasoning fidelity and consistency. We open-source the dataset and code to enable further research and adoption.
Authors: Emma Kondrup, Sebastian Sabry, Hussein Abdallah, Zachary Yang, James Zhou, Kellin Pelrine, Jean-Fran\c{c}ois Godbout, Michael M. Bronstein, Reihaneh Rabbany, Shenyang Huang
Abstract: Online misinformation poses an escalating threat, amplified by the Internet's open nature and increasingly capable LLMs that generate persuasive yet deceptive content. Existing misinformation detection methods typically focus on either textual content or network structure in isolation, failing to leverage the rich, dynamic interplay between website content and hyperlink relationships that characterizes real-world misinformation ecosystems. We introduce CrediBench: a large-scale data processing pipeline for constructing temporal web graphs that jointly model textual content and hyperlink structure for misinformation detection. Unlike prior work, our approach captures the dynamic evolution of general misinformation domains, including changes in both content and inter-site references over time. Our processed one-month snapshot extracted from the Common Crawl archive in December 2024 contains 45 million nodes and 1 billion edges, representing the largest web graph dataset made publicly available for misinformation research to date. From our experiments on this graph snapshot, we demonstrate the strength of both structural and webpage content signals for learning credibility scores, which measure source reliability. The pipeline and experimentation code are all available here, and the dataset is in this folder.
Authors: Ting-Kang Wang, Yueh-Po Peng, Li Su, Vincent K. M. Cheung
Abstract: While automatic music transcription is well-established in music information retrieval, most models are limited to transcribing pitch and timing information from audio, and thus omit crucial expressive and instrument-specific nuances. One example is playing technique on the violin, which affords its distinct palette of timbres for maximal emotional impact. Here, we propose VioPTT (Violin Playing Technique-aware Transcription), a lightweight, end-to-end model that directly transcribes violin playing technique in addition to pitch onset and offset. Furthermore, we release MOSA-VPT, a novel, high-quality synthetic violin playing technique dataset to circumvent the need for manually labeled annotations. Leveraging this dataset, our model demonstrated strong generalization to real-world note-level violin technique recordings in addition to achieving state-of-the-art transcription performance. To our knowledge, VioPTT is the first to jointly combine violin transcription and playing technique prediction within a unified framework.
Authors: Zeyuan Zhao, Chaoran Li, Shao Zhang, Ying Wen
Abstract: Multi-Agent Pickup and Delivery (MAPD) is a challenging extension of Multi-Agent Path Finding (MAPF), where agents are required to sequentially complete tasks with fixed-location pickup and delivery demands. Although learning-based methods have made progress in MAPD, they often perform poorly in warehouse-like environments with narrow pathways and long corridors when relying only on local observations for distributed decision-making. Communication learning can alleviate the lack of global information but introduce high computational complexity due to point-to-point communication. To address this challenge, we formulate MAPF as a sequence modeling problem and prove that path-finding policies under sequence modeling possess order-invariant optimality, ensuring its effectiveness in MAPD. Building on this, we propose the Sequential Pathfinder (SePar), which leverages the Transformer paradigm to achieve implicit information exchange, reducing decision-making complexity from exponential to linear while maintaining efficiency and global awareness. Experiments demonstrate that SePar consistently outperforms existing learning-based methods across various MAPF tasks and their variants, and generalizes well to unseen environments. Furthermore, we highlight the necessity of integrating imitation learning in complex maps like warehouses.
Authors: Melika Honarmand, Ayati Sharma, Badr AlKhamissi, Johannes Mehrer, Martin Schrimpf
Abstract: Dyslexia, a neurodevelopmental disorder characterized by persistent reading difficulties, is often linked to reduced activity of the visual word form area in the ventral occipito-temporal cortex. Traditional approaches to studying dyslexia, such as behavioral and neuroimaging methods, have provided valuable insights but remain limited in their ability to test causal hypotheses about the underlying mechanisms of reading impairments. In this study, we use large-scale vision-language models (VLMs) to simulate dyslexia by functionally identifying and perturbing artificial analogues of word processing. Using stimuli from cognitive neuroscience, we identify visual-word-form-selective units within VLMs and demonstrate that targeted ablation of these units, unlike ablation of random units, leads to selective impairments in reading tasks while general visual and language comprehension abilities remain intact. In particular, the resulting model matches dyslexic humans' phonological deficits without a significant change in orthographic processing. Taken together, our modeling results replicate key characteristics of dyslexia and establish a computational framework for investigating reading disorders.
Authors: Zhen Bi, Zhenlin Hu, Jinnan Yang, Mingyang Chen, Cheng Deng, Yida Xue, Zeyu Yang, Qing Shen, Zhenfang Liu, Kang Zhao, Ningyu Zhang, Jungang Lou
Abstract: Recent advances in large language models (LLMs) highlight the importance of training data structure and quality in shaping reasoning behavior. However, most existing approaches focus on transforming data formats while neglecting the internal reasoning complexity of training samples, leaving the reasoning potential of data under-explored and underutilized. In this work, we posit that LLM logical reasoning performance is jointly constrained by the potential of the training data and the cognitive capacity of the model. To make this relationship measurable, we introduce Data Reasoning Intensity (DRI), a novel metric that quantifies the latent logical reasoning complexity of samples by decomposing and aggregating their logical structures. This allows us to analyze how well current LLMs utilize logical reasoning signals and identify performance gaps relative to data potential. Based on this insight, we introduce a re-cognizing optimization strategy that systematically enhances the logical reasoning intensity of training data. Rather than increasing data volume, our method re-optimizes existing samples to better align with the LLM's logical reasoning boundary. Extensive experiments show that our approach significantly improves performance and generalization over data-centric strategies. We further validate our method under a reinforcement learning framework. Our results indicate that prioritizing reasoning complexity in data rather than sheer scale or superficial form is essential to realizing LLMs' full cognitive potential.