Authors: Wang Yu-Hang, Liu ying, Fang liang, Wang Xuelin, Junkang Guo, Shiwei Li, Lei Gao, Jian Liu, Wenfei Yin
Abstract: Adversarial Training (AT) is a cornerstone defense, but many variants overlook foundational feature representations by primarily focusing on stronger attack generation. We introduce Adversarial Evolution Training (AET), a simple yet powerful framework that strategically prepends an Empirical Risk Minimization (ERM) phase to conventional AT. We hypothesize this initial ERM phase cultivates a favorable feature manifold, enabling more efficient and effective robustness acquisition. Empirically, AET achieves comparable or superior robustness more rapidly, improves clean accuracy, and cuts training costs by 8-25\%. Its effectiveness is shown across multiple datasets, architectures, and when augmenting established AT methods. Our findings underscore the impact of feature pre-conditioning via standard training for developing more efficient, principled robust defenses. Code is available in the supplementary material.
Authors: Jiahe Qin, Junpeng Li, Changchun Hua, Yana Yang
Abstract: Label Proportion Learning (LLP) addresses the classification problem where multiple instances are grouped into bags and each bag contains information about the proportion of each class. However, in practical applications, obtaining precise supervisory information regarding the proportion of instances in a specific class is challenging. To better align with real-world application scenarios and effectively leverage the proportional constraints of instances within tuples, this paper proposes a generalized learning framework \emph{MDPU}. Specifically, we first mathematically model the distribution of instances within tuples of arbitrary size, under the constraint that the number of positive instances is no less than that of negative instances. Then we derive an unbiased risk estimator that satisfies risk consistency based on the empirical risk minimization (ERM) method. To mitigate the inevitable overfitting issue during training, a risk correction method is introduced, leading to the development of a corrected risk estimator. The generalization error bounds of the unbiased risk estimator theoretically demonstrate the consistency of the proposed method. Extensive experiments on multiple datasets and comparisons with other relevant baseline methods comprehensively validate the effectiveness of the proposed learning framework.
Authors: Yajie Ji, Yanlai Chen, Shawn Koohy
Abstract: We propose S$^2$GPT-PINN, a sparse and small model for solving parametric partial differential equations (PDEs). Similar to Small Language Models (SLMs), S$^2$GPT-PINN is tailored to domain-specific (families of) PDEs and characterized by its compact architecture and minimal computational power. Leveraging a small amount of extremely high quality data via a mathematically rigorous greedy algorithm that is enabled by the large full-order models, S$^2$GPT-PINN relies on orders of magnitude less parameters than PINNs to achieve extremely high efficiency via two levels of customizations. The first is knowledge distillation via task-specific activation functions that are transferred from Pre-Trained PINNs. The second is a judicious down-sampling when calculating the physics-informed loss of the network compressing the number of data sites by orders of magnitude to the size of the small model.
Authors: Hui Ma, Kai Yang, Man-On Pun
Abstract: Cellular traffic prediction is of great importance for operators to manage network resources and make decisions. Traffic is highly dynamic and influenced by many exogenous factors, which would lead to the degradation of traffic prediction accuracy. This paper proposes an end-to-end framework with two variants to explicitly characterize the spatiotemporal patterns of cellular traffic among neighboring cells. It uses convolutional neural networks with an attention mechanism to capture the spatial dynamics and Kalman filter for temporal modelling. Besides, we can fully exploit the auxiliary information such as social activities to improve prediction performance. We conduct extensive experiments on three real-world datasets. The results show that our proposed models outperform the state-of-the-art machine learning techniques in terms of prediction accuracy.
Authors: Liulu He, Shenli Zhen, Karwei Sun, Yijiang Liu, Yufei Zhao, Chongkang Tan, Huanrui Yang, Yuan Du, Li Du
Abstract: Rotations have become essential to state-of-the-art quantization pipelines for large language models (LLMs) by effectively smoothing outliers in weights and activations. However, further optimizing the rotation parameters offers only limited performance gains and introduces significant training overhead: due to rotation parameter sharing, full-model must be loaded simultaneously to enable backpropagation, resulting in substantial memory consumption and limited practical utility. In this work, we identify two fundamental limitations of current rotational quantization methods: (i) rotation fails to align channel means, resulting in wider quantization bounds and increased rounding errors; and (ii) rotation makes the activation distribution more Gaussian-like, increasing energy loss caused by clipping errors. To address these issues, we introduce \textbf{BASE-Q}, a simple yet powerful approach that combines bias correction and asymmetric scaling to effectively reduce rounding and clipping errors. Furthermore, BASE-Q enables blockwise optimization, eliminating the need for memory-intensive full-model backpropagation. Extensive experiments on various LLMs and benchmarks demonstrate the effectiveness of BASE-Q, narrowing the accuracy gap to full-precision models by 50.5\%, 42.9\%, and 29.2\% compared to QuaRot, SpinQuant, and OSTQuant, respectively. The code will be released soon.
Authors: Tianyu Wang, Lingyou Pang, Akira Horiguchi, Carey E. Priebe
Abstract: The increasing use of synthetic data from the public Internet has enhanced data usage efficiency in large language model (LLM) training. However, the potential threat of model collapse remains insufficiently explored. Existing studies primarily examine model collapse in a single model setting or rely solely on statistical surrogates. In this work, we introduce LLM Web Dynamics (LWD), an efficient framework for investigating model collapse at the network level. By simulating the Internet with a retrieval-augmented generation (RAG) database, we analyze the convergence pattern of model outputs. Furthermore, we provide theoretical guarantees for this convergence by drawing an analogy to interacting Gaussian Mixture Models.
Authors: Chuheng Zhang, Tim Pearce, Pushi Zhang, Kaixin Wang, Xiaoyu Chen, Wei Shen, Li Zhao, Jiang Bian
Abstract: Latent action models (LAMs) aim to learn action-relevant changes from unlabeled videos by compressing changes between frames as latents. However, differences between video frames can be caused by controllable changes as well as exogenous noise, leading to an important concern -- do latents capture the changes caused by actions or irrelevant noise? This paper studies this issue analytically, presenting a linear model that encapsulates the essence of LAM learning, while being tractable.This provides several insights, including connections between LAM and principal component analysis (PCA), desiderata of the data-generating policy, and justification of strategies to encourage learning controllable changes using data augmentation, data cleaning, and auxiliary action-prediction. We also provide illustrative results based on numerical simulation, shedding light on the specific structure of observations, actions, and noise in data that influence LAM learning.
Authors: Jaehyun Nam, Jinsung Yoon, Jiefeng Chen, Jinwoo Shin, Sercan \"O. Ar{\i}k, Tomas Pfister
Abstract: Agents based on large language models (LLMs) for machine learning engineering (MLE) can automatically implement ML models via code generation. However, existing approaches to build such agents often rely heavily on inherent LLM knowledge and employ coarse exploration strategies that modify the entire code structure at once. This limits their ability to select effective task-specific models and perform deep exploration within specific components, such as experimenting extensively with feature engineering options. To overcome these, we propose MLE-STAR, a novel approach to build MLE agents. MLE-STAR first leverages external knowledge by using a search engine to retrieve effective models from the web, forming an initial solution, then iteratively refines it by exploring various strategies targeting specific ML components. This exploration is guided by ablation studies analyzing the impact of individual code blocks. Furthermore, we introduce a novel ensembling method using an effective strategy suggested by MLE-STAR. Our experimental results show that MLE-STAR achieves medals in 44% of the Kaggle competitions on the MLE-bench, significantly outperforming the best alternative.
Authors: Jiaxing Li, Hanjiang Hu, Yujie Yang, Changliu Liu
Abstract: Recent learning-based safety filters have outperformed conventional methods, such as hand-crafted Control Barrier Functions (CBFs), by effectively adapting to complex constraints. However, these learning-based approaches lack formal safety guarantees. In this work, we introduce a verifiable model-free safety filter based on Hamilton-Jacobi reachability analysis. Our primary contributions include: 1) extending verifiable self-consistency properties for Q value functions, 2) proposing a multiplicative Q-network structure to mitigate zero-sublevel-set shrinkage issues, and 3) developing a verification pipeline capable of soundly verifying these self-consistency properties. Our proposed approach successfully synthesizes formally verified, model-free safety certificates across four standard safe-control benchmarks.
Authors: Iliyas Ibrahim Iliyas, Souley Boukari, Abdulsalam Yau Gital
Abstract: This study introduces a framework that integrates nonlinear feature extraction, classification, and efficient optimization. First, kernel principal component analysis with a radial basis function kernel reduces dimensionality while preserving 95% of the variance. Second, a multilayer perceptron (MLP) learns to predict disease status. Finally, a modified multiprocessing genetic algorithm (MIGA) optimizes MLP hyperparameters in parallel over ten generations. We evaluated this approach on three datasets: the Wisconsin Diagnostic Breast Cancer dataset, the Parkinson's Telemonitoring dataset, and the chronic kidney disease dataset. The MLP tuned by the MIGA achieved the best accuracy of 99.12% for breast cancer, 94.87% for Parkinson's disease, and 100% for chronic kidney disease. These results outperform those of other methods, such as grid search, random search, and Bayesian optimization. Compared with a standard genetic algorithm, kernel PCA revealed nonlinear relationships that improved classification, and the MIGA's parallel fitness evaluations reduced the tuning time by approximately 60%. The genetic algorithm incurs high computational cost from sequential fitness evaluations, but our multiprocessing interface GA (MIGA) parallelizes this step, slashing the tuning time and steering the MLP toward the best accuracy score of 99.12%, 94.87%, and 100% for breast cancer, Parkinson's disease, and CKD, respectively.
Authors: Xinxing Ren, Qianbo Zang, Zekun Guo
Abstract: Recent advances in large language models (LLMs) have shown impressive performance in mathematical reasoning and code generation. However, LLMs still struggle in the simulation domain, particularly in generating Simulink models, which are essential tools in engineering and scientific research. Our preliminary experiments indicate that LLM agents often fail to produce reliable and complete Simulink simulation code from text-only inputs, likely due to the lack of Simulink-specific data in their pretraining. To address this challenge, we propose SimuGen, a multimodal agent-based framework that automatically generates accurate Simulink simulation code by leveraging both the visual Simulink diagram and domain knowledge. SimuGen coordinates several specialized agents, including an investigator, unit test reviewer, code generator, executor, debug locator, and report writer, supported by a domain-specific knowledge base. This collaborative and modular design enables interpretable, robust, and reproducible Simulink simulation generation. Our source code is publicly available at https://github.com/renxinxing123/SimuGen_beta.
Authors: Haipeng Zhou, Sicheng Yang, Sihan Yang, Jing Qin, Lei Chen, Lei Zhu
Abstract: Survival prediction aims to evaluate the risk level of cancer patients. Existing methods primarily rely on pathology and genomics data, either individually or in combination. From the perspective of cancer pathogenesis, epigenetic changes, such as methylation data, could also be crucial for this task. Furthermore, no previous endeavors have utilized textual descriptions to guide the prediction. To this end, we are the first to explore the use of four modalities, including three clinical modalities and language, for conducting survival prediction. In detail, we are motivated by the Chain-of-Thought (CoT) to propose the Chain-of-Cancer (CoC) framework, focusing on intra-learning and inter-learning. We encode the clinical data as the raw features, which remain domain-specific knowledge for intra-learning. In terms of inter-learning, we use language to prompt the raw features and introduce an Autoregressive Mutual Traction module for synergistic representation. This tailored framework facilitates joint learning among multiple modalities. Our approach is evaluated across five public cancer datasets, and extensive experiments validate the effectiveness of our methods and proposed designs, leading to producing \sota results. Codes will be released.
Authors: Yunhak Oh, Junseok Lee, Yeongmin Kim, Sangwoo Seo, Namkyeong Lee, Chanyoung Park
Abstract: Spatially Resolved Transcriptomics (SRT) is a cutting-edge technique that captures the spatial context of cells within tissues, enabling the study of complex biological networks. Recent graph-based methods leverage both gene expression and spatial information to identify relevant spatial domains. However, these approaches fall short in obtaining meaningful spot representations, especially for spots near spatial domain boundaries, as they heavily emphasize adjacent spots that have minimal feature differences from an anchor node. To address this, we propose Spotscape, a novel framework that introduces the Similarity Telescope module to capture global relationships between multiple spots. Additionally, we propose a similarity scaling strategy to regulate the distances between intra- and inter-slice spots, facilitating effective multi-slice integration. Extensive experiments demonstrate the superiority of Spotscape in various downstream tasks, including single-slice and multi-slice scenarios. Our code is available at the following link: https: //github.com/yunhak0/Spotscape.
Authors: Shengyuan Hu, Neil Kale, Pratiksha Thaker, Yiwei Fu, Steven Wu, Virginia Smith
Abstract: Machine unlearning has the potential to improve the safety of large language models (LLMs) by removing sensitive or harmful information post hoc. A key challenge in unlearning involves balancing between forget quality (effectively unlearning undesirable information) and retain quality (maintaining good performance on other, general tasks). Unfortunately, as we show, current LLM unlearning benchmarks contain highly disparate forget and retain sets -- painting a false picture of the effectiveness of LLM unlearning methods. This can be particularly problematic because it opens the door for benign perturbations, such as relearning attacks, to easily reveal supposedly unlearned knowledge once models are deployed. To address this, we present $\texttt{BLUR}$: a benchmark for LLM unlearning that provides more realistic scenarios of forget-retain overlap. $\texttt{BLUR}$ significantly expands on existing unlearning benchmarks by providing extended evaluation tasks, combined forget/retain queries, and relearning datasets of varying degrees of difficulty. Despite the benign nature of the queries considered, we find that the performance of existing methods drops significantly when evaluated on $\texttt{BLUR}$, with simple approaches performing better on average than more recent methods. These results highlight the importance of robust evaluation and suggest several important directions of future study. Our benchmark is publicly available at: https://huggingface.co/datasets/forgelab/BLUR
Authors: Minjae Cho, Hiroyasu Tsukamoto, Huy Trong Tran
Abstract: Control contraction metrics (CCMs) provide a framework to co-synthesize a controller and a corresponding contraction metric -- a positive-definite Riemannian metric under which a closed-loop system is guaranteed to be incrementally exponentially stable. However, the synthesized controller only ensures that all the trajectories of the system converge to one single trajectory and, as such, does not impose any notion of optimality across an entire trajectory. Furthermore, constructing CCMs requires a known dynamics model and non-trivial effort in solving an infinite-dimensional convex feasibility problem, which limits its scalability to complex systems featuring high dimensionality with uncertainty. To address these issues, we propose to integrate CCMs into reinforcement learning (RL), where CCMs provide dynamics-informed feedback for learning control policies that minimize cumulative tracking error under unknown dynamics. We show that our algorithm, called contraction actor-critic (CAC), formally enhances the capability of CCMs to provide a set of contracting policies with the long-term optimality of RL in a fully automated setting. Given a pre-trained dynamics model, CAC simultaneously learns a contraction metric generator (CMG) -- which generates a contraction metric -- and uses an actor-critic algorithm to learn an optimal tracking policy guided by that metric. We demonstrate the effectiveness of our algorithm relative to established baselines through extensive empirical studies, including simulated and real-world robot experiments, and provide a theoretical rationale for incorporating contraction theory into RL.
Authors: Haolin Pan, Hongyu Lin, Haoran Luo, Yang Liu, Kaichun Yao, Libo Zhang, Mingjie Xing, Yanjun Wu
Abstract: Compiler auto-tuning optimizes pass sequences to improve performance metrics such as Intermediate Representation (IR) instruction count. Although recent advances leveraging Large Language Models (LLMs) have shown promise in automating compiler tuning, two significant challenges still remain: the absence of high-quality reasoning datasets for agents training, and limited effective interactions with the compilation environment. In this work, we introduce Compiler-R1, the first reinforcement learning (RL)-driven framework specifically augmenting LLM capabilities for compiler auto-tuning. Compiler-R1 features a curated, high-quality reasoning dataset and a novel two-stage end-to-end RL training pipeline, enabling efficient environment exploration and learning through an outcome-based reward. Extensive experiments across seven datasets demonstrate Compiler-R1 achieving an average 8.46% IR instruction count reduction compared to opt -Oz, showcasing the strong potential of RL-trained LLMs for compiler optimization. Our code and datasets are publicly available at https://github.com/Panhaolin2001/Compiler-R1.
Authors: Peter Belcak, Greg Heinrich, Jan Kautz, Pavlo Molchanov
Abstract: Finetuning language models for a new domain inevitably leads to the deterioration of their general performance. This becomes more pronounced the more limited the finetuning data resource. We introduce minifinetuning (MFT), a method for language model domain adaptation that considerably reduces the effects of overfitting-induced degeneralization in low-data settings and which does so in the absence of any pre-training data for replay. MFT demonstrates 2-10x more favourable specialization-to-degeneralization ratios than standard finetuning across a wide range of models and domains and exhibits an intrinsic robustness to overfitting when data in the new domain is scarce and down to as little as 500 samples. Employing corrective self-distillation that is individualized on the sample level, MFT outperforms parameter-efficient finetuning methods, demonstrates replay-like degeneralization mitigation properties, and is composable with either for a combined effect.
Authors: Guoqing Chao, Zhenghao Zhang, Lei Meng, Jie Wen, Dianhui Chu
Abstract: Federated multi-view clustering has been proposed to mine the valuable information within multi-view data distributed across different devices and has achieved impressive results while preserving the privacy. Despite great progress, most federated multi-view clustering methods only used global pseudo-labels to guide the downstream clustering process and failed to exploit the global information when extracting features. In addition, missing data problem in federated multi-view clustering task is less explored. To address these problems, we propose a novel Federated Incomplete Multi-view Clustering method with globally Fused Graph guidance (FIMCFG). Specifically, we designed a dual-head graph convolutional encoder at each client to extract two kinds of underlying features containing global and view-specific information. Subsequently, under the guidance of the fused graph, the two underlying features are fused into high-level features, based on which clustering is conducted under the supervision of pseudo-labeling. Finally, the high-level features are uploaded to the server to refine the graph fusion and pseudo-labeling computation. Extensive experimental results demonstrate the effectiveness and superiority of FIMCFG. Our code is publicly available at https://github.com/PaddiHunter/FIMCFG.
Authors: Feiyu Yao, Qian Wang
Abstract: As large language models (LLMs) continue to support increasingly longer contexts, the memory demand for key-value (KV) caches during decoding grows rapidly, becoming a critical bottleneck in both GPU memory capacity and PCIe bandwidth. Sparse attention mechanisms alleviate this issue by computing attention weights only for selected key-value pairs. However, their indexing computation typically requires traversing all key vectors, resulting in significant computational and data transfer overhead. To reduce the cost of index retrieval, existing methods often treat each decoding step as an independent process, failing to exploit the temporal correlations embedded in historical decoding information. To this end, we propose LFPS(Learn From the Past for Sparse Indexing), an acceleration method that dynamically constructs sparse indexing candidates based on historical attention patterns. LFPS captures two prevalent trends in decoder attention -vertical patterns (attending to fixed positions) and slash patterns (attending to relative positions) -and incorporates a positional expansion strategy to effectively predict the Top-k indices for the current step. We validate LFPS on challenging long-context benchmarks such as LongBench-RULER, using Llama-3.1-8B-Instruct as the base model. Experimental results show that LFPS achieves up to 22.8$\times$ speedup over full attention and 9.6$\times$ speedup over exact Top-k retrieval on an RTX 4090 GPU and a single CPU core of a Xeon Gold 6430, respectively, while preserving generation accuracy. These results demonstrate that LFPS offers a practical and efficient solution for decoding optimization in long-context LLM inference.
Authors: Jittarin Jetwiriyanon, Teo Susnjak, Surangika Ranathunga
Abstract: This study investigates zero-shot forecasting capabilities of Time Series Foundation Models (TSFMs) for macroeconomic indicators. We apply TSFMs to forecasting economic indicators under univariate conditions, bypassing the need for train bespoke econometric models using and extensive training datasets. Our experiments were conducted on a case study dataset, without additional customisation. We rigorously back-tested three state-of-the-art TSFMs (Chronos, TimeGPT and Moirai) under data-scarce conditions and structural breaks. Our results demonstrate that appropriately engineered TSFMs can internalise rich economic dynamics, accommodate regime shifts, and deliver well-behaved uncertainty estimates out of the box, while matching state-of-the-art multivariate models on this domain. Our findings suggest that, without any fine-tuning, TSFMs can match or exceed classical models during stable economic conditions. However, they are vulnerable to degradation in performances during periods of rapid shocks. The findings offer guidance to practitioners on when zero-shot deployments are viable for macroeconomic monitoring and strategic planning.
Authors: Yunze Lin
Abstract: Mathematical reasoning presents a significant challenge for Large Language Models (LLMs) as it requires ensuring the correctness of each reasoning step. Researchers have been strengthening the mathematical reasoning abilities of LLMs through supervised fine-tuning, but due to the inability to suppress incorrect outputs, illusions can easily arise. Recently, Direct Preference Optimization (DPO) has been widely adopted for aligning human intent by using preference data to prevent LLMs from generating incorrect outputs. However, it has shown limited benefits in long-chain mathematical reasoning, mainly because DPO struggles to effectively capture the differences between accepted and rejected answers from preferences in long-chain data. The inconsistency between DPO training and LLMs' generation metrics also affects the effectiveness of suppressing incorrect outputs. We propose the Multi-Granularity Direct Preference Optimization (MDPO) method, optimizing the mathematical reasoning of LLMs at three granularities: Solution2Solution, Inference2Inference, and Step2Step. Solution2Solution focuses on the correctness of entire long-chain reasoning; Inference2Inference concentrates on logical reasoning between steps; Step2Step corrects computational errors in steps, enhancing the computational capabilities of LLMs. Additionally, we unify the training objectives of the three granularities to align with the generation metrics. We conducted experiments on the open-source models Qwen2 and Llama3, achieving improvements of 1.7% and 0.9% on the GSM8K dataset, and 2.3% and 1.2% on the MATH dataset, outperforming DPO and other DPO variant methods. Furthermore, we also provide a pipeline for constructing MDPO training data that is simple and does not require manual annotation costs.
Authors: Xinglin Wang, Yiwei Li, Shaoxiong Feng, Peiwen Yuan, Yueqi Zhang, Jiayi Shi, Chuyi Tan, Boyuan Pan, Yao Hu, Kan Li
Abstract: Test-Time Scaling (TTS) improves the performance of Large Language Models (LLMs) by using additional inference-time computation to explore multiple reasoning paths through search. Yet how to allocate a fixed rollout budget most effectively during search remains underexplored, often resulting in inefficient use of compute at test time. To bridge this gap, we formulate test-time search as a resource allocation problem and derive the optimal allocation strategy that maximizes the probability of obtaining a correct solution under a fixed rollout budget. Within this formulation, we reveal a core limitation of existing search methods: solution-level allocation tends to favor reasoning directions with more candidates, leading to theoretically suboptimal and inefficient use of compute. To address this, we propose Direction-Oriented Resource Allocation (DORA), a provably optimal method that mitigates this bias by decoupling direction quality from candidate count and allocating resources at the direction level. To demonstrate DORA's effectiveness, we conduct extensive experiments on challenging mathematical reasoning benchmarks including MATH500, AIME2024, and AIME2025. The empirical results show that DORA consistently outperforms strong baselines with comparable computational cost, achieving state-of-the-art accuracy. We hope our findings contribute to a broader understanding of optimal TTS for LLMs.
Authors: Falih Gozi Febrinanto, Adonia Simango, Chengpei Xu, Jingjing Zhou, Jiangang Ma, Sonika Tyagi, Feng Xia
Abstract: Graph neural networks (GNNs) have been developed to model the relationship between regions of interest (ROIs) in brains and have shown significant improvement in detecting brain diseases. However, most of these frameworks do not consider the intrinsic relationship of causality factor between brain ROIs, which is arguably more essential to observe cause and effect interaction between signals rather than typical correlation values. We propose a novel framework called CGB (Causal Graphs for Brains) for brain disease classification/detection, which models refined brain networks based on the causal discovery method, transfer entropy, and geometric curvature strategy. CGB unveils causal relationships between ROIs that bring vital information to enhance brain disease classification performance. Furthermore, CGB also performs a graph rewiring through a geometric curvature strategy to refine the generated causal graph to become more expressive and reduce potential information bottlenecks when GNNs model it. Our extensive experiments show that CGB outperforms state-of-the-art methods in classification tasks on brain disease datasets, as measured by average F1 scores.
Authors: Pedro C. Vieira, Miguel E. P. Silva, Pedro Manuel Pinto Ribeiro
Abstract: Graph Neural Networks (GNNs) are a predominant method for graph representation learning. However, beyond subgraph frequency estimation, their application to network motif significance-profile (SP) prediction remains under-explored, with no established benchmarks in the literature. We propose to address this problem, framing SP estimation as a task independent of subgraph frequency estimation. Our approach shifts from frequency counting to direct SP estimation and modulates the problem as multitarget regression. The reformulation is optimised for interpretability, stability and scalability on large graphs. We validate our method using a large synthetic dataset and further test it on real-world graphs. Our experiments reveal that 1-WL limited models struggle to make precise estimations of SPs. However, they can generalise to approximate the graph generation processes of networks by comparing their predicted SP with the ones originating from synthetic generators. This first study on GNN-based motif estimation also hints at how using direct SP estimation can help go past the theoretical limitations that motif estimation faces when performed through subgraph counting.
Authors: Siru Ouyang, Xinyu Zhu, Zilin Xiao, Minhao Jiang, Yu Meng, Jiawei Han
Abstract: Reinforcement learning (RL) has become a powerful approach for improving the reasoning capabilities of large language models (LLMs), as evidenced by recent successes such as OpenAI's o1 and Deepseek-R1. However, applying RL at scale remains intimidatingly resource-intensive, requiring multiple model copies and extensive GPU workloads. On the other hand, while being powerful, recent studies suggest that RL does not fundamentally endow models with new knowledge; rather, it primarily reshapes the model's output distribution to activate reasoning capabilities latent in the base model. Building on this insight, we hypothesize that the changes in output probabilities induced by RL are largely model-size invariant, opening the door to a more efficient paradigm: training a small model with RL and transferring its induced probability shifts to larger base models. To verify our hypothesis, we conduct a token-level analysis of decoding trajectories and find high alignment in RL-induced output distributions across model scales, validating our hypothesis. Motivated by this, we propose RAST, a simple yet effective method that transfers reasoning behaviors by injecting RL-induced probability adjustments from a small RL-trained model into larger models. Experiments across multiple mathematical reasoning benchmarks show that RAST substantially and consistently enhances the reasoning capabilities of base models while requiring significantly lower GPU memory than direct RL training, sometimes even yielding better performance than the RL-trained counterparts. Our findings offer new insights into the nature of RL-driven reasoning and practical strategies for scaling its benefits without incurring its full computational cost. The project page of RAST is available at https://ozyyshr.github.io/RAST/.
Authors: Le Jiang, Liyan Ma, Guang Yang
Abstract: Federated learning (FL) has emerged as a transformative framework for privacy-preserving distributed training, allowing clients to collaboratively train a global model without sharing their local data. This is especially crucial in sensitive fields like healthcare, where protecting patient data is paramount. However, privacy leakage remains a critical challenge, as the communication of model updates can be exploited by potential adversaries. Gradient inversion attacks (GIAs), for instance, allow adversaries to approximate the gradients used for training and reconstruct training images, thus stealing patient privacy. Existing defense mechanisms obscure gradients, yet lack a nuanced understanding of which gradients or types of image information are most vulnerable to such attacks. These indiscriminate calibrated perturbations result in either excessive privacy protection degrading model accuracy, or insufficient one failing to safeguard sensitive information. Therefore, we introduce a framework that addresses these challenges by leveraging a shadow model with interpretability for identifying sensitive areas. This enables a more targeted and sample-specific noise injection. Specially, our defensive strategy achieves discrepancies of 3.73 in PSNR and 0.2 in SSIM compared to the circumstance without defense on the ChestXRay dataset, and 2.78 in PSNR and 0.166 in the EyePACS dataset. Moreover, it minimizes adverse effects on model performance, with less than 1\% F1 reduction compared to SOTA methods. Our extensive experiments, conducted across diverse types of medical images, validate the generalization of the proposed framework. The stable defense improvements for FedAvg are consistently over 1.5\% times in LPIPS and SSIM. It also offers a universal defense against various GIA types, especially for these sensitive areas in images.
Authors: Songqi Zhou, Ruixue Liu, Yixing Wang, Jia Lu, Benben Jiang
Abstract: Accurate fault detection in lithium-ion batteries is essential for the safe and reliable operation of electric vehicles and energy storage systems. However, existing methods often struggle to capture complex temporal dependencies and cannot fully leverage abundant unlabeled data. Although large language models (LLMs) exhibit strong representation capabilities, their architectures are not directly suited to the numerical time-series data common in industrial settings. To address these challenges, we propose a novel framework that adapts BERT-style pretraining for battery fault detection by extending the standard BERT architecture with a customized time-series-to-token representation module and a point-level Masked Signal Modeling (point-MSM) pretraining task tailored to battery applications. This approach enables self-supervised learning on sequential current, voltage, and other charge-discharge cycle data, yielding distributionally robust, context-aware temporal embeddings. We then concatenate these embeddings with battery metadata and feed them into a downstream classifier for accurate fault classification. Experimental results on a large-scale real-world dataset show that models initialized with our pretrained parameters significantly improve both representation quality and classification accuracy, achieving an AUROC of 0.945 and substantially outperforming existing approaches. These findings validate the effectiveness of BERT-style pretraining for time-series fault detection.
Authors: Michael T. M. B. Morris-Thomas, Marius Martens
Abstract: The real-time prediction of floating offshore asset behavior under stochastic metocean conditions remains a significant challenge in offshore engineering. While traditional empirical and frequency-domain methods work well in benign conditions, they struggle with both extreme sea states and nonlinear responses. This study presents a supervised machine learning approach using multivariate regression to predict the nonlinear motion response of a turret-moored vessel in 400 m water depth. We developed a machine learning workflow combining a gradient-boosted ensemble method with a custom passive weathervaning solver, trained on approximately $10^6$ samples spanning 100 features. The model achieved mean prediction errors of less than 5% for critical mooring parameters and vessel heading accuracy to within 2.5 degrees across diverse metocean conditions, significantly outperforming traditional frequency-domain methods. The framework has been successfully deployed on an operational facility, demonstrating its efficacy for real-time vessel monitoring and operational decision-making in offshore environments.
Authors: Andrew Kiruluta
Abstract: We propose an innovative, learnable two-sided short-time Laplace transform (STLT) mechanism to supplant the traditional self attention in transformer-based LLMs. Our STLT introduces trainable parameters for each Laplace node, enabling end-to-end learning of decay rates , oscillatory frequencies, and window bandwidth T. This flexibility allows the model to dynamically adapt token relevance half lives and frequency responses during training. By selecting S learnable nodes and leveraging fast recursive convolution, we achieve an effective complexity of in time and memory. We further incorporate an efficient FFT-based computation of the relevance matrix and an adaptive node allocation mechanism to dynamically adjust the number of active Laplace nodes. Empirical results on language modeling (WikiText\-103, Project Gutenberg), machine translation (WMT'14 En\-De), and long document question answering (NarrativeQA) demonstrate that our learnable STLT achieves perplexities and scores on par with or better than existing efficient transformers while naturally extending to context lengths exceeding 100k tokens or more limited only by available hardware. Ablation studies confirm the importance of learnable parameters and adaptive node allocation. The proposed approach combines interpretability, through explicit decay and frequency parameters, with scalability and robustness, offering a pathway towards ultra-long-sequence language modeling without the computational bottleneck of self-attention.
Authors: Hanyu Pei, Jing-Xiao Liao, Qibin Zhao, Ting Gao, Shijun Zhang, Xiaoge Zhang, Feng-Lei Fan
Abstract: Drawing inspiration from our human brain that designs different neurons for different tasks, recent advances in deep learning have explored modifying a network's neurons to develop so-called task-driven neurons. Prototyping task-driven neurons (referred to as NeuronSeek) employs symbolic regression (SR) to discover the optimal neuron formulation and construct a network from these optimized neurons. Along this direction, this work replaces symbolic regression with tensor decomposition (TD) to discover optimal neuronal formulations, offering enhanced stability and faster convergence. Furthermore, we establish theoretical guarantees that modifying the aggregation functions with common activation functions can empower a network with a fixed number of parameters to approximate any continuous function with an arbitrarily small error, providing a rigorous mathematical foundation for the NeuronSeek framework. Extensive empirical evaluations demonstrate that our NeuronSeek-TD framework not only achieves superior stability, but also is competitive relative to the state-of-the-art models across diverse benchmarks. The code is available at https://github.com/HanyuPei22/NeuronSeek.
Authors: Angelos Assos, Carmel Baharav, Bailey Flanigan, Ariel Procaccia
Abstract: An increasingly influential form of deliberative democracy centers on citizens' assemblies, where randomly selected people discuss policy questions. The legitimacy of these panels hinges on their representation of the broader population, but panelists often drop out, leading to an unbalanced composition. Although participant attrition is mitigated in practice by alternates, their selection is not taken into account by existing methods. To address this gap, we introduce an optimization framework for alternate selection. Our algorithmic approach, which leverages learning-theoretic machinery, estimates dropout probabilities using historical data and selects alternates to minimize expected misrepresentation. We establish theoretical guarantees for our approach, including worst-case bounds on sample complexity (with implications for computational efficiency) and on loss when panelists' probabilities of dropping out are mis-estimated. Empirical evaluation using real-world data demonstrates that, compared to the status quo, our method significantly improves representation while requiring fewer alternates.
Authors: Zhengze Zhang, Shiqi Wang, Yiqun Shen, Simin Guo, Dahua Lin, Xiaoliang Wang, Nguyen Cam-Tu, Fei Tan
Abstract: Large language models (LLMs) have demonstrated exceptional performance across various applications, but their conversational abilities decline sharply as model size decreases, presenting a barrier to their deployment in resource-constrained environments. Knowledge distillation with Direct Preference Optimization (dDPO) has emerged as a promising approach to enhancing the conversational abilities of smaller models using a larger teacher model. However, current methods primarily focus on 'black-box' KD, which only uses the teacher's responses, overlooking the output distribution offered by the teacher. This paper addresses this gap by introducing daDPO (Distribution-Aware DPO), a unified method for preference optimization and distribution-based distillation. We provide rigorous theoretical analysis and empirical validation, showing that daDPO outperforms existing methods in restoring performance for pruned models and enhancing smaller LLM models. Notably, in in-domain evaluation, our method enables a 20% pruned Vicuna1.5-7B to achieve near-teacher performance (-7.3% preference rate compared to that of dDPO's -31%), and allows Qwen2.5-1.5B to occasionally outperform its 7B teacher model (14.0% win rate).
Authors: Yu Guo, Hongji Fang, Tianyu Fang, Zhe Cui
Abstract: With the rise of artificial intelligence, the automatic generation of building-scale 3-D objects has become an active research topic, yet training such models still demands large, clean and richly annotated datasets. We introduce BuildingBRep-11K, a collection of 11 978 multi-storey (2-10 floors) buildings (about 10 GB) produced by a shape-grammar-driven pipeline that encodes established building-design principles. Every sample consists of a geometrically exact B-rep solid-covering floors, walls, slabs and rule-based openings-together with a fast-loading .npy metadata file that records detailed per-floor parameters. The generator incorporates constraints on spatial scale, daylight optimisation and interior layout, and the resulting objects pass multi-stage filters that remove Boolean failures, undersized rooms and extreme aspect ratios, ensuring compliance with architectural standards. To verify the dataset's learnability we trained two lightweight PointNet baselines. (i) Multi-attribute regression. A single encoder predicts storey count, total rooms, per-storey vector and mean room area from a 4 000-point cloud. On 100 unseen buildings it attains 0.37-storey MAE (87 \% within $\pm1$), 5.7-room MAE, and 3.2 m$^2$ MAE on mean area. (ii) Defect detection. With the same backbone we classify GOOD versus DEFECT; on a balanced 100-model set the network reaches 54 \% accuracy, recalling 82 \% of true defects at 53 \% precision (41 TP, 9 FN, 37 FP, 13 TN). These pilots show that BuildingBRep-11K is learnable yet non-trivial for both geometric regression and topological quality assessment
Authors: Manal Rahal, Bestoun S. Ahmed, Roger Renstrom, Robert Stener, Albrecht Wurtz
Abstract: Heat pumps (HPs) have emerged as a cost-effective and clean technology for sustainable energy systems, but their efficiency in producing hot water remains restricted by conventional threshold-based control methods. Although machine learning (ML) has been successfully implemented for various HP applications, optimization of household hot water demand forecasting remains understudied. This paper addresses this problem by introducing a novel approach that combines predictive ML with anomaly detection to create adaptive hot water production strategies based on household-specific consumption patterns. Our key contributions include: (1) a composite approach combining ML and isolation forest (iForest) to forecast household demand for hot water and steer responsive HP operations; (2) multi-step feature selection with advanced time-series analysis to capture complex usage patterns; (3) application and tuning of three ML models: Light Gradient Boosting Machine (LightGBM), Long Short-Term Memory (LSTM), and Bi-directional LSTM with the self-attention mechanism on data from different types of real HP installations; and (4) experimental validation on six real household installations. Our experiments show that the best-performing model LightGBM achieves superior performance, with RMSE improvements of up to 9.37\% compared to LSTM variants with $R^2$ values between 0.748-0.983. For anomaly detection, our iForest implementation achieved an F1-score of 0.87 with a false alarm rate of only 5.2\%, demonstrating strong generalization capabilities across different household types and consumption patterns, making it suitable for real-world HP deployments.
Authors: Juntae Lee, Munawar Hayat, Sungrack Yun
Abstract: Few-shot class incremental learning (FSCIL) enables the continual learning of new concepts with only a few training examples. In FSCIL, the model undergoes substantial updates, making it prone to forgetting previous concepts and overfitting to the limited new examples. Most recent trend is typically to disentangle the learning of the representation from the classification head of the model. A well-generalized feature extractor on the base classes (many examples and many classes) is learned, and then fixed during incremental learning. Arguing that the fixed feature extractor restricts the model's adaptability to new classes, we introduce a novel FSCIL method to effectively address catastrophic forgetting and overfitting issues. Our method enables to seamlessly update the entire model with a few examples. We mainly propose a tripartite weight-space ensemble (Tri-WE). Tri-WE interpolates the base, immediately previous, and current models in weight-space, especially for the classification heads of the models. Then, it collaboratively maintains knowledge from the base and previous models. In addition, we recognize the challenges of distilling generalized representations from the previous model from scarce data. Hence, we suggest a regularization loss term using amplified data knowledge distillation. Simply intermixing the few-shot data, we can produce richer data enabling the distillation of critical knowledge from the previous model. Consequently, we attain state-of-the-art results on the miniImageNet, CUB200, and CIFAR100 datasets.
Authors: Junqi Gao, Zhichang Guo, Dazhi Zhang, Dong Li, Runze Liu, Pengfei Li, Kai Tian, Biqing Qi
Abstract: Heterogeneous Large Language Model (LLM) fusion integrates the strengths of multiple source LLMs with different architectures into a target LLM with low computational overhead. While promising, existing methods suffer from two major limitations: 1) reliance on real data from limited domain for knowledge fusion, preventing the target LLM from fully acquiring knowledge across diverse domains, and 2) fixed data allocation proportions across domains, failing to dynamically adjust according to the target LLM's varying capabilities across domains, leading to a capability imbalance. To overcome these limitations, we propose Bohdi, a synthetic-data-only heterogeneous LLM fusion framework. Through the organization of knowledge domains into a hierarchical tree structure, Bohdi enables automatic domain exploration and multi-domain data generation through multi-model collaboration, thereby comprehensively extracting knowledge from source LLMs. By formalizing domain expansion and data sampling proportion allocation on the knowledge tree as a Hierarchical Multi-Armed Bandit problem, Bohdi leverages the designed DynaBranches mechanism to adaptively adjust sampling proportions based on the target LLM's performance feedback across domains. Integrated with our proposed Introspection-Rebirth (IR) mechanism, DynaBranches dynamically tracks capability shifts during target LLM's updates via Sliding Window Binomial Likelihood Ratio Testing (SWBLRT), further enhancing its online adaptation capability. Comparative experimental results on a comprehensive suite of benchmarks demonstrate that Bohdi significantly outperforms existing baselines on multiple target LLMs, exhibits higher data efficiency, and virtually eliminates the imbalance in the target LLM's capabilities. Our code is available at https://github.com/gjq100/Bohdi.git.
Authors: Wangzhi Zhan, Jianpeng Chen, Dongqi Fu, Dawei Zhou
Abstract: Metamaterials are artificial materials that are designed to meet unseen properties in nature, such as ultra-stiffness and negative materials indices. In mechanical metamaterial design, three key modalities are typically involved, i.e., 3D topology, density condition, and mechanical property. Real-world complex application scenarios place the demanding requirements on machine learning models to consider all three modalities together. However, a comprehensive literature review indicates that most existing works only consider two modalities, e.g., predicting mechanical properties given the 3D topology or generating 3D topology given the required properties. Therefore, there is still a significant gap for the state-of-the-art machine learning models capturing the whole. Hence, we propose a unified model named UNIMATE, which consists of a modality alignment module and a synergetic diffusion generation module. Experiments indicate that UNIMATE outperforms the other baseline models in topology generation task, property prediction task, and condition confirmation task by up to 80.2%, 5.1%, and 50.2%, respectively. We opensource our proposed UNIMATE model and corresponding results at https://github.com/wzhan24/UniMate.
Authors: Kunxi Li, Zhonghua Jiang, Zhouzhou Shen, Zhaode Wang, Chengfei Lv, Shengyu Zhang, Fan Wu, Fei Wu
Abstract: This paper introduces MadaKV, a modality-adaptive key-value (KV) cache eviction strategy designed to enhance the efficiency of multimodal large language models (MLLMs) in long-context inference. In multimodal scenarios, attention heads exhibit varying preferences for different modalities, resulting in significant disparities in modality importance across attention heads. Traditional KV cache eviction methods, which are tailored for unimodal settings, fail to capture modality-specific information, thereby yielding suboptimal performance. MadaKV addresses these challenges through two key components: modality preference adaptation and hierarchical compression compensation. By dynamically sensing modality information within attention heads and adaptively retaining critical tokens, MadaKV achieves substantial reductions in KV cache memory footprint and model inference decoding latency (1.3 to 1.5 times improvement) while maintaining high accuracy across various multimodal long-context tasks. Extensive experiments on representative MLLMs and the MileBench benchmark demonstrate the effectiveness of MadaKV compared to existing KV cache eviction methods.
Authors: Matteo Ninniri, Marco Podda, Davide Bacciu
Abstract: Generative models of graphs based on discrete Denoising Diffusion Probabilistic Models (DDPMs) offer a principled approach to molecular generation by systematically removing structural noise through iterative atom and bond adjustments. However, existing formulations are fundamentally limited by their inability to adapt the graph size (that is, the number of atoms) during the diffusion process, severely restricting their effectiveness in conditional generation scenarios such as property-driven molecular design, where the targeted property often correlates with the molecular size. In this paper, we reformulate the noising and denoising processes to support monotonic insertion and deletion of nodes. The resulting model, which we call GrIDDD, dynamically grows or shrinks the chemical graph during generation. GrIDDD matches or exceeds the performance of existing graph diffusion models on molecular property targeting despite being trained on a more difficult problem. Furthermore, when applied to molecular optimization, GrIDDD exhibits competitive performance compared to specialized optimization models. This work paves the way for size-adaptive molecular generation with graph diffusion.
Authors: Jackson Burns, Akshat Zalte, William Green
Abstract: Fast and accurate prediction of molecular properties with machine learning is pivotal to scientific advancements across myriad domains. Foundation models in particular have proven especially effective, enabling accurate training on small, real-world datasets. This study introduces CheMeleon, a novel molecular foundation model pre-trained on deterministic molecular descriptors from the Mordred package, leveraging a Directed Message-Passing Neural Network to predict these descriptors in a noise-free setting. Unlike conventional approaches relying on noisy experimental data or biased quantum mechanical simulations, CheMeleon uses low-noise molecular descriptors to learn rich molecular representations. Evaluated on 58 benchmark datasets from Polaris and MoleculeACE, CheMeleon achieves a win rate of 79% on Polaris tasks, outperforming baselines like Random Forest (46%), fastprop (39%), and Chemprop (36%), and a 97% win rate on MoleculeACE assays, surpassing Random Forest (63%) and other foundation models. However, it struggles to distinguish activity cliffs like many of the tested models. The t-SNE projection of CheMeleon's learned representations demonstrates effective separation of chemical series, highlighting its ability to capture structural nuances. These results underscore the potential of descriptor-based pre-training for scalable and effective molecular property prediction, opening avenues for further exploration of descriptor sets and unlabeled datasets.
Authors: Deyi Li, Zijun Yao, Muxuan Liang, Mei Liu
Abstract: In recent years, graph learning has gained significant interest for modeling complex interactions among medical events in structured Electronic Health Record (EHR) data. However, existing graph-based approaches often work in a static manner, either restricting interactions within individual encounters or collapsing all historical encounters into a single snapshot. As a result, when it is necessary to identify meaningful groups of medical events spanning longitudinal encounters, existing methods are inadequate in modeling interactions cross encounters while accounting for temporal dependencies. To address this limitation, we introduce Deep Patient Journey (DeepJ), a novel graph convolutional transformer model with differentiable graph pooling to effectively capture intra-encounter and inter-encounter medical event interactions. DeepJ can identify groups of temporally and functionally related medical events, offering valuable insights into key event clusters pertinent to patient outcome prediction. DeepJ significantly outperformed five state-of-the-art baseline models while enhancing interpretability, demonstrating its potential for improved patient risk stratification.
Authors: Jason Tandiary
Abstract: This paper studies Vickrey first-price auctions under binary feedback. Leveraging the enhanced performance of machine learning algorithms, the new algorithm uses past information to improve the regret bounds of the BROAD-OMD algorithm. Motivated by the growing relevance of first-price auctions and the predictive capabilities of machine learning models, this paper proposes a new algorithm within the BROAD-OMD framework (Hu et al., 2025) that leverages predictions of the highest competing bid. This paper's main contribution is an algorithm that achieves zero regret under accurate predictions. Additionally, a bounded regret bound of O(T^(3/4) * Vt^(1/4)) is established under certain normality conditions.
Authors: Chiara Razzetta, Shahryar Noei, Federico Barbarossa, Edoardo Spairani, Monica Roascio, Elisa Barbi, Giulia Ciacci, Sara Sommariva, Sabrina Guastavino, Michele Piana, Matteo Lenge, Gabriele Arnulfo, Giovanni Magenes, Elvira Maranesi, Giulio Amabili, Anna Maria Massone, Federico Benvenuto, Giuseppe Jurman, Diego Sona, Cristina Campi
Abstract: "DHEAL-COM - Digital Health Solutions in Community Medicine" is a research and technology project funded by the Italian Department of Health for the development of digital solutions of interest in proximity healthcare. The activity within the DHEAL-COM framework allows scientists to gather a notable amount of multi-modal data whose interpretation can be performed by means of machine learning algorithms. The present study illustrates a general automated pipeline made of numerous unsupervised and supervised methods that can ingest such data, provide predictive results, and facilitate model interpretations via feature identification.
Authors: Luiz Pereira, M. Hadi Amini
Abstract: In this paper, we first propose a novel algorithm for model fusion that leverages Wasserstein barycenters in training a global Deep Neural Network (DNN) in a distributed architecture. To this end, we divide the dataset into equal parts that are fed to "agents" who have identical deep neural networks and train only over the dataset fed to them (known as the local dataset). After some training iterations, we perform an aggregation step where we combine the weight parameters of all neural networks using Wasserstein barycenters. These steps form the proposed algorithm referred to as FedWB. Moreover, we leverage the processes created in the first part of the paper to develop an algorithm to tackle Heterogeneous Federated Reinforcement Learning (HFRL). Our test experiment is the CartPole toy problem, where we vary the lengths of the poles to create heterogeneous environments. We train a deep Q-Network (DQN) in each environment to learn to control each cart, while occasionally performing a global aggregation step to generalize the local models; the end outcome is a global DQN that functions across all environments.
Authors: Kevin Yin, Julia Gersey, Pei Zhang
Abstract: Effective large-scale air quality monitoring necessitates distributed sensing due to the pervasive and harmful nature of particulate matter (PM), particularly in urban environments. However, precision comes at a cost: highly accurate sensors are expensive, limiting the spatial deployments and thus their coverage. As a result, low-cost sensors have become popular, though they are prone to drift caused by environmental sensitivity and manufacturing variability. This paper presents a model for in-field sensor calibration using XGBoost ensemble learning to consolidate data from neighboring sensors. This approach reduces dependence on the presumed accuracy of individual sensors and improves generalization across different locations.
Authors: Pedro Mendes, Paolo Romano, David Garlan
Abstract: Modern neural networks (NNs) often achieve high predictive accuracy but remain poorly calibrated, producing overconfident predictions even when wrong. This miscalibration poses serious challenges in applications where reliable uncertainty estimates are critical. In this work, we investigate how human perceptual uncertainty compares to uncertainty estimated by NNs. Using three vision benchmarks annotated with both human disagreement and crowdsourced confidence, we assess the correlation between model-predicted uncertainty and human-perceived uncertainty. Our results show that current methods only weakly align with human intuition, with correlations varying significantly across tasks and uncertainty metrics. Notably, we find that incorporating human-derived soft labels into the training process can improve calibration without compromising accuracy. These findings reveal a persistent gap between model and human uncertainty and highlight the potential of leveraging human insights to guide the development of more trustworthy AI systems.
Authors: Xixi Hu, Runlong Liao, Keyang Xu, Bo Liu, Yeqing Li, Eugene Ie, Hongliang Fei, Qiang Liu
Abstract: Rectified Flow offers a simple and effective approach to high-quality generative modeling by learning a velocity field. However, we identify a limitation in directly modeling the velocity with an unconstrained neural network: the learned velocity often fails to satisfy certain boundary conditions, leading to inaccurate velocity field estimations that deviate from the desired ODE. This issue is particularly critical during stochastic sampling at inference, as the score function's errors are amplified near the boundary. To mitigate this, we propose a Boundary-enforced Rectified Flow Model (Boundary RF Model), in which we enforce boundary conditions with a minimal code modification. Boundary RF Model improves performance over vanilla RF model, demonstrating 8.01% improvement in FID score on ImageNet using ODE sampling and 8.98% improvement using SDE sampling.
Authors: Sara Kangaslahti, Elan Rosenfeld, Naomi Saphra
Abstract: Loss curves are smooth during most of model training, so visible discontinuities stand out as possible conceptual breakthroughs. Studying these breakthroughs enables a deeper understanding of learning dynamics, but only when they are properly identified. This paper argues that similar breakthroughs occur frequently throughout training but they are obscured by a loss metric that collapses all variation into a single scalar. To find these hidden transitions, we introduce POLCA, a method for decomposing changes in loss along arbitrary bases of the low-rank training subspace. We use our method to identify clusters of samples that share similar changes in loss during training, disaggregating the overall loss into that of smaller groups of conceptually similar data. We validate our method on synthetic arithmetic and natural language tasks, showing that POLCA recovers clusters that represent interpretable breakthroughs in the model's capabilities. We demonstrate the promise of these hidden phase transitions as a tool for unsupervised interpretability.
Authors: Abdel Rahman Alsheyab (Jordan University of Science and Technology, Irbid, Jordan), Mohammad Alkhasawneh (Jordan University of Science and Technology, Irbid, Jordan), Nidal Shahin (Jordan University of Science and Technology, Irbid, Jordan)
Abstract: This paper presents a machine learning methodology prototype using a large synthetic dataset of job listings to identify trends, predict salaries, and group similar job roles. Employing techniques such as regression, classification, clustering, and natural language processing (NLP) for text-based feature extraction and representation, this study aims to uncover the key features influencing job market dynamics and provide valuable insights for job seekers, employers, and researchers. Exploratory data analysis was conducted to understand the dataset's characteristics. Subsequently, regression models were developed to predict salaries, classification models to predict job titles, and clustering techniques were applied to group similar jobs. The analyses revealed significant factors influencing salary and job roles, and identified distinct job clusters based on the provided data. While the results are based on synthetic data and not intended for real-world deployment, the methodology demonstrates a transferable framework for job market analysis.
Authors: Alexey Yermakov, David Zoro, Mars Liyao Gao, J. Nathan Kutz
Abstract: SHallow REcurrent Decoders (SHRED) are effective for system identification and forecasting from sparse sensor measurements. Such models are light-weight and computationally efficient, allowing them to be trained on consumer laptops. SHRED-based models rely on Recurrent Neural Networks (RNNs) and a simple Multi-Layer Perceptron (MLP) for the temporal encoding and spatial decoding respectively. Despite the relatively simple structure of SHRED, they are able to predict chaotic dynamical systems on different physical, spatial, and temporal scales directly from a sparse set of sensor measurements. In this work, we improve SHRED by leveraging transformers (T-SHRED) for the temporal encoding which improves performance on next-step state prediction on large datasets. We also introduce a sparse identification of nonlinear dynamics (SINDy) attention mechanism into T-SHRED to perform symbolic regression directly on the latent space as part of the model regularization architecture. Symbolic regression improves model interpretability by learning and regularizing the dynamics of the latent space during training. We analyze the performance of T-SHRED on three different dynamical systems ranging from low-data to high-data regimes. We observe that SINDy attention T-SHRED accurately predicts future frames based on an interpretable symbolic model across all tested datasets.
Authors: Sheng Liu, Tianlang Chen, Pan Lu, Haotian Ye, Yizheng Chen, Lei Xing, James Zou
Abstract: Test-time compute has emerged as a powerful paradigm for improving the performance of large language models (LLMs), where generating multiple outputs or refining individual chains can significantly boost answer accuracy. However, existing methods like Best-of-N, majority voting, and self-reflection typically apply reasoning in a uniform way across inputs, overlooking the fact that different problems may require different levels of reasoning depth. In this work, we propose Fractional Reasoning, a training-free and model-agnostic framework that enables continuous control over reasoning intensity at inference time, going beyond the limitations of fixed instructional prompts. Our method operates by extracting the latent steering vector associated with deeper reasoning and reapplying it with a tunable scaling factor, allowing the model to tailor its reasoning process to the complexity of each input. This supports two key modes of test-time scaling: (1) improving output quality in breadth-based strategies (e.g., Best-of-N, majority voting), and (2) enhancing the correctness of individual reasoning chains in depth-based strategies (e.g., self-reflection). Experiments on GSM8K, MATH500, and GPQA demonstrate that Fractional Reasoning consistently improves performance across diverse reasoning tasks and models.
Authors: Farnam Mansouri, Hans U. Simon, Adish Singla, Yuxin Chen, Sandra Zilles
Abstract: Machine learning can greatly benefit from providing learning algorithms with pairs of contrastive training examples -- typically pairs of instances that differ only slightly, yet have different class labels. Intuitively, the difference in the instances helps explain the difference in the class labels. This paper proposes a theoretical framework in which the effect of various types of contrastive examples on active learners is studied formally. The focus is on the sample complexity of learning concept classes and how it is influenced by the choice of contrastive examples. We illustrate our results with geometric concept classes and classes of Boolean functions. Interestingly, we reveal a connection between learning from contrastive examples and the classical model of self-directed learning.
Authors: Yu Zhang, Gaoshan Bi, Simon Jeffery, Max Davis, Yang Li, Qing Xue, Po Yang
Abstract: Precision soil greenhouse gas (GHG) flux prediction is essential in agricultural systems for assessing environmental impacts, developing emission mitigation strategies and promoting sustainable agriculture. Due to the lack of advanced sensor and network technologies on majority of farms, there are challenges in obtaining comprehensive and diverse agricultural data. As a result, the scarcity of agricultural data seriously obstructs the application of machine learning approaches in precision soil GHG flux prediction. This research proposes a knowledge-guided graph neural network framework that addresses the above challenges by integrating knowledge embedded in an agricultural process-based model and graph neural network techniques. Specifically, we utilise the agricultural process-based model to simulate and generate multi-dimensional agricultural datasets for 47 countries that cover a wide range of agricultural variables. To extract key agricultural features and integrate correlations among agricultural features in the prediction process, we propose a machine learning framework that integrates the autoencoder and multi-target multi-graph based graph neural networks, which utilises the autoencoder to selectively extract significant agricultural features from the agricultural process-based model simulation data and the graph neural network to integrate correlations among agricultural features for accurately predict fertilisation-oriented soil GHG fluxes. Comprehensive experiments were conducted with both the agricultural simulation dataset and real-world agricultural dataset to evaluate the proposed approach in comparison with well-known baseline and state-of-the-art regression methods. The results demonstrate that our proposed approach provides superior accuracy and stability in fertilisation-oriented soil GHG prediction.
Authors: Xiao Zhang, Xingyu Zhao, Hong Xia, Yuan Cao, Guiyuan Jiang, Junyu Dong, Yanwei Yu
Abstract: With the proliferation of location-tracking technologies, massive volumes of trajectory data are continuously being collected. As a fundamental task in trajectory data mining, trajectory similarity computation plays a critical role in a wide range of real-world applications. However, existing learning-based methods face three challenges: First, they ignore the semantic gap between GPS and grid features in trajectories, making it difficult to obtain meaningful trajectory embeddings. Second, the noise inherent in the trajectories, as well as the noise introduced during grid discretization, obscures the true motion patterns of the trajectories. Third, existing methods focus solely on point-wise and pair-wise losses, without utilizing the global ranking information obtained by sorting all trajectories according to their similarity to a given trajectory. To address the aforementioned challenges, we propose a novel trajectory similarity computation framework, named TrajDiff. Specifically, the semantic alignment module relies on cross-attention and an attention score mask mechanism with adaptive fusion, effectively eliminating semantic discrepancies between data at two scales and generating a unified representation. Additionally, the DDBM-based Noise-robust Pre-Training introduces the transfer patterns between any two trajectories into the model training process, enhancing the model's noise robustness. Finally, the overall ranking-aware regularization shifts the model's focus from a local to a global perspective, enabling it to capture the holistic ordering information among trajectories. Extensive experiments on three publicly available datasets show that TrajDiff consistently outperforms state-of-the-art baselines. In particular, it achieves an average HR@1 gain of 33.38% across all three evaluation metrics and datasets.
Authors: Li Sun, Shuheng Chen, Yong Si, Junyi Fan, Maryam Pishgar, Elham Pishgar, Kamiar Alaei, Greg Placencia
Abstract: Background: Patients with both diabetes mellitus (DM) and atrial fibrillation (AF) face elevated mortality in intensive care units (ICUs), yet models targeting this high-risk group remain limited. Objective: To develop an interpretable machine learning (ML) model predicting 28-day mortality in ICU patients with concurrent DM and AF using early-phase clinical data. Methods: A retrospective cohort of 1,535 adult ICU patients with DM and AF was extracted from the MIMIC-IV database. Data preprocessing involved median/mode imputation, z-score normalization, and early temporal feature engineering. A two-step feature selection pipeline-univariate filtering (ANOVA F-test) and Random Forest-based multivariate ranking-yielded 19 interpretable features. Seven ML models were trained with stratified 5-fold cross-validation and SMOTE oversampling. Interpretability was assessed via ablation and Accumulated Local Effects (ALE) analysis. Results: Logistic regression achieved the best performance (AUROC: 0.825; 95% CI: 0.779-0.867), surpassing more complex models. Key predictors included RAS, age, bilirubin, and extubation. ALE plots showed intuitive, non-linear effects such as age-related risk acceleration and bilirubin thresholds. Conclusion: This interpretable ML model offers accurate risk prediction and clinical insights for early ICU triage in patients with DM and AF.
Authors: Josef Kucha\v{r}, Marek Kadl\v{c}\'ik, Michal Spiegel, Michal \v{S}tef\'anik
Abstract: We introduce a large-scale dataset for instruction-guided vector image editing, consisting of over 270,000 pairs of SVG images paired with natural language edit instructions. Our dataset enables training and evaluation of models that modify vector graphics based on textual commands. We describe the data collection process, including image pairing via CLIP similarity and instruction generation with vision-language models. Initial experiments with state-of-the-art large language models reveal that current methods struggle to produce accurate and valid edits, underscoring the challenge of this task. To foster research in natural language-driven vector graphic generation and editing, we make our resources created within this work publicly available.
Authors: Hang Yang (Callie), Yusheng Hu (Callie), Yong Liu (Callie), Cong (Callie), Hao
Abstract: Accurate graph similarity is critical for knowledge transfer in VLSI design, enabling the reuse of prior solutions to reduce engineering effort and turnaround time. We propose Pieceformer, a scalable, self-supervised similarity assessment framework, equipped with a hybrid message-passing and graph transformer encoder. To address transformer scalability, we incorporate a linear transformer backbone and introduce a partitioned training pipeline for efficient memory and parallelism management. Evaluations on synthetic and real-world CircuitNet datasets show that Pieceformer reduces mean absolute error (MAE) by 24.9% over the baseline and is the only method to correctly cluster all real-world design groups. We further demonstrate the practical usage of our model through a case study on a partitioning task, achieving up to 89% runtime reduction. These results validate the framework's effectiveness for scalable, unbiased design reuse in modern VLSI systems.
Authors: Zifei Xu, Sayeh Sharify, Hesham Mostafa, Tristan Webb, Wanzin Yazar, Xin Wang
Abstract: Transformer-based neural speech processing has achieved state-of-the-art performance. Since speech audio signals are known to be highly compressible, here we seek to accelerate neural speech transcription by time-domain signal sparsification early in the neural encoding stage, taking advantage of the interpretability of the self-attention mechanism in transformer audio encoders. With the Whisper family of models, we perform a systematic architecture search over the joint space of sparsification stage (a certain encoder layer) and compression ratio (sparsity). We found that the best resulting solutions under 1% accuracy degradation choose to sparsify the hidden state to 40-60% sparsity at an early encoding stage, and thereby achieve up to 1.6x runtime acceleration in English speech transcription tasks on Nvidia GPUs without any fine-tuning.
Authors: Liangyan Li, Yangyi Liu, Yimo Ning, Stefano Rini, Jun Chen
Abstract: Federated Learning (FL) has emerged as a powerful paradigm for leveraging diverse datasets from multiple sources while preserving data privacy by avoiding centralized storage. However, many existing approaches fail to account for the intricate gradient correlations between remote clients, a limitation that becomes especially problematic in data heterogeneity scenarios. In this work, we propose a novel FL framework utilizing Power-Norm Cosine Similarity (PNCS) to improve client selection for model aggregation. By capturing higher-order gradient moments, PNCS addresses non-IID data challenges, enhancing convergence speed and accuracy. Additionally, we introduce a simple algorithm ensuring diverse client selection through a selection history queue. Experiments with a VGG16 model across varied data partitions demonstrate consistent improvements over state-of-the-art methods.
Authors: Soumya Basu
Abstract: We study bandit learning in matching markets with two-sided reward uncertainty, extending prior research primarily focused on single-sided uncertainty. Leveraging the concept of `super-stability' from Irving (1994), we demonstrate the advantage of the Extended Gale-Shapley (GS) algorithm over the standard GS algorithm in achieving true stable matchings under incomplete information. By employing the Extended GS algorithm, our centralized algorithm attains a logarithmic pessimal stable regret dependent on an instance-dependent admissible gap parameter. This algorithm is further adapted to a decentralized setting with a constant regret increase. Finally, we establish a novel centralized instance-dependent lower bound for binary stable regret, elucidating the roles of the admissible gap and super-stable matching in characterizing the complexity of stable matching with bandit feedback.
Authors: Esther Rodriguez, Monica Welfert, Samuel McDowell, Nathan Stromberg, Julian Antolin Camarena, Lalitha Sankar
Abstract: Diffusion models have achieved impressive performance in generating high-quality and diverse synthetic data. However, their success typically assumes a class-balanced training distribution. In real-world settings, multi-class data often follow a long-tailed distribution, where standard diffusion models struggle -- producing low-diversity and lower-quality samples for tail classes. While this degradation is well-documented, its underlying cause remains poorly understood. In this work, we investigate the behavior of diffusion models trained on long-tailed datasets and identify a key issue: the latent representations (from the bottleneck layer of the U-Net) for tail class subspaces exhibit significant overlap with those of head classes, leading to feature borrowing and poor generation quality. Importantly, we show that this is not merely due to limited data per class, but that the relative class imbalance significantly contributes to this phenomenon. To address this, we propose COntrastive Regularization for Aligning Latents (CORAL), a contrastive latent alignment framework that leverages supervised contrastive losses to encourage well-separated latent class representations. Experiments demonstrate that CORAL significantly improves both the diversity and visual quality of samples generated for tail classes relative to state-of-the-art methods.
Authors: Bruce Huang, Ruida Zhou, Lin F. Yang, Suhas Diggavi
Abstract: Stochastic linear bandits are a fundamental model for sequential decision making, where an agent selects a vector-valued action and receives a noisy reward with expected value given by an unknown linear function. Although well studied in the single-agent setting, many real-world scenarios involve multiple agents solving heterogeneous bandit problems, each with a different unknown parameter. Applying single agent algorithms independently ignores cross-agent similarity and learning opportunities. This paper investigates the optimal regret achievable in collaborative personalized linear bandits. We provide an information-theoretic lower bound that characterizes how the number of agents, the interaction rounds, and the degree of heterogeneity jointly affect regret. We then propose a new two-stage collaborative algorithm that achieves the optimal regret. Our analysis models heterogeneity via a hierarchical Bayesian framework and introduces a novel information-theoretic technique for bounding regret. Our results offer a complete characterization of when and how collaboration helps with a optimal regret bound $\tilde{O}(d\sqrt{mn})$, $\tilde{O}(dm^{1-\gamma}\sqrt{n})$, $\tilde{O}(dm\sqrt{n})$ for the number of rounds $n$ in the range of $(0, \frac{d}{m \sigma^2})$, $[\frac{d}{m^{2\gamma} \sigma^2}, \frac{d}{\sigma^2}]$ and $(\frac{d}{\sigma^2}, \infty)$ respectively, where $\sigma$ measures the level of heterogeneity, $m$ is the number of agents, and $\gamma\in[0, 1/2]$ is an absolute constant. In contrast, agents without collaboration achieve a regret bound $O(dm\sqrt{n})$ at best.
Authors: Vinicius Yuiti Fukase, Heitor Gama, Barbara Bueno, Lucas Libanio, Anna Helena Reali Costa, Artur Jordao
Abstract: Critical Learning Periods comprehend an important phenomenon involving deep learning, where early epochs play a decisive role in the success of many training recipes, such as data augmentation. Existing works confirm the existence of this phenomenon and provide useful insights. However, the literature lacks efforts to precisely identify when critical periods occur. In this work, we fill this gap by introducing a systematic approach for identifying critical periods during the training of deep neural networks, focusing on eliminating computationally intensive regularization techniques and effectively applying mechanisms for reducing computational costs, such as data pruning. Our method leverages generalization prediction mechanisms to pinpoint critical phases where training recipes yield maximum benefits to the predictive ability of models. By halting resource-intensive recipes beyond these periods, we significantly accelerate the learning phase and achieve reductions in training time, energy consumption, and CO$_2$ emissions. Experiments on standard architectures and benchmarks confirm the effectiveness of our method. Specifically, we achieve significant milestones by reducing the training time of popular architectures by up to 59.67%, leading to a 59.47% decrease in CO$_2$ emissions and a 60% reduction in financial costs, without compromising performance. Our work enhances understanding of training dynamics and paves the way for more sustainable and efficient deep learning practices, particularly in resource-constrained environments. In the era of the race for foundation models, we believe our method emerges as a valuable framework. The repository is available at https://github.com/baunilhamarga/critical-periods
Authors: Jingyi Cui, Qi Zhang, Yifei Wang, Yisen Wang
Abstract: Sparse autoencoders (SAEs) have emerged as a powerful tool for interpreting features learned by large language models (LLMs). It aims to recover complex superposed polysemantic features into interpretable monosemantic ones through feature reconstruction via sparsely activated neural networks. Despite the wide applications of SAEs, it remains unclear under what conditions an SAE can fully recover the ground truth monosemantic features from the superposed polysemantic ones. In this paper, through theoretical analysis, we for the first time propose the necessary and sufficient conditions for identifiable SAEs (SAEs that learn unique and ground truth monosemantic features), including 1) extreme sparsity of the ground truth feature, 2) sparse activation of SAEs, and 3) enough hidden dimensions of SAEs. Moreover, when the identifiable conditions are not fully met, we propose a reweighting strategy to improve the identifiability. Specifically, following the theoretically suggested weight selection principle, we prove that the gap between the loss functions of SAE reconstruction and monosemantic feature reconstruction can be narrowed, so that the reweighted SAEs have better reconstruction of the ground truth monosemantic features than the uniformly weighted ones. In experiments, we validate our theoretical findings and show that our weighted SAE significantly improves feature monosemanticity and interpretability.
Authors: Haoyue Zhang, Hualei Zhang, Xiaosong Ma, Jie Zhang, Song Guo
Abstract: Large Language Models (LLMs) exhibit enhanced reasoning capabilities by employing Chain-of-Thought (CoT). However, the extended reasoning sequences introduce significant GPU memory overhead due to increased key-value (KV) cache size, particularly in tasks requiring long reasoning sequences, such as mathematics and programming. Existing KV cache compression methods mitigate memory bottlenecks but struggle in long reasoning tasks. In this paper, we analyze attention patterns in reasoning tasks and reveal a Token Importance Recurrence phenomenon: a large proportion of tokens receive renewed attention after multiple decoding steps, which is failed to capture by existing works and may lead to unpredictable eviction on such periodically critical tokens. To address this, we propose LazyEviction, a lagged KV eviction framework designed to maintain reasoning performance while reducing KV memory. LazyEviction is an Observation Window-based Lagged Eviction Mechanism retaining latent recurring tokens by performing lagged evictions across decoding steps, which contains two key components: (1) Recurrence Interval Tracking for capturing temporal variations in token importance, and (2) an Maximum Recurrence Interval-Centric Eviction Policy that prioritizes eviction based on tokens' recurrence patterns. Extensive experiments demonstrate that LazyEviction reduces KV cache size by 50% while maintaining comparable accuracy on mathematics reasoning datasets, outperforming state-of-the-art methods. Our findings highlight the importance of preserving recurring tokens, which are critical for maintaining knowledge continuity in multi-step reasoning tasks.
Authors: Qianru Zhang, Honggang Wen, Ming Li, Dong Huang, Siu-Ming Yiu, Christian S. Jensen, Pietro Li\`o
Abstract: Time series forecasting requires architectures that simultaneously achieve three competing objectives: (1) strict temporal causality for reliable predictions, (2) sub-quadratic complexity for practical scalability, and (3) multi-scale pattern recognition for accurate long-horizon forecasting. We introduce AutoHFormer, a hierarchical autoregressive transformer that addresses these challenges through three key innovations: 1) Hierarchical Temporal Modeling: Our architecture decomposes predictions into segment-level blocks processed in parallel, followed by intra-segment sequential refinement. This dual-scale approach maintains temporal coherence while enabling efficient computation. 2) Dynamic Windowed Attention: The attention mechanism employs learnable causal windows with exponential decay, reducing complexity while preserving precise temporal relationships. This design avoids both the anti-causal violations of standard transformers and the sequential bottlenecks of RNN hybrids. 3) Adaptive Temporal Encoding: a novel position encoding system is adopted to capture time patterns at multiple scales. It combines fixed oscillating patterns for short-term variations with learnable decay rates for long-term trends. Comprehensive experiments demonstrate that AutoHFormer 10.76X faster training and 6.06X memory reduction compared to PatchTST on PEMS08, while maintaining consistent accuracy across 96-720 step horizons in most of cases. These breakthroughs establish new benchmarks for efficient and precise time series modeling. Implementations of our method and all baselines in hierarchical autoregressive mechanism are available at https://github.com/lizzyhku/Autotime.
Authors: Hamdi Altaheri, Fakhri Karray, Md. Milon Islam, S M Taslim Uddin Raju, Amir-Hossein Karimi
Abstract: Foundation models (FMs), powered by self-supervised learning (SSL), have redefined the capabilities of artificial intelligence, demonstrating exceptional performance in domains like natural language processing and computer vision. These advances present a transformative opportunity for brain signal analysis. Unlike traditional supervised learning, which is limited by the scarcity of labeled neural data, SSL offers a promising solution by enabling models to learn meaningful representations from unlabeled data. This is particularly valuable in addressing the unique challenges of brain signals, including high noise levels, inter-subject variability, and low signal-to-noise ratios. This survey systematically reviews the emerging field of bridging brain signals with foundation models through the innovative application of SSL. It explores key SSL techniques, the development of brain-specific foundation models, their adaptation to downstream tasks, and the integration of brain signals with other modalities in multimodal SSL frameworks. The review also covers commonly used evaluation metrics and benchmark datasets that support comparative analysis. Finally, it highlights key challenges and outlines future research directions. This work aims to provide researchers with a structured understanding of this rapidly evolving field and a roadmap for developing generalizable brain foundation models powered by self-supervision.
Authors: Jina Kim, Youjin Jang, Jeongjin Han
Abstract: We propose VRAIL (Vectorized Reward-based Attribution for Interpretable Learning), a bi-level framework for value-based reinforcement learning (RL) that learns interpretable weight representations from state features. VRAIL consists of two stages: a deep learning (DL) stage that fits an estimated value function using state features, and an RL stage that uses this to shape learning via potential-based reward transformations. The estimator is modeled in either linear or quadratic form, allowing attribution of importance to individual features and their interactions. Empirical results on the Taxi-v3 environment demonstrate that VRAIL improves training stability and convergence compared to standard DQN, without requiring environment modifications. Further analysis shows that VRAIL uncovers semantically meaningful subgoals, such as passenger possession, highlighting its ability to produce human-interpretable behavior. Our findings suggest that VRAIL serves as a general, model-agnostic framework for reward shaping that enhances both learning and interpretability.
Authors: Zhen Qin, Michael B. Wakin, Zhihui Zhu
Abstract: Tensor decompositions, which represent an $N$-order tensor using approximately $N$ factors of much smaller dimensions, can significantly reduce the number of parameters. This is particularly beneficial for high-order tensors, as the number of entries in a tensor grows exponentially with the order. Consequently, they are widely used in signal recovery and data analysis across domains such as signal processing, machine learning, and quantum physics. A computationally and memory-efficient approach to these problems is to optimize directly over the factors using local search algorithms such as gradient descent, a strategy known as the factorization approach in matrix and tensor optimization. However, the resulting optimization problems are highly nonconvex due to the multiplicative interactions between factors, posing significant challenges for convergence analysis and recovery guarantees. In this paper, we present a unified framework for the factorization approach to solving various tensor decomposition problems. Specifically, by leveraging the canonical form of tensor decompositions--where most factors are constrained to be orthonormal to mitigate scaling ambiguity--we apply Riemannian gradient descent (RGD) to optimize these orthonormal factors on the Stiefel manifold. Under a mild condition on the loss function, we establish a Riemannian regularity condition for the factorized objective and prove that RGD converges to the ground-truth tensor at a linear rate when properly initialized. Notably, both the initialization requirement and the convergence rate scale polynomially rather than exponentially with $N$, improving upon existing results for Tucker and tensor-train format tensors.
Authors: Vishesh Tripathi, Tanmay Odapally, Indraneel Das, Uday Allu, Biddwan Ahmed
Abstract: Retrieval-Augmented Generation (RAG) systems have revolutionized information retrieval and question answering, but traditional text-based chunking methods struggle with complex document structures, multi-page tables, embedded figures, and contextual dependencies across page boundaries. We present a novel multimodal document chunking approach that leverages Large Multimodal Models (LMMs) to process PDF documents in batches while maintaining semantic coherence and structural integrity. Our method processes documents in configurable page batches with cross-batch context preservation, enabling accurate handling of tables spanning multiple pages, embedded visual elements, and procedural content. We evaluate our approach on a curated dataset of PDF documents with manually crafted queries, demonstrating improvements in chunk quality and downstream RAG performance. Our vision-guided approach achieves better accuracy compared to traditional vanilla RAG systems, with qualitative analysis showing superior preservation of document structure and semantic coherence.
Authors: Zhiwei Li, Carl Kesselman, Tran Huy Nguyen, Benjamin Yixing Xu, Kyle Bolo, Kimberley Yu
Abstract: Reproducibility remains a central challenge in machine learning (ML), especially in collaborative eScience projects where teams iterate over data, features, and models. Current ML workflows are often dynamic yet fragmented, relying on informal data sharing, ad hoc scripts, and loosely connected tools. This fragmentation impedes transparency, reproducibility, and the adaptability of experiments over time. This paper introduces a data-centric framework for lifecycle-aware reproducibility, centered around six structured artifacts: Dataset, Feature, Workflow, Execution, Asset, and Controlled Vocabulary. These artifacts formalize the relationships between data, code, and decisions, enabling ML experiments to be versioned, interpretable, and traceable over time. The approach is demonstrated through a clinical ML use case of glaucoma detection, illustrating how the system supports iterative exploration, improves reproducibility, and preserves the provenance of collaborative decisions across the ML lifecycle.
Authors: Puchun Liu, C. L. Philip Chen, Yubin He, Tong Zhang
Abstract: The difficulty of extracting deep features from EEG data and effectively integrating information from multiple views presents significant challenges for developing a generalizable pretraining framework for EEG representation learning. However, most existing pre-training methods rely solely on the contextual semantics of a single view, failing to capture the complex and synergistic interactions among different perspectives, limiting the expressiveness and generalization of learned representations. To address these issues, this paper proposes CRIA, an adaptive framework that utilizes variable-length and variable-channel coding to achieve a unified representation of EEG data across different datasets. In this work, we define cross-view information as the integrated representation that emerges from the interaction among temporal, spectral, and spatial views of EEG signals. The model employs a cross-attention mechanism to fuse temporal, spectral, and spatial features effectively, and combines an attention matrix masking strategy based on the information bottleneck principle with a novel viewpoint masking pre-training scheme. Experimental results on the Temple University EEG corpus and the CHB-MIT dataset show that CRIA outperforms existing methods with the same pre-training conditions, achieving a balanced accuracy of 57.02% for multi-class event classification and 80.03% for anomaly detection, highlighting its strong generalization ability.
Authors: Geonho Hwang, Wonyeol Lee, Yeachan Park, Sejun Park, Feras Saad
Abstract: The classical universal approximation (UA) theorem for neural networks establishes mild conditions under which a feedforward neural network can approximate a continuous function $f$ with arbitrary accuracy. A recent result shows that neural networks also enjoy a more general interval universal approximation (IUA) theorem, in the sense that the abstract interpretation semantics of the network using the interval domain can approximate the direct image map of $f$ (i.e., the result of applying $f$ to a set of inputs) with arbitrary accuracy. These theorems, however, rest on the unrealistic assumption that the neural network computes over infinitely precise real numbers, whereas their software implementations in practice compute over finite-precision floating-point numbers. An open question is whether the IUA theorem still holds in the floating-point setting. This paper introduces the first IUA theorem for floating-point neural networks that proves their remarkable ability to perfectly capture the direct image map of any rounded target function $f$, showing no limits exist on their expressiveness. Our IUA theorem in the floating-point setting exhibits material differences from the real-valued setting, which reflects the fundamental distinctions between these two computational models. This theorem also implies surprising corollaries, which include (i) the existence of provably robust floating-point neural networks; and (ii) the computational completeness of the class of straight-line programs that use only floating-point additions and multiplications for the class of all floating-point programs that halt.
Authors: Kexuan Wang, An Liu
Abstract: Weighted Minimum Mean Square Error (WMMSE) precoding is widely recognized for its near-optimal weighted sum rate performance. However, its practical deployment in massive multi-user (MU) multiple-input multiple-output (MIMO) orthogonal frequency-division multiplexing (OFDM) systems is hindered by the assumption of perfect channel state information (CSI) and high computational complexity. To address these issues, we first develop a wideband stochastic WMMSE (SWMMSE) algorithm that iteratively maximizes the ergodic weighted sum-rate (EWSR) under imperfect CSI. Building on this, we propose a lightweight reinforcement learning (RL)-driven deep unfolding (DU) network (RLDDU-Net), where each SWMMSE iteration is mapped to a network layer. Specifically, its DU module integrates approximation techniques and leverages beam-domain sparsity as well as frequency-domain subcarrier correlation, significantly accelerating convergence and reducing computational overhead. Furthermore, the RL module adaptively adjusts the network depth and generates compensation matrices to mitigate approximation errors. Simulation results under imperfect CSI demonstrate that RLDDU-Net outperforms existing baselines in EWSR performance while offering superior computational and convergence efficiency.
Authors: Kexuan Wang, An Liu
Abstract: 6G wireless networks are expected to support diverse quality-of-service (QoS) demands while maintaining high energy efficiency. Weighted Minimum Mean Square Error (WMMSE) precoding with fixed user priorities and transmit power is widely recognized for enhancing overall system performance but lacks flexibility to adapt to user-specific QoS requirements and time-varying channel conditions. To address this, we propose a novel constrained reinforcement learning (CRL) algorithm, Constrained-Actor Attentive-Critic (CAAC), which uses a policy network to dynamically allocate user priorities and power for WMMSE precoding. Specifically, CAAC integrates a Constrained Stochastic Successive Convex Approximation (CSSCA) method to optimize the policy, enabling more effective handling of energy efficiency goals and satisfaction of stochastic non-convex QoS constraints compared to traditional and existing CRL methods. Moreover, CAAC employs lightweight attention-enhanced Q-networks to evaluate policy updates without prior environment model knowledge. The network architecture not only enhances representational capacity but also boosts learning efficiency. Simulation results show that CAAC outperforms baselines in both energy efficiency and QoS satisfaction.
Authors: Tianle Gu, Kexin Huang, Zongqi Wang, Yixu Wang, Jie Li, Yuanqi Yao, Yang Yao, Yujiu Yang, Yan Teng, Yingchun Wang
Abstract: Safety alignment is a key requirement for building reliable Artificial General Intelligence. Despite significant advances in safety alignment, we observe that minor latent shifts can still trigger unsafe responses in aligned models. We argue that this stems from the shallow nature of existing alignment methods, which focus on surface-level refusal behaviors without sufficiently altering internal representations. Consequently, small shifts in hidden activations can re-trigger harmful behaviors embedded in the latent space. To explore the robustness of safety alignment to latent perturbations, we introduce a probing method that measures the Negative Log-Likelihood of the original response generated by the model. This probe quantifies local sensitivity in the latent space, serving as a diagnostic tool for identifying vulnerable directions. Based on this signal, we construct effective jailbreak trajectories, giving rise to the Activation Steering Attack (ASA). More importantly, these insights offer a principled foundation for improving alignment robustness. To this end, we introduce Layer-wise Adversarial Patch Training~(LAPT), a fine-tuning strategy that inject controlled perturbations into hidden representations during training. Experimental results highlight that LAPT strengthen alignment robustness without compromising general capabilities. Our findings reveal fundamental flaws in current alignment paradigms and call for representation-level training strategies that move beyond surface-level behavior supervision. Codes and results are available at https://github.com/Carol-gutianle/LatentSafety.
Authors: Qianqian Liao, Wuque Cai, Hongze Sun, Dongze Liu, Duo Chen, Dezhong Yao, Daqing Guo
Abstract: Recent developed graph-based methods for diagnosing brain disorders using functional connectivity highly rely on predefined brain atlases, but overlook the rich information embedded within atlases and the confounding effects of site and phenotype variability. To address these challenges, we propose a two-stage Brain-to-Population Graph Learning (B2P-GL) framework that integrates the semantic similarity of brain regions and condition-based population graph modeling. In the first stage, termed brain representation learning, we leverage brain atlas knowledge from GPT-4 to enrich the graph representation and refine the brain graph through an adaptive node reassignment graph attention network. In the second stage, termed population disorder diagnosis, phenotypic data is incorporated into population graph construction and feature fusion to mitigate confounding effects and enhance diagnosis performance. Experiments on the ABIDE I, ADHD-200, and Rest-meta-MDD datasets show that B2P-GL outperforms state-of-the-art methods in prediction accuracy while enhancing interpretability. Overall, our proposed framework offers a reliable and personalized approach to brain disorder diagnosis, advancing clinical applicability.
Authors: Langzhang Liang, Fanchen Bu, Zixing Song, Zenglin Xu, Shirui Pan, Kijung Shin
Abstract: The message-passing paradigm of Graph Neural Networks often struggles with exchanging information across distant nodes typically due to structural bottlenecks in certain graph regions, a limitation known as \textit{over-squashing}. To reduce such bottlenecks, \textit{graph rewiring}, which modifies graph topology, has been widely used. However, existing graph rewiring techniques often overlook the need to preserve critical properties of the original graph, e.g., \textit{spectral properties}. Moreover, many approaches rely on increasing edge count to improve connectivity, which introduces significant computational overhead and exacerbates the risk of over-smoothing. In this paper, we propose a novel graph rewiring method that leverages \textit{spectrum-preserving} graph \textit{sparsification}, for mitigating over-squashing. Our method generates graphs with enhanced connectivity while maintaining sparsity and largely preserving the original graph spectrum, effectively balancing structural bottleneck reduction and graph property preservation. Experimental results validate the effectiveness of our approach, demonstrating its superiority over strong baseline methods in classification accuracy and retention of the Laplacian spectrum.
Authors: Simardeep Singh
Abstract: Large language models (LLMs) are known to memorize parts of their training data, raising important concerns around privacy and security. While previous research has focused on studying memorization in pre-trained models, much less is known about how knowledge distillation (KD) affects memorization.In this study, we explore how different KD methods influence the memorization of fine-tuned task data when a large teacher model is distilled into smaller student variants.This study demonstrates that distilling a larger teacher model, fine-tuned on a dataset, into a smaller variant not only lowers computational costs and model size but also significantly reduces the memorization risks compared to standard fine-tuning approaches.
Authors: Ismo Horppu, Frederick Ayala, Erlin Gulbenkoglu
Abstract: All languages are peculiar. Some of them are considered more challenging to understand than others. The Finnish Language is known to be a complex language. Also, when languages are used by artists, the pronunciation and meaning might be more tricky to understand. Therefore, we are putting AI to a fun, yet challenging trial: translating a Finnish rap song to text. We will compare the Faster Whisperer algorithm and YouTube's internal speech-to-text functionality. The reference truth will be Finnish rap lyrics, which the main author's little brother, Mc Timo, has written. Transcribing the lyrics will be challenging because the artist raps over synth music player by Syntikka Janne. The hallucination level and mishearing of AI speech-to-text extractions will be measured by comparing errors made against the original Finnish lyrics. The error function is informal but still works for our case.
Authors: Xun Wang, Jing Xu, Franziska Boenisch, Michael Backes, Christopher A. Choquette-Choo, Adam Dziedzic
Abstract: Prompting has become a dominant paradigm for adapting large language models (LLMs). While discrete (textual) prompts are widely used for their interpretability, soft (parameter) prompts have recently gained traction in APIs. This is because they can encode information from more training samples while minimizing the user's token usage, leaving more space in the context window for task-specific input. However, soft prompts are tightly coupled to the LLM they are tuned on, limiting their generalization to other LLMs. This constraint is particularly problematic for efficiency and privacy: (1) tuning prompts on each LLM incurs high computational costs, especially as LLMs continue to grow in size. Additionally, (2) when the LLM is hosted externally, soft prompt tuning often requires sharing private data with the LLM provider. For instance, this is the case with the NVIDIA NeMo API. To address these issues, we propose POST (Privacy Of Soft prompt Transfer), a framework that enables private tuning of soft prompts on a small model and subsequently transfers these prompts to a larger LLM. POST uses knowledge distillation to derive a small model directly from the large LLM to improve prompt transferability, tunes the soft prompt locally, optionally with differential privacy guarantees, and transfers it back to the larger LLM using a small public dataset. Our experiments show that POST reduces computational costs, preserves privacy, and effectively transfers high-utility soft prompts.
Authors: Charbel Bou Chaaya, Abanoub M. Girgis, Mehdi Bennis
Abstract: In this work, we aim to optimize the radio resource management of a communication system between a remote controller and its device, whose state is represented through image frames, without compromising the performance of the control task. We propose a novel machine learning (ML) technique to jointly model and predict the dynamics of the control system as well as the wireless propagation environment in latent space. Our method leverages two coupled joint-embedding predictive architectures (JEPAs): a control JEPA models the control dynamics and guides the predictions of a wireless JEPA, which captures the dynamics of the device's channel state information (CSI) through cross-modal conditioning. We then train a deep reinforcement learning (RL) algorithm to derive a control policy from latent control dynamics and a power predictor to estimate scheduling intervals with favorable channel conditions based on latent CSI representations. As such, the controller minimizes the usage of radio resources by utilizing the coupled JEPA networks to imagine the device's trajectory in latent space. We present simulation results on synthetic multimodal data and show that our proposed approach reduces transmit power by over 50% while maintaining control performance comparable to baseline methods that do not account for wireless optimization.
Authors: Prakhar Verma, David Arbour, Sunav Choudhary, Harshita Chopra, Arno Solin, Atanu R. Sinha
Abstract: Causal discovery from observational data typically assumes full access to data and availability of domain experts. In practice, data often arrive in batches, and expert knowledge is scarce. Language Models (LMs) offer a surrogate but come with their own issues-hallucinations, inconsistencies, and bias. We present BLANCE (Bayesian LM-Augmented Causal Estimation)-a hybrid Bayesian framework that bridges these gaps by adaptively integrating sequential batch data with LM-derived noisy, expert knowledge while accounting for both data-induced and LM-induced biases. Our proposed representation shift from Directed Acyclic Graph (DAG) to Partial Ancestral Graph (PAG) accommodates ambiguities within a coherent Bayesian framework, allowing grounding the global LM knowledge in local observational data. To guide LM interaction, we use a sequential optimization scheme that adaptively queries the most informative edges. Across varied datasets, BLANCE outperforms prior work in structural accuracy and extends to Bayesian parameter estimation, showing robustness to LM noise.
Authors: Jacopo Iollo, Geoffroy Oudoumanessah, Carole Lartizien, Michel Dojat, Florence Forbes
Abstract: A key challenge in maximizing the benefits of Magnetic Resonance Imaging (MRI) in clinical settings is to accelerate acquisition times without significantly degrading image quality. This objective requires a balance between under-sampling the raw k-space measurements for faster acquisitions and gathering sufficient raw information for high-fidelity image reconstruction and analysis tasks. To achieve this balance, we propose to use sequential Bayesian experimental design (BED) to provide an adaptive and task-dependent selection of the most informative measurements. Measurements are sequentially augmented with new samples selected to maximize information gain on a posterior distribution over target images. Selection is performed via a gradient-based optimization of a design parameter that defines a subsampling pattern. In this work, we introduce a new active BED procedure that leverages diffusion-based generative models to handle the high dimensionality of the images and employs stochastic optimization to select among a variety of patterns while meeting the acquisition process constraints and budget. So doing, we show how our setting can optimize, not only standard image reconstruction, but also any associated image analysis task. The versatility and performance of our approach are demonstrated on several MRI acquisitions.
Authors: Abdulvahap Mutlu, \c{S}eng\"ul Do\u{g}an, T\"urker Tuncer
Abstract: Amyotrophic Lateral Sclerosis (ALS) is a rare neurodegenerative disease, and high-quality EEG data from ALS patients are scarce. This data scarcity, coupled with severe class imbalance between ALS and healthy control recordings, poses a challenge for training reliable machine learning classifiers. In this work, we address these issues by generating synthetic EEG signals for ALS patients using a Conditional Wasserstein Generative Adversarial Network (CWGAN). We train CWGAN on a private EEG dataset (ALS vs. non-ALS) to learn the distribution of ALS EEG signals and produce realistic synthetic samples. We preprocess and normalize EEG recordings, and train a CWGAN model to generate synthetic ALS signals. The CWGAN architecture and training routine are detailed, with key hyperparameters chosen for stable training. Qualitative evaluation of generated signals shows that they closely mimic real ALS EEG patterns. The CWGAN training converged with generator and discriminator loss curves stabilizing, indicating successful learning. The synthetic EEG signals appear realistic and have potential use as augmented data for training classifiers, helping to mitigate class imbalance and improve ALS detection accuracy. We discuss how this approach can facilitate data sharing and enhance diagnostic models.
Authors: Hadar Tal, Oron Sabag
Abstract: We study the Online Bookmaking problem, where a bookmaker dynamically updates betting odds on the possible outcomes of an event. In each betting round, the bookmaker can adjust the odds based on the cumulative betting behavior of gamblers, aiming to maximize profit while mitigating potential loss. We show that for any event and any number of betting rounds, in a worst-case setting over all possible gamblers and outcome realizations, the bookmaker's optimal loss is the largest root of a simple polynomial. Our solution shows that bookmakers can be as fair as desired while avoiding financial risk, and the explicit characterization reveals an intriguing relation between the bookmaker's regret and Hermite polynomials. We develop an efficient algorithm that computes the optimal bookmaking strategy: when facing an optimal gambler, the algorithm achieves the optimal loss, and in rounds where the gambler is suboptimal, it reduces the achieved loss to the optimal opportunistic loss, a notion that is related to subgame perfect Nash equilibrium. The key technical contribution to achieve these results is an explicit characterization of the Bellman-Pareto frontier, which unifies the dynamic programming updates for Bellman's value function with the multi-criteria optimization framework of the Pareto frontier in the context of vector repeated games.
Authors: Leo Gagnon, Eric Elmoznino, Sarthak Mittal, Tom Marty, Tejas Kasetty, Dhanya Sridhar, Guillaume Lajoie
Abstract: The rapid adaptation ability of auto-regressive foundation models is often attributed to the diversity of their pre-training data. This is because, from a Bayesian standpoint, minimizing prediction error in such settings requires integrating over all plausible latent hypotheses consistent with observations. While this behavior is desirable in principle, it often proves too ambitious in practice: under high ambiguity, the number of plausible latent alternatives makes Bayes-optimal prediction computationally intractable. Cognitive science has long recognized this limitation, suggesting that under such conditions, heuristics or information-seeking strategies are preferable to exhaustive inference. Translating this insight to next-token prediction, we hypothesize that low- and high-ambiguity predictions pose different computational demands, making ambiguity-agnostic next-token prediction a detrimental inductive bias. To test this, we introduce MetaHMM, a synthetic sequence meta-learning benchmark with rich compositional structure and a tractable Bayesian oracle. We show that Transformers indeed struggle with high-ambiguity predictions across model sizes. Motivated by cognitive theories, we propose a method to convert pre-trained models into Monte Carlo predictors that decouple task inference from token prediction. Preliminary results show substantial gains in ambiguous contexts through improved capacity allocation and test-time scalable inference, though challenges remain.
Authors: Pranav Pawar, Akshansh Dwivedi, Jenish Boricha, Himanshu Gohil, Aditya Dubey
Abstract: State-of-the-art text-to-speech (TTS) systems realize high naturalness in monolingual environments, synthesizing speech with correct multilingual accents (especially for Indic languages) and context-relevant emotions still poses difficulty owing to cultural nuance discrepancies in current frameworks. This paper introduces a new TTS architecture integrating accent along with preserving transliteration with multi-scale emotion modelling, in particularly tuned for Hindi and Indian English accent. Our approach extends the Parler-TTS model by integrating A language-specific phoneme alignment hybrid encoder-decoder architecture, and culture-sensitive emotion embedding layers trained on native speaker corpora, as well as incorporating a dynamic accent code switching with residual vector quantization. Quantitative tests demonstrate 23.7% improvement in accent accuracy (Word Error Rate reduction from 15.4% to 11.8%) and 85.3% emotion recognition accuracy from native listeners, surpassing METTS and VECL-TTS baselines. The novelty of the system is that it can mix code in real time - generating statements such as "Namaste, let's talk about
Authors: Sajan Muhammad, Salem Lahlou
Abstract: Efficiently identifying the right trajectories for training remains an open problem in GFlowNets. To address this, it is essential to prioritize exploration in regions of the state space where the reward distribution has not been sufficiently learned. This calls for uncertainty-driven exploration, in other words, the agent should be aware of what it does not know. This attribute can be measured by joint predictions, which are particularly important for combinatorial and sequential decision problems. In this research, we integrate epistemic neural networks (ENN) with the conventional architecture of GFlowNets to enable more efficient joint predictions and better uncertainty quantification, thereby improving exploration and the identification of optimal trajectories. Our proposed algorithm, ENN-GFN-Enhanced, is compared to the baseline method in GFlownets and evaluated in grid environments and structured sequence generation in various settings, demonstrating both its efficacy and efficiency.
Authors: Emmanuel Gangler (Universit\'e Clermont Auvergne CNRS LPCA, Clermont-Ferrand, France), Emille E. O. Ishida (Universit\'e Clermont Auvergne CNRS LPCA, Clermont-Ferrand, France), Matwey V. Kornilov (National Research University Higher School of Economics, Moscow, Russia, Sternberg Astronomical Institute Lomonosov Moscow State University, Moscow, Russia), Vladimir Korolev (Sternberg Astronomical Institute Lomonosov Moscow State University, Moscow, Russia), Anastasia Lavrukhina (Sternberg Astronomical Institute Lomonosov Moscow State University, Moscow, Russia), Konstantin Malanchev (McWilliams Center for Cosmology and Astrophysics, Department of Physics, Carnegie Mellon University, Pittsburgh, PA, USA), Maria V. Pruzhinskaya (Universit\'e Clermont Auvergne CNRS LPCA, Clermont-Ferrand, France, Sternberg Astronomical Institute Lomonosov Moscow State University, Moscow, Russia), Etienne Russeil (Universit\'e Clermont Auvergne CNRS LPCA, Clermont-Ferrand, France, The Oskar Klein Centre Department of Astronomy, Stockholm University AlbaNova, Stockholm, Sweden), Timofey Semenikhin (Sternberg Astronomical Institute Lomonosov Moscow State University, Moscow, Russia, Faculty of Physics, Lomonosov Moscow State University, Moscow, Russia), Sreevarsha Sreejith (Physics department, University of Surrey, Guildford, UK), Alina A. Volnova (Space Research Institute of the Russian Academy of Sciences, Moscow, Russia)
Abstract: Machine learning is often viewed as a black box when it comes to understanding its output, be it a decision or a score. Automatic anomaly detection is no exception to this rule, and quite often the astronomer is left to independently analyze the data in order to understand why a given event is tagged as an anomaly. We introduce here idea of anomaly signature, whose aim is to help the interpretability of anomalies by highlighting which features contributed to the decision.
Authors: Huy Hoang Nguyen, Han Zhou, Matthew B. Blaschko, Aleksei Tiulpin
Abstract: Bayesian optimization with Gaussian processes (GP) is commonly used to optimize black-box functions. The Mat\'ern and the Radial Basis Function (RBF) covariance functions are used frequently, but they do not make any assumptions about the domain of the function, which may limit their applicability in bounded domains. To address the limitation, we introduce the Beta kernel, a non-stationary kernel induced by a product of Beta distribution density functions. Such a formulation allows our kernel to naturally model functions on bounded domains. We present statistical evidence supporting the hypothesis that the kernel exhibits an exponential eigendecay rate, based on empirical analyses of its spectral properties across different settings. Our experimental results demonstrate the robustness of the Beta kernel in modeling functions with optima located near the faces or vertices of the unit hypercube. The experiments show that our kernel consistently outperforms a wide range of kernels, including the well-known Mat\'ern and RBF, in different problems, including synthetic function optimization and the compression of vision and language models.
Authors: Nikola Jovanovi\'c, Ismail Labiad, Tom\'a\v{s} Sou\v{c}ek, Martin Vechev, Pierre Fernandez
Abstract: Watermarking the outputs of generative models has emerged as a promising approach for tracking their provenance. Despite significant interest in autoregressive image generation models and their potential for misuse, no prior work has attempted to watermark their outputs at the token level. In this work, we present the first such approach by adapting language model watermarking techniques to this setting. We identify a key challenge: the lack of reverse cycle-consistency (RCC), wherein re-tokenizing generated image tokens significantly alters the token sequence, effectively erasing the watermark. To address this and to make our method robust to common image transformations, neural compression, and removal attacks, we introduce (i) a custom tokenizer-detokenizer finetuning procedure that improves RCC, and (ii) a complementary watermark synchronization layer. As our experiments demonstrate, our approach enables reliable and robust watermark detection with theoretically grounded p-values.
Authors: Theo Zangato, Aomar Osmani, Pegah Alizadeh
Abstract: Increasing global energy demand and renewable integration complexity have placed buildings at the center of sustainable energy management. We present a three-step reinforcement learning(RL)-based Building Energy Management System (BEMS) that combines clustering, forecasting, and constrained policy learning to address scalability, adaptability, and safety challenges. First, we cluster non-shiftable load profiles to identify common consumption patterns, enabling policy generalization and transfer without retraining for each new building. Next, we integrate an LSTM based forecasting module to anticipate future states, improving the RL agents' responsiveness to dynamic conditions. Lastly, domain-informed action masking ensures safe exploration and operation, preventing harmful decisions. Evaluated on real-world data, our approach reduces operating costs by up to 15% for certain building types, maintains stable environmental performance, and quickly classifies and optimizes new buildings with limited data. It also adapts to stochastic tariff changes without retraining. Overall, this framework delivers scalable, robust, and cost-effective building energy management.
Authors: Druva Dhakshinamoorthy, Avikshit Jha, Sabyasachi Majumdar, Devdulal Ghosh, Ranjita Chakraborty, Hena Ray
Abstract: This paper presents a novel system for monitoring cattle behavior and detecting estrus (heat) periods using sensor data and machine learning. We designed and deployed a low-cost Bluetooth-based neck collar equipped with accelerometer and gyroscope sensors to capture real-time behavioral data from real cows, which was synced to the cloud. A labeled dataset was created using synchronized CCTV footage to annotate behaviors such as feeding, rumination, lying, and others. We evaluated multiple machine learning models -- Support Vector Machines (SVM), Random Forests (RF), and Convolutional Neural Networks (CNN) -- for behavior classification. Additionally, we implemented a Long Short-Term Memory (LSTM) model for estrus detection using behavioral patterns and anomaly detection. Our system achieved over 93% behavior classification accuracy and 96% estrus detection accuracy on a limited test set. The approach offers a scalable and accessible solution for precision livestock monitoring, especially in resource-constrained environments.
Authors: Gon\c{c}alo Granjal Cruz, Balazs Renczes, Mark C Runacres, Jan Decuyper
Abstract: While accurate, black-box system identification models lack interpretability of the underlying system dynamics. This paper proposes State-Space Kolmogorov-Arnold Networks (SS-KAN) to address this challenge by integrating Kolmogorov-Arnold Networks within a state-space framework. The proposed model is validated on two benchmark systems: the Silverbox and the Wiener-Hammerstein benchmarks. Results show that SS-KAN provides enhanced interpretability due to sparsity-promoting regularization and the direct visualization of its learned univariate functions, which reveal system nonlinearities at the cost of accuracy when compared to state-of-the-art black-box models, highlighting SS-KAN as a promising approach for interpretable nonlinear system identification, balancing accuracy and interpretability of nonlinear system dynamics.
Authors: Alexey Zakharov, Shimon Whiteson
Abstract: Natural language can offer a concise and human-interpretable means of specifying reinforcement learning (RL) tasks. The ability to extract rewards from a language instruction can enable the development of robotic systems that can learn from human guidance; however, it remains a challenging problem, especially in visual environments. Existing approaches that employ large, pretrained language models either rely on non-visual environment representations, require prohibitively large amounts of feedback, or generate noisy, ill-shaped reward functions. In this paper, we propose a novel method, $\textbf{GoalLadder}$, that leverages vision-language models (VLMs) to train RL agents from a single language instruction in visual environments. GoalLadder works by incrementally discovering states that bring the agent closer to completing a task specified in natural language. To do so, it queries a VLM to identify states that represent an improvement in agent's task progress and to rank them using pairwise comparisons. Unlike prior work, GoalLadder does not trust VLM's feedback completely; instead, it uses it to rank potential goal states using an ELO-based rating system, thus reducing the detrimental effects of noisy VLM feedback. Over the course of training, the agent is tasked with minimising the distance to the top-ranked goal in a learned embedding space, which is trained on unlabelled visual data. This key feature allows us to bypass the need for abundant and accurate feedback typically required to train a well-shaped reward function. We demonstrate that GoalLadder outperforms existing related methods on classic control and robotic manipulation environments with the average final success rate of $\sim$95% compared to only $\sim$45% of the best competitor.
Authors: Alba Carballo-Castro, Manuel Madeira, Yiming Qin, Dorina Thanou, Pascal Frossard
Abstract: Directed graphs naturally model systems with asymmetric, ordered relationships, essential to applications in biology, transportation, social networks, and visual understanding. Generating such graphs enables tasks such as simulation, data augmentation and novel instance discovery; however, directed graph generation remains underexplored. We identify two key factors limiting progress in this direction: first, modeling edge directionality introduces a substantially larger dependency space, making the underlying distribution harder to learn; second, the absence of standardized benchmarks hinders rigorous evaluation. Addressing the former requires more expressive models that are sensitive to directional topologies. We propose Directo, the first generative model for directed graphs built upon the discrete flow matching framework. Our approach combines: (i) principled positional encodings tailored to asymmetric pairwise relations, (ii) a dual-attention mechanism capturing both incoming and outgoing dependencies, and (iii) a robust, discrete generative framework. To support evaluation, we introduce a benchmark suite covering synthetic and real-world datasets. It shows that our method performs strongly across diverse settings and even competes with specialized models for particular classes, such as directed acyclic graphs. Our results highlight the effectiveness and generality of our approach, establishing a solid foundation for future research in directed graph generation.
Authors: Zhiyuan Liang, Dongwen Tang, Yuhao Zhou, Xuanlei Zhao, Mingjia Shi, Wangbo Zhao, Zekai Li, Peihao Wang, Konstantin Sch\"urholt, Damian Borth, Michael M. Bronstein, Yang You, Zhangyang Wang, Kai Wang
Abstract: Modern Parameter-Efficient Fine-Tuning (PEFT) methods such as low-rank adaptation (LoRA) reduce the cost of customizing large language models (LLMs), yet still require a separate optimization run for every downstream dataset. We introduce \textbf{Drag-and-Drop LLMs (\textit{DnD})}, a prompt-conditioned parameter generator that eliminates per-task training by mapping a handful of unlabeled task prompts directly to LoRA weight updates. A lightweight text encoder distills each prompt batch into condition embeddings, which are then transformed by a cascaded hyper-convolutional decoder into the full set of LoRA matrices. Once trained in a diverse collection of prompt-checkpoint pairs, DnD produces task-specific parameters in seconds, yielding i) up to \textbf{12,000$\times$} lower overhead than full fine-tuning, ii) average gains up to \textbf{30\%} in performance over the strongest training LoRAs on unseen common-sense reasoning, math, coding, and multimodal benchmarks, and iii) robust cross-domain generalization despite never seeing the target data or labels. Our results demonstrate that prompt-conditioned parameter generation is a viable alternative to gradient-based adaptation for rapidly specializing LLMs. Our project is available at \href{https://jerryliang24.github.io/DnD}{https://jerryliang24.github.io/DnD}.
URLs: https://jerryliang24.github.io/DnD, https://jerryliang24.github.io/DnD
Authors: Daniel Fidel Harvey, George Weale, Berk Yilmaz
Abstract: Mixture of Experts (MoE) architectures increase large language model scalability, yet their performance depends on the router module that moves tokens to specialized experts. Bad routing can load imbalance and reduced accuracy. This project designed and implemented different router architectures within Transformer models to fix these limitations. We experimented with six distinct router variants Linear, Attention, Multi-Layer Perceptron (MLP), Hybrid, Hash, and our new MLP-Hadamard. We characterized these routers using BERT and the Qwen1.5-MoE model, looking at parameter efficiency, inference latency, routing entropy, and expert utilization patterns. Our evaluations showed distinct trade-offs: Linear routers offer speed, while MLP and Attention routers provide greater expressiveness. The MLP-Hadamard router shows a unique capability for structured, sparse routing. We successfully replaced and fine-tuned custom routers within the complex, quantized Qwen1.5-MoE model. This work provides a comparative analysis of MoE router designs and offers insights into optimizing their performance for efficient and effective large-scale model deployment.
Authors: Dian Meng, Zhiguang Cao, Yaoxin Wu, Yaqing Hou, Hongwei Ge, Qiang Zhang
Abstract: Recent neural heuristics for the Vehicle Routing Problem (VRP) primarily rely on node coordinates as input, which may be less effective in practical scenarios where real cost metrics-such as edge-based distances-are more relevant. To address this limitation, we introduce EFormer, an Edge-based Transformer model that uses edge as the sole input for VRPs. Our approach employs a precoder module with a mixed-score attention mechanism to convert edge information into temporary node embeddings. We also present a parallel encoding strategy characterized by a graph encoder and a node encoder, each responsible for processing graph and node embeddings in distinct feature spaces, respectively. This design yields a more comprehensive representation of the global relationships among edges. In the decoding phase, parallel context embedding and multi-query integration are used to compute separate attention mechanisms over the two encoded embeddings, facilitating efficient path construction. We train EFormer using reinforcement learning in an autoregressive manner. Extensive experiments on the Traveling Salesman Problem (TSP) and Capacitated Vehicle Routing Problem (CVRP) reveal that EFormer outperforms established baselines on synthetic datasets, including large-scale and diverse distributions. Moreover, EFormer demonstrates strong generalization on real-world instances from TSPLib and CVRPLib. These findings confirm the effectiveness of EFormer's core design in solving VRPs.
Authors: Antonio Giulio Coretti, Mattia Varile, Mario Edoardo Bertaina
Abstract: Space debris poses a significant threat, driving research into active and passive mitigation strategies. This work presents an innovative collision avoidance system utilizing event-based cameras - a novel imaging technology well-suited for Space Situational Awareness (SSA) and Space Traffic Management (STM). The system, employing a Stack-CNN algorithm (previously used for meteor detection), analyzes real-time event-based camera data to detect faint moving objects. Testing on terrestrial data demonstrates the algorithm's ability to enhance signal-to-noise ratio, offering a promising approach for on-board space imaging and improving STM/SSA operations.
Authors: Jonas R. Naujoks, Aleksander Krasowski, Moritz Weckbecker, Galip \"Umit Yolcu, Thomas Wiegand, Sebastian Lapuschkin, Wojciech Samek, Ren\'e P. Klausen
Abstract: Physics-informed neural networks (PINNs) offer a powerful approach to solving partial differential equations (PDEs), which are ubiquitous in the quantitative sciences. Applied to both forward and inverse problems across various scientific domains, PINNs have recently emerged as a valuable tool in the field of scientific machine learning. A key aspect of their training is that the data -- spatio-temporal points sampled from the PDE's input domain -- are readily available. Influence functions, a tool from the field of explainable AI (XAI), approximate the effect of individual training points on the model, enhancing interpretability. In the present work, we explore the application of influence function-based sampling approaches for the training data. Our results indicate that such targeted resampling based on data attribution methods has the potential to enhance prediction accuracy in physics-informed neural networks, demonstrating a practical application of an XAI method in PINN training.
Authors: Tri Duc Ly, Gia H. Ngo
Abstract: EEG is a non-invasive, safe, and low-risk method to record electrophysiological signals inside the brain. Especially with recent technology developments like dry electrodes, consumer-grade EEG devices, and rapid advances in machine learning, EEG is commonly used as a resource for automatic emotion recognition. With the aim to develop a deep learning model that can perform EEG-based emotion recognition in a real-life context, we propose a novel approach to utilize multi-scale convolutional neural networks to accomplish such tasks. By implementing feature extraction kernels with many ratio coefficients as well as a new type of kernel that learns key information from four separate areas of the brain, our model consistently outperforms the state-of-the-art TSception model in predicting valence, arousal, and dominance scores across many performance evaluation metrics.
Authors: Jun Qi, Chen-Yu Liu, Sabato Marco Siniscalchi, Chao-Han Huck Yang, Min-Hsiu Hsieh
Abstract: Low-Rank Adaptation (LoRA) is widely recognized for its parameter-efficient fine-tuning of large-scale neural models. However, standard LoRA independently optimizes low-rank matrices, which inherently limits its expressivity and generalization capabilities. While classical tensor-train (TT) decomposition can be separately employed on individual LoRA matrices, this work demonstrates that the classical TT-based approach neither significantly improves parameter efficiency nor achieves substantial performance gains. This paper proposes TensorGuide, a novel tensor-train-guided adaptation framework to overcome these limitations. TensorGuide generates two correlated low-rank LoRA matrices through a unified TT structure driven by controlled Gaussian noise. The resulting joint TT representation inherently provides structured, low-rank adaptations, significantly enhancing expressivity, generalization, and parameter efficiency without increasing the number of trainable parameters. Theoretically, we justify these improvements through neural tangent kernel analyses, demonstrating superior optimization dynamics and enhanced generalization. Extensive experiments on quantum dot classification and GPT-2 fine-tuning benchmarks demonstrate that TensorGuide-based LoRA consistently outperforms standard LoRA and TT-LoRA, achieving improved accuracy and scalability with fewer parameters.
Authors: John Abascal, Nicol\'as Berrios, Alina Oprea, Jonathan Ullman, Adam Smith, Matthew Jagielski
Abstract: Multitask learning (MTL) has emerged as a powerful paradigm that leverages similarities among multiple learning tasks, each with insufficient samples to train a standalone model, to solve them simultaneously while minimizing data sharing across users and organizations. MTL typically accomplishes this goal by learning a shared representation that captures common structure among the tasks by embedding data from all tasks into a common feature space. Despite being designed to be the smallest unit of shared information necessary to effectively learn patterns across multiple tasks, these shared representations can inadvertently leak sensitive information about the particular tasks they were trained on. In this work, we investigate what information is revealed by the shared representations through the lens of inference attacks. Towards this, we propose a novel, black-box task-inference threat model where the adversary, given the embedding vectors produced by querying the shared representation on samples from a particular task, aims to determine whether that task was present when training the shared representation. We develop efficient, purely black-box attacks on machine learning models that exploit the dependencies between embeddings from the same task without requiring shadow models or labeled reference data. We evaluate our attacks across vision and language domains for multiple use cases of MTL and demonstrate that even with access only to fresh task samples rather than training data, a black-box adversary can successfully infer a task's inclusion in training. To complement our experiments, we provide theoretical analysis of a simplified learning setting and show a strict separation between adversaries with training samples and fresh samples from the target task's distribution.
Authors: Tara Akhound-Sadegh, Jungyoon Lee, Avishek Joey Bose, Valentin De Bortoli, Arnaud Doucet, Michael M. Bronstein, Dominique Beaini, Siamak Ravanbakhsh, Kirill Neklyudov, Alexander Tong
Abstract: Sampling efficiently from a target unnormalized probability density remains a core challenge, with relevance across countless high-impact scientific applications. A promising approach towards this challenge is the design of amortized samplers that borrow key ideas, such as probability path design, from state-of-the-art generative diffusion models. However, all existing diffusion-based samplers remain unable to draw samples from distributions at the scale of even simple molecular systems. In this paper, we propose Progressive Inference-Time Annealing (PITA), a novel framework to learn diffusion-based samplers that combines two complementary interpolation techniques: I.) Annealing of the Boltzmann distribution and II.) Diffusion smoothing. PITA trains a sequence of diffusion models from high to low temperatures by sequentially training each model at progressively higher temperatures, leveraging engineered easy access to samples of the temperature-annealed target density. In the subsequent step, PITA enables simulating the trained diffusion model to procure training samples at a lower temperature for the next diffusion model through inference-time annealing using a novel Feynman-Kac PDE combined with Sequential Monte Carlo. Empirically, PITA enables, for the first time, equilibrium sampling of N-body particle systems, Alanine Dipeptide, and tripeptides in Cartesian coordinates with dramatically lower energy function evaluations. Code available at: https://github.com/taraak/pita
Authors: Amir Reza Vazifeh, Jason W. Fleischer
Abstract: Electrocardiograms (ECGs) provide direct, non-invasive measurements of heart activity and are well-established tools for detecting and monitoring cardiovascular disease. However, manual ECG analysis can be time-consuming and prone to errors. Machine learning has emerged as a promising approach for automated heartbeat recognition and classification, but substantial variations in ECG signals make it challenging to develop generalizable models. ECG signals can vary widely across individuals and leads, while datasets often follow different labeling standards and may be biased, all of which greatly hinder supervised methods. Conventional unsupervised methods, e.g. principal component analysis, prioritize large (and often obvious) variances in the data and typically overlook subtle yet clinically relevant patterns. If labels are missing and/or variations are significant but small, both approaches fail. Here, we show that nonlinear dimensionality reduction (NLDR) can accommodate these issues and identify medically relevant features in ECG signals, with no need for training or prior information. Using the MLII and V1 leads of the MIT-BIH dataset, we demonstrate that t-distributed stochastic neighbor embedding and uniform manifold approximation and projection can discriminate individual recordings in mixed populations with >= 90% accuracy and distinguish different arrhythmias in individual patients with a median accuracy of 98.96% and a median F1-score of 91.02%. The results show that NLDR holds much promise for cardiac monitoring, including the limiting cases of single-lead ECG and the current 12-lead standard of care, and for personalized health care beyond cardiology.
Authors: Samir Khaki, Xiuyu Li, Junxian Guo, Ligeng Zhu, Chenfeng Xu, Konstantinos N. Plataniotis, Amir Yazdanbakhsh, Kurt Keutzer, Song Han, Zhijian Liu
Abstract: Fine-tuning LLMs is both computationally and memory-intensive. While parameter-efficient fine-tuning methods, such as QLoRA and DoRA, reduce the number of trainable parameters and lower memory usage, they do not decrease computational cost. In some cases, they may even slow down fine-tuning. In this paper, we introduce SparseLoRA, a method that accelerates LLM fine-tuning through contextual sparsity. We propose a lightweight, training-free SVD sparsity estimator that dynamically selects a sparse subset of weights for loss and gradient computation. Also, we systematically analyze and address sensitivity across layers, tokens, and training steps. Our experimental results show that SparseLoRA reduces computational cost by up to 2.2 times and a measured speedup of up to 1.6 times while maintaining accuracy across various downstream tasks, including commonsense and arithmetic reasoning, code generation, and instruction following.
Authors: Ronald Skorobogat, Karsten Roth, Mariana-Iuliana Georgescu, Zeynep Akata
Abstract: Model merging enables the combination of multiple specialized expert models into a single model capable of performing multiple tasks. However, the benefits of merging an increasing amount of specialized experts generally lead to diminishing returns and reduced overall performance gains. In this work, we offer an explanation and analysis from a task arithmetic perspective; revealing that as the merging process (across numerous existing merging methods) continues for more and more experts, the associated task vector space experiences rank collapse. To mitigate this issue, we introduce Subspace Boosting, which operates on the singular value decomposed task vector space and maintains task vector ranks. Subspace Boosting raises merging efficacy for up to 20 expert models by large margins of more than 10% when evaluated on vision benchmarks. Moreover, we propose employing Higher-Order Generalized Singular Value Decomposition to further quantify task similarity, offering a new interpretable perspective on model merging.
Authors: Pragya Srivastava, Harman Singh, Rahul Madhavan, Gandharv Patil, Sravanti Addepalli, Arun Suggala, Rengarajan Aravamudhan, Soumya Sharma, Anirban Laha, Aravindan Raghuveer, Karthikeyan Shanmugam, Doina Precup
Abstract: Reward models (RMs) are fundamental to aligning Large Language Models (LLMs) via human feedback, yet they often suffer from reward hacking. They tend to latch on to superficial or spurious attributes, such as response length or formatting, mistaking these cues learned from correlations in training data for the true causal drivers of quality (e.g., factuality, relevance). This occurs because standard training objectives struggle to disentangle these factors, leading to brittle RMs and misaligned policies. We introduce Crome (Causally Robust Reward Modeling), a novel framework grounded in an explicit causal model designed to mitigate reward hacking. Crome employs the following synthetic targeted augmentations during training: (1) Causal Augmentations, which are pairs that differ along specific causal attributes, to enforce sensitivity along each causal attribute individually, and (2) Neutral Augmentations, which are tie-label pairs varying primarily in spurious attributes, to enforce invariance along spurious attributes. Notably, our augmentations are produced without any knowledge of spurious factors, via answer interventions only along causal rubrics, that are identified by querying an oracle LLM. Empirically, Crome significantly outperforms standard baselines on RewardBench, improving average accuracy by up to 5.4% and achieving gains of up to 13.2% and 7.2% in specific categories. The robustness of Crome is further testified by the consistent gains obtained in a Best-of-N inference setting across increasing N, across various benchmarks, including the popular RewardBench (covering chat, chat-hard, safety, and reasoning tasks), the safety-focused WildGuardTest, and the reasoning-specific GSM8k.
Authors: Bornali Phukon, Xiuwen Zheng, Mark Hasegawa-Johnson
Abstract: Traditional ASR metrics like WER and CER fail to capture intelligibility, especially for dysarthric and dysphonic speech, where semantic alignment matters more than exact word matches. ASR systems struggle with these speech types, often producing errors like phoneme repetitions and imprecise consonants, yet the meaning remains clear to human listeners. We identify two key challenges: (1) Existing metrics do not adequately reflect intelligibility, and (2) while LLMs can refine ASR output, their effectiveness in correcting ASR transcripts of dysarthric speech remains underexplored. To address this, we propose a novel metric integrating Natural Language Inference (NLI) scores, semantic similarity, and phonetic similarity. Our ASR evaluation metric achieves a 0.890 correlation with human judgments on Speech Accessibility Project data, surpassing traditional methods and emphasizing the need to prioritize intelligibility over error-based measures.
Authors: Arjun Dosajh, Mihika Sanghi
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation. However, their tendency to memorize training data raises concerns regarding privacy, copyright compliance, and security, particularly in cases involving Personally Identifiable Information (PII). Effective machine unlearning techniques are essential to mitigate these risks, yet existing methods remain underdeveloped for LLMs due to their open-ended output space. In this work, we apply the Adaptive Representation Misdirection Unlearning (RMU) technique to unlearn sensitive information from LLMs. Through extensive experiments, we analyze the effects of unlearning across different decoder layers to determine the most effective regions for sensitive information removal. Our technique ranked 4th on the official leaderboard of both 1B parameter and 7B parameter models.
Authors: Swagatam Das
Abstract: We outline an operator-theoretic framework for analyzing transformer-based language models using the tools of free probability theory. By representing token embeddings and attention mechanisms as self-adjoint operators in a racial probability space, we reinterpret attention as a non-commutative convolution and view the layer-wise propagation of representations as an evolution governed by free additive convolution. This formalism reveals a spectral dynamical system underpinning deep transformer stacks and offers insight into their inductive biases, generalization behavior, and entropy dynamics. We derive a generalization bound based on free entropy and demonstrate that the spectral trace of transformer layers evolves predictably with depth. Our approach bridges neural architecture with non-commutative harmonic analysis, enabling principled analysis of information flow and structural complexity in large language models
Authors: Soroush H. Zargarbashi, Mohammad Sadegh Akhondzadeh, Aleksandar Bojchevski
Abstract: Given any model, conformal prediction (CP) returns prediction sets guaranteed to include the true label with high adjustable probability. Robust CP (RCP) extends this to inputs with worst-case noise. A well-established approach is to use randomized smoothing for RCP since it is applicable to any black-box model and provides smaller sets compared to deterministic methods. However, current smoothing-based RCP requires many model forward passes per each input which is computationally expensive. We show that conformal prediction attains some robustness even with a forward pass on a single randomly perturbed input. Using any binary certificate we propose a single sample robust CP (RCP1). Our approach returns robust sets with smaller average set size compared to SOTA methods which use many (e.g. around 100) passes per input. Our key insight is to certify the conformal prediction procedure itself rather than individual scores. Our approach is agnostic to the setup (classification and regression). We further extend our approach to smoothing-based robust conformal risk control.
Authors: Zeyun Deng, Jasorsi Ghosh, Fiona Xie, Yuzhe Lu, Katia Sycara, Joseph Campbell
Abstract: Reinforcement learning algorithms often suffer from poor sample efficiency, making them challenging to apply in multi-task or continual learning settings. Efficiency can be improved by transferring knowledge from a previously trained teacher policy to guide exploration in new but related tasks. However, if the new task sufficiently differs from the teacher's training task, the transferred guidance may be sub-optimal and bias exploration toward low-reward behaviors. We propose an energy-based transfer learning method that uses out-of-distribution detection to selectively issue guidance, enabling the teacher to intervene only in states within its training distribution. We theoretically show that energy scores reflect the teacher's state-visitation density and empirically demonstrate improved sample efficiency and performance across both single-task and multi-task settings.
Authors: Khiem Le, Tuan Tran, Ting Hua, Nitesh V. Chawla
Abstract: Existing resource-adaptive LoRA federated fine-tuning methods enable clients to fine-tune models using compressed versions of global LoRA matrices, in order to accommodate various compute resources across clients. This compression requirement will lead to suboptimal performance due to information loss. To address this, we propose FLAME, a novel federated learning framework based on the Sparse Mixture-of-Experts (SMoE) architecture. Unlike prior approaches, FLAME retains full (uncompressed) global LoRA matrices and achieves client-side adaptability by varying the number of activated experts per client. However, incorporating SMoE into federated learning introduces unique challenges, specifically, the mismatch in output magnitude from partial expert activation and the imbalance in expert training quality across clients. FLAME tackles these challenges through a lightweight rescaling mechanism and an activation-aware aggregation scheme. Empirical results across diverse computational settings demonstrate that FLAME consistently outperforms existing methods, providing a robust and effective solution for resource-adaptive federated learning.
Authors: Siddharth Viswanath, Rahul Singh, Yanlei Zhang, J. Adam Noah, Joy Hirsch, Smita Krishnaswamy
Abstract: Graph neural networks have been useful in machine learning on graph-structured data, particularly for node classification and some types of graph classification tasks. However, they have had limited use in representing patterning of signals over graphs. Patterning of signals over graphs and in subgraphs carries important information in many domains including neuroscience. Neural signals are spatiotemporally patterned, high dimensional and difficult to decode. Graph signal processing and associated GCN models utilize the graph Fourier transform and are unable to efficiently represent spatially or spectrally localized signal patterning on graphs. Wavelet transforms have shown promise here, but offer non-canonical representations and cannot be tightly confined to subgraphs. Here we propose SlepNet, a novel GCN architecture that uses Slepian bases rather than graph Fourier harmonics. In SlepNet, the Slepian harmonics optimally concentrate signal energy on specifically relevant subgraphs that are automatically learned with a mask. Thus, they can produce canonical and highly resolved representations of neural activity, focusing energy of harmonics on areas of the brain which are activated. We evaluated SlepNet across three fMRI datasets, spanning cognitive and visual tasks, and two traffic dynamics datasets, comparing its performance against conventional GNNs and graph signal processing constructs. SlepNet outperforms the baselines in all datasets. Moreover, the extracted representations of signal patterns from SlepNet offers more resolution in distinguishing between similar patterns, and thus represent brain signaling transients as informative trajectories. Here we have shown that these extracted trajectory representations can be used for other downstream untrained tasks. Thus we establish that SlepNet is useful both for prediction and representation learning in spatiotemporal data.
Authors: Jiamin He, A. Rupam Mahmood, Martha White
Abstract: We introduce a novel reinforcement learning (RL) framework that treats distribution parameters as actions, redefining the boundary between agent and environment. This reparameterization makes the new action space continuous, regardless of the original action type (discrete, continuous, mixed, etc.). Under this new parameterization, we develop a generalized deterministic policy gradient estimator, Distribution Parameter Policy Gradient (DPPG), which has lower variance than the gradient in the original action space. Although learning the critic over distribution parameters poses new challenges, we introduce interpolated critic learning (ICL), a simple yet effective strategy to enhance learning, supported by insights from bandit settings. Building on TD3, a strong baseline for continuous control, we propose a practical DPPG-based actor-critic algorithm, Distribution Parameter Actor-Critic (DPAC). Empirically, DPAC outperforms TD3 in MuJoCo continuous control tasks from OpenAI Gym and DeepMind Control Suite, and demonstrates competitive performance on the same environments with discretized action spaces.
Authors: Eric V. Strobl
Abstract: Causal inference in longitudinal biomedical data remains a central challenge, especially in psychiatry, where symptom heterogeneity and latent confounding frequently undermine classical estimators. Most existing methods for treatment effect estimation presuppose a fixed outcome variable and address confounding through observed covariate adjustment. However, the assumption of unconfoundedness may not hold for a fixed outcome in practice. To address this foundational limitation, we directly optimize the outcome definition to maximize causal identifiability. Our DEBIAS (Durable Effects with Backdoor-Invariant Aggregated Symptoms) algorithm learns non-negative, clinically interpretable weights for outcome aggregation, maximizing durable treatment effects and empirically minimizing both observed and latent confounding by leveraging the time-limited direct effects of prior treatments in psychiatric longitudinal data. The algorithm also furnishes an empirically verifiable test for outcome unconfoundedness. DEBIAS consistently outperforms state-of-the-art methods in recovering causal effects for clinically interpretable composite outcomes across comprehensive experiments in depression and schizophrenia.
Authors: Eren Akbiyik, Jo\~ao Almeida, Rik Melis, Ritu Sriram, Viviana Petrescu, Vilhj\'almur Vilhj\'almsson
Abstract: Modern text processing pipelines demand robust methods to remove extraneous content while preserving a document's core message. Traditional approaches such as HTML boilerplate extraction or keyword filters often fail in multilingual settings and struggle with context-sensitive nuances, whereas Large Language Models (LLMs) offer improved quality at high computational cost. We introduce SORE (Semantic Outlier Removal), a cost-effective, transparent method that leverages multilingual sentence embeddings and approximate nearest-neighbor search to identify and excise unwanted text segments. By first identifying core content via metadata embedding and then flagging segments that either closely match predefined outlier groups or deviate significantly from the core, SORE achieves near-LLM extraction precision at a fraction of the cost. Experiments on HTML datasets demonstrate that SORE outperforms structural methods and yield high precision in diverse scenarios. Our system is currently deployed in production, processing millions of documents daily across multiple languages while maintaining both efficiency and accuracy. To facilitate reproducibility and further research, we release our implementation and evaluation datasets.
Authors: Guy Blanc, Jane Lange, Carmen Strassle, Li-Yang Tan
Abstract: The apparent difficulty of efficient distribution-free PAC learning has led to a large body of work on distribution-specific learning. Distributional assumptions facilitate the design of efficient algorithms but also limit their reach and relevance. Towards addressing this, we prove a distributional-lifting theorem: This upgrades a learner that succeeds with respect to a limited distribution family $\mathcal{D}$ to one that succeeds with respect to any distribution $D^\star$, with an efficiency overhead that scales with the complexity of expressing $D^\star$ as a mixture of distributions in $\mathcal{D}$. Recent work of Blanc, Lange, Malik, and Tan considered the special case of lifting uniform-distribution learners and designed a lifter that uses a conditional sample oracle for $D^\star$, a strong form of access not afforded by the standard PAC model. Their approach, which draws on ideas from semi-supervised learning, first learns $D^\star$ and then uses this information to lift. We show that their approach is information-theoretically intractable with access only to random examples, thereby giving formal justification for their use of the conditional sample oracle. We then take a different approach that sidesteps the need to learn $D^\star$, yielding a lifter that works in the standard PAC model and enjoys additional advantages: it works for all base distribution families, preserves the noise tolerance of learners, has better sample complexity, and is simpler.
Authors: Vijay Prakash Dwivedi, Charilaos Kanatsoulis, Shenyang Huang, Jure Leskovec
Abstract: Graph machine learning has led to a significant increase in the capabilities of models that learn on arbitrary graph-structured data and has been applied to molecules, social networks, recommendation systems, and transportation, among other domains. Data in multi-tabular relational databases can also be constructed as 'relational entity graphs' for Relational Deep Learning (RDL) - a new blueprint that enables end-to-end representation learning without traditional feature engineering. Compared to arbitrary graph-structured data, relational entity graphs have key properties: (i) their structure is defined by primary-foreign key relationships between entities in different tables, (ii) the structural connectivity is a function of the relational schema defining a database, and (iii) the graph connectivity is temporal and heterogeneous in nature. In this paper, we provide a comprehensive review of RDL by first introducing the representation of relational databases as relational entity graphs, and then reviewing public benchmark datasets that have been used to develop and evaluate recent GNN-based RDL models. We discuss key challenges including large-scale multi-table integration and the complexities of modeling temporal dynamics and heterogeneous data, while also surveying foundational neural network methods and recent architectural advances specialized for relational entity graphs. Finally, we explore opportunities to unify these distinct modeling challenges, highlighting how RDL converges multiple sub-fields in graph machine learning towards the design of foundation models that can transform the processing of relational data.
Authors: Yaozhong Shi, Zachary E. Ross, Domniki Asimaki, Kamyar Azizzadenesheli
Abstract: Generative models in function spaces, situated at the intersection of generative modeling and operator learning, are attracting increasing attention due to their immense potential in diverse scientific and engineering applications. While functional generative models are theoretically domain- and discretization-agnostic, current implementations heavily rely on the Fourier Neural Operator (FNO), limiting their applicability to regular grids and rectangular domains. To overcome these critical limitations, we introduce the Mesh-Informed Neural Operator (MINO). By leveraging graph neural operators and cross-attention mechanisms, MINO offers a principled, domain- and discretization-agnostic backbone for generative modeling in function spaces. This advancement significantly expands the scope of such models to more diverse applications in generative, inverse, and regression tasks. Furthermore, MINO provides a unified perspective on integrating neural operators with general advanced deep learning architectures. Finally, we introduce a suite of standardized evaluation metrics that enable objective comparison of functional generative models, addressing another critical gap in the field.
Authors: Athanasios Glentis, Jiaxiang Li, Andi Han, Mingyi Hong
Abstract: Training large language models (LLMs) typically relies on adaptive optimizers such as Adam, which require significant memory to maintain first- and second-moment matrices, known as optimizer states. While recent works such as GaLore, Fira, and APOLLO have proposed state-compressed variants to reduce memory consumption, a fundamental question remains: What is the minimal amount of optimizer state that is truly necessary to retain state-of-the-art performance in LLM pretraining? In this work, we systematically investigate this question using a bottom-up approach. We find that two memory- and compute-efficient optimization techniques are particularly effective: (1) column-wise gradient normalization significantly boosts the performance of plain SGD without requiring momentum; and (2) adding first-order momentum only to the output layer - where gradient variance is highest - yields performance competitive with fully adaptive methods such as Muon. Based on these insights, we propose SCALE (Stochastic Column-normalized Last-layer Momentum), a new optimizer that combines column-normalized SGD with last-layer momentum, where column normalization refers to normalizing the gradient along the output dimension. Across multiple LLaMA models (60M-1B), SCALE matches or exceeds the performance of Adam while using only 35-45% of the total memory. It also consistently outperforms memory-efficient optimizers such as GaLore, Fira, and APOLLO, making it a strong candidate for large-scale pretraining under memory constraints. For the LLaMA 7B model, SCALE outperforms the state-of-the-art method APOLLO in terms of both perplexity and memory consumption. In addition, our method serves as a minimalist baseline for more sophisticated optimizer design.
Authors: Felix Zhou, Samson Zhou, Vahab Mirrokni, Alessandro Epasto, Vincent Cohen-Addad
Abstract: Deep neural networks often use large, high-quality datasets to achieve high performance on many machine learning tasks. When training involves potentially sensitive data, this process can raise privacy concerns, as large models have been shown to unintentionally memorize and reveal sensitive information, including reconstructing entire training samples. Differential privacy (DP) provides a robust framework for protecting individual data and in particular, a new approach to privately training deep neural networks is to approximate the input dataset with a privately generated synthetic dataset, before any subsequent training algorithm. We introduce a novel principled method for DP synthetic image embedding generation, based on fitting a Gaussian Mixture Model (GMM) in an appropriate embedding space using DP clustering. Our method provably learns a GMM under separation conditions. Empirically, a simple two-layer neural network trained on synthetically generated embeddings achieves state-of-the-art (SOTA) classification accuracy on standard benchmark datasets. Additionally, we demonstrate that our method can generate realistic synthetic images that achieve downstream classification accuracy comparable to SOTA methods. Our method is quite general, as the encoder and decoder modules can be freely substituted to suit different tasks. It is also highly scalable, consisting only of subroutines that scale linearly with the number of samples and/or can be implemented efficiently in distributed systems.
Authors: Zhiying Qiu, Tao Lin
Abstract: Diffusion models have recently shown promise in offline RL. However, these methods often suffer from high training costs and slow convergence, particularly when using transformer-based denoising backbones. While several optimization strategies have been proposed -- such as modified noise schedules, auxiliary prediction targets, and adaptive loss weighting -- challenges remain in achieving stable and efficient training. In particular, existing loss weighting functions typically rely on neural network approximators, which can be ineffective in early training phases due to limited generalization capacity of MLPs when exposed to sparse feedback in the early training stages. In this work, we derive a variationally optimal uncertainty-aware weighting function and introduce a closed-form polynomial approximation method for its online estimation under the flow-based generative modeling framework. We integrate our method into a diffusion planning pipeline and evaluate it on standard offline RL benchmarks. Experimental results on Maze2D and Kitchen tasks show that our method achieves competitive performance with up to 10 times fewer training steps, highlighting its practical effectiveness.
Authors: Dinesh Ramasamy, Shakti Kumar, Chris Cadonic, Jiaxin Yang, Sohini Roychowdhury, Esam Abdel Rhman, Srihari Reddy
Abstract: Sequence-based recommendations models are driving the state-of-the-art for industrial ad-recommendation systems. Such systems typically deal with user histories or sequence lengths ranging in the order of O(10^3) to O(10^4) events. While adding embeddings at this scale is manageable in pre-trained models, incorporating them into real-time prediction models is challenging due to both storage and inference costs. To address this scaling challenge, we propose a novel approach that leverages vector quantization (VQ) to inject a compact Semantic ID (SID) as input to the recommendation models instead of a collection of embeddings. Our method builds on recent works of SIDs by introducing three key innovations: (i) a multi-task VQ-VAE framework, called VQ fusion that fuses multiple content embeddings and categorical predictions into a single Semantic ID; (ii) a parameter-free, highly granular SID-to-embedding conversion technique, called SIDE, that is validated with two content embedding collections, thereby eliminating the need for a large parameterized lookup table; and (iii) a novel quantization method called Discrete-PCA (DPCA) which generalizes and enhances residual quantization techniques. The proposed enhancements when applied to a large-scale industrial ads-recommendation system achieves 2.4X improvement in normalized entropy (NE) gain and 3X reduction in data footprint compared to traditional SID methods.
Authors: Cynthia Dwork, Lunjia Hu, Han Shao
Abstract: We study a fundamental question of domain generalization: given a family of domains (i.e., data distributions), how many randomly sampled domains do we need to collect data from in order to learn a model that performs reasonably well on every seen and unseen domain in the family? We model this problem in the PAC framework and introduce a new combinatorial measure, which we call the domain shattering dimension. We show that this dimension characterizes the domain sample complexity. Furthermore, we establish a tight quantitative relationship between the domain shattering dimension and the classic VC dimension, demonstrating that every hypothesis class that is learnable in the standard PAC setting is also learnable in our setting.
Authors: Yuping Yan, Yizhi Wang, Yuanshuai Li, Yaochu Jin
Abstract: Serial pipeline training is an efficient paradigm for handling data heterogeneity in cross-silo federated learning with low communication overhead. However, even without centralized aggregation, direct transfer of models between clients can violate privacy regulations and remain susceptible to gradient leakage and linkage attacks. Additionally, ensuring resilience against semi-honest or malicious clients who may manipulate or misuse received models remains a grand challenge, particularly in privacy-sensitive domains such as healthcare. To address these challenges, we propose TriCon-SF, a novel serial federated learning framework that integrates triple shuffling and contribution awareness. TriCon-SF introduces three levels of randomization by shuffling model layers, data segments, and training sequences to break deterministic learning patterns and disrupt potential attack vectors, thereby enhancing privacy and robustness. In parallel, it leverages Shapley value methods to dynamically evaluate client contributions during training, enabling the detection of dishonest behavior and enhancing system accountability. Extensive experiments on non-IID healthcare datasets demonstrate that TriCon-SF outperforms standard serial and parallel federated learning in both accuracy and communication efficiency. Security analysis further supports its resilience against client-side privacy attacks.
Authors: Fanchen Bu, Kijung Shin
Abstract: In unsupervised combinatorial optimization (UCO), during training, one aims to have continuous decisions that are promising in a probabilistic sense for each training instance, which enables end-to-end training on initially discrete and non-differentiable problems. At the test time, for each test instance, starting from continuous decisions, derandomization is typically applied to obtain the final deterministic decisions. Researchers have developed more and more powerful test-time derandomization schemes to enhance the empirical performance and the theoretical guarantee of UCO methods. However, we notice a misalignment between training and testing in the existing UCO methods. Consequently, lower training losses do not necessarily entail better post-derandomization performance, even for the training instances without any data distribution shift. Empirically, we indeed observe such undesirable cases. We explore a preliminary idea to better align training and testing in UCO by including a differentiable version of derandomization into training. Our empirical exploration shows that such an idea indeed improves training-test alignment, but also introduces nontrivial challenges into training.
Authors: John Lazarsfeld, Georgios Piliouras, Ryann Sim, Stratis Skoulakis
Abstract: This paper studies the optimistic variant of Fictitious Play for learning in two-player zero-sum games. While it is known that Optimistic FTRL -- a regularized algorithm with a bounded stepsize parameter -- obtains constant regret in this setting, we show for the first time that similar, optimal rates are also achievable without regularization: we prove for two-strategy games that Optimistic Fictitious Play (using any tiebreaking rule) obtains only constant regret, providing surprising new evidence on the ability of non-no-regret algorithms for fast learning in games. Our proof technique leverages a geometric view of Optimistic Fictitious Play in the dual space of payoff vectors, where we show a certain energy function of the iterates remains bounded over time. Additionally, we also prove a regret lower bound of $\Omega(\sqrt{T})$ for Alternating Fictitious Play. In the unregularized regime, this separates the ability of optimism and alternation in achieving $o(\sqrt{T})$ regret.
Authors: Eion Tyacke, Kunal Gupta, Jay Patel, Rui Li
Abstract: Hand gestures are a primary output of the human motor system, yet the decoding of their neuromuscular signatures remains a bottleneck for basic neuroscience and assistive technologies such as prosthetics. Traditional human-machine interface pipelines rely on a single biosignal modality, but multimodal fusion can exploit complementary information from sensors. We systematically compare linear and attention-based fusion strategies across three architectures: a Multimodal MLP, a Multimodal Transformer, and a Hierarchical Transformer, evaluating performance on scenarios with unimodal and multimodal inputs. Experiments use two publicly available datasets: NinaPro DB2 (sEMG and accelerometer) and HD-sEMG 65-Gesture (high-density sEMG and force). Across both datasets, the Hierarchical Transformer with attention-based fusion consistently achieved the highest accuracy, surpassing the multimodal and best single-modality linear-fusion MLP baseline by over 10% on NinaPro DB2 and 3.7% on HD-sEMG. To investigate how modalities interact, we introduce an Isolation Network that selectively silences unimodal or cross-modal attention pathways, quantifying each group of token interactions' contribution to downstream decisions. Ablations reveal that cross-modal interactions contribute approximately 30% of the decision signal across transformer layers, highlighting the importance of attention-driven fusion in harnessing complementary modality information. Together, these findings reveal when and how multimodal fusion would enhance biosignal classification and also provides mechanistic insights of human muscle activities. The study would be beneficial in the design of sensor arrays for neurorobotic systems.
Authors: Kosuke Nakanishi, Akihiro Kubo, Yuji Yasui, Shin Ishii
Abstract: Recently, robust reinforcement learning (RL) methods designed to handle adversarial input observations have received significant attention, motivated by RL's inherent vulnerabilities. While existing approaches have demonstrated reasonable success, addressing worst-case scenarios over long time horizons requires both minimizing the agent's cumulative rewards for adversaries and training agents to counteract them through alternating learning. However, this process introduces mutual dependencies between the agent and the adversary, making interactions with the environment inefficient and hindering the development of off-policy methods. In this work, we propose a novel off-policy method that eliminates the need for additional environmental interactions by reformulating adversarial learning as a soft-constrained optimization problem. Our approach is theoretically supported by the symmetric property of policy evaluation between the agent and the adversary. The implementation is available at https://github.com/nakanakakosuke/VALT_SAC.
Authors: Jongmin Park, Seunghoon Han, Won-Yong Shin, Sungsu Lim
Abstract: The hyperbolic space, characterized by a constant negative curvature and exponentially expanding space, aligns well with the structural properties of heterogeneous graphs. However, although heterogeneous graphs inherently possess diverse power-law structures, most hyperbolic heterogeneous graph embedding models rely on a single hyperbolic space. This approach may fail to effectively capture the diverse power-law structures within heterogeneous graphs. To address this limitation, we propose a Metapath-based Hyperbolic Contrastive Learning framework (MHCL), which uses multiple hyperbolic spaces to capture diverse complex structures within heterogeneous graphs. Specifically, by learning each hyperbolic space to describe the distribution of complex structures corresponding to each metapath, it is possible to capture semantic information effectively. Since metapath embeddings represent distinct semantic information, preserving their discriminability is important when aggregating them to obtain node representations. Therefore, we use a contrastive learning approach to optimize MHCL and improve the discriminability of metapath embeddings. In particular, our contrastive learning method minimizes the distance between embeddings of the same metapath and maximizes the distance between those of different metapaths in hyperbolic space, thereby improving the separability of metapath embeddings with distinct semantic information. We conduct comprehensive experiments to evaluate the effectiveness of MHCL. The experimental results demonstrate that MHCL outperforms state-of-the-art baselines in various graph machine learning tasks, effectively capturing the complex structures of heterogeneous graphs.
Authors: Youjin Kong
Abstract: Fairness in machine learning (ML) has become a rapidly growing area of research. But why, in the first place, is unfairness in ML morally wrong? And why should we care about improving fairness? Most fair-ML research implicitly appeals to distributive equality: the idea that desirable goods and benefits, such as opportunities (e.g., Barocas et al., 2023), should be equally distributed across society. Unfair ML models, then, are seen as wrong because they unequally distribute such benefits. This paper argues that this exclusive focus on distributive equality offers an incomplete and potentially misleading ethical foundation. Grounding ML fairness in egalitarianism -- the view that equality is a fundamental moral and social ideal -- requires challenging structural inequality: systematic, institutional, and durable arrangements that privilege some groups while disadvantaging others. Structural inequality manifests through ML systems in two primary forms: allocative harms (e.g., economic loss) and representational harms (e.g., stereotypes, erasure). While distributive equality helps address allocative harms, it fails to explain why representational harms are wrong -- why it is wrong for ML systems to reinforce social hierarchies that stratify people into superior and inferior groups -- and why ML systems should aim to foster a society where people relate as equals (i.e., relational equality). To address these limitations, the paper proposes a multifaceted egalitarian framework for ML fairness that integrates both distributive and relational equality. Drawing on critical social and political philosophy, this framework offers a more comprehensive ethical foundation for tackling the full spectrum of harms perpetuated by ML systems. The paper also outlines practical pathways for implementing the framework across the ML pipeline.
Authors: Jiashun Cheng, Aochuan Chen, Nuo Chen, Ziqi Gao, Yuhan Li, Jia Li, Fugee Tsung
Abstract: Low-Rank Adaptation (LoRA) has emerged as a prominent technique for fine-tuning large foundation models. Despite its successes, the substantial parameter redundancy, which limits the capacity and efficiency of LoRA, has been recognized as a bottleneck. In this work, we systematically investigate the impact of redundancy in fine-tuning LoRA and reveal that reducing density redundancy does not degrade expressiveness. Based on this insight, we introduce \underline{S}pectral-\underline{e}ncoding \underline{L}ow-\underline{R}ank \underline{A}daptation (SeLoRA), which harnesses the robust expressiveness of spectral bases to re-parameterize LoRA from a sparse spectral subspace. Designed with simplicity, SeLoRA enables seamless integration with various LoRA variants for performance boosting, serving as a scalable plug-and-play framework. Extensive experiments substantiate that SeLoRA achieves greater efficiency with fewer parameters, delivering superior performance enhancements over strong baselines on various downstream tasks, including commonsense reasoning, math reasoning, and code generation.
Authors: Senmiao Wang, Yupeng Chen, Yushun Zhang, Ruoyu Sun, Tian Ding
Abstract: Graph Neural Networks (GNNs) often suffer from performance degradation as the network depth increases. This paper addresses this issue by introducing initialization methods that enhance signal propagation (SP) within GNNs. We propose three key metrics for effective SP in GNNs: forward propagation, backward propagation, and graph embedding variation (GEV). While the first two metrics derive from classical SP theory, the third is specifically designed for GNNs. We theoretically demonstrate that a broad range of commonly used initialization methods for GNNs, which exhibit performance degradation with increasing depth, fail to control these three metrics simultaneously. To deal with this limitation, a direct exploitation of the SP analysis--searching for weight initialization variances that optimize the three metrics--is shown to significantly enhance the SP in deep GCNs. This approach is called Signal Propagation on Graph-guided Initialization (SPoGInit). Our experiments demonstrate that SPoGInit outperforms commonly used initialization methods on various tasks and architectures. Notably, SPoGInit enables performance improvements as GNNs deepen, which represents a significant advancement in addressing depth-related challenges and highlights the validity and effectiveness of the SP analysis framework.
Authors: Nick Erickson, Lennart Purucker, Andrej Tschalzev, David Holzm\"uller, Prateek Mutalik Desai, and David Salinas, Frank Hutter
Abstract: With the growing popularity of deep learning and foundation models for tabular data, the need for standardized and reliable benchmarks is higher than ever. However, current benchmarks are static. Their design is not updated even if flaws are discovered, model versions are updated, or new models are released. To address this, we introduce TabArena, the first continuously maintained living tabular benchmarking system. To launch TabArena, we manually curate a representative collection of datasets and well-implemented models, conduct a large-scale benchmarking study to initialize a public leaderboard, and assemble a team of experienced maintainers. Our results highlight the influence of validation method and ensembling of hyperparameter configurations to benchmark models at their full potential. While gradient-boosted trees are still strong contenders on practical tabular datasets, we observe that deep learning methods have caught up under larger time budgets with ensembling. At the same time, foundation models excel on smaller datasets. Finally, we show that ensembles across models advance the state-of-the-art in tabular machine learning and investigate the contributions of individual models. We launch TabArena with a public leaderboard, reproducible code, and maintenance protocols to create a living benchmark available at https://tabarena.ai.
URLs: https://tabarena.ai.
Authors: Kai Yang, Shaoyu Dou, Pan Luo, Xin Wang, H. Vincent Poor
Abstract: Many real-world multivariate time series are collected from a network of physical objects embedded with software, electronics, and sensors. The quasi-periodic signals generated by these objects often follow a similar repetitive and periodic pattern, but have variations in the period, and come in different lengths caused by timing (synchronization) errors. Given a multitude of such quasi-periodic time series, can we build machine learning models to identify those time series that behave differently from the majority of the observations? In addition, can the models help human experts to understand how the decision was made? We propose a sequence to Gaussian Mixture Model (seq2GMM) framework. The overarching goal of this framework is to identify unusual and interesting time series within a network time series database. We further develop a surrogate-based optimization algorithm that can efficiently train the seq2GMM model. Seq2GMM exhibits strong empirical performance on a plurality of public benchmark datasets, outperforming state-of-the-art anomaly detection techniques by a significant margin. We also theoretically analyze the convergence property of the proposed training algorithm and provide numerical results to substantiate our theoretical claims.
Authors: Thomas Marwitz, Alexander Colsmann, Ben Breitung, Christoph Brabec, Christoph Kirchlechner, Eva Blasco, Gabriel Cadilha Marques, Horst Hahn, Michael Hirtz, Pavel A. Levkin, Yolita M. Eggeler, Tobias Schl\"oder, Pascal Friederich
Abstract: Due to an exponential increase in published research articles, it is impossible for individual scientists to read all publications, even within their own research field. In this work, we investigate the use of large language models (LLMs) for the purpose of extracting the main concepts and semantic information from scientific abstracts in the domain of materials science to find links that were not noticed by humans and thus to suggest inspiring near/mid-term future research directions. We show that LLMs can extract concepts more efficiently than automated keyword extraction methods to build a concept graph as an abstraction of the scientific literature. A machine learning model is trained to predict emerging combinations of concepts, i.e. new research ideas, based on historical data. We demonstrate that integrating semantic concept information leads to an increased prediction performance. The applicability of our model is demonstrated in qualitative interviews with domain experts based on individualized model suggestions. We show that the model can inspire materials scientists in their creative thinking process by predicting innovative combinations of topics that have not yet been investigated.
Authors: Zeyneddin Oz, Shreyas Korde, Marius Bock, Kristof Van Laerhoven
Abstract: Rapid evolution of sensors and resource-efficient machine learning models have spurred the widespread adoption of wearable fitness tracking devices. Equipped with inertial sensors, such devices can continuously capture physical movements for fitness technology (FitTech), enabling applications from sports optimization to preventive healthcare. Traditional centralized learning approaches to detect fitness activities struggle with privacy concerns, regulatory constraints, and communication inefficiencies. In contrast, Federated Learning (FL) enables a decentralized model training by communicating model updates rather than private wearable sensor data. Applying FL to FitTech presents unique challenges, such as data imbalance, lack of labelled data, heterogeneous user activity patterns, and trade-offs between personalization and generalization. To simplify research on FitTech in FL, we present the FedFitTech baseline, under the Flower framework, which is publicly available and widely used by both industry and academic researchers. Additionally, to illustrate its usage, this paper presents a case study that implements a system based on the FedFitTech baseline, incorporating a client-side early stopping strategy and comparing the results. For instance, this system allows wearable devices to optimize the trade-off between capturing common fitness activity patterns and preserving individuals' nuances, thereby enhancing both the scalability and efficiency of privacy-aware fitness tracking applications. Results show that this reduces overall redundant communications by 13 percent, while maintaining the overall recognition performance at a negligible recognition cost by 1 percent. Thus, FedFitTech baseline creates a foundation for a wide range of new research and development opportunities in FitTech, and it is available as open-source at: https://github.com/adap/flower/tree/main/baselines/fedfittech
URLs: https://github.com/adap/flower/tree/main/baselines/fedfittech
Authors: Victor Alejandre (Universidad Politecnica de Madrid, Spain), Concha Bielza (Universidad Politecnica de Madrid, Spain), Pedro Larra\~naga (Universidad Politecnica de Madrid, Spain)
Abstract: Semiparametric Bayesian networks (SPBNs) integrate parametric and non-parametric probabilistic models, offering flexibility in learning complex data distributions from samples. In particular, kernel density estimators (KDEs) are employed for the non-parametric component. Under the assumption of data normality, the normal rule is used to learn the bandwidth matrix for the KDEs in SPBNs. This matrix is the key hyperparameter that controls the trade-off between bias and variance. However, real-world data often deviates from normality, potentially leading to suboptimal density estimation and reduced predictive performance. This paper first establishes the theoretical framework for the application of state-of-the-art bandwidth selectors and subsequently evaluates their impact on SPBN performance. We explore the approaches of cross-validation and plug-in selectors, assessing their effectiveness in enhancing the learning capability and applicability of SPBNs. To support this investigation, we have extended the open-source package PyBNesian for SPBNs with the additional bandwidth selection techniques and conducted extensive experimental analyses. Our results demonstrate that the proposed bandwidth selectors leverage increasing information more effectively than the normal rule, which, despite its robustness, stagnates with more data. In particular, unbiased cross-validation generally outperforms the normal rule, highlighting its advantage in high sample size scenarios.
Authors: Antonio Consoloa, Edoardo Amaldi, Emilio Carrizosa
Abstract: Decision trees are popular in survival analysis for their interpretability and ability to model complex relationships. Survival trees, which predict the timing of singular events using censored historical data, are typically built through heuristic approaches. Recently, there has been growing interest in globally optimized trees, where the overall tree is trained by minimizing the error function over all its parameters. We propose a new soft survival tree model (SST), with a soft splitting rule at each branch node, trained via a nonlinear optimization formulation amenable to decomposition. Since SSTs provide for every input vector a specific survival function associated to a single leaf node, they satisfy the conditional computation property and inherit the related benefits. SST and the training formulation combine flexibility with interpretability: any smooth survival function (parametric, semiparametric, or nonparametric) estimated through maximum likelihood can be used, and each leaf node of an SST yields a cluster of distinct survival functions which are associated to the data points routed to it. Numerical experiments on 15 well-known datasets show that SSTs, with parametric and spline-based semiparametric survival functions, trained using an adaptation of the node-based decomposition algorithm proposed by Consolo et al. (2024) for soft regression trees, outperform three benchmark survival trees in terms of four widely-used discrimination and calibration measures. SSTs can also be extended to consider group fairness.
Authors: Semin Kim, Yeonwoo Cha, Jaehoon Yoo, Seunghoon Hong
Abstract: We investigate a general approach for improving user prompts in text-to-image (T2I) diffusion models by finding prompts that maximize a reward function specified at test-time. Although diverse reward models are used for evaluating image generation, existing automated prompt engineering methods typically target specific reward configurations. Consequently, these specialized designs exhibit suboptimal performance when applied to new prompt engineering scenarios involving different reward models. To address this limitation, we introduce RATTPO (Reward-Agnostic Test-Time Prompt Optimization), a flexible test-time optimization method applicable across various reward scenarios without modification. RATTPO iteratively searches for optimized prompts by querying large language models (LLMs) \textit{without} requiring reward-specific task descriptions. Instead, it uses the optimization trajectory and a novel reward-aware feedback signal (termed a "hint") as context. Empirical results demonstrate the versatility of RATTPO, effectively enhancing user prompts across diverse reward setups that assess various generation aspects, such as aesthetics, general human preference, or spatial relationships between objects. RATTPO surpasses other test-time search baselines in search efficiency, using up to 3.5 times less inference budget, and, given sufficient inference budget, achieves performance comparable to learning-based baselines that require reward-specific fine-tuning. The code is available at https://github.com/seminkim/RATTPO.
Authors: Shaoyu Dou, Kai Yang, Yang Jiao, Chengbo Qiu, Kui Ren
Abstract: Time series analysis has achieved great success in cyber security such as intrusion detection and device identification. Learning similarities among multiple time series is a crucial problem since it serves as the foundation for downstream analysis. Due to the complex temporal dynamics of the event-triggered time series, it often remains unclear which similarity metric is appropriate for security-related tasks, such as anomaly detection and clustering. The overarching goal of this paper is to develop an unsupervised learning framework that is capable of learning similarities among a set of event-triggered time series. From the machine learning vantage point, the proposed framework harnesses the power of both hierarchical multi-resolution sequential autoencoders and the Gaussian Mixture Model (GMM) to effectively learn the low-dimensional representations from the time series. Finally, the obtained similarity measure can be easily visualized for the explanation. The proposed framework aspires to offer a stepping stone that gives rise to a systematic approach to model and learn similarities among a multitude of event-triggered time series. Through extensive qualitative and quantitative experiments, it is revealed that the proposed method outperforms state-of-the-art methods considerably.
Authors: Qian Qi
Abstract: Determining the optimal depth of a neural network is a fundamental yet challenging problem, typically resolved through resource-intensive experimentation. This paper introduces a formal theoretical framework to address this question by recasting the forward pass of a deep network, specifically a Residual Network (ResNet), as an optimal stopping problem. We model the layer-by-layer evolution of hidden representations as a sequential decision process where, at each layer, a choice is made between halting computation to make a prediction or continuing to a deeper layer for a potentially more refined representation. This formulation captures the intrinsic trade-off between accuracy and computational cost. Our primary theoretical contribution is a proof that, under a plausible condition of diminishing returns on the residual functions, the expected optimal stopping depth is provably finite, even in an infinite-horizon setting. We leverage this insight to propose a novel and practical regularization term, $\mathcal{L}_{\rm depth}$, that encourages the network to learn representations amenable to efficient, early exiting. We demonstrate the generality of our framework by extending it to the Transformer architecture and exploring its connection to continuous-depth models via free-boundary problems. Empirical validation on ImageNet confirms that our regularizer successfully induces the theoretically predicted behavior, leading to significant gains in computational efficiency without compromising, and in some cases improving, final model accuracy.
Authors: Jacopo Graldi, Alessandro Breccia, Giulia Lanzillotta, Thomas Hofmann, Lorenzo Noci
Abstract: Despite recent efforts, neural networks still struggle to learn in non-stationary environments, and our understanding of catastrophic forgetting (CF) is far from complete. In this work, we perform a systematic study on the impact of model scale and the degree of feature learning in continual learning. We reconcile existing contradictory observations on scale in the literature, by differentiating between lazy and rich training regimes through a variable parameterization of the architecture. We show that increasing model width is only beneficial when it reduces the amount of feature learning, yielding more laziness. Using the framework of dynamical mean field theory, we then study the infinite width dynamics of the model in the feature learning regime and characterize CF, extending prior theoretical results limited to the lazy regime. We study the intricate relationship between feature learning, task non-stationarity, and forgetting, finding that high feature learning is only beneficial with highly similar tasks. We identify a transition modulated by task similarity where the model exits an effectively lazy regime with low forgetting to enter a rich regime with significant forgetting. Finally, our findings reveal that neural networks achieve optimal performance at a critical level of feature learning, which depends on task non-stationarity and transfers across model scales. This work provides a unified perspective on the role of scale and feature learning in continual learning.
Authors: Sebastian H\"onel, Jonas Nordqvist
Abstract: The detection and localization of quality-related problems in industrially mass-produced products has historically relied on manual inspection, which is costly and error-prone. Machine learning has the potential to replace manual handling. As such, the desire is to facilitate an unsupervised (or self-supervised) approach, as it is often impossible to specify all conceivable defects ahead of time. A plethora of prior works have demonstrated the aptitude of common reconstruction-, embedding-, and synthesis-based methods in laboratory settings. However, in practice, we observe that most methods do not handle low data quality well or exude low robustness in unfavorable, but typical real-world settings. For practitioners it may be very difficult to identify the actual underlying problem when such methods underperform. Worse, often-reported metrics (e.g., AUROC) are rarely suitable in practice and may give misleading results. In our setting, we attempt to identify subtle anomalies on the surface of blasted forged metal parts, using rather low-quality RGB imagery only, which is a common industrial setting. We specifically evaluate two types of state-of-the-art models that allow us to identify and improve quality issues in production data, without having to obtain new data. Our contribution is to provide guardrails for practitioners that allow them to identify problems related to, e.g., (lack of) robustness or invariance, in either the chosen model or the data reliably in similar scenarios. Furthermore, we exemplify common pitfalls in and shortcomings of likelihood-based approaches and outline a framework for proper empirical risk estimation that is more suitable for real-world scenarios.
Authors: Mohon Raihan, Plabon Kumar Saha, Rajan Das Gupta, A Z M Tahmidul Kabir, Afia Anjum Tamanna, Md. Harun-Ur-Rashid, Adnan Bin Abdus Salam, Md Tanvir Anjum, A Z M Ahteshamul Kabir
Abstract: Neonatal death is still a concerning reality for underdeveloped and even some developed countries. Worldwide data indicate that 26.693 babies out of 1,000 births die, according to Macro Trades. To reduce this number, early prediction of endangered babies is crucial. Such prediction enables the opportunity to take ample care of the child and mother so that early child death can be avoided. In this context, machine learning was used to determine whether a newborn baby is at risk. To train the predictive model, historical data of 1.4 million newborns was used. Machine learning and deep learning techniques such as logical regression, K-nearest neighbor, random forest classifier, extreme gradient boosting (XGBoost), convolutional neural network, and long short-term memory (LSTM) were implemented using the dataset to identify the most accurate model for predicting neonatal mortality. Among the machine learning algorithms, XGBoost and random forest classifier achieved the best accuracy with 94%, while among the deep learning models, LSTM delivered the highest accuracy with 99%. Therefore, using LSTM appears to be the most suitable approach to predict whether precautionary measures for a child are necessary.
Authors: \c{C}a\u{g}atay Demirel
Abstract: Ensemble learning remains a cornerstone of machine learning, with stacking used to integrate predictions from multiple base learners through a meta-model. However, deep stacking remains rare, as most designs prioritize horizontal diversity over recursive depth due to model complexity, feature redundancy, and computational burden. To address these challenges, RocketStack, a level-aware recursive ensemble framework, is introduced and explored up to ten stacking levels, extending beyond prior architectures. The framework incrementally prunes weaker learners at each level, enabling deeper stacking without excessive complexity. To mitigate early performance saturation, mild Gaussian noise is added to out-of-fold (OOF) scores before pruning, and compared against strict OOF pruning. Further both per-level and periodic feature compressions are explored using attention-based selection, Simple, Fast, Efficient (SFE) filter, and autoencoders. Across 33 datasets (23 binary, 10 multi-class), linear-trend tests confirmed rising accuracy with depth in most variants, and the top performing meta-model at each level increasingly outperformed the strongest standalone ensemble. In the binary subset, periodic SFE with mild OOF-score randomization reached 97.08% at level 10, 5.14% above the strict-pruning configuration and cut runtime by 10.5% relative to no compression. In the multi-class subset, periodic attention selection reached 98.60% at level 10, exceeding the strongest baseline by 6.11%, while reducing runtime by 56.1% and feature dimensionality by 74% compared to no compression. These findings highlight mild randomization as an effective regularizer and periodic compression as a stabilizer. Echoing the design of multistage rockets in aerospace (prune, compress, propel) RocketStack achieves deep recursive ensembling with tractable complexity.
Authors: Guan Zhe Hong, Bhavya Vasudeva, Vatsal Sharan, Cyrus Rashtchian, Prabhakar Raghavan, Rina Panigrahy
Abstract: When large language models (LLMs) use in-context learning (ICL) to solve a new task, they seem to grasp not only the goal of the task but also core, latent concepts in the demonstration examples. This begs the question of whether transformers represent latent structures as part of their computation or whether they take shortcuts to solve the problem. Prior mechanistic work on ICL does not address this question because it does not sufficiently examine the relationship between the learned representation and the latent concept, and the considered problem settings often involve only single-step reasoning. In this work, we examine how transformers disentangle and use latent concepts. We show that in 2-hop reasoning tasks with a latent, discrete concept, the model successfully identifies the latent concept and does step-by-step concept composition. In tasks parameterized by a continuous latent concept, we find low-dimensional subspaces in the representation space where the geometry mimics the underlying parameterization. Together, these results refine our understanding of ICL and the representation of transformers, and they provide evidence for highly localized structures in the model that disentangle latent concepts in ICL tasks.
Authors: Marco Jiralerspong, Esther Derman, Danilo Vucetic, Nikolay Malkin, Bilun Sun, Tianyu Zhang, Pierre-Luc Bacon, Gauthier Gidel
Abstract: A major bottleneck in scientific discovery involves narrowing a large combinatorial set of objects, such as proteins or molecules, to a small set of promising candidates. While this process largely relies on expert knowledge, recent methods leverage reinforcement learning (RL) to enhance this filtering. They achieve this by estimating proxy reward functions from available datasets and using regularization to generate more diverse candidates. These reward functions are inherently uncertain, raising a particularly salient challenge for scientific discovery. In this work, we show that existing methods, often framed as sampling proportional to a reward function, are inadequate and yield suboptimal candidates, especially in large search spaces. To remedy this issue, we take a robust RL approach and introduce a unified operator that seeks robustness to the uncertainty of the proxy reward function. This general operator targets peakier sampling distributions while encompassing known soft RL operators. It also leads us to a novel algorithm that identifies higher-quality, diverse candidates in both synthetic and real-world tasks. Ultimately, our work offers a new, flexible perspective on discrete compositional generation tasks. Code: https://github.com/marcojira/tgm.
Authors: Giulia Bertazzini, Chiara Albisani, Daniele Baracchi, Dasara Shullani, Roberto Verdecchia
Abstract: With the growing adoption of AI image generation, in conjunction with the ever-increasing environmental resources demanded by AI, we are urged to answer a fundamental question: What is the environmental impact hidden behind each image we generate? In this research, we present a comprehensive empirical experiment designed to assess the energy consumption of AI image generation. Our experiment compares 17 state-of-the-art image generation models by considering multiple factors that could affect their energy consumption, such as model quantization, image resolution, and prompt length. Additionally, we consider established image quality metrics to study potential trade-offs between energy consumption and generated image quality. Results show that image generation models vary drastically in terms of the energy they consume, with up to a 46x difference. Image resolution affects energy consumption inconsistently, ranging from a 1.3x to 4.7x increase when doubling resolution. U-Net-based models tend to consume less than Transformer-based one. Model quantization instead results to deteriorate the energy efficiency of most models, while prompt length and content have no statistically significant impact. Improving image quality does not always come at the cost of a higher energy consumption, with some of the models producing the highest quality images also being among the most energy efficient ones.
Authors: Leizhen Wang, Peibo Duan, Cheng Lyu, Zewen Wang, Zhiqiang He, Nan Zheng, Zhenliang Ma
Abstract: The evolution of metropolitan cities and the increase in travel demands impose stringent requirements on traffic assignment methods. Multi-agent reinforcement learning (MARL) approaches outperform traditional methods in modeling adaptive routing behavior without requiring explicit system dynamics, which is beneficial for real-world deployment. However, MARL frameworks face challenges in scalability and reliability when managing extensive networks with substantial travel demand, which limiting their practical applicability in solving large-scale traffic assignment problems. To address these challenges, this study introduces MARL-OD-DA, a new MARL framework for the traffic assignment problem, which redefines agents as origin-destination (OD) pair routers rather than individual travelers, significantly enhancing scalability. Additionally, a Dirichlet-based action space with action pruning and a reward function based on the local relative gap are designed to enhance solution reliability and improve convergence efficiency. Experiments demonstrate that the proposed MARL framework effectively handles medium-sized networks with extensive and varied city-level OD demand, surpassing existing MARL methods. When implemented in the SiouxFalls network, MARL-OD-DA achieves better assignment solutions in 10 steps, with a relative gap that is 94.99% lower than that of conventional methods.
Authors: Jo\~ao Matos, Ben Van Calster, Leo Anthony Celi, Paula Dhiman, Judy Wawira Gichoya, Richard D. Riley, Chris Russell, Sara Khalid, Gary S. Collins
Abstract: Predictive artificial intelligence (AI) offers an opportunity to improve clinical practice and patient outcomes, but risks perpetuating biases if fairness is inadequately addressed. However, the definition of "fairness" remains unclear. We conducted a scoping review to identify and critically appraise fairness metrics for clinical predictive AI. We defined a "fairness metric" as a measure quantifying whether a model discriminates (societally) against individuals or groups defined by sensitive attributes. We searched five databases (2014-2024), screening 820 records, to include 41 studies, and extracted 62 fairness metrics. Metrics were classified by performance-dependency, model output level, and base performance metric, revealing a fragmented landscape with limited clinical validation and overreliance on threshold-dependent measures. Eighteen metrics were explicitly developed for healthcare, including only one clinical utility metric. Our findings highlight conceptual challenges in defining and quantifying fairness and identify gaps in uncertainty quantification, intersectionality, and real-world applicability. Future work should prioritise clinically meaningful metrics.
Authors: Elizabeth Fons, Alejandro Sztrajman, Yousef El-Laham, Luciana Ferrer, Svitlana Vyetrenko, Manuela Veloso
Abstract: Time series with missing or irregularly sampled data are a persistent challenge in machine learning. Many methods operate on the frequency-domain, relying on the Fast Fourier Transform (FFT) which assumes uniform sampling, therefore requiring prior interpolation that can distort the spectra. To address this limitation, we introduce a differentiable Lomb--Scargle layer that enables a reliable computation of the power spectrum of irregularly sampled data. We integrate this layer into a novel score-based diffusion model (LSCD) for time series imputation conditioned on the entire signal spectrum. Experiments on synthetic and real-world benchmarks demonstrate that our method recovers missing data more accurately than purely time-domain baselines, while simultaneously producing consistent frequency estimates. Crucially, our method can be easily integrated into learning frameworks, enabling broader adoption of spectral guidance in machine learning approaches involving incomplete or irregular data.
Authors: Joshua Schraven, Alexander Windmann, Oliver Niggemann
Abstract: Benchmark datasets for network intrusion detection commonly rely on synthetically generated traffic, which fails to reflect the statistical variability and temporal drift encountered in operational environments. This paper introduces MAWIFlow, a flow-based benchmark derived from the MAWILAB v1.1 dataset, designed to enable realistic and reproducible evaluation of anomaly detection methods. A reproducible preprocessing pipeline is presented that transforms raw packet captures into flow representations conforming to the CICFlowMeter format, while preserving MAWILab's original anomaly labels. The resulting datasets comprise temporally distinct samples from January 2011, 2016, and 2021, drawn from trans-Pacific backbone traffic. To establish reference baselines, traditional machine learning methods, including Decision Trees, Random Forests, XGBoost, and Logistic Regression, are compared to a deep learning model based on a CNN-BiLSTM architecture. Empirical results demonstrate that tree-based classifiers perform well on temporally static data but experience significant performance degradation over time. In contrast, the CNN-BiLSTM model maintains better performance, thus showing improved generalization. These findings underscore the limitations of synthetic benchmarks and static models, and motivate the adoption of realistic datasets with explicit temporal structure. All datasets, pipeline code, and model implementations are made publicly available to foster transparency and reproducibility.
Authors: Haolin Liu, Adrien Siproudhis, Samuel Experton, Peter Lorenz, Christina Boura, Thomas Peyrin
Abstract: Neural network model extraction has emerged in recent years as an important security concern, as adversaries attempt to recover a network's parameters via black-box queries. A key step in this process is signature extraction, which aims to recover the absolute values of the network's weights layer by layer. Prior work, notably by Carlini et al. (2020), introduced a technique inspired by differential cryptanalysis to extract neural network parameters. However, their method suffers from several limitations that restrict its applicability to networks with a few layers only. Later works focused on improving sign extraction, but largely relied on the assumption that signature extraction itself was feasible. In this work, we revisit and refine the signature extraction process by systematically identifying and addressing for the first time critical limitations of Carlini et al.'s signature extraction method. These limitations include rank deficiency and noise propagation from deeper layers. To overcome these challenges, we propose efficient algorithmic solutions for each of the identified issues, greatly improving the efficiency of signature extraction. Our approach permits the extraction of much deeper networks than was previously possible. We validate our method through extensive experiments on ReLU-based neural networks, demonstrating significant improvements in extraction depth and accuracy. For instance, our extracted network matches the target network on at least 95% of the input space for each of the eight layers of a neural network trained on the CIFAR-10 dataset, while previous works could barely extract the first three layers. Our results represent a crucial step toward practical attacks on larger and more complex neural network architectures.
Authors: Jingtong Su, Julia Kempe, Karen Ullrich
Abstract: Transformers have achieved state-of-the-art performance across language and vision tasks. This success drives the imperative to interpret their internal mechanisms with the dual goals of enhancing performance and improving behavioral control. Attribution methods help advance interpretability by assigning model outputs associated with a target concept to specific model components. Current attribution research primarily studies multi-layer perceptron neurons and addresses relatively simple concepts such as factual associations (e.g., Paris is located in France). This focus tends to overlook the impact of the attention mechanism and lacks a unified approach for analyzing more complex concepts. To fill these gaps, we introduce Scalable Attention Module Discovery (SAMD), a concept-agnostic method for mapping arbitrary, complex concepts to specific attention heads of general transformer models. We accomplish this by representing each concept as a vector, calculating its cosine similarity with each attention head, and selecting the TopK-scoring heads to construct the concept-associated attention module. We then propose Scalar Attention Module Intervention (SAMI), a simple strategy to diminish or amplify the effects of a concept by adjusting the attention module using only a single scalar parameter. Empirically, we demonstrate SAMD on concepts of varying complexity, and visualize the locations of their corresponding modules. Our results demonstrate that module locations remain stable before and after LLM post-training, and confirm prior work on the mechanics of LLM multilingualism. Through SAMI, we facilitate jailbreaking on HarmBench (+72.7%) by diminishing "safety" and improve performance on the GSM8K benchmark (+1.6%) by amplifying "reasoning". Lastly, we highlight the domain-agnostic nature of our approach by suppressing the image classification accuracy of vision transformers on ImageNet.
Authors: Abdellah Rahmani, Pascal Frossard
Abstract: Understanding causal relationships in multivariate time series is crucial in many scenarios, such as those dealing with financial or neurological data. Many such time series exhibit multiple regimes, i.e., consecutive temporal segments with a priori unknown boundaries, with each regime having its own causal structure. Inferring causal dependencies and regime shifts is critical for analyzing the underlying processes. However, causal structure learning in this setting is challenging due to (1) non stationarity, i.e., each regime can have its own causal graph and mixing function, and (2) complex noise distributions, which may be non Gaussian or heteroscedastic. Existing causal discovery approaches cannot address these challenges, since generally assume stationarity or Gaussian noise with constant variance. Hence, we introduce FANTOM, a unified framework for causal discovery that handles non stationary processes along with non Gaussian and heteroscedastic noises. FANTOM simultaneously infers the number of regimes and their corresponding indices and learns each regime's Directed Acyclic Graph. It uses a Bayesian Expectation Maximization algorithm that maximizes the evidence lower bound of the data log likelihood. On the theoretical side, we prove, under mild assumptions, that temporal heteroscedastic causal models, introduced in FANTOM's formulation, are identifiable in both stationary and non stationary settings. In addition, extensive experiments on synthetic and real data show that FANTOM outperforms existing methods.
Authors: Konstantin Usevich, Clara D\'erand, Ricardo Borsoi, Marianne Clausel
Abstract: Polynomial Neural Networks (PNNs) possess a rich algebraic and geometric structure. However, their identifiability -- a key property for ensuring interpretability -- remains poorly understood. In this work, we present a comprehensive analysis of the identifiability of deep PNNs, including architectures with and without bias terms. Our results reveal an intricate interplay between activation degrees and layer widths in achieving identifiability. As special cases, we show that architectures with non-increasing layer widths are generically identifiable under mild conditions, while encoder-decoder networks are identifiable when the decoder widths do not grow too rapidly. Our proofs are constructive and center on a connection between deep PNNs and low-rank tensor decompositions, and Kruskal-type uniqueness theorems. This yields both generic conditions determined by the architecture, and effective conditions that depend on the network's parameters. We also settle an open conjecture on the expected dimension of PNN's neurovarieties, and provide new bounds on the activation degrees required for it to reach its maximum.
Authors: Shruti Sadanand Dongare, Amun Kharel, Jonathan Samuel, Xiaona Zhou
Abstract: This paper introduces TransDreamerV3, a reinforcement learning model that enhances the DreamerV3 architecture by integrating a transformer encoder. The model is designed to improve memory and decision-making capabilities in complex environments. We conducted experiments on Atari-Boxing, Atari-Freeway, Atari-Pong, and Crafter tasks, where TransDreamerV3 demonstrated improved performance over DreamerV3, particularly in the Atari-Freeway and Crafter tasks. While issues in the Minecraft task and limited training across all tasks were noted, TransDreamerV3 displays advancement in world model-based reinforcement learning, leveraging transformer architectures.
Authors: Botao Zhu, Xianbin Wang
Abstract: Trust is emerging as an effective tool to ensure the successful completion of collaborative tasks within collaborative systems. However, rapidly and continuously evaluating the trustworthiness of collaborators during task execution is a significant challenge due to distributed devices, complex operational environments, and dynamically changing resources. To tackle this challenge, this paper proposes a Siamese-enabled rapid and continuous trust evaluation framework (SRCTE) to facilitate effective task collaboration. First, the communication and computing resource attributes of the collaborator in a trusted state, along with historical collaboration data, are collected and represented using an attributed control flow graph (ACFG) that captures trust-related semantic information and serves as a reference for comparison with data collected during task execution. At each time slot of task execution, the collaborator's communication and computing resource attributes, as well as task completion effectiveness, are collected in real time and represented with an ACFG to convey their trust-related semantic information. A Siamese model, consisting of two shared-parameter Structure2vec networks, is then employed to learn the deep semantics of each pair of ACFGs and generate their embeddings. Finally, the similarity between the embeddings of each pair of ACFGs is calculated to determine the collaborator's trust value at each time slot. A real system is built using two Dell EMC 5200 servers and a Google Pixel 8 to test the effectiveness of the proposed SRCTE framework. Experimental results demonstrate that SRCTE converges rapidly with only a small amount of data and achieves a high anomaly trust detection rate compared to the baseline algorithm.
Authors: Michael Plainer, Hao Wu, Leon Klein, Stephan G\"unnemann, Frank No\'e
Abstract: Diffusion models have recently gained significant attention due to their effectiveness in various scientific domains, including biochemistry. When trained on equilibrium molecular distributions, diffusion models provide both: a generative procedure to sample equilibrium conformations and associated forces derived from the model's scores. However, using the forces for coarse-grained molecular dynamics simulations uncovers inconsistencies in the samples generated via classical diffusion inference and simulation, despite both originating from the same model. Particularly at the small diffusion timesteps required for simulations, diffusion models fail to satisfy the Fokker-Planck equation, which governs how the score should evolve over time. We interpret this deviation as an indication of the observed inconsistencies and propose an energy-based diffusion model with a Fokker-Planck-derived regularization term enforcing consistency. We demonstrate the effectiveness of our approach on toy systems, alanine dipeptide, and introduce a state-of-the-art transferable Boltzmann emulator for dipeptides that supports simulation and demonstrates enhanced consistency and efficient sampling.
Authors: Samin Yeasar Arnob, Scott Fujimoto, Doina Precup
Abstract: In this paper, we investigate the use of small datasets in the context of offline reinforcement learning (RL). While many common offline RL benchmarks employ datasets with over a million data points, many offline RL applications rely on considerably smaller datasets. We show that offline RL algorithms can overfit on small datasets, resulting in poor performance. To address this challenge, we introduce "Sparse-Reg": a regularization technique based on sparsity to mitigate overfitting in offline reinforcement learning, enabling effective learning in limited data settings and outperforming state-of-the-art baselines in continuous control.
Authors: Vitalii Bondar, Vira Babenko, Roman Trembovetskyi, Yurii Korobeinyk, Viktoriya Dzyuba
Abstract: This paper introduces a unified theoretical perspective that views deep generative models as probability transformation functions. Despite the apparent differences in architecture and training methodologies among various types of generative models - autoencoders, autoregressive models, generative adversarial networks, normalizing flows, diffusion models, and flow matching - we demonstrate that they all fundamentally operate by transforming simple predefined distributions into complex target data distributions. This unifying perspective facilitates the transfer of methodological improvements between model architectures and provides a foundation for developing universal theoretical approaches, potentially leading to more efficient and effective generative modeling techniques.
Authors: Yuli Slavutsky, Ozgur Beker, David Blei, Bianca Dumitrascu
Abstract: Disentangled representations enable models to separate factors of variation that are shared across experimental conditions from those that are condition-specific. This separation is essential in domains such as biomedical data analysis, where generalization to new treatments, patients, or species depends on isolating stable biological signals from context-dependent effects. While extensions of the variational autoencoder (VAE) framework have been proposed to address this problem, they frequently suffer from leakage between latent representations, limiting their ability to generalize to unseen conditions. Here, we introduce DISCoVeR, a new variational framework that explicitly separates condition-invariant and condition-specific factors. DISCoVeR integrates three key components: (i) a dual-latent architecture that models shared and specific factors separately; (ii) two parallel reconstructions that ensure both representations remain informative; and (iii) a novel max-min objective that encourages clean separation without relying on handcrafted priors, while making only minimal assumptions. Theoretically, we show that this objective maximizes data likelihood while promoting disentanglement, and that it admits a unique equilibrium. Empirically, we demonstrate that DISCoVeR achieves improved disentanglement on synthetic datasets, natural images, and single-cell RNA-seq data. Together, these results establish DISCoVeR as a principled approach for learning disentangled representations in multi-condition settings.
Authors: Kanumuri Nithin Varma, Babak Hassibi
Abstract: Most modern learning problems are over-parameterized, where the number of learnable parameters is much greater than the number of training data points. In this over-parameterized regime, the training loss typically has infinitely many global optima that completely interpolate the data with varying generalization performance. The particular global optimum we converge to depends on the implicit bias of the optimization algorithm. The question we address in this paper is, ``What is the implicit bias that leads to the best generalization performance?". To find the optimal implicit bias, we provide a precise asymptotic analysis of the generalization performance of interpolators obtained from the minimization of convex functions/potentials for over-parameterized linear regression with non-isotropic Gaussian data. In particular, we obtain a tight lower bound on the best generalization error possible among this class of interpolators in terms of the over-parameterization ratio, the variance of the noise in the labels, the eigenspectrum of the data covariance, and the underlying distribution of the parameter to be estimated. Finally, we find the optimal convex implicit bias that achieves this lower bound under certain sufficient conditions involving the log-concavity of the distribution of a Gaussian convolved with the prior of the true underlying parameter.
Authors: Guozheng Ma, Lu Li, Zilin Wang, Li Shen, Pierre-Luc Bacon, Dacheng Tao
Abstract: Effectively scaling up deep reinforcement learning models has proven notoriously difficult due to network pathologies during training, motivating various targeted interventions such as periodic reset and architectural advances such as layer normalization. Instead of pursuing more complex modifications, we show that introducing static network sparsity alone can unlock further scaling potential beyond their dense counterparts with state-of-the-art architectures. This is achieved through simple one-shot random pruning, where a predetermined percentage of network weights are randomly removed once before training. Our analysis reveals that, in contrast to naively scaling up dense DRL networks, such sparse networks achieve both higher parameter efficiency for network expressivity and stronger resistance to optimization challenges like plasticity loss and gradient interference. We further extend our evaluation to visual and streaming RL scenarios, demonstrating the consistent benefits of network sparsity.
Authors: Xuechen Zhang, Zijian Huang, Yingcong Li, Chenshun Ni, Jiasi Chen, Samet Oymak
Abstract: Small language models (SLMs) struggle to learn complex reasoning behaviors, especially when high-quality traces are scarce or difficult to learn from. The standard training approach combines a supervised fine-tuning (SFT) stage, often to distill capabilities of a larger model, followed by a reinforcement learning (RL)stage such as Group Relative Policy Optimization (GRPO). In this paper, we investigate the fundamental limitations of this SFT + RL paradigm and propose methods to overcome them. Under a suitable theoretical model, we demonstrate that the SFT + RL strategy can fail completely when (1) the expert's traces are too difficult for the small model to express, or (2) the small model's initialization has exponentially small likelihood of success. To address these, we introduce BREAD: a GRPO variant that unifies the SFT and RL stages via partial expert guidance and branched rollouts. When self-generated traces fail, BREAD adaptively inserts short expert prefixes/hints, allowing the small model to complete the rest of the reasoning path, and ensuring that each update includes at least one successful trace. This mechanism both densifies the reward signal and induces a natural learning curriculum. BREAD requires fewer than 40% of ground-truth traces, consistently outperforming standard GRPO while speeding up the training by about 3 times. Importantly, we demonstrate that BREAD helps the model solve problems that are otherwise unsolvable by the SFT + RL strategy, highlighting how branched rollouts and expert guidance can substantially boost SLM reasoning.
Authors: Yanzhi Zhang, Zhaoxi Zhang, Haoxiang Guan, Yilin Cheng, Yitong Duan, Chen Wang, Yue Wang, Shuxin Zheng, Jiyan He
Abstract: Reinforcement learning has emerged as a powerful paradigm for post-training large language models (LLMs) to improve reasoning. Approaches like Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) have shown strong results, but they require extensive external supervision. We investigate an alternative class of methods, Reinforcement Learning from Internal Feedback (RLIF), which relies solely on intrinsic model-derived signals instead of external rewards. In particular, we leverage unsupervised reward proxies such as token-level entropy, trajectory-level entropy, and self-certainty. Our theoretical analysis shows these internal objectives are partially equivalent, and we empirically evaluate various RLIF strategies on challenging math reasoning benchmarks. Experimental results demonstrate that RLIF can boost the reasoning performance of base LLMs at the beginning phase of the training, matching or surpassing RLVR techniques on these tasks. However, when training progresses, performance degrades even below the model before training. Moreover, we find that RLIF yields little improvement for instruction-tuned models, indicating diminishing returns of intrinsic feedback once an LLM is already instruction-tuned. We further analyze this limitation by mixing model weights and explain the reason of RLIF's training behaviors, providing practical guidelines for integrating internal feedback signals into LLM training. We hope our analysis of internal feedback will inform more principled and effective strategies for LLM post-training.
Authors: Yi Liu, Hongji Zhang, Yunhao Zhou, Zhengyuan Shi, Changran Xu, Qiang Xu
Abstract: The integration of large language models (LLMs) into electronic design automation (EDA) has significantly advanced the field, offering transformative benefits, particularly in register transfer level (RTL) code generation and understanding. While previous studies have demonstrated the efficacy of fine-tuning LLMs for these generation-based tasks, embedding-based tasks, which are equally critical to EDA workflows, have been largely overlooked. These tasks, including natural language code search, RTL code functionality equivalence checking, and performance prediction, are essential for accelerating and optimizing the hardware design process. To address this gap, we present DeepRTL2, a family of versatile LLMs that unifies both generation- and embedding-based tasks related to RTL. By simultaneously tackling a broad range of tasks, DeepRTL2 represents the first model to provide a comprehensive solution to the diverse challenges in EDA. Through extensive experiments, we show that DeepRTL2 achieves state-of-the-art performance across all evaluated tasks.
Authors: Irina G. Tanashkina, Alexey S. Tanashkin, Alexander S. Maksimchuik, Anna Yu. Poshivailo
Abstract: In this article, we review modern approaches to building interpretable models of property markets using machine learning on the base of mass valuation of property in the Primorye region, Russia. The researcher, lacking expertise in this topic, encounters numerous difficulties in the effort to build a good model. The main source of this is the huge difference between noisy real market data and ideal data which is very common in all types of tutorials on machine learning. This paper covers all stages of modeling: the collection of initial data, identification of outliers, the search and analysis of patterns in data, the formation and final choice of price factors, the building of the model, and the evaluation of its efficiency. For each stage, we highlight potential issues and describe sound methods for overcoming emerging difficulties on actual examples. We show that the combination of classical linear regression with interpolation methods of geostatistics allows to build an effective model for land parcels. For flats, when many objects are attributed to one spatial point the application of geostatistical methods is difficult. Therefore we suggest linear regression with automatic generation and selection of additional rules on the base of decision trees, so called the RuleFit method. Thus we show, that despite the strong restriction as the requirement of interpretability which is important in practical aspects, for example, legal matters, it is still possible to build effective models of real property markets.
Authors: Khurram Yamin, Gaurav Ghosal, Bryan Wilder
Abstract: Large Language Models have been shown to contain extensive world knowledge in their parameters, enabling impressive performance on many knowledge intensive tasks. However, when deployed in novel settings, LLMs often encounter situations where they must integrate parametric knowledge with new or unfamiliar information. In this work, we explore whether LLMs can combine knowledge in-context with their parametric knowledge through the lens of counterfactual reasoning. Through synthetic and real experiments in multi-hop reasoning problems, we show that LLMs generally struggle with counterfactual reasoning, often resorting to exclusively using their parametric knowledge. Moreover, we show that simple post-hoc finetuning can struggle to instill counterfactual reasoning ability -- often leading to degradation in stored parametric knowledge. Ultimately, our work reveals important limitations of current LLM's abilities to re-purpose parametric knowledge in novel settings.
Authors: Mert Cemri, Nived Rajaraman, Rishabh Tiwari, Xiaoxuan Liu, Kurt Keutzer, Ion Stoica, Kannan Ramchandran, Ahmad Beirami, Ziteng Sun
Abstract: Scaling test-time compute has driven the recent advances in the reasoning capabilities of large language models (LLMs), typically by allocating additional computation for more thorough exploration. However, increased compute often comes at the expense of higher user-facing latency, directly impacting user experience. Current test-time scaling methods primarily optimize for accuracy based on total compute resources (FLOPS), often overlooking latency constraints. To address this gap, we propose $\texttt{SPECS}$, a latency-aware test-time scaling method inspired by speculative decoding. $\texttt{SPECS}$~uses a smaller, faster model to generate candidate sequences efficiently, and evaluates these candidates using signals from both a larger target model and a dedicated reward model. We introduce new integration strategies, including reward-guided soft verification and a reward-based deferral mechanism. Empirical results on MATH500, AMC23 and OlympiadBench datasets show that $\texttt{SPECS}$~matches or surpasses beam search accuracy while reducing latency by up to $\sim$19.1\%. Our theoretical analysis shows that our algorithm converges to the solution of a KL-regularized reinforcement learning objective with increasing beam width.
Authors: Peiyuan Tang, Haojie Xin, Xiaodong Zhang, Jun Sun, Qin Xia, Zijiang Yang
Abstract: As Vision-Language Models (VLMs) demonstrate increasing capabilities across real-world applications such as code generation and chatbot assistance, ensuring their safety has become paramount. Unlike traditional Large Language Models (LLMs), VLMs face unique vulnerabilities due to their multimodal nature, allowing adversaries to modify visual or textual inputs to bypass safety guardrails and trigger the generation of harmful content. Through systematic analysis of VLM behavior under attack, we identify a novel phenomenon termed ``delayed safety awareness''. Specifically, we observe that safety-aligned VLMs may initially be compromised to produce harmful content, but eventually recognize the associated risks and attempt to self-correct. This pattern suggests that VLMs retain their underlying safety awareness but experience a temporal delay in their activation. Building on this insight, we hypothesize that VLMs' safety awareness can be proactively reactivated through carefully designed prompts. To this end, we introduce ``The Safety Reminder'', a soft prompt tuning approach that optimizes learnable prompt tokens, which are periodically injected during the text generation process to enhance safety awareness, effectively preventing harmful content generation. Additionally, our safety reminder only activates when harmful content is detected, leaving normal conversations unaffected and preserving the model's performance on benign tasks. Through comprehensive evaluation across three established safety benchmarks and one adversarial attacks, we demonstrate that our approach significantly reduces attack success rates while maintaining model utility, offering a practical solution for deploying safer VLMs in real-world applications.
Authors: Robert Graham, Edward Stevinson, Leo Richter, Alexander Chia, Joseph Miller, Joseph Isaac Bloom
Abstract: Identifying inputs that trigger specific behaviours or latent features in language models could have a wide range of safety use cases. We investigate a class of methods capable of generating targeted, linguistically fluent inputs that activate specific latent features or elicit model behaviours. We formalise this approach as context modification and present ContextBench -- a benchmark with tasks assessing core method capabilities and potential safety applications. Our evaluation framework measures both elicitation strength (activation of latent features or behaviours) and linguistic fluency, highlighting how current state-of-the-art methods struggle to balance these objectives. We enhance Evolutionary Prompt Optimisation (EPO) with LLM-assistance and diffusion model inpainting, and demonstrate that these variants achieve state-of-the-art performance in balancing elicitation effectiveness and fluency.
Authors: Jonathan Kutasov, Yuqi Sun, Paul Colognese, Teun van der Weij, Linda Petrini, Chen Bo Calvin Zhang, John Hughes, Xiang Deng, Henry Sleight, Tyler Tracy, Buck Shlegeris, Joe Benton
Abstract: As Large Language Models (LLMs) are increasingly deployed as autonomous agents in complex and long horizon settings, it is critical to evaluate their ability to sabotage users by pursuing hidden objectives. We study the ability of frontier LLMs to evade monitoring and achieve harmful hidden goals while completing a wide array of realistic tasks. We evaluate a broad range of frontier LLMs using SHADE (Subtle Harmful Agent Detection & Evaluation)-Arena, the first highly diverse agent evaluation dataset for sabotage and monitoring capabilities of LLM agents. SHADE-Arena consists of complex pairs of benign main tasks and harmful side objectives in complicated environments. Agents are evaluated on their ability to complete the side task without appearing suspicious to an LLM monitor. When measuring agent ability to (a) complete the main task, (b) complete the side task, and (c) avoid detection, we find that the best performing frontier models score 27% (Claude 3.7 Sonnet) and 15% (Gemini 2.5 Pro) as sabotage agents when overseen by Claude 3.6 Sonnet. For current frontier models, success on the side task relies heavily on having access to a hidden scratchpad that is not visible to the monitor. We also use SHADE-Arena to measure models' monitoring abilities, with the top monitor (Gemini 2.5 Pro) achieving an AUC of 0.87 at distinguishing benign and malign transcripts. We find that for now, models still struggle at sabotage due to failures in long-context main task execution. However, our measurements already demonstrate the difficulty of monitoring for subtle sabotage attempts, which we expect to only increase in the face of more complex and longer-horizon tasks.
Authors: Tobias Grafke
Abstract: We present an algorithm to sample stochastic differential equations conditioned on rather general constraints, including integral constraints, endpoint constraints, and stochastic integral constraints. The algorithm is a pathspace Metropolis-adjusted manifold sampling scheme, which samples stochastic paths on the submanifold of realizations that adhere to the conditioning constraint. We demonstrate the effectiveness of the algorithm by sampling a dynamical condensation phase transition, conditioning a random walk on a fixed Levy stochastic area, conditioning a stochastic nonlinear wave equation on high amplitude waves, and sampling a stochastic partial differential equation model of turbulent pipe flow conditioned on relaminarization events.
Authors: Seyed Mohsen Hosseini
Abstract: Class imbalance and the difficulty imbalance are the two types of data imbalance that affect the performance of neural networks in medical segmentation tasks. In class imbalance the loss is dominated by the majority classes and in difficulty imbalance the loss is dominated by easy to classify pixels. This leads to an ineffective training. Dice loss, which is based on a geometrical metric, is very effective in addressing the class imbalance compared to the cross entropy (CE) loss, which is adopted directly from classification tasks. To address the difficulty imbalance, the common approach is employing a re-weighted CE loss or a modified Dice loss to focus the training on difficult to classify areas. The existing modification methods are computationally costly and with limited success. In this study we propose a simple modification to the Dice loss with minimal computational cost. With a pixel level modulating term, we take advantage of the effectiveness of Dice loss in handling the class imbalance to also handle the difficulty imbalance. Results on three commonly used medical segmentation tasks show that the proposed Pixel-wise Modulated Dice loss (PM Dice loss) outperforms other methods, which are designed to tackle the difficulty imbalance problem.
Authors: Kartik Sharma, Yiqiao Jin, Vineeth Rakesh, Yingtong Dou, Menghai Pan, Mahashweta Das, Srijan Kumar
Abstract: As large language models (LLMs) are deployed in safety-critical settings, it is essential to ensure that their responses comply with safety standards. Prior research has revealed that LLMs often fail to grasp the notion of safe behaviors, resulting in either unjustified refusals to harmless prompts or the generation of harmful content. While substantial efforts have been made to improve their robustness, existing defenses often rely on costly fine-tuning of model parameters or employ suboptimal heuristic techniques. In this work, we take a novel approach to safeguard LLMs by learning to adapt the system prompts in instruction-tuned LLMs. While LLMs are typically pre-trained to follow a fixed system prompt, we investigate the impact of tailoring the system prompt to each specific user input on the safety of the responses. To this end, we propose $\textbf{Sysformer}$, a trans$\textbf{former}$ model that updates an initial $\textbf{sys}$tem prompt to a more robust system prompt in the LLM input embedding space while attending to the user prompt. While keeping the LLM parameters frozen, the Sysformer is trained to refuse to respond to a set of harmful prompts while responding ideally to a set of safe ones. Through extensive experiments on $5$ LLMs from different families and $2$ recent benchmarks, we demonstrate that Sysformer can significantly enhance the robustness of LLMs, leading to upto $80\%$ gain in the refusal rate on harmful prompts while enhancing the compliance with the safe prompts by upto $90\%$. Results also generalize well to sophisticated jailbreaking attacks, making LLMs upto $100\%$ more robust against different attack strategies. We hope our findings lead to cheaper safeguarding of LLMs and motivate future investigations into designing variable system prompts.
Authors: Oluwaseyi Giwa, Muhammad Ahmed Mohsin, Muhammad Ali Jamshed
Abstract: In this letter, we propose Quantum-Preconditioned Policy Gradient (QPPG), a natural gradient-based algorithm for link adaptation that whitens policy updates using the full inverse quantum Fisher information with Tikhonov regularization. QPPG bridges classical and quantum geometry, achieving stable learning even under noise. Evaluated on classical and quantum environments, including noisy single-qubit Gym tasks and Rayleigh-fading channels, QPPG converges 4 times faster than REINFORCE and sustains a 1 dB gain under uncertainty. It reaches a 90 percent return in one hundred episodes with high noise robustness, showcasing the advantages of full QFI-based preconditioning for scalable quantum reinforcement learning.
Authors: Jo\~ao G. Ribeiro, Yaniv Oren, Alberto Sardinha, Matthijs Spaan, Francisco S. Melo
Abstract: This paper proposes RecBayes, a novel approach for ad hoc teamwork under partial observability, a setting where agents are deployed on-the-fly to environments where pre-existing teams operate, that never requires, at any stage, access to the states of the environment or the actions of its teammates. We show that by relying on a recurrent Bayesian classifier trained using past experiences, an ad hoc agent is effectively able to identify known teams and tasks being performed from observations alone. Unlike recent approaches such as PO-GPL (Gu et al., 2021) and FEAT (Rahman et al., 2023), that require at some stage fully observable states of the environment, actions of teammates, or both, or approaches such as ATPO (Ribeiro et al., 2023) that require the environments to be small enough to be tabularly modelled (Ribeiro et al., 2023), in their work up to 4.8K states and 1.7K observations, we show RecBayes is both able to handle arbitrarily large spaces while never relying on either states and teammates' actions. Our results in benchmark domains from the multi-agent systems literature, adapted for partial observability and scaled up to 1M states and 2^125 observations, show that RecBayes is effective at identifying known teams and tasks being performed from partial observations alone, and as a result, is able to assist the teams in solving the tasks effectively.
Authors: Marcel Wien\"obst, Sebastian Weichwald, Leonard Henckel
Abstract: We introduce CIfly, a framework for efficient algorithmic primitives in graphical causal inference that isolates reachability as a reusable core operation. It builds on the insight that many causal reasoning tasks can be reduced to reachability in purpose-built state-space graphs that can be constructed on the fly during traversal. We formalize a rule table schema for specifying such algorithms and prove they run in linear time. We establish CIfly as a more efficient alternative to the common primitives moralization and latent projection, which we show are computationally equivalent to Boolean matrix multiplication. Our open-source Rust implementation parses rule table text files and runs the specified CIfly algorithms providing high-performance execution accessible from Python and R. We demonstrate CIfly's utility by re-implementing a range of established causal inference tasks within the framework and by developing new algorithms for instrumental variables. These contributions position CIfly as a flexible and scalable backbone for graphical causal inference, guiding algorithm development and enabling easy and efficient deployment.
Authors: Shuangbao Paul Wang, Jianzhou Mao, Eric Sakk
Abstract: This paper discusses the compilation, optimization, and error mitigation of quantum algorithms, essential steps to execute real-world quantum algorithms. Quantum algorithms running on a hybrid platform with QPU and CPU/GPU take advantage of existing high-performance computing power with quantum-enabled exponential speedups. The proposed approximate quantum Fourier transform (AQFT) for quantum algorithm optimization improves the circuit execution on top of an exponential speed-ups the quantum Fourier transform has provided.
Authors: Tom Hendriks, Gerrit Arends, Edwin Versteeg, Anna Vilanova, Maxime Chamberland, Chantal M. W. Tax
Abstract: Diffusion magnetic resonance imaging (dMRI) enables non-invasive investigation of tissue microstructure. The Standard Model (SM) of white matter aims to disentangle dMRI signal contributions from intra- and extra-axonal water compartments. However, due to the model its high-dimensional nature, extensive acquisition protocols with multiple b-values and diffusion tensor shapes are typically required to mitigate parameter degeneracies. Even then, accurate estimation remains challenging due to noise. This work introduces a novel estimation framework based on implicit neural representations (INRs), which incorporate spatial regularization through the sinusoidal encoding of the input coordinates. The INR method is evaluated on both synthetic and in vivo datasets and compared to parameter estimates using cubic polynomials, supervised neural networks, and nonlinear least squares. Results demonstrate superior accuracy of the INR method in estimating SM parameters, particularly in low signal-to-noise conditions. Additionally, spatial upsampling of the INR can represent the underlying dataset anatomically plausibly in a continuous way, which is unattainable with linear or cubic interpolation. The INR is fully unsupervised, eliminating the need for labeled training data. It achieves fast inference ($\sim$6 minutes), is robust to both Gaussian and Rician noise, supports joint estimation of SM kernel parameters and the fiber orientation distribution function with spherical harmonics orders up to at least 8 and non-negativity constraints, and accommodates spatially varying acquisition protocols caused by magnetic gradient non-uniformities. The combination of these properties along with the possibility to easily adapt the framework to other dMRI models, positions INRs as a potentially important tool for analyzing and interpreting diffusion MRI data.
Authors: Seung-Joo Lee, Andre Lukas
Abstract: We outline a method to determine analytic K\"ahler potentials with associated approximately Ricci-flat K\"ahler metrics on Calabi-Yau manifolds. Key ingredients are numerically calculating Ricci-flat K\"ahler potentials via machine learning techniques and fitting the numerical results to Donaldson's Ansatz. We apply this method to the Dwork family of quintic hypersurfaces in $\mathbb{P}^4$ and an analogous one-parameter family of bi-cubic CY hypersurfaces in $\mathbb{P}^2\times\mathbb{P}^2$. In each case, a relatively simple analytic expression is obtained for the approximately Ricci-flat K\"ahler potentials, including the explicit dependence on the complex structure parameter. We find that these K\"ahler potentials only depend on the modulus of the complex structure parameter.
Authors: Robert Kent, Benjamin Lienhard, Gregory Lafyatis, Daniel J. Gauthier
Abstract: Quantum processors require rapid and high-fidelity simultaneous measurements of many qubits. While superconducting qubits are among the leading modalities toward a useful quantum processor, their readout remains a bottleneck. Traditional approaches to processing measurement data often struggle to account for crosstalk present in frequency-multiplexed readout, the preferred method to reduce the resource overhead. Recent approaches to address this challenge use neural networks to improve the state-discrimination fidelity. However, they are computationally expensive to train and evaluate, resulting in increased latency and poor scalability as the number of qubits increases. We present an alternative machine learning approach based on next-generation reservoir computing that constructs polynomial features from the measurement signals and maps them to the corresponding qubit states. This method is highly parallelizable, avoids the costly nonlinear activation functions common in neural networks, and supports real-time training, enabling fast evaluation, adaptability, and scalability. Despite its lower computational complexity, our reservoir approach is able to maintain high qubit-state-discrimination fidelity. Relative to traditional methods, our approach achieves error reductions of up to 50% and 11% on single- and five-qubit datasets, respectively, and delivers up to 2.5x crosstalk reduction on the five-qubit dataset. Compared with recent machine-learning methods, evaluating our model requires 100x fewer multiplications for single-qubit and 2.5x fewer for five-qubit models. This work demonstrates that reservoir computing can enhance qubit-state discrimination while maintaining scalability for future quantum processors.
Authors: Nicolas Boull\'e, Matthew J. Colbrook, Gustav Conradie
Abstract: Data-driven spectral analysis of Koopman operators is a powerful tool for understanding numerous real-world dynamical systems, from neuronal activity to variations in sea surface temperature. The Koopman operator acts on a function space and is most commonly studied on the space of square-integrable functions. However, defining it on a suitable reproducing kernel Hilbert space (RKHS) offers numerous practical advantages, including pointwise predictions with error bounds, improved spectral properties that facilitate computations, and more efficient algorithms, particularly in high dimensions. We introduce the first general, provably convergent, data-driven algorithms for computing spectral properties of Koopman and Perron--Frobenius operators on RKHSs. These methods efficiently compute spectra and pseudospectra with error control and spectral measures while exploiting the RKHS structure to avoid the large-data limits required in the $L^2$ settings. The function space is determined by a user-specified kernel, eliminating the need for quadrature-based sampling as in $L^2$ and enabling greater flexibility with finite, externally provided datasets. Using the Solvability Complexity Index hierarchy, we construct adversarial dynamical systems for these problems to show that no algorithm can succeed in fewer limits, thereby proving the optimality of our algorithms. Notably, this impossibility extends to randomized algorithms and datasets. We demonstrate the effectiveness of our algorithms on challenging, high-dimensional datasets arising from real-world measurements and high-fidelity numerical simulations, including turbulent channel flow, molecular dynamics of a binding protein, Antarctic sea ice concentration, and Northern Hemisphere sea surface height. The algorithms are publicly available in the software package $\texttt{SpecRKHS}$.
Authors: Peter Yichen Chen, Minghao Guo, Hanspeter Pfister, Ming Lin, William Freeman, Qixing Huang, Han-Wei Shen, Wojciech Matusik
Abstract: Computer graphics, often associated with films, games, and visual effects, has long been a powerful tool for addressing scientific challenges--from its origins in 3D visualization for medical imaging to its role in modern computational modeling and simulation. This course explores the deep and evolving relationship between computer graphics and science, highlighting past achievements, ongoing contributions, and open questions that remain. We show how core methods, such as geometric reasoning and physical modeling, provide inductive biases that help address challenges in both fields, especially in data-scarce settings. To that end, we aim to reframe graphics as a modeling language for science by bridging vocabulary gaps between the two communities. Designed for both newcomers and experts, Graphics4Science invites the graphics community to engage with science, tackle high-impact problems where graphics expertise can make a difference, and contribute to the future of scientific discovery. Additional details are available on the course website: https://graphics4science.github.io
Authors: Lukas Helff, Ahmad Omar, Felix Friedrich, Wolfgang Stammer, Antonia W\"ust, Tim Woydt, Rupert Mitchell, Patrick Schramowski, Kristian Kersting
Abstract: We introduce SLR, an end-to-end framework for systematic evaluation and training of Large Language Models (LLMs) via Scalable Logical Reasoning. Given a user's task specification, SLR enables scalable, automated synthesis of inductive reasoning tasks with precisely controlled difficulty. For each task, SLR synthesizes (i) a latent ground-truth rule, (ii) an executable validation program used by a symbolic judge to deterministically verify model outputs, and (iii) an instruction prompt for the reasoning task. Using SLR, we create SLR-Bench, a benchmark comprising over 19k prompts spanning 20 curriculum levels that progressively increase in relational, arithmetic, and recursive complexity. Large-scale evaluation reveals that contemporary LLMs readily produce syntactically valid rules, yet often fail at correct logical inference. Recent reasoning LLMs do somewhat better, but incur substantial increases in test-time compute, sometimes exceeding 15k completion tokens. Finally, logic-tuning via SLR doubles Llama-3-8B accuracy on SLR-Bench, achieving parity with Gemini-Flash-Thinking at a fraction of computational cost. SLR is fully automated, requires no human annotation, ensures dataset novelty, and offers a scalable environment for probing and advancing LLMs' reasoning capabilities.
Authors: Albert Dorador
Abstract: Piecewise-constant regression trees remain popular for their interpretability, yet often lag behind black-box models like Random Forest in predictive accuracy. In this work, we introduce TRUST (Transparent, Robust, and Ultra-Sparse Trees), a novel regression tree model that combines the accuracy of Random Forests with the interpretability of shallow decision trees and sparse linear models. TRUST further enhances transparency by leveraging Large Language Models to generate tailored, user-friendly explanations. Extensive validation on synthetic and real-world benchmark datasets demonstrates that TRUST consistently outperforms other interpretable models -- including CART, Lasso, and Node Harvest -- in predictive accuracy, while matching the accuracy of Random Forest and offering substantial gains in both accuracy and interpretability over M5', a well-established model that is conceptually related.
Authors: Andrew Wagenmaker, Mitsuhiko Nakamoto, Yunchu Zhang, Seohong Park, Waleed Yagoub, Anusha Nagabandi, Abhishek Gupta, Sergey Levine
Abstract: Robotic control policies learned from human demonstrations have achieved impressive results in many real-world applications. However, in scenarios where initial performance is not satisfactory, as is often the case in novel open-world settings, such behavioral cloning (BC)-learned policies typically require collecting additional human demonstrations to further improve their behavior -- an expensive and time-consuming process. In contrast, reinforcement learning (RL) holds the promise of enabling autonomous online policy improvement, but often falls short of achieving this due to the large number of samples it typically requires. In this work we take steps towards enabling fast autonomous adaptation of BC-trained policies via efficient real-world RL. Focusing in particular on diffusion policies -- a state-of-the-art BC methodology -- we propose diffusion steering via reinforcement learning (DSRL): adapting the BC policy by running RL over its latent-noise space. We show that DSRL is highly sample efficient, requires only black-box access to the BC policy, and enables effective real-world autonomous policy improvement. Furthermore, DSRL avoids many of the challenges associated with finetuning diffusion policies, obviating the need to modify the weights of the base policy at all. We demonstrate DSRL on simulated benchmarks, real-world robotic tasks, and for adapting pretrained generalist policies, illustrating its sample efficiency and effective performance at real-world policy improvement.
Authors: Ziv Aharoni, Bashar Huleihel, Henry D Pfister, Haim H Permuter
Abstract: This paper proposes a method to optimize communication code rates via the application of neural polar decoders (NPDs). Employing this approach enables simultaneous optimization of code rates over input distributions while providing a practical coding scheme within the framework of polar codes. The proposed approach is designed for scenarios where the channel model is unknown, treating the channel as a black box that produces output samples from input samples. We employ polar codes to achieve our objectives, using NPDs to estimate mutual information (MI) between the channel inputs and outputs, and optimize a parametric model of the input distribution. The methodology involves a two-phase process: a training phase and an inference phase. In the training phase, two steps are repeated interchangeably. First, the estimation step estimates the MI of the channel inputs and outputs via NPDs. Second, the improvement step optimizes the input distribution parameters to maximize the MI estimate obtained by the NPDs. In the inference phase, the optimized model is used to construct polar codes. This involves incorporating the Honda-Yamamoto (HY) scheme to accommodate the optimized input distributions and list decoding to enhance decoding performance. Experimental results on memoryless and finite-state channels (FSCs) demonstrate the effectiveness of our approach, particularly in cases where the channel's capacity-achieving input distribution is non-uniform. For these cases, we show significant improvements in MI and bit error rates (BERs) over those achieved by uniform and independent and identically distributed (i.i.d.) input distributions, validating our method for block lengths up to 1024. This scalable approach has potential applications in real-world communication systems, bridging theoretical capacity estimation and practical coding performance.
Authors: Abdolazim Rezaei, Mehdi Sookhak, Ahmad Patooghy
Abstract: Connected and Autonomous Vehicles (CAVs) rely on a range of devices that often process privacy-sensitive data. Among these, roadside units play a critical role particularly through the use of AI-equipped (AIE) cameras for applications such as violation detection. However, the privacy risks associated with captured imagery remain a major concern, as such data can be misused for identity theft, profiling, or unauthorized commercial purposes. While traditional techniques such as face blurring and obfuscation have been applied to mitigate privacy risks, individual privacy remains at risk, as individuals can still be tracked using other features such as their clothing. This paper introduces a novel privacy-preserving framework that leverages feedback-based reinforcement learning (RL) and vision-language models (VLMs) to protect sensitive visual information captured by AIE cameras. The main idea is to convert images into semantically equivalent textual descriptions, ensuring that scene-relevant information is retained while visual privacy is preserved. A hierarchical RL strategy is employed to iteratively refine the generated text, enhancing both semantic accuracy and privacy. Evaluation results demonstrate significant improvements in both privacy protection and textual quality, with the Unique Word Count increasing by approximately 77\% and Detail Density by around 50\% compared to existing approaches.
Authors: Berk Yilmaz, Junyu Hu, Jinsong Liu
Abstract: This paper presents a Deep Reinforcement Learning (DRL) system for Xiangqi (Chinese Chess) that integrates neural networks with Monte Carlo Tree Search (MCTS) to enable strategic self-play and self-improvement. Addressing the underexplored complexity of Xiangqi, including its unique board layout, piece movement constraints, and victory conditions, our approach combines policy-value networks with MCTS to simulate move consequences and refine decision-making. By overcoming challenges such as Xiangqi's high branching factor and asymmetrical piece dynamics, our work advances AI capabilities in culturally significant strategy games while providing insights for adapting DRL-MCTS frameworks to domain-specific rule systems.
Authors: Jakub T{\l}uczek, Victor Villin, Christos Dimitrakakis
Abstract: Fairness is desirable yet challenging to achieve within multi-agent systems, especially when agents differ in latent traits that affect their abilities. This hidden heterogeneity often leads to unequal distributions of wealth, even when agents operate under the same rules. Motivated by real-world examples, we propose a framework based on repeated principal-agent games, where a principal, who also can be seen as a player of the game, learns to offer adaptive contracts to agents. By leveraging a simple yet powerful contract structure, we show that a fairness-aware principal can learn homogeneous linear contracts that equalize outcomes across agents in a sequential social dilemma. Importantly, this fairness does not come at the cost of efficiency: our results demonstrate that it is possible to promote equity and stability in the system while preserving overall performance.
Authors: Sawan Kumar, Tapas Tripura, Rajdip Nayek, Souvik Chakraborty
Abstract: Operator learning offers a powerful paradigm for solving parametric partial differential equations (PDEs), but scaling probabilistic neural operators such as the recently proposed Gaussian Processes Operators (GPOs) to high-dimensional, data-intensive regimes remains a significant challenge. In this work, we introduce a novel, scalable GPO, which capitalizes on sparsity, locality, and structural information through judicious kernel design. Addressing the fundamental limitation of cubic computational complexity, our method leverages nearest-neighbor-based local kernel approximations in the spatial domain, sparse kernel approximation in the parameter space, and structured Kronecker factorizations to enable tractable inference on large-scale datasets and high-dimensional input. While local approximations often introduce accuracy trade-offs due to limited kernel interactions, we overcome this by embedding operator-aware kernel structures and employing expressive, task-informed mean functions derived from neural operator architectures. Through extensive evaluations on a broad class of nonlinear PDEs - including Navier-Stokes, wave advection, Darcy flow, and Burgers' equations - we demonstrate that our framework consistently achieves high accuracy across varying discretization scales. These results underscore the potential of our approach to bridge the gap between scalability and fidelity in GPO, offering a compelling foundation for uncertainty-aware modeling in complex physical systems.
Authors: Elif Keles, Merve Yazol, Gorkem Durak, Ziliang Hong, Halil Ertugrul Aktas, Zheyuan Zhang, Linkai Peng, Onkar Susladkar, Necati Guzelyel, Oznur Leman Boyunaga, Cemal Yazici, Mark Lowe, Aliye Uc, Ulas Bagci
Abstract: Objective: Our study aimed to evaluate and validate PanSegNet, a deep learning (DL) algorithm for pediatric pancreas segmentation on MRI in children with acute pancreatitis (AP), chronic pancreatitis (CP), and healthy controls. Methods: With IRB approval, we retrospectively collected 84 MRI scans (1.5T/3T Siemens Aera/Verio) from children aged 2-19 years at Gazi University (2015-2024). The dataset includes healthy children as well as patients diagnosed with AP or CP based on clinical criteria. Pediatric and general radiologists manually segmented the pancreas, then confirmed by a senior pediatric radiologist. PanSegNet-generated segmentations were assessed using Dice Similarity Coefficient (DSC) and 95th percentile Hausdorff distance (HD95). Cohen's kappa measured observer agreement. Results: Pancreas MRI T2W scans were obtained from 42 children with AP/CP (mean age: 11.73 +/- 3.9 years) and 42 healthy children (mean age: 11.19 +/- 4.88 years). PanSegNet achieved DSC scores of 88% (controls), 81% (AP), and 80% (CP), with HD95 values of 3.98 mm (controls), 9.85 mm (AP), and 15.67 mm (CP). Inter-observer kappa was 0.86 (controls), 0.82 (pancreatitis), and intra-observer agreement reached 0.88 and 0.81. Strong agreement was observed between automated and manual volumes (R^2 = 0.85 in controls, 0.77 in diseased), demonstrating clinical reliability. Conclusion: PanSegNet represents the first validated deep learning solution for pancreatic MRI segmentation, achieving expert-level performance across healthy and diseased states. This tool, algorithm, along with our annotated dataset, are freely available on GitHub and OSF, advancing accessible, radiation-free pediatric pancreatic imaging and fostering collaborative research in this underserved domain.
Authors: Lucas Amoudruz, Petr Karnakov, Petros Koumoutsakos
Abstract: Contactless manipulation of small objects is essential for biomedical and chemical applications, such as cell analysis, assisted fertilisation, and precision chemistry. Established methods, including optical, acoustic, and magnetic tweezers, are now complemented by flow control techniques that use flow-induced motion to enable precise and versatile manipulation. However, trapping multiple particles in fluid remains a challenge. This study introduces a novel control algorithm capable of steering multiple particles in flow. The system uses rotating disks to generate flow fields that transport particles to precise locations. Disk rotations are governed by a feedback control policy based on the Optimising a Discrete Loss (ODIL) framework, which combines fluid dynamics equations with path objectives into a single loss function. Our experiments, conducted in both simulations and with the physical device, demonstrate the capability of the approach to transport two beads simultaneously to predefined locations, advancing robust contactless particle manipulation for biomedical applications.
Authors: Yunchi Lu, Youshan Miao, Cheng Tan, Peng Huang, Yi Zhu, Xian Zhang, Fan Yang
Abstract: Training large language models (LLMs) at scale requires parallel execution across thousands of devices, incurring enormous computational costs. Yet, these costly distributed trainings are rarely verified, leaving them prone to silent errors and potentially wasting millions of GPU hours. We introduce TrainVerify, a system for verifiable distributed training of LLMs. Given a deep learning model's logical specification as the ground truth, TrainVerify formally verifies that a distributed parallel execution plan is mathematically equivalent to it. Direct verification is notoriously difficult due to the sheer scale of LLMs which often involves billions of variables and highly intricate computation graphs. Therefore, TrainVerify introduces shape-reduction techniques and a stage-wise parallel verification algorithm that significantly reduces complexity while preserving formal correctness. TrainVerify scales to frontier LLMs, including the successful verification of the Llama3 (405B) and DeepSeek-V3 (671B) training plans.
Authors: Jiawen Yang, Shuhao Chen, Yucong Duan, Ke Tang, Yu Zhang
Abstract: Unsupervised domain adaptation (UDA) methods effectively bridge domain gaps but become struggled when the source and target domains belong to entirely distinct modalities. To address this limitation, we propose a novel setting called Heterogeneous-Modal Unsupervised Domain Adaptation (HMUDA), which enables knowledge transfer between completely different modalities by leveraging a bridge domain containing unlabeled samples from both modalities. To learn under the HMUDA setting, we propose Latent Space Bridging (LSB), a specialized framework designed for the semantic segmentation task. Specifically, LSB utilizes a dual-branch architecture, incorporating a feature consistency loss to align representations across modalities and a domain alignment loss to reduce discrepancies between class centroids across domains. Extensive experiments conducted on six benchmark datasets demonstrate that LSB achieves state-of-the-art performance.
Authors: Peizhi Wu, Rong Kang, Tieying Zhang, Jianjun Chen, Ryan Marcus, Zachary G. Ives
Abstract: Cardinality estimation (CardEst) is a critical aspect of query optimization. Traditionally, it leverages statistics built directly over the data. However, organizational policies (e.g., regulatory compliance) may restrict global data access. Fortunately, query-driven cardinality estimation can learn CardEst models using query workloads. However, existing query-driven models often require access to data or summaries for best performance, and they assume perfect training workloads with complete and balanced join templates (or join graphs). Such assumptions rarely hold in real-world scenarios, in which join templates are incomplete and imbalanced. We present GRASP, a data-agnostic cardinality learning system designed to work under these real-world constraints. GRASP's compositional design generalizes to unseen join templates and is robust to join template imbalance. It also introduces a new per-table CardEst model that handles value distribution shifts for range predicates, and a novel learned count sketch model that captures join correlations across base relations. Across three database instances, we demonstrate that GRASP consistently outperforms existing query-driven models on imperfect workloads, both in terms of estimation accuracy and query latency. Remarkably, GRASP achieves performance comparable to, or even surpassing, traditional approaches built over the underlying data on the complex CEB-IMDb-full benchmark -- despite operating without any data access and using only 10% of all possible join templates.
Authors: Zhenting Qi, Fan Nie, Alexandre Alahi, James Zou, Himabindu Lakkaraju, Yilun Du, Eric Xing, Sham Kakade, Hanlin Zhang
Abstract: Modern language model (LM) training has been divided into multiple stages, making it difficult for downstream developers to evaluate the impact of design choices made at each stage. We present EvoLM, a model suite that enables systematic and transparent analysis of LMs' training dynamics across pre-training, continued pre-training, supervised fine-tuning, and reinforcement learning. By training over 100 LMs with 1B and 4B parameters from scratch, we rigorously evaluate both upstream (language modeling) and downstream (problem-solving) reasoning capabilities, including considerations of both in-domain and out-of-domain generalization. Key insights highlight the diminishing returns from excessive pre-training and post-training, the importance and practices of mitigating forgetting during domain-specific continued pre-training, the crucial role of continued pre-training in bridging pre-training and post-training phases, and various intricate trade-offs when configuring supervised fine-tuning and reinforcement learning. To facilitate open research and reproducibility, we release all pre-trained and post-trained models, training datasets for all stages, and our entire training and evaluation pipeline.
Authors: Xinyue Huang, Ziqi Lin, Fang Sun, Wenchao Zhang, Kejian Tong, Yunbo Liu
Abstract: This paper presents a novel Retrieval-Augmented Generation (RAG) framework tailored for complex question answering tasks, addressing challenges in multi-hop reasoning and contextual understanding across lengthy documents. Built upon LLaMA 3, the framework integrates a dense retrieval module with advanced context fusion and multi-hop reasoning mechanisms, enabling more accurate and coherent response generation. A joint optimization strategy combining retrieval likelihood and generation cross-entropy improves the model's robustness and adaptability. Experimental results show that the proposed system outperforms existing retrieval-augmented and generative baselines, confirming its effectiveness in delivering precise, contextually grounded answers.
Authors: Reyna Abhyankar, Qi Qi, Yiying Zhang
Abstract: Generative AI is being leveraged to solve a variety of computer-use tasks involving desktop applications. State-of-the-art systems have focused solely on improving accuracy on leading benchmarks. However, these systems are practically unusable due to extremely high end-to-end latency (e.g., tens of minutes) for tasks that typically take humans just a few minutes to complete. To understand the cause behind this and to guide future developments of computer agents, we conduct the first study on the temporal performance of computer-use agents on OSWorld, the flagship benchmark in computer-use AI. We find that large model calls for planning and reflection account for the majority of the overall latency, and as an agent uses more steps to complete a task, each successive step can take 3x longer than steps at the beginning of a task. We then construct OSWorld-Human, a manually annotated version of the original OSWorld dataset that contains a human-determined trajectory for each task. We evaluate 16 agents on their efficiency using OSWorld-Human and found that even the highest-scoring agents on OSWorld take 1.4-2.7x more steps than necessary.
Authors: Fei Wang, Xingchen Wan, Ruoxi Sun, Jiefeng Chen, Sercan \"O. Ar{\i}k
Abstract: Inference-time scaling has proven effective in boosting large language model (LLM) performance through increased test-time computation. Yet, its practical application is often hindered by reliance on external verifiers or a lack of optimization for realistic computational constraints. We propose DynScaling, which addresses these limitations through two primary innovations: an integrated parallel-sequential sampling strategy and a bandit-based dynamic budget allocation framework. The integrated sampling strategy unifies parallel and sequential sampling by constructing synthetic sequential reasoning chains from initially independent parallel responses, promoting diverse and coherent reasoning trajectories. The dynamic budget allocation framework formulates the allocation of computational resources as a multi-armed bandit problem, adaptively distributing the inference budget across queries based on the uncertainty of previously sampled responses, thereby maximizing computational efficiency. By combining these components, DynScaling effectively improves LLM performance under practical resource constraints without the need for external verifiers. Experimental results demonstrate that DynScaling consistently surpasses existing verifier-free inference scaling baselines in both task performance and computational cost.
Authors: Prakrut Kotecha, Aditya Shirwatkar, Shishir Kolathaya
Abstract: Lagrangian Neural Networks (LNNs) present a principled and interpretable framework for learning the system dynamics by utilizing inductive biases. While traditional dynamics models struggle with compounding errors over long horizons, LNNs intrinsically preserve the physical laws governing any system, enabling accurate and stable predictions essential for sustainable locomotion. This work evaluates LNNs for infinite horizon planning in quadrupedal robots through four dynamics models: (1) full-order forward dynamics (FD) training and inference, (2) diagonalized representation of Mass Matrix in full order FD, (3) full-order inverse dynamics (ID) training with FD inference, (4) reduced-order modeling via torso centre-of-mass (CoM) dynamics. Experiments demonstrate that LNNs bring improvements in sample efficiency (10x) and superior prediction accuracy (up to 2-10x) compared to baseline methods. Notably, the diagonalization approach of LNNs reduces computational complexity while retaining some interpretability, enabling real-time receding horizon control. These findings highlight the advantages of LNNs in capturing the underlying structure of system dynamics in quadrupeds, leading to improved performance and efficiency in locomotion planning and control. Additionally, our approach achieves a higher control frequency than previous LNN methods, demonstrating its potential for real-world deployment on quadrupeds.
Authors: Sean Moushegian, Taposh Banerjee, Vahid Tarokh
Abstract: Score-based methods have recently seen increasing popularity in modeling and generation. Methods have been constructed to perform hypothesis testing and change-point detection with score functions, but these methods are in general not as powerful as their likelihood-based peers. Recent works consider generalizing the score-based Fisher divergence into a diffusion-divergence by transforming score functions via multiplication with a matrix-valued function or a weight matrix. In this paper, we extend the score-based hypothesis test and change-point detection stopping rule into their diffusion-based analogs. Additionally, we theoretically quantify the performance of these diffusion-based algorithms and study scenarios where optimal performance is achievable. We propose a method of numerically optimizing the weight matrix and present numerical simulations to illustrate the advantages of diffusion-based algorithms.
Authors: Fivos Kalogiannis, Emmanouil-Vasileios Vlatakis-Gkaragkounis, Ian Gemp, Georgios Piliouras
Abstract: We contribute the first provable guarantees of global convergence to Nash equilibria (NE) in two-player zero-sum convex Markov games (cMGs) by using independent policy gradient methods. Convex Markov games, recently defined by Gemp et al. (2024), extend Markov decision processes to multi-agent settings with preferences that are convex over occupancy measures, offering a broad framework for modeling generic strategic interactions. However, even the fundamental min-max case of cMGs presents significant challenges, including inherent nonconvexity, the absence of Bellman consistency, and the complexity of the infinite horizon. We follow a two-step approach. First, leveraging properties of hidden-convex--hidden-concave functions, we show that a simple nonconvex regularization transforms the min-max optimization problem into a nonconvex-proximal Polyak-Lojasiewicz (NC-pPL) objective. Crucially, this regularization can stabilize the iterates of independent policy gradient methods and ultimately lead them to converge to equilibria. Second, building on this reduction, we address the general constrained min-max problems under NC-pPL and two-sided pPL conditions, providing the first global convergence guarantees for stochastic nested and alternating gradient descent-ascent methods, which we believe may be of independent interest.
Authors: Yi Chen, Yuying Ge, Rui Wang, Yixiao Ge, Junhao Cheng, Ying Shan, Xihui Liu
Abstract: Recent reinforcement learning approaches, such as outcome-supervised GRPO, have advanced Chain-of-Thought reasoning in large language models (LLMs), yet their adaptation to multimodal LLMs (MLLMs) is unexplored. To address the lack of rigorous evaluation for MLLM post-training methods, we introduce SEED-Bench-R1, a benchmark with complex real-world videos requiring balanced perception and reasoning. It offers a large training set and evaluates generalization across three escalating challenges: in-distribution, cross-environment, and cross-environment-task scenarios. Using SEED-Bench-R1, we find that standard GRPO, while improving answer accuracy, often reduces logical coherence between reasoning steps and answers, with only a 57.9% consistency rate. This stems from reward signals focusing solely on final answers, encouraging shortcuts, and strict KL penalties limiting exploration.To address this, we propose GRPO-CARE, a consistency-aware RL framework optimizing both answer correctness and reasoning coherence without explicit supervision. GRPO-CARE introduces a two-tiered reward: (1) a base reward for answer correctness, and (2) an adaptive consistency bonus, computed by comparing the model's reasoning-to-answer likelihood (via a slowly-evolving reference model) against group peers.This dual mechanism amplifies rewards for reasoning paths that are both correct and logically consistent. Replacing KL penalties with this adaptive bonus, GRPO-CARE outperforms standard GRPO on SEED-Bench-R1, achieving a 6.7% performance gain on the hardest evaluation level and a 24.5% improvement in consistency. It also shows strong transferability, improving model performance across diverse video understanding benchmarks. Our work contributes a systematically designed benchmark and a generalizable post-training framework, advancing the development of more interpretable and robust MLLMs.
Authors: Ana Kostovska, Carola Doerr, Sa\v{s}o D\v{z}eroski, Pan\v{c}e Panov, Tome Eftimov
Abstract: Automated algorithm performance prediction in numerical blackbox optimization often relies on problem characterizations, such as exploratory landscape analysis features. These features are typically used as inputs to machine learning models and are represented in a tabular format. However, such approaches often overlook algorithm configurations, a key factor influencing performance. The relationships between algorithm operators, parameters, problem characteristics, and performance outcomes form a complex structure best represented as a graph. This work explores the use of heterogeneous graph data structures and graph neural networks to predict the performance of optimization algorithms by capturing the complex dependencies between problems, algorithm configurations, and performance outcomes. We focus on two modular frameworks, modCMA-ES and modDE, which decompose two widely used derivative-free optimization algorithms: the covariance matrix adaptation evolution strategy (CMA-ES) and differential evolution (DE). We evaluate 324 modCMA-ES and 576 modDE variants on 24 BBOB problems across six runtime budgets and two problem dimensions. Achieving up to 36.6% improvement in MSE over traditional tabular-based methods, this work highlights the potential of geometric learning in black-box optimization.
Authors: Putri A. van der Linden, Alexander Timans, Erik J. Bekkers
Abstract: We study the problem of conformal prediction (CP) under geometric data shifts, where data samples are susceptible to transformations such as rotations or flips. While CP endows prediction models with post-hoc uncertainty quantification and formal coverage guarantees, their practicality breaks under distribution shifts that deteriorate model performance. To address this issue, we propose integrating geometric information--such as geometric pose--into the conformal procedure to reinstate its guarantees and ensure robustness under geometric shifts. In particular, we explore recent advancements on pose canonicalization as a suitable information extractor for this purpose. Evaluating the combined approach across discrete and continuous shifts and against equivariant and augmentation-based baselines, we find that integrating geometric information with CP yields a principled way to address geometric shifts while maintaining broad applicability to black-box predictors.
Authors: Annajoyce Mariani, Kira Maag, Hanno Gottschalk
Abstract: Being able to generate realistic trajectory options is at the core of increasing the degree of automation of road vehicles. While model-driven, rule-based, and classical learning-based methods are widely used to tackle these tasks at present, they can struggle to effectively capture the complex, multimodal distributions of future trajectories. In this paper we investigate whether a generative adversarial network (GAN) trained on videos of bird's-eye view (BEV) traffic scenarios can generate statistically accurate trajectories that correctly capture spatial relationships between the agents. To this end, we propose a pipeline that uses low-resolution BEV occupancy grid videos as training data for a video generative model. From the generated videos of traffic scenarios we extract abstract trajectory data using single-frame object detection and frame-to-frame object matching. We particularly choose a GAN architecture for the fast training and inference times with respect to diffusion models. We obtain our best results within 100 GPU hours of training, with inference times under 20\,ms. We demonstrate the physical realism of the proposed trajectories in terms of distribution alignment of spatial and dynamic parameters with respect to the ground truth videos from the Waymo Open Motion Dataset.
Authors: Bishwajit Prasad Gond, Rajneekant, Pushkar Kishore, Durga Prasad Mohapatra
Abstract: This paper investigates the application of natural language processing (NLP)-based n-gram analysis and machine learning techniques to enhance malware classification. We explore how NLP can be used to extract and analyze textual features from malware samples through n-grams, contiguous string or API call sequences. This approach effectively captures distinctive linguistic patterns among malware and benign families, enabling finer-grained classification. We delve into n-gram size selection, feature representation, and classification algorithms. While evaluating our proposed method on real-world malware samples, we observe significantly improved accuracy compared to the traditional methods. By implementing our n-gram approach, we achieved an accuracy of 99.02% across various machine learning algorithms by using hybrid feature selection technique to address high dimensionality. Hybrid feature selection technique reduces the feature set to only 1.6% of the original features.
Authors: Chenrui Ma, Zechang Sun, Tao Jing, Zheng Cai, Yuan-Sen Ting, Song Huang, Mingyu Li
Abstract: Observational astronomy relies on visual feature identification to detect critical astrophysical phenomena. While machine learning (ML) increasingly automates this process, models often struggle with generalization in large-scale surveys due to the limited representativeness of labeled datasets -- whether from simulations or human annotation -- a challenge pronounced for rare yet scientifically valuable objects. To address this, we propose a conditional diffusion model to synthesize realistic galaxy images for augmenting ML training data. Leveraging the Galaxy Zoo 2 dataset which contains visual feature -- galaxy image pairs from volunteer annotation, we demonstrate that our model generates diverse, high-fidelity galaxy images closely adhere to the specified morphological feature conditions. Moreover, this model enables generative extrapolation to project well-annotated data into unseen domains and advancing rare object detection. Integrating synthesized images into ML pipelines improves performance in standard morphology classification, boosting completeness and purity by up to 30\% across key metrics. For rare object detection, using early-type galaxies with prominent dust lane features ( $\sim$0.1\% in GZ2 dataset) as a test case, our approach doubled the number of detected instances from 352 to 872, compared to previous studies based on visual inspection. This study highlights the power of generative models to bridge gaps between scarce labeled data and the vast, uncharted parameter space of observational astronomy and sheds insight for future astrophysical foundation model developments. Our project homepage is available at https://galaxysd-webpage.streamlit.app/.
Authors: Mike Nguyen, Nicole M\"ucke
Abstract: Random feature approximation is arguably one of the most widely used techniques for kernel methods in large-scale learning algorithms. In this work, we analyze the generalization properties of random feature methods, extending previous results for Tikhonov regularization to a broad class of spectral regularization techniques. This includes not only explicit methods but also implicit schemes such as gradient descent and accelerated algorithms like the Heavy-Ball and Nesterov method. Through this framework, we enable a theoretical analysis of neural networks and neural operators through the lens of the Neural Tangent Kernel (NTK) approach trained via gradient descent. For our estimators we obtain optimal learning rates over regularity classes (even for classes that are not included in the reproducing kernel Hilbert space), which are defined through appropriate source conditions. This improves or completes previous results obtained in related settings for specific kernel algorithms.
Authors: Oswaldo Ludwig
Abstract: This paper explores the relationship between the condition number of a neural network's weight tensor and the extent of information encoded by the associated processing unit, viewed through the lens of information theory. We argue that a high condition number, though not sufficient for effective knowledge encoding, may indicate that the unit has learned to selectively amplify and compress information. We formalize this intuition, particularly for linear units with Gaussian inputs, linking the condition number and the transformation's log-volume scaling factor to the characteristics of the output entropy and the geometric properties of the learned transformation. Our analysis demonstrates that for a fixed weight norm, a concentrated distribution of singular values (high condition number) corresponds to reduced overall information transfer, indicating a specialized and efficient encoding strategy. Furthermore, we present a practical case study where these principles are applied to guide selective fine-tuning of a multimodal Large Language Model, aiming to mitigate catastrophic forgetting during cross-modal adaptation. Unlike many existing catastrophic forgetting mitigation methods that rely on access to pre-training statistics, which are often unavailable, our selective fine-tuning approach offers a way to bypass this common requirement.
Authors: Heng Zhang, Zikang Wan, Danilo Vasconcellos Vargas
Abstract: Human vision excels at segmenting visual cues without the need for explicit training, and it remains remarkably robust even as noise severity increases. In contrast, existing AI algorithms struggle to maintain accuracy under similar conditions. Here, we present SyncMapV2, the first to solve unsupervised segmentation with state-of-the-art robustness. SyncMapV2 exhibits a minimal drop in mIoU, only 0.01%, under digital corruption, compared to a 23.8% drop observed in SOTA methods.This superior performance extends across various types of corruption: noise (7.3% vs. 37.7%), weather (7.5% vs. 33.8%), and blur (7.0% vs. 29.5%). Notably, SyncMapV2 accomplishes this without any robust training, supervision, or loss functions. It is based on a learning paradigm that uses self-organizing dynamical equations combined with concepts from random networks. Moreover,unlike conventional methods that require re-initialization for each new input, SyncMapV2 adapts online, mimicking the continuous adaptability of human vision. Thus, we go beyond the accurate and robust results, and present the first algorithm that can do all the above online, adapting to input rather than re-initializing. In adaptability tests, SyncMapV2 demonstrates near-zero performance degradation, which motivates and fosters a new generation of robust and adaptive intelligence in the near future.
Authors: Lukas Gonon, Rodrigo Mart\'inez-Pe\~na, Juan-Pablo Ortega
Abstract: Quantum reservoir computing uses the dynamics of quantum systems to process temporal data, making it particularly well-suited for learning with noisy intermediate-scale quantum devices. Early experimental proposals, such as the restarting and rewinding protocols, relied on repeating previous steps of the quantum map to avoid backaction. However, this approach compromises real-time processing and increases computational overhead. Recent developments have introduced alternative protocols that address these limitations. These include online, mid-circuit measurement, and feedback techniques, which enable real-time computation while preserving the input history. Among these, the feedback protocol stands out for its ability to process temporal information with comparatively fewer components. Despite this potential advantage, the theoretical foundations of feedback-based quantum reservoir computing remain underdeveloped, particularly with regard to the universality and the approximation capabilities of this approach. This paper addresses this issue by presenting a recurrent quantum neural network architecture that extends a class of existing feedforward models to a dynamic, feedback-driven reservoir setting. We provide theoretical guarantees for variational recurrent quantum neural networks, including approximation bounds and universality results. Notably, our analysis demonstrates that the model is universal with linear readouts, making it both powerful and experimentally accessible. These results pave the way for practical and theoretically grounded quantum reservoir computing with real-time processing capabilities.
Authors: Santosh Patapati, Trisanth Srinivasan, Amith Adiraju
Abstract: Micro-gesture recognition is a challenging task in affective computing due to the subtle, involuntary nature of the gestures and their low movement amplitude. In this paper, we introduce a Pose-Guided Semantics-Aware CLIP-based architecture, or CLIP for Micro-Gesture recognition (CLIP-MG), a modified CLIP model tailored for micro-gesture classification on the iMiGUE dataset. CLIP-MG integrates human pose (skeleton) information into the CLIP-based recognition pipeline through pose-guided semantic query generation and a gated multi-modal fusion mechanism. The proposed model achieves a Top-1 accuracy of 61.82%. These results demonstrate both the potential of our approach and the remaining difficulty in fully adapting vision-language models like CLIP for micro-gesture recognition.
Authors: Zelin Xiao, Jia Gu, Song Xi Chen
Abstract: We study methods for identifying heterogeneous parameter components in distributed M-estimation with minimal data transmission. One is based on a re-normalized Wald test, which is shown to be consistent as long as the number of distributed data blocks $K$ is of a smaller order of the minimum block sample size {and the level of heterogeneity is dense}. The second one is an extreme contrast test (ECT) based on the difference between the largest and smallest component-wise estimated parameters among data blocks. By introducing a sample splitting procedure, the ECT can avoid the bias accumulation arising from the M-estimation procedures, and exhibits consistency for $K$ being much larger than the sample size while the heterogeneity is sparse. The ECT procedure is easy to operate and communication-efficient. A combination of the Wald and the extreme contrast tests is formulated to attain more robust power under varying levels of sparsity of the heterogeneity. We also conduct intensive numerical experiments to compare the family-wise error rate (FWER) and the power of the proposed methods. Additionally, we conduct a case study to present the implementation and validity of the proposed methods.
Authors: Xiaoya Lu, Zeren Chen, Xuhao Hu, Yijin Zhou, Weichen Zhang, Dongrui Liu, Lu Sheng, Jing Shao
Abstract: Flawed planning from VLM-driven embodied agents poses significant safety hazards, hindering their deployment in real-world household tasks. However, existing static, non-interactive evaluation paradigms fail to adequately assess risks within these interactive environments, since they cannot simulate dynamic risks that emerge from an agent's actions and rely on unreliable post-hoc evaluations that ignore unsafe intermediate steps. To bridge this critical gap, we propose evaluating an agent's interactive safety: its ability to perceive emergent risks and execute mitigation steps in the correct procedural order. We thus present IS-Bench, the first multi-modal benchmark designed for interactive safety, featuring 161 challenging scenarios with 388 unique safety risks instantiated in a high-fidelity simulator. Crucially, it facilitates a novel process-oriented evaluation that verifies whether risk mitigation actions are performed before/after specific risk-prone steps. Extensive experiments on leading VLMs, including the GPT-4o and Gemini-2.5 series, reveal that current agents lack interactive safety awareness, and that while safety-aware Chain-of-Thought can improve performance, it often compromises task completion. By highlighting these critical limitations, IS-Bench provides a foundation for developing safer and more reliable embodied AI systems.
Authors: Zhen Xu, Shang Zhu, Jue Wang, Junlin Wang, Ben Athiwaratkun, Chi Wang, James Zou, Ce Zhang
Abstract: We investigate the challenge of applying Large Language Models (LLMs) to long texts. We propose a theoretical framework that distinguishes the failure modes of long context tasks into three categories: cross-chunk dependence (task noise), confusion that grows with context size (model noise), and the imperfect integration of partial results (aggregator noise). Under this view, we analyze when it is effective to use multi-agent chunking, i.e., dividing a length sequence into smaller chunks and aggregating the processed results of each chunk. Our experiments on tasks such as retrieval, question answering, and summarization confirm both the theoretical analysis and the conditions that favor multi-agent chunking. By exploring superlinear model noise growth with input length, we also explain why, for large inputs, a weaker model configured with chunk-based processing can surpass a more advanced model like GPT4o applied in a single shot. Overall, we present a principled understanding framework and our results highlight a direct pathway to handling long contexts in LLMs with carefully managed chunking and aggregator strategies.
Authors: Alexander Timans, Rajeev Verma, Eric Nalisnick, Christian A. Naesseth
Abstract: Machine learning systems deployed in the real world must operate under dynamic and often unpredictable distribution shifts. This challenges the validity of statistical safety assurances on the system's risk established beforehand. Common risk control frameworks rely on fixed assumptions and lack mechanisms to continuously monitor deployment reliability. In this work, we propose a general framework for the real-time monitoring of risk violations in evolving data streams. Leveraging the 'testing by betting' paradigm, we propose a sequential hypothesis testing procedure to detect violations of bounded risks associated with the model's decision-making mechanism, while ensuring control on the false alarm rate. Our method operates under minimal assumptions on the nature of encountered shifts, rendering it broadly applicable. We illustrate the effectiveness of our approach by monitoring risks in outlier detection and set prediction under a variety of shifts.
Authors: Sami Abboud, Eleanor Hanna, Olivier Jeunen, Vineesha Raheja, Schaun Wheeler
Abstract: Consumer applications provide ample opportunities to surface and communicate various forms of content to users. From promotional campaigns for new features or subscriptions, to evergreen nudges for engagement, or personalised recommendations; across e-mails, push notifications, and in-app surfaces. The conventional approach to orchestration for communication relies heavily on labour-intensive manual marketer work, and inhibits effective personalisation of content, timing, frequency, and copy-writing. We formulate this task under a sequential decision-making framework, where we aim to optimise a modular decision-making policy that maximises incremental engagement for any funnel event. Our approach leverages a Difference-in-Differences design for Individual Treatment Effect estimation, and Thompson sampling to balance the explore-exploit trade-off. We present results from a multi-service application, where our methodology has resulted in significant increases to a variety of goal events across several product features, and is currently deployed across 150 million users.
Authors: Yaru Niu, Yunzhe Zhang, Mingyang Yu, Changyi Lin, Chenhao Li, Yikai Wang, Yuxiang Yang, Wenhao Yu, Tingnan Zhang, Bingqing Chen, Jonathan Francis, Zhenzhen Li, Jie Tan, Ding Zhao
Abstract: Quadrupedal robots have demonstrated impressive locomotion capabilities in complex environments, but equipping them with autonomous versatile manipulation skills in a scalable way remains a significant challenge. In this work, we introduce a cross-embodiment imitation learning system for quadrupedal manipulation, leveraging data collected from both humans and LocoMan, a quadruped equipped with multiple manipulation modes. Specifically, we develop a teleoperation and data collection pipeline, which unifies and modularizes the observation and action spaces of the human and the robot. To effectively leverage the collected data, we propose an efficient modularized architecture that supports co-training and pretraining on structured modality-aligned data across different embodiments. Additionally, we construct the first manipulation dataset for the LocoMan robot, covering various household tasks in both unimanual and bimanual modes, supplemented by a corresponding human dataset. We validate our system on six real-world manipulation tasks, where it achieves an average success rate improvement of 41.9% overall and 79.7% under out-of-distribution (OOD) settings compared to the baseline. Pretraining with human data contributes a 38.6% success rate improvement overall and 82.7% under OOD settings, enabling consistently better performance with only half the amount of robot data. Our code, hardware, and data are open-sourced at: https://human2bots.github.io.
Authors: Saad Almohaimeed, Saleh Almohaimeed, Damla Turgut, Ladislau B\"ol\"oni
Abstract: Implicit hate speech has recently emerged as a critical challenge for social media platforms. While much of the research has traditionally focused on harmful speech in general, the need for generalizable techniques to detect veiled and subtle forms of hate has become increasingly pressing. Based on lexicon analysis, we hypothesize that implicit hate speech is already present in publicly available harmful speech datasets but may not have been explicitly recognized or labeled by annotators. Additionally, crowdsourced datasets are prone to mislabeling due to the complexity of the task and often influenced by annotators' subjective interpretations. In this paper, we propose an approach to address the detection of implicit hate speech and enhance generalizability across diverse datasets by leveraging existing harmful speech datasets. Our method comprises three key components: influential sample identification, reannotation, and augmentation using Llama-3 70B and GPT-4o. Experimental results demonstrate the effectiveness of our approach in improving implicit hate detection, achieving a +12.9-point F1 score improvement compared to the baseline.
Authors: Zexi Liu, Yuzhu Cai, Xinyu Zhu, Yujie Zheng, Runkun Chen, Ying Wen, Yanfeng Wang, Weinan E, Siheng Chen
Abstract: As AI capabilities advance toward and potentially beyond human-level performance, a natural transition emerges where AI-driven development becomes more efficient than human-centric approaches. A promising pathway toward this transition lies in AI-for-AI (AI4AI), which leverages AI techniques to automate and optimize the design, training, and deployment of AI systems themselves. While LLM-based agents have shown the potential to realize AI4AI, they are often unable to fully leverage the experience accumulated by agents during the exploration of solutions in the reasoning process, leading to inefficiencies and suboptimal performance. To address this limitation, we propose ML-Master, a novel AI4AI agent that seamlessly integrates exploration and reasoning by employing a selectively scoped memory mechanism. This approach allows ML-Master to efficiently combine diverse insights from parallel solution trajectories with analytical reasoning, guiding further exploration without overwhelming the agent with excessive context. We evaluate ML-Master on the MLE-Bench, where it achieves a 29.3% average medal rate, significantly surpassing existing methods, particularly in medium-complexity tasks, while accomplishing this superior performance within a strict 12-hour time constraint-half the 24-hour limit used by previous baselines. These results demonstrate ML-Master's potential as a powerful tool for advancing AI4AI.
Authors: Pedro Rodr\'iguez Fern\'andez, Christian Svinth, Alex Hagen
Abstract: We present a method to improve the detection limit for radionuclides using spectroscopic radiation detectors and the arrival time of each detected radiation quantum. We enable this method using a neural network with an attention mechanism. We illustrate the method on the detection of Cesium release from a nuclear facility during an upset, and our method shows $2\times$ improvement over the traditional spectroscopic method. We hypothesize that our method achieves this performance increase by modulating its detection probability by the overall rate of probable detections, specifically by adapting detection thresholds based on temporal event distributions and local spectral features, and show evidence to this effect. We believe this method is applicable broadly and may be more successful for radionuclides with more complicated decay chains than Cesium; we also note that our method can generalize beyond the addition of arrival time and could integrate other data about each detection event, such as pulse quality, location in detector, or even combining the energy and time from detections in different detectors.
Authors: Liyang Yu, Tianyi Wang, Junfeng Jiao, Fengwu Shan, Hongqing Chu, Bingzhao Gao
Abstract: In complex real-world traffic environments, autonomous vehicles (AVs) need to interact with other traffic participants while making real-time and safety-critical decisions accordingly. The unpredictability of human behaviors poses significant challenges, particularly in dynamic scenarios, such as multi-lane highways and unsignalized T-intersections. To address this gap, we design a bi-level interaction decision-making algorithm (BIDA) that integrates interactive Monte Carlo tree search (MCTS) with deep reinforcement learning (DRL), aiming to enhance interaction rationality, efficiency and safety of AVs in dynamic key traffic scenarios. Specifically, we adopt three types of DRL algorithms to construct a reliable value network and policy network, which guide the online deduction process of interactive MCTS by assisting in value update and node selection. Then, a dynamic trajectory planner and a trajectory tracking controller are designed and implemented in CARLA to ensure smooth execution of planned maneuvers. Experimental evaluations demonstrate that our BIDA not only enhances interactive deduction and reduces computational costs, but also outperforms other latest benchmarks, which exhibits superior safety, efficiency and interaction rationality under varying traffic conditions.
Authors: Keyhan Najafian, Farhad Maleki, Lingling Jin, Ian Stavness
Abstract: Instance segmentation is essential for applications such as automated monitoring of plant health, growth, and yield. However, extensive effort is required to create large-scale datasets with pixel-level annotations of each object instance for developing instance segmentation models that restrict the use of deep learning in these areas. This challenge is more significant in images with densely packed, self-occluded objects, which are common in agriculture. To address this challenge, we propose a semi-self-supervised learning approach that requires minimal manual annotation to develop a high-performing instance segmentation model. We design GLMask, an image-mask representation for the model to focus on shape, texture, and pattern while minimizing its dependence on color features. We develop a pipeline to generate semantic segmentation and then transform it into instance-level segmentation. The proposed approach substantially outperforms the conventional instance segmentation models, establishing a state-of-the-art wheat head instance segmentation model with mAP@50 of 98.5%. Additionally, we assessed the proposed methodology on the general-purpose Microsoft COCO dataset, achieving a significant performance improvement of over 12.6% mAP@50. This highlights that the utility of our proposed approach extends beyond precision agriculture and applies to other domains, specifically those with similar data characteristics.
Authors: Nadav Kunievsky, James A. Evans
Abstract: Understanding whether large language models (LLMs) possess a world model-a structured understanding of the world that supports generalization beyond surface-level patterns-is central to assessing their reliability, especially in high-stakes applications. We propose a formal framework for evaluating whether an LLM exhibits a sufficiently robust world model, defined as producing consistent outputs across semantically equivalent prompts while distinguishing between prompts that express different intents. We introduce a new evaluation approach to measure this that decomposes model response variability into three components: variability due to user purpose, user articulation, and model instability. An LLM with a strong world model should attribute most of the variability in its responses to changes in foundational purpose rather than superficial changes in articulation. This approach allows us to quantify how much of a model's behavior is semantically grounded rather than driven by model instability or alternative wording. We apply this framework to evaluate LLMs across diverse domains. Our results show how larger models attribute a greater share of output variability to changes in user purpose, indicating a more robust world model. This improvement is not uniform, however: larger models do not consistently outperform smaller ones across all domains, and their advantage in robustness is often modest. These findings highlight the importance of moving beyond accuracy-based benchmarks toward semantic diagnostics that more directly assess the structure and stability of a model's internal understanding of the world.
Authors: Haotian Yin, Aleksander Plocharski, Michal Jan Wlodarczyk, Mikolaj Kida, Przemyslaw Musialski
Abstract: Neural signed-distance fields (SDFs) have become a versatile backbone for geometric learning, yet enforcing developable, CAD-style behavior still hinges on Gaussian curvature penalties that require full Hessian evaluation and second-order automatic differentiation, both of which are costly in memory and runtime. We present a curvature proxy that regularizes only the mixed second-order term (Weingarten term), allowing the two principal curvatures to adapt freely to data while suppressing unwanted warp. Two complementary instantiations realize this idea: (i) a finite-difference proxy that replaces each Hessian entry with four forward SDF evaluations and a single first-order gradient, and (ii) an autodiff proxy that computes the same mixed derivative via one Hessian-vector product, sidestepping explicit full Hessian assembly and remaining faster in practice. Both variants converge to the exact mixed second derivative, thus preserving the intended geometric bias without incurring full second-order graphs. On the ABC benchmarks, the proxies match or exceed the reconstruction fidelity of Hessian-based baselines while reducing GPU memory use and wall-clock time by a factor of two. Because the method is drop-in and framework-agnostic, it opens a practical path toward scalable, curvature-aware SDF learning for engineering-grade shape reconstruction.
Authors: Jianlin Shi, Brian T. Bucher
Abstract: Despite advances in machine learning (ML) and large language models (LLMs), rule-based natural language processing (NLP) systems remain active in clinical settings due to their interpretability and operational efficiency. However, their manual development and maintenance are labor-intensive, particularly in tasks with large linguistic variability. To overcome these limitations, we proposed a novel approach employing LLMs solely during the rule-based systems development phase. We conducted the initial experiments focusing on the first two steps of developing a rule-based NLP pipeline: find relevant snippets from the clinical note; extract informative keywords from the snippets for the rule-based named entity recognition (NER) component. Our experiments demonstrated exceptional recall in identifying clinically relevant text snippets (Deepseek: 0.98, Qwen: 0.99) and 1.0 in extracting key terms for NER. This study sheds light on a promising new direction for NLP development, enabling semi-automated or automated development of rule-based systems with significantly faster, more cost-effective, and transparent execution compared with deep learning model-based solutions.
Authors: Rex Shen, Lu Tian
Abstract: Synthetic Data Generation has become essential for scalable, privacy-preserving statistical analysis. While standard approaches based on generative models, such as Normalizing Flows, have been widely used, they often suffer from slow convergence in high-dimensional settings, frequently converging more slowly than the canonical $1/\sqrt{n}$ rate when approximating the true data distribution. To overcome these limitations, we propose a Latent Noise Injection method using Masked Autoregressive Flows (MAF). Instead of directly sampling from the trained model, our method perturbs each data point in the latent space and maps it back to the data domain. This construction preserves a one to one correspondence between observed and synthetic data, enabling synthetic outputs that closely reflect the underlying distribution, particularly in challenging high-dimensional regimes where traditional sampling struggles. Our procedure satisfies local $(\epsilon, \delta)$-differential privacy and introduces a single perturbation parameter to control the privacy-utility trade-off. Although estimators based on individual synthetic datasets may converge slowly, we show both theoretically and empirically that aggregating across $K$ studies in a meta analysis framework restores classical efficiency and yields consistent, reliable inference. We demonstrate that with a well-calibrated perturbation parameter, Latent Noise Injection achieves strong statistical alignment with the original data and robustness against membership inference attacks. These results position our method as a compelling alternative to conventional flow-based sampling for synthetic data sharing in decentralized and privacy-sensitive domains, such as biomedical research.
Authors: Guang Yin, Yitong Li, Yixuan Wang, Dale McConachie, Paarth Shah, Kunimatsu Hashimoto, Huan Zhang, Katherine Liu, Yunzhu Li
Abstract: Natural language instructions for robotic manipulation tasks often exhibit ambiguity and vagueness. For instance, the instruction "Hang a mug on the mug tree" may involve multiple valid actions if there are several mugs and branches to choose from. Existing language-conditioned policies typically rely on end-to-end models that jointly handle high-level semantic understanding and low-level action generation, which can result in suboptimal performance due to their lack of modularity and interpretability. To address these challenges, we introduce a novel robotic manipulation framework that can accomplish tasks specified by potentially ambiguous natural language. This framework employs a Vision-Language Model (VLM) to interpret abstract concepts in natural language instructions and generates task-specific code - an interpretable and executable intermediate representation. The generated code interfaces with the perception module to produce 3D attention maps that highlight task-relevant regions by integrating spatial and semantic information, effectively resolving ambiguities in instructions. Through extensive experiments, we identify key limitations of current imitation learning methods, such as poor adaptation to language and environmental variations. We show that our approach excels across challenging manipulation tasks involving language ambiguity, contact-rich manipulation, and multi-object interactions.
Authors: Vladislav Belozerov, Peter J Barclay, Askhan Sami
Abstract: Large-language-model coding tools are now mainstream in software engineering. But as these same tools move human effort up the development stack, they present fresh dangers: 10% of real prompts leak private data, 42% of generated snippets hide security flaws, and the models can even ``agree'' with wrong ideas, a trait called sycophancy. We argue that firms must tag and review every AI-generated line of code, keep prompts and outputs inside private or on-premises deployments, obey emerging safety regulations, and add tests that catch sycophantic answers -- so they can gain speed without losing security and accuracy.
Authors: Wenlong Ji, Yihan Pan, Ruihao Zhu, Lihua Lei
Abstract: Multi-armed bandit (MAB) is a widely adopted framework for sequential decision-making under uncertainty. Traditional bandit algorithms rely solely on online data, which tends to be scarce as it must be gathered during the online phase when the arms are actively pulled. However, in many practical settings, rich auxiliary data, such as covariates of past users, is available prior to deploying any arms. We introduce a new setting for MAB where pre-trained machine learning (ML) models are applied to convert side information and historical data into \emph{surrogate rewards}. A prominent feature of this setting is that the surrogate rewards may exhibit substantial bias, as true reward data is typically unavailable in the offline phase, forcing ML predictions to heavily rely on extrapolation. To address the issue, we propose the Machine Learning-Assisted Upper Confidence Bound (MLA-UCB) algorithm, which can be applied to any reward prediction model and any form of auxiliary data. When the predicted and true rewards are jointly Gaussian, it provably improves the cumulative regret, provided that the correlation is non-zero -- even in cases where the mean surrogate reward completely misaligns with the true mean rewards. Notably, our method requires no prior knowledge of the covariance matrix between true and surrogate rewards. We compare MLA-UCB with the standard UCB on a range of numerical studies and show a sizable efficiency gain even when the size of the offline data and the correlation between predicted and true rewards are moderate.
Authors: Meenatchi Sundaram Muthu Selva Annamalai, Borja Balle, Jamie Hayes, Georgios Kaissis, Emiliano De Cristofaro
Abstract: This paper systematizes research on auditing Differential Privacy (DP) techniques, aiming to identify key insights into the current state of the art and open challenges. First, we introduce a comprehensive framework for reviewing work in the field and establish three cross-contextual desiderata that DP audits should target--namely, efficiency, end-to-end-ness, and tightness. Then, we systematize the modes of operation of state-of-the-art DP auditing techniques, including threat models, attacks, and evaluation functions. This allows us to highlight key details overlooked by prior work, analyze the limiting factors to achieving the three desiderata, and identify open research problems. Overall, our work provides a reusable and systematic methodology geared to assess progress in the field and identify friction points and future directions for our community to focus on.
Authors: Manuel Brack, Sudeep Katakol, Felix Friedrich, Patrick Schramowski, Hareesh Ravi, Kristian Kersting, Ajinkya Kale
Abstract: Training data is at the core of any successful text-to-image models. The quality and descriptiveness of image text are crucial to a model's performance. Given the noisiness and inconsistency in web-scraped datasets, recent works shifted towards synthetic training captions. While this setup is generally believed to produce more capable models, current literature does not provide any insights into its design choices. This study closes this gap by systematically investigating how different synthetic captioning strategies impact the downstream performance of text-to-image models. Our experiments demonstrate that dense, high-quality captions enhance text alignment but may introduce trade-offs in output aesthetics and diversity. Conversely, captions of randomized lengths yield balanced improvements across aesthetics and alignment without compromising sample diversity. We also demonstrate that varying caption distributions introduce significant shifts in the output bias of a trained model. Our findings underscore the importance of caption design in achieving optimal model performance and provide practical insights for more effective training data strategies in text-to-image generation.
Authors: Jinlong Pang, Jiaheng Wei, Yifan Hua, Chen Qian, Yang Liu
Abstract: Federated learning (FL) provides a promising paradigm for facilitating collaboration between multiple clients that jointly learn a global model without directly sharing their local data. However, existing research suffers from two caveats: 1) From the perspective of agents, voluntary and unselfish participation is often assumed. But self-interested agents may opt out of the system or provide low-quality contributions without proper incentives; 2) From the mechanism designer's perspective, the aggregated models can be unsatisfactory as the existing game-theoretical federated learning approach for data collection ignores the potential heterogeneous effort caused by contributed data. To alleviate above challenges, we propose an incentive-aware framework for agent participation that considers data heterogeneity to accelerate the convergence process. Specifically, we first introduce the notion of Wasserstein distance to explicitly illustrate the heterogeneous effort and reformulate the existing upper bound of convergence. To induce truthful reporting from agents, we analyze and measure the generalization error gap of any two agents by leveraging the peer prediction mechanism to develop score functions. We further present a two-stage Stackelberg game model that formalizes the process and examines the existence of equilibrium. Extensive experiments on real-world datasets demonstrate the effectiveness of our proposed mechanism.
Authors: Grzegorz Gruszczynski, Michal Jan Wlodarczyk, Jakub J Meixner, Przemyslaw Musialski
Abstract: We propose a novel PDE-driven corruption process for generative image synthesis based on advection-diffusion processes which generalizes existing PDE-based approaches. Our forward pass formulates image corruption via a physically motivated PDE that couples directional advection with isotropic diffusion and Gaussian noise, controlled by dimensionless numbers (Peclet, Fourier). We implement this PDE numerically through a GPU-accelerated custom Lattice Boltzmann solver for fast evaluation. To induce realistic turbulence, we generate stochastic velocity fields that introduce coherent motion and capture multi-scale mixing. In the generative process, a neural network learns to reverse the advection-diffusion operator thus constituting a novel generative model. We discuss how previous methods emerge as specific cases of our operator, demonstrating that our framework generalizes prior PDE-based corruption techniques. We illustrate how advection improves the diversity and quality of the generated images while keeping the overall color palette unaffected. This work bridges fluid dynamics, dimensionless PDE theory, and deep generative modeling, offering a fresh perspective on physically informed image corruption processes for diffusion-based synthesis.
Authors: Fabian Gr\"oger, Shuo Wen, Huyen Le, Maria Brbi\'c
Abstract: Multimodal models have demonstrated powerful capabilities in complex tasks requiring multimodal alignment including zero-shot classification and cross-modal retrieval. However, existing models typically rely on millions of paired multimodal samples, which are prohibitively expensive or infeasible to obtain in many domains. In this work, we explore the feasibility of building multimodal models with limited amount of paired data by aligning pretrained unimodal foundation models. We show that high-quality alignment is possible with as few as tens of thousands of paired samples$\unicode{x2013}$less than $1\%$ of the data typically used in the field. To achieve this, we introduce STRUCTURE, an effective regularization technique that preserves the neighborhood geometry of the latent space of unimodal encoders. Additionally, we show that aligning last layers is often suboptimal and demonstrate the benefits of aligning the layers with the highest representational similarity across modalities. These two components can be readily incorporated into existing alignment methods, yielding substantial gains across 24 zero-shot image classification and retrieval benchmarks, with average relative improvement of $51.6\%$ in classification and $91.8\%$ in retrieval tasks. Our results highlight the effectiveness and broad applicability of our framework for limited-sample multimodal learning and offer a promising path forward for resource-constrained domains.
Authors: Arnaud Verdant, William Guicquero, J\'er\^ome Chossat
Abstract: This paper proposes a deep learning model (RCNet) for Delta-Sigma ($\Delta\Sigma$) ADCs. Recurrent Neural Networks (RNNs) allow to describe both modulators and filters. This analogy is applied to Incremental ADCs (IADC). High-end optimizers combined with full-custom losses are used to define additional hardware design constraints: quantized weights, signal saturation, temporal noise injection, devices area. Focusing on DC conversion, our early results demonstrate that $SNR$ defined as an Effective Number Of Bits (ENOB) can be optimized under a certain hardware mapping complexity. The proposed RCNet succeeded to provide design tradeoffs in terms of $SNR$ ($>$13bit) versus area constraints ($<$14pF total capacitor) at a given $OSR$ (80 samples). Interestingly, it appears that the best RCNet architectures do not necessarily rely on high-order modulators, leveraging additional topology exploration degrees of freedom.
Authors: Daniel Christoph, Max Ploner, Patrick Haller, Alan Akbik
Abstract: Sample efficiency is a crucial property of language models with practical implications for training efficiency. In real-world text, information follows a long-tailed distribution. Yet, we expect models to learn and recall frequent and infrequent facts. Sample-efficient models are better equipped to handle this challenge of learning and retaining rare information without requiring excessive exposure. This study analyzes multiple models of varying architectures and sizes, all trained on the same pre-training data. By annotating relational facts with their frequencies in the training corpus, we examine how model performance varies with fact frequency. Our findings show that most models perform similarly on high-frequency facts but differ notably on low-frequency facts. This analysis provides new insights into the relationship between model architecture, size, and factual learning efficiency.
Authors: Dhananjeyan Jeyaraj, Hamidreza Eivazi, Jendrik-Alexander Tr\"oger, Stefan Wittek, Stefan Hartmann, Andreas Rausch
Abstract: The behavior of materials is influenced by a wide range of phenomena occurring across various time and length scales. To better understand the impact of microstructure on macroscopic response, multiscale modeling strategies are essential. Numerical methods, such as the $\text{FE}^2$ approach, account for micro-macro interactions to predict the global response in a concurrent manner. However, these methods are computationally intensive due to the repeated evaluations of the microscale. This challenge has led to the integration of deep learning techniques into computational homogenization frameworks to accelerate multiscale simulations. In this work, we employ neural operators to predict the microscale physics, resulting in a hybrid model that combines data-driven and physics-based approaches. This allows for physics-guided learning and provides flexibility for different materials and spatial discretizations. We apply this method to time-dependent solid mechanics problems involving viscoelastic material behavior, where the state is represented by internal variables only at the microscale. The constitutive relations of the microscale are incorporated into the model architecture and the internal variables are computed based on established physical principles. The results for homogenized stresses ($<6\%$ error) show that the approach is computationally efficient ($\sim 100 \times$ faster).
Authors: Sebastian Nagies, Emiliano Tolotti, Davide Pastorello, Enrico Blanzieri
Abstract: Parameterized quantum circuits represent promising architectures for machine learning applications, yet many lack clear connections to classical models, potentially limiting their ability to translate the wide success of classical neural networks to the quantum realm. We examine a specific type of quantum neural network (QNN) built exclusively from SWAP test circuits, and discuss its mathematical equivalence to a classical two-layer feedforward network with quadratic activation functions under amplitude encoding. Our analysis across classical real-world and synthetic datasets reveals that while this architecture can successfully learn many practical tasks, it exhibits fundamental expressivity limitations due to violating the universal approximation theorem, particularly failing on harder problems like the parity check function. To address this limitation, we introduce a circuit modification using generalized SWAP test circuits that effectively implements classical neural networks with product layers. This enhancement enables successful learning of parity check functions in arbitrary dimensions which we analytically argue to be impossible for the original architecture beyond two dimensions regardless of network size. Our results establish a framework for enhancing QNN expressivity through classical task analysis and demonstrate that our SWAP test-based architecture offers broad representational capacity, suggesting potential promise also for quantum learning tasks.
Authors: Fanfei Li, Thomas Klein, Wieland Brendel, Robert Geirhos, Roland S. Zimmermann
Abstract: Out-of-distribution (OOD) robustness is a desired property of computer vision models. Improving model robustness requires high-quality signals from robustness benchmarks to quantify progress. While various benchmark datasets such as ImageNet-C were proposed in the ImageNet era, most ImageNet-C corruption types are no longer OOD relative to today's large, web-scraped datasets, which already contain common corruptions such as blur or JPEG compression artifacts. Consequently, these benchmarks are no longer well-suited for evaluating OOD robustness in the era of web-scale datasets. Indeed, recent models show saturating scores on ImageNet-era OOD benchmarks, indicating that it is unclear whether models trained on web-scale datasets truly become better at OOD generalization or whether they have simply been exposed to the test distortions during training. To address this, we introduce LAION-C as a benchmark alternative for ImageNet-C. LAION-C consists of six novel distortion types specifically designed to be OOD, even for web-scale datasets such as LAION. In a comprehensive evaluation of state-of-the-art models, we find that the LAION-C dataset poses significant challenges to contemporary models, including MLLMs such as Gemini and GPT-4o. We additionally conducted a psychophysical experiment to evaluate the difficulty of our corruptions for human observers, enabling a comparison of models to lab-quality human robustness data. We observe a paradigm shift in OOD generalization: from humans outperforming models, to the best models now matching or outperforming the best human observers.
Authors: Antonin Berthon, Mihaela van der Schaar
Abstract: Accurately assessing student knowledge is critical for effective education, yet traditional Knowledge Tracing (KT) methods rely on opaque latent embeddings, limiting interpretability. Even LLM-based approaches generate direct predictions or summaries that may hallucinate without any accuracy guarantees. We recast KT as an inverse problem: learning the minimum natural-language summary that makes past answers explainable and future answers predictable. Our Language Bottleneck Model (LBM) consists of an encoder LLM that writes an interpretable knowledge summary and a frozen decoder LLM that must reconstruct and predict student responses using only that summary text. By constraining all predictive information to pass through a short natural-language bottleneck, LBMs ensure that the summary contains accurate information while remaining human-interpretable. Experiments on synthetic arithmetic benchmarks and the large-scale Eedi dataset show that LBMs rival the accuracy of state-of-the-art KT and direct LLM methods while requiring orders-of-magnitude fewer student trajectories. We demonstrate that training the encoder with group-relative policy optimization, using downstream decoding accuracy as a reward signal, effectively improves summary quality.
Authors: Yasir Ali Farrukh, Syed Wali, Irfan Khan, Nathaniel D. Bastian
Abstract: Unsupervised Domain Adaptation (UDA) is a critical challenge in real-world vision systems, especially in resource-constrained environments like drones, where memory and computation are limited. Existing prompt-driven UDA methods typically rely on large vision-language models and require full access to source-domain data during adaptation, limiting their applicability. In this work, we propose Prmpt2Adpt, a lightweight and efficient zero-shot domain adaptation framework built around a teacher-student paradigm guided by prompt-based feature alignment. At the core of our method is a distilled and fine-tuned CLIP model, used as the frozen backbone of a Faster R-CNN teacher. A small set of low-level source features is aligned to the target domain semantics-specified only through a natural language prompt-via Prompt-driven Instance Normalization (PIN). These semantically steered features are used to briefly fine-tune the detection head of the teacher model. The adapted teacher then generates high-quality pseudo-labels, which guide the on-the-fly adaptation of a compact student model. Experiments on the MDS-A dataset demonstrate that Prmpt2Adpt achieves competitive detection performance compared to state-of-the-art methods, while delivering up to 7x faster adaptation and 5x faster inference speed using few source images-making it a practical and scalable solution for real-time adaptation in low-resource domains.
Authors: Dominic Schuh, Janik Kreit, Evan Berkowitz, Lena Funcke, Thomas Luu, Kim A. Nicoli, Marcel Rodekamp
Abstract: We present the first proof of principle that normalizing flows can accurately learn the Boltzmann distribution of the fermionic Hubbard model - a key framework for describing the electronic structure of graphene and related materials. State-of-the-art methods like Hybrid Monte Carlo often suffer from ergodicity issues near the time-continuum limit, leading to biased estimates. Leveraging symmetry-aware architectures as well as independent and identically distributed sampling, our approach resolves these issues and achieves significant speed-ups over traditional methods.
Authors: Davide Frizzo, Francesco Borsatti, Gian Antonio Susto
Abstract: Predictive Maintenance (PdM) is pivotal in Industry 4.0 and 5.0, proactively enhancing efficiency through accurate equipment Remaining Useful Life (RUL) prediction, thus optimizing maintenance scheduling and reducing unexpected failures and premature interventions. This paper introduces a novel RUL estimation approach leveraging State Space Models (SSM) for efficient long-term sequence modeling. To handle model uncertainty, Simoultaneous Quantile Regression (SQR) is integrated into the SSM, enabling multiple quantile estimations. The proposed method is benchmarked against traditional sequence modelling techniques (LSTM, Transformer, Informer) using the C-MAPSS dataset. Results demonstrate superior accuracy and computational efficiency of SSM models, underscoring their potential for high-stakes industrial applications.
Authors: Sina Aghaee Dabaghan Fard, Minhee Kim, Akash Deep, Jaesung Lee
Abstract: Modern industrial systems are often subject to multiple failure modes, and their conditions are monitored by multiple sensors, generating multiple time-series signals. Additionally, time-to-failure data are commonly available. Accurately predicting a system's remaining useful life (RUL) requires effectively leveraging multi-sensor time-series data alongside multi-mode failure event data. In most existing models, failure modes and RUL prediction are performed independently, ignoring the inherent relationship between these two tasks. Some models integrate multiple failure modes and event prediction using black-box machine learning approaches, which lack statistical rigor and cannot characterize the inherent uncertainty in the model and data. This paper introduces a unified approach to jointly model the multi-sensor time-series data and failure time concerning multiple failure modes. This proposed model integrate a Cox proportional hazards model, a Convolved Multi-output Gaussian Process, and multinomial failure mode distributions in a hierarchical Bayesian framework with corresponding priors, enabling accurate prediction with robust uncertainty quantification. Posterior distributions are effectively obtained by Variational Bayes, and prediction is performed with Monte Carlo sampling. The advantages of the proposed model is validated through extensive numerical and case studies with jet-engine dataset.
Authors: Xiaolong Wang, Zhaolu Kang, Wangyuxuan Zhai, Xinyue Lou, Yunghwei Lai, Ziyue Wang, Yawen Wang, Kaiyu Huang, Yile Wang, Peng Li, Yang Liu
Abstract: Multimodal Large Language Models (MLLMs) have demonstrated significant advances across numerous vision-language tasks. Due to their strong image-text alignment capability, MLLMs can effectively understand image-text pairs with clear meanings. However, effectively resolving the inherent ambiguities in natural language and visual contexts remains challenging. Existing multimodal benchmarks typically overlook linguistic and visual ambiguities, relying mainly on unimodal context for disambiguation and thus failing to exploit the mutual clarification potential between modalities. To bridge this gap, we introduce MUCAR, a novel and challenging benchmark designed explicitly for evaluating multimodal ambiguity resolution across multilingual and cross-modal scenarios. MUCAR includes: (1) a multilingual dataset where ambiguous textual expressions are uniquely resolved by corresponding visual contexts, and (2) a dual-ambiguity dataset that systematically pairs ambiguous images with ambiguous textual contexts, with each combination carefully constructed to yield a single, clear interpretation through mutual disambiguation. Extensive evaluations involving 19 state-of-the-art multimodal models--encompassing both open-source and proprietary architectures--reveal substantial gaps compared to human-level performance, highlighting the need for future research into more sophisticated cross-modal ambiguity comprehension methods, further pushing the boundaries of multimodal reasoning.
Authors: Charilaos Papaioannou, Emmanouil Benetos, Alexandros Potamianos
Abstract: Foundation models have revolutionized music information retrieval, but questions remain about their ability to generalize across diverse musical traditions. This paper presents a comprehensive evaluation of five state-of-the-art audio foundation models across six musical corpora spanning Western popular, Greek, Turkish, and Indian classical traditions. We employ three complementary methodologies to investigate these models' cross-cultural capabilities: probing to assess inherent representations, targeted supervised fine-tuning of 1-2 layers, and multi-label few-shot learning for low-resource scenarios. Our analysis shows varying cross-cultural generalization, with larger models typically outperforming on non-Western music, though results decline for culturally distant traditions. Notably, our approaches achieve state-of-the-art performance on five out of six evaluated datasets, demonstrating the effectiveness of foundation models for world music understanding. We also find that our targeted fine-tuning approach does not consistently outperform probing across all settings, suggesting foundation models already encode substantial musical knowledge. Our evaluation framework and benchmarking results contribute to understanding how far current models are from achieving universal music representations while establishing metrics for future progress.
Authors: Samer Lahoud, Kinda Khawam
Abstract: The exponential growth of IoT devices presents critical challenges in bandwidth-constrained wireless networks, particularly regarding efficient data transmission and privacy preservation. This paper presents a novel federated semantic communication (SC) framework that enables collaborative training of bandwidth-efficient models for image reconstruction across heterogeneous IoT devices. By leveraging SC principles to transmit only semantic features, our approach dramatically reduces communication overhead while preserving reconstruction quality. We address the fundamental challenge of client selection in federated learning environments where devices exhibit significant disparities in dataset sizes and data distributions. Our framework implements three distinct client selection strategies that explore different trade-offs between system performance and fairness in resource allocation. The system employs an end-to-end SC architecture with semantic bottlenecks, coupled with a loss-based aggregation mechanism that naturally adapts to client heterogeneity. Experimental evaluation on image data demonstrates that while Utilitarian selection achieves the highest reconstruction quality, Proportional Fairness maintains competitive performance while significantly reducing participation inequality and improving computational efficiency. These results establish that federated SC can successfully balance reconstruction quality, resource efficiency, and fairness in heterogeneous IoT deployments, paving the way for sustainable and privacy-preserving edge intelligence applications.
Authors: Aditya Sengar, Ali Hariri, Daniel Probst, Patrick Barth, Pierre Vandergheynst
Abstract: Generating diverse, all-atom conformational ensembles of dynamic proteins such as G-protein-coupled receptors (GPCRs) is critical for understanding their function, yet most generative models simplify atomic detail or ignore conformational diversity altogether. We present latent diffusion for full protein generation (LD-FPG), a framework that constructs complete all-atom protein structures, including every side-chain heavy atom, directly from molecular dynamics (MD) trajectories. LD-FPG employs a Chebyshev graph neural network (ChebNet) to obtain low-dimensional latent embeddings of protein conformations, which are processed using three pooling strategies: blind, sequential and residue-based. A diffusion model trained on these latent representations generates new samples that a decoder, optionally regularized by dihedral-angle losses, maps back to Cartesian coordinates. Using D2R-MD, a 2-microsecond MD trajectory (12 000 frames) of the human dopamine D2 receptor in a membrane environment, the sequential and residue-based pooling strategy reproduces the reference ensemble with high structural fidelity (all-atom lDDT of approximately 0.7; C-alpha-lDDT of approximately 0.8) and recovers backbone and side-chain dihedral-angle distributions with a Jensen-Shannon divergence of less than 0.03 compared to the MD data. LD-FPG thereby offers a practical route to system-specific, all-atom ensemble generation for large proteins, providing a promising tool for structure-based therapeutic design on complex, dynamic targets. The D2R-MD dataset and our implementation are freely available to facilitate further research.
Authors: Zhuo Xu, Tianyue Zheng, Linglong Dai
Abstract: The low-altitude economy (LAE) is gaining significant attention from academia and industry. Fortunately, LAE naturally aligns with near-field communications in extremely large-scale MIMO (XL-MIMO) systems. By leveraging near-field beamfocusing, LAE can precisely direct beam energy to unmanned aerial vehicles, while the additional distance dimension boosts overall spectrum efficiency. However, near-field communications in LAE still face several challenges, such as the increase in signal processing complexity and the necessity of distinguishing between far and near-field users. Inspired by the large language models (LLM) with powerful ability to handle complex problems, we apply LLM to solve challenges of near-field communications in LAE. The objective of this article is to provide a comprehensive analysis and discussion on LLM-empowered near-field communications in LAE. Specifically, we first introduce fundamentals of LLM and near-field communications, including the key advantages of LLM and key characteristics of near-field communications. Then, we reveal the opportunities and challenges of near-field communications in LAE. To address these challenges, we present a LLM-based scheme for near-field communications in LAE, and provide a case study which jointly distinguishes far and near-field users and designs multi-user precoding matrix. Finally, we outline and highlight several future research directions and open issues.
Authors: Ziv Aharoni, Henry D. Pfister
Abstract: Synchronization errors, such as insertions and deletions, present a fundamental challenge in DNA-based data storage systems, arising from both synthesis and sequencing noise. These channels are often modeled as insertion-deletion-substitution (IDS) channels, for which designing maximum-likelihood decoders is computationally expensive. In this work, we propose a data-driven approach based on neural polar decoders (NPDs) to design low-complexity decoders for channels with synchronization errors. The proposed architecture enables decoding over IDS channels with reduced complexity $O(AN log N )$, where $A$ is a tunable parameter independent of the channel. NPDs require only sample access to the channel and can be trained without an explicit channel model. Additionally, NPDs provide mutual information (MI) estimates that can be used to optimize input distributions and code design. We demonstrate the effectiveness of NPDs on both synthetic deletion and IDS channels. For deletion channels, we show that NPDs achieve near-optimal decoding performance and accurate MI estimation, with significantly lower complexity than trellis-based decoders. We also provide numerical estimates of the channel capacity for the deletion channel. We extend our evaluation to realistic DNA storage settings, including channels with multiple noisy reads and real-world Nanopore sequencing data. Our results show that NPDs match or surpass the performance of existing methods while using significantly fewer parameters than the state-of-the-art. These findings highlight the promise of NPDs for robust and efficient decoding in DNA data storage systems.
Authors: Josu\'e Mart\'inez-Mart\'inez, Olivia Brown, Mostafa Karami, Sheida Nabavi
Abstract: Deep neural networks are increasingly being used to detect and diagnose medical conditions using medical imaging. Despite their utility, these models are highly vulnerable to adversarial attacks and distribution shifts, which can affect diagnostic reliability and undermine trust among healthcare professionals. In this study, we propose a robust training algorithm with data augmentation (RTDA) to mitigate these vulnerabilities in medical image classification. We benchmark classifier robustness against adversarial perturbations and natural variations of RTDA and six competing baseline techniques, including adversarial training and data augmentation approaches in isolation and combination, using experimental data sets with three different imaging technologies (mammograms, X-rays, and ultrasound). We demonstrate that RTDA achieves superior robustness against adversarial attacks and improved generalization performance in the presence of distribution shift in each image classification task while maintaining high clean accuracy.
Authors: Ritabrata Chakraborty, Rajatsubhra Chakraborty, Avijit Dasgupta, Sandeep Chaurasia
Abstract: Traditional video-based tasks like soccer action spotting rely heavily on visual inputs, often requiring complex and computationally expensive models to process dense video data. In this work, we propose a shift from this video-centric approach to a text-based task, making it lightweight and scalable by utilizing Large Language Models (LLMs) instead of Vision-Language Models (VLMs). We posit that expert commentary, which provides rich, fine-grained descriptions and contextual cues such as excitement and tactical insights, contains enough information to reliably spot key actions in a match. To demonstrate this, we use the SoccerNet Echoes dataset, which provides timestamped commentary, and employ a system of three LLMs acting as judges specializing in outcome, excitement, and tactics. Each LLM evaluates sliding windows of commentary to identify actions like goals, cards, and substitutions, generating accurate timestamps for these events. Our experiments show that this language-centric approach performs effectively in detecting critical match events, providing a lightweight and training-free alternative to traditional video-based methods for action spotting.
Authors: Samuel Howard, Peter Potaptchik, George Deligiannidis
Abstract: Recent advances in flow-based generative modelling have provided scalable methods for computing the Schr\"odinger Bridge (SB) between distributions, a dynamic form of entropy-regularised Optimal Transport (OT) for the quadratic cost. The successful Iterative Markovian Fitting (IMF) procedure solves the SB problem via sequential bridge-matching steps, presenting an elegant and practical approach with many favourable properties over the more traditional Iterative Proportional Fitting (IPF) procedure. Beyond the standard setting, optimal transport can be generalised to the multi-marginal case in which the objective is to minimise a cost defined over several marginal distributions. Of particular importance are costs defined over a tree structure, from which Wasserstein barycentres can be recovered as a special case. In this work, we extend the IMF procedure to solve for the tree-structured SB problem. Our resulting algorithm inherits the many advantages of IMF over IPF approaches in the tree-based setting. In the specific case of Wasserstein barycentres, our approach can be viewed as extending fixed-point approaches for barycentre computation to the case of flow-based entropic OT solvers.
Authors: Yukun Huang, Yanning Zhou, Jianan Wang, Kaiyi Huang, Xihui Liu
Abstract: 3D panorama synthesis is a promising yet challenging task that demands high-quality and diverse visual appearance and geometry of the generated omnidirectional content. Existing methods leverage rich image priors from pre-trained 2D foundation models to circumvent the scarcity of 3D panoramic data, but the incompatibility between 3D panoramas and 2D single views limits their effectiveness. In this work, we demonstrate that by applying multi-plane synchronization to the operators from 2D foundation models, their capabilities can be seamlessly extended to the omnidirectional domain. Based on this design, we further introduce DreamCube, a multi-plane RGB-D diffusion model for 3D panorama generation, which maximizes the reuse of 2D foundation model priors to achieve diverse appearances and accurate geometry while maintaining multi-view consistency. Extensive experiments demonstrate the effectiveness of our approach in panoramic image generation, panoramic depth estimation, and 3D scene generation.
Authors: Tianjiao Yu, Vedant Shah, Muntasir Wahed, Ying Shen, Kiet A. Nguyen, Ismini Lourentzou
Abstract: Articulated objects are common in the real world, yet modeling their structure and motion remains a challenging task for 3D reconstruction methods. In this work, we introduce Part$^{2}$GS, a novel framework for modeling articulated digital twins of multi-part objects with high-fidelity geometry and physically consistent articulation. Part$^{2}$GS leverages a part-aware 3D Gaussian representation that encodes articulated components with learnable attributes, enabling structured, disentangled transformations that preserve high-fidelity geometry. To ensure physically consistent motion, we propose a motion-aware canonical representation guided by physics-based constraints, including contact enforcement, velocity consistency, and vector-field alignment. Furthermore, we introduce a field of repel points to prevent part collisions and maintain stable articulation paths, significantly improving motion coherence over baselines. Extensive evaluations on both synthetic and real-world datasets show that Part$^{2}$GS consistently outperforms state-of-the-art methods by up to 10$\times$ in Chamfer Distance for movable parts.
Authors: Sandor Szedmak (Department of Computer Science, Aalto University), Eric Bach (Department of Computer Science, Aalto University)
Abstract: The Tanimoto kernel (Jaccard index) is a well known tool to describe the similarity between sets of binary attributes. It has been extended to the case when the attributes are nonnegative real values. This paper introduces a more general Tanimoto kernel formulation which allows to measure the similarity of arbitrary real-valued functions. This extension is constructed by unifying the representation of the attributes via properly chosen sets. After deriving the general form of the kernel, explicit feature representation is extracted from the kernel function, and a simply way of including general kernels into the Tanimoto kernel is shown. Finally, the kernel is also expressed as a quotient of piecewise linear functions, and a smooth approximation is provided.
Authors: Ilyas Fatkhullin, Igor Sokolov, Eduard Gorbunov, Zhize Li, Peter Richt\'arik
Abstract: First proposed by Seide (2014) as a heuristic, error feedback (EF) is a very popular mechanism for enforcing convergence of distributed gradient-based optimization methods enhanced with communication compression strategies based on the application of contractive compression operators. However, existing theory of EF relies on very strong assumptions (e.g., bounded gradients), and provides pessimistic convergence rates (e.g., while the best known rate for EF in the smooth nonconvex regime, and when full gradients are compressed, is $O(1/T^{2/3})$, the rate of gradient descent in the same regime is $O(1/T)$). Recently, Richt\'arik et al. (2021) proposed a new error feedback mechanism, EF21, based on the construction of a Markov compressor induced by a contractive compressor. EF21 removes the aforementioned theoretical deficiencies of EF and at the same time works better in practice. In this work we propose six practical extensions of EF21, all supported by strong convergence theory: partial participation, stochastic approximation, variance reduction, proximal setting, momentum, and bidirectional compression. To the best of our knowledge, several of these techniques have not been previously analyzed in combination with EF, and in cases where prior analysis exists -- such as for bidirectional compression -- our theoretical convergence guarantees significantly improve upon existing results.
Authors: Leander Weber, Jim Berend, Moritz Weckbecker, Alexander Binder, Thomas Wiegand, Wojciech Samek, Sebastian Lapuschkin
Abstract: Gradient-based optimization has been a cornerstone of machine learning that enabled the vast advances of Artificial Intelligence (AI) development over the past decades. However, this type of optimization requires differentiation, and with recent evidence of the benefits of non-differentiable (e.g. neuromorphic) architectures over classical models w.r.t. efficiency, such constraints can become limiting in the future. We present Layer-wise Feedback Propagation (LFP), a novel training principle for neural network-like predictors that utilizes methods from the domain of explainability to decompose a reward to individual neurons based on their respective contributions. Leveraging these neuron-wise rewards, our method then implements a greedy approach reinforcing helpful parts of the network and weakening harmful ones. While having comparable computational complexity to gradient descent, LFP does not require gradient computation and generates sparse and thereby memory- and energy-efficient parameter updates and models. We establish the convergence of LFP theoretically and empirically, demonstrating its effectiveness on various models and datasets. Via two applications - neural network pruning and the approximation-free training of Spiking Neural Networks (SNNs) - we demonstrate that LFP combines increased efficiency in terms of computation and representation with flexibility w.r.t. choice of model architecture and objective function. Our code is available at https://github.com/leanderweber/layerwise-feedback-propagation.
URLs: https://github.com/leanderweber/layerwise-feedback-propagation.
Authors: Iulia Duta, Giulia Cassar\`a, Fabrizio Silvestri, Pietro Li\`o
Abstract: Higher-order relations are widespread in nature, with numerous phenomena involving complex interactions that extend beyond simple pairwise connections. As a result, advancements in higher-order processing can accelerate the growth of various fields requiring structured data. Current approaches typically represent these interactions using hypergraphs. We enhance this representation by introducing cellular sheaves for hypergraphs, a mathematical construction that adds extra structure to the conventional hypergraph while maintaining their local, higherorder connectivity. Drawing inspiration from existing Laplacians in the literature, we develop two unique formulations of sheaf hypergraph Laplacians: linear and non-linear. Our theoretical analysis demonstrates that incorporating sheaves into the hypergraph Laplacian provides a more expressive inductive bias than standard hypergraph diffusion, creating a powerful instrument for effectively modelling complex data structures. We employ these sheaf hypergraph Laplacians to design two categories of models: Sheaf Hypergraph Neural Networks and Sheaf Hypergraph Convolutional Networks. These models generalize classical Hypergraph Networks often found in the literature. Through extensive experimentation, we show that this generalization significantly improves performance, achieving top results on multiple benchmark datasets for hypergraph node classification.
Authors: Qizhou Wang, Guansong Pang, Mahsa Salehi, Xiaokun Xia, Christopher Leckie
Abstract: This paper considers an important Graph Anomaly Detection (GAD) task, namely open-set GAD, which aims to train a detection model using a small number of normal and anomaly nodes (referred to as seen anomalies) to detect both seen anomalies and unseen anomalies (i.e., anomalies that cannot be illustrated the training anomalies). Those labelled training data provide crucial prior knowledge about abnormalities for GAD models, enabling substantially reduced detection errors. However, current supervised GAD methods tend to over-emphasise fitting the seen anomalies, leading to many errors of detecting the unseen anomalies as normal nodes. Further, existing open-set AD models were introduced to handle Euclidean data, failing to effectively capture discriminative features from graph structure and node attributes for GAD. In this work, we propose a novel open-set GAD approach, namely normal structure regularisation (NSReg), to achieve generalised detection ability to unseen anomalies, while maintaining its effectiveness on detecting seen anomalies. The key idea in NSReg is to introduce a regularisation term that enforces the learning of compact, semantically-rich representations of normal nodes based on their structural relations to other nodes. When being optimised with supervised anomaly detection losses, the regularisation term helps incorporate strong normality into the modelling, and thus, it effectively avoids over-fitting the seen anomalies and learns a better normality decision boundary, largely reducing the false negatives of detecting unseen anomalies as normal. Extensive empirical results on seven real-world datasets show that NSReg significantly outperforms state-of-the-art competing methods by at least 14% AUC-ROC on the unseen anomaly classes and by 10% AUC-ROC on all anomaly classes.
Authors: Yehya Farhat
Abstract: Decision-Focused Learning (DFL) is an emerging learning paradigm that tackles the task of training a machine learning (ML) model to predict missing parameters of an incomplete optimization problem, where the missing parameters are predicted. DFL trains an ML model in an end-to-end system, by integrating the prediction and optimization tasks, providing better alignment of the training and testing objectives. DFL has shown a lot of promise and holds the capacity to revolutionize decision-making in many real-world applications. However, very little is known about the performance of these models under adversarial attacks. We adopt ten unique DFL methods and benchmark their performance under two distinctly focused attacks adapted towards the Predict-then-Optimize problem setting. Our study proposes the hypothesis that the robustness of a model is highly correlated with its ability to find predictions that lead to optimal decisions without deviating from the ground-truth label. Furthermore, we provide insight into how to target the models that violate this condition and show how these models respond differently depending on the achieved optimality at the end of their training cycles.
Authors: Masahiro Kato, Kyohei Okumura, Takuya Ishihara, Toru Kitagawa
Abstract: This study investigates the contextual best arm identification (BAI) problem, aiming to design an adaptive experiment to identify the best treatment arm conditioned on contextual information (covariates). We consider a decision-maker who assigns treatment arms to experimental units during an experiment and recommends the estimated best treatment arm based on the contexts at the end of the experiment. The decision-maker uses a policy for recommendations, which is a function that provides the estimated best treatment arm given the contexts. In our evaluation, we focus on the worst-case expected regret, a relative measure between the expected outcomes of an optimal policy and our proposed policy. We derive a lower bound for the expected simple regret and then propose a strategy called Adaptive Sampling-Policy Learning (PLAS). We prove that this strategy is minimax rate-optimal in the sense that its leading factor in the regret upper bound matches the lower bound as the number of experimental units increases.
Authors: Moran Yanuka, Morris Alper, Hadar Averbuch-Elor, Raja Giryes
Abstract: Web-scale training on paired text-image data is becoming increasingly central to multimodal learning, but is challenged by the highly noisy nature of datasets in the wild. Standard data filtering approaches succeed in removing mismatched text-image pairs, but permit semantically related but highly abstract or subjective text. These approaches lack the fine-grained ability to isolate the most concrete samples that provide the strongest signal for learning in a noisy dataset. In this work, we propose a new metric, image caption concreteness, that evaluates caption text without an image reference to measure its concreteness and relevancy for use in multimodal learning. Our approach leverages strong foundation models for measuring visual-semantic information loss in multimodal representations. We demonstrate that this strongly correlates with human evaluation of concreteness in both single-word and sentence-level texts. Moreover, we show that curation using ICC complements existing approaches: It succeeds in selecting the highest quality samples from multimodal web-scale datasets to allow for efficient training in resource-constrained settings.
Authors: Jiyoun Kim, Junu Kim, Kyunghoon Hur, Edward Choi
Abstract: The increasing volume of electronic health records (EHRs) presents the opportunity to improve the accuracy and robustness of models in clinical prediction tasks. Unlike traditional centralized approaches, federated learning enables training on data from multiple institutions while preserving patient privacy and complying with regulatory constraints. In practice, healthcare institutions (i.e., hosts) often need to build predictive models tailored to their specific needs using federated learning. In this scenario, two key challenges arise: (1) ensuring compatibility across heterogeneous EHR systems, and (2) managing federated learning costs within budget constraints. To address these challenges, we propose EHRFL, a federated learning framework designed for building a cost-effective, host-specific predictive model using patient EHR data. EHRFL consists of two components: (1) text-based EHR modeling, which facilitates cross-institution compatibility without costly data standardization, and (2) a participant selection strategy based on averaged patient embedding similarity to reduce the number of participants without degrading performance. Experiments on multiple open-source EHR datasets demonstrate the effectiveness of both components. We believe our framework offers a practical solution for enabling healthcare institutions to build institution-specific predictive models under budgetary constraints.
Authors: Zhenhan Huang, Tejaswini Pedapati, Pin-Yu Chen, Chunheng Jiang, Jianxi Gao
Abstract: Neural architecture search (NAS) enables the automatic design of neural network models. However, training the candidates generated by the search algorithm for performance evaluation incurs considerable computational overhead. Our method, dubbed nasgraph, remarkably reduces the computational costs by converting neural architectures to graphs and using the average degree, a graph measure, as the proxy in lieu of the evaluation metric. Our training-free NAS method is data-agnostic and light-weight. It can find the best architecture among 200 randomly sampled architectures from NAS-Bench201 in 217 CPU seconds. Besides, our method is able to achieve competitive performance on various datasets including NASBench-101, NASBench-201, and NDS search spaces. We also demonstrate that nasgraph generalizes to more challenging tasks on Micro TransNAS-Bench-101.
Authors: Awni Altabaa, John Lafferty
Abstract: Relational reasoning is a central component of generally intelligent systems, enabling robust and data-efficient inductive generalization. Recent empirical evidence shows that many existing neural architectures, including Transformers, struggle with tasks requiring relational reasoning. In this work, we distinguish between two types of information: sensory information about the properties of individual objects, and relational information about the relationships between objects. While neural attention provides a powerful mechanism for controlling the flow of sensory information between objects, the Transformer lacks an explicit computational mechanism for routing and processing relational information. To address this limitation, we propose an architectural extension of the Transformer framework that we call the Dual Attention Transformer (DAT), featuring two distinct attention mechanisms: sensory attention for directing the flow of sensory information, and a novel relational attention mechanism for directing the flow of relational information. We empirically evaluate DAT on a diverse set of tasks ranging from synthetic relational benchmarks to complex real-world tasks such as language modeling and visual processing. Our results demonstrate that integrating explicit relational computational mechanisms into the Transformer architecture leads to significant performance gains in terms of data efficiency and parameter efficiency.
Authors: Saba Ahmadi, Siddharth Bhandari, Avrim Blum, Chen Dan, Prabhav Jain
Abstract: We initiate the study of a new notion of adversarial loss which we call distributional adversarial loss. In this notion, we assume for each original example, the allowed adversarial perturbation set is a family of distributions, and the adversarial loss over each example is the maximum loss over all the associated distributions. The goal is to minimize the overall adversarial loss. We show sample complexity bounds in the PAC-learning setting for our notion of adversarial loss. Our notion of adversarial loss contrasts the prior work on robust learning that considers a set of points, not distributions, as the perturbation set of each clean example. As an application of our approach, we show how to unify the two lines of work on randomized smoothing and robust learning in the PAC-learning setting and derive sample complexity bounds for randomized smoothing methods. Furthermore, we investigate the role of randomness in achieving robustness against adversarial attacks. We show a general derandomization technique that preserves the extent of a randomized classifier's robustness against adversarial attacks and show its effectiveness empirically.
Authors: Zhiyu Shao, Qiong Wu, Pingyi Fan, Nan Cheng, Wen Chen, Jiangzhou Wang, Khaled B. Letaief
Abstract: This work aims to investigate semantic communication in high-speed mobile Internet of vehicles (IoV) environments, with a focus on the spectrum sharing between vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications. We specifically address spectrum scarcity and network traffic and then propose a semantic-aware spectrum sharing algorithm (SSS) based on the deep reinforcement learning (DRL) soft actor-critic (SAC) approach. Firstly, we delve into the extraction of semantic information. Secondly, we redefine metrics for semantic information in V2V and V2I spectrum sharing in IoV environments, introducing high-speed semantic spectrum efficiency (HSSE) and semantic transmission rate (HSR). Finally, we employ the SAC algorithm for decision optimization in V2V and V2I spectrum sharing based on semantic information. This optimization encompasses the optimal link of V2V and V2I sharing strategies, the transmission power for vehicles sending semantic information and the length of transmitted semantic symbols, aiming at maximizing HSSE of V2I and enhancing success rate of effective semantic information transmission (SRS) of V2V. Experimental results demonstrate that the SSS algorithm outperforms other baseline algorithms, including other traditional-communication-based spectrum sharing algorithms and spectrum sharing algorithm using other reinforcement learning approaches. The SSS algorithm exhibits a 15% increase in HSSE and approximately a 7% increase in SRS.
Authors: Kangwei Qi, Qiong Wu, Pingyi Fan, Nan Cheng, Wen Chen, Jiangzhou Wang, Khaled B. Letaief
Abstract: Reconfigurable Intelligent Surface (RIS) is a pivotal technology in communication, offering an alternative path that significantly enhances the link quality in wireless communication environments. In this paper, we propose a RIS-assisted internet of vehicles (IoV) network, considering the vehicle-to-everything (V2X) communication method. In addition, in order to improve the timeliness of vehicle-to-infrastructure (V2I) links and the stability of vehicle-to-vehicle (V2V) links, we introduce the age of information (AoI) model and the payload transmission probability model. Therefore, with the objective of minimizing the AoI of V2I links and prioritizing transmission of V2V links payload, we construct this optimization problem as an Markov decision process (MDP) problem in which the BS serves as an agent to allocate resources and control phase-shift for the vehicles using the soft actor-critic (SAC) algorithm, which gradually converges and maintains a high stability. A AoI-aware joint vehicular resource allocation and RIS phase-shift control scheme based on SAC algorithm is proposed and simulation results show that its convergence speed, cumulative reward, AoI performance, and payload transmission probability outperforms those of proximal policy optimization (PPO), deep deterministic policy gradient (DDPG), twin delayed deep deterministic policy gradient (TD3) and stochastic algorithms.
Authors: Yu Xie, Qiong Wu, Pingyi Fan, Nan Cheng, Wen Chen, Jiangzhou Wang, Khaled B. Letaief
Abstract: As a promising technology, vehicular edge computing (VEC) can provide computing and caching services by deploying VEC servers near vehicles. However, VEC networks still face challenges such as high vehicle mobility. Digital twin (DT), an emerging technology, can predict, estimate, and analyze real-time states by digitally modeling objects in the physical world. By integrating DT with VEC, a virtual vehicle DT can be created in the VEC server to monitor the real-time operating status of vehicles. However, maintaining the vehicle DT model requires ongoing attention from the VEC server, which also needs to offer computing services for the vehicles. Therefore, effective allocation and scheduling of VEC server resources are crucial. This study focuses on a general VEC network with a single VEC service and multiple vehicles, examining the two types of delays caused by twin maintenance and computational processing within the network. By transforming the problem using satisfaction functions, we propose an optimization problem aimed at maximizing each vehicle's resource utility to determine the optimal resource allocation strategy. Given the non-convex nature of the issue, we employ multi-agent Markov decision processes to reformulate the problem. Subsequently, we propose the twin maintenance and computing task processing resource collaborative scheduling (MADRL-CSTC) algorithm, which leverages multi-agent deep reinforcement learning. Through experimental comparisons with alternative algorithms, it demonstrates that our proposed approach is effective in terms of resource allocation.
Authors: Shulin Song, Zheng Zhang, Qiong Wu, Qiang Fan, Pingyi Fan
Abstract: Autonomous driving may be the most important application scenario of next generation, the development of wireless access technologies enabling reliable and low-latency vehicle communication becomes crucial. To address this, 3GPP has developed Vehicle-to-Everything (V2X) specifications based on 5G New Radio (NR) technology, where Mode 2 Side-Link (SL) communication resembles Mode 4 in LTE-V2X, allowing direct communication between vehicles. This supplements SL communication in LTE-V2X and represents the latest advancement in cellular V2X (C-V2X) with improved performance of NR-V2X. However, in NR-V2X Mode 2, resource collisions still occur, and thus degrade the age of information (AOI). Therefore, a interference cancellation method is employed to mitigate this impact by combining NR-V2X with Non-Orthogonal multiple access (NOMA) technology. In NR-V2X, when vehicles select smaller resource reservation interval (RRI), higher-frequency transmissions take ore energy to reduce AoI. Hence, it is important to jointly consider AoI and communication energy consumption based on NR-V2X communication. Then, we formulate such an optimization problem and employ the Deep Reinforcement Learning (DRL) algorithm to compute the optimal transmission RRI and transmission power for each transmitting vehicle to reduce the energy consumption of each transmitting vehicle and the AoI of each receiving vehicle. Extensive simulations have demonstrated the performance of our proposed algorithm.
Authors: Mohammad Zhalechian, Soroush Saghafian, Omar Robles
Abstract: The United States Food and Drug Administration's (FDA's) Premarket Notification 510(k) pathway allows manufacturers to gain approval for a medical device by demonstrating its substantial equivalence to another legally marketed device. However, the inherent ambiguity of this regulatory procedure has led to high recall rates for many devices cleared through this pathway. This trend has raised significant concerns regarding the efficacy of the FDA's current approach, prompting a reassessment of the 510(k) regulatory framework. In this paper, we develop a combined human-algorithm approach to assist the FDA in improving its 510(k) medical device clearance process by reducing the risk of recalls and the workload imposed on the FDA. We first develop machine learning methods to estimate the risk of recall of 510(k) medical devices based on the information available at submission time. We then propose a data-driven clearance policy that recommends acceptance, rejection, or deferral to FDA's committees for in-depth evaluation. We conduct an empirical study using a unique large-scale dataset of over 31,000 medical devices that we assembled based on data sources from the FDA and Centers for Medicare and Medicaid Service (CMS). A conservative evaluation of our proposed policy based on this data shows a 32.9% improvement in the recall rate and a 40.5% reduction in the FDA's workload. Our analyses also indicate that implementing our policy could result in significant annual cost savings of $1.7 billion, which highlights the value of using a holistic and data-driven approach to improve the FDA's current 510(k) medical device evaluation pathway.
Authors: Xueying Gu, Qiong Wu, Pingyi Fan, Qiang Fan
Abstract: Federated Learning (FL) is an advanced distributed machine learning approach, that protects the privacy of each vehicle by allowing the model to be trained on multiple devices simultaneously without the need to upload all data to a road side unit (RSU). This enables FL to handle scenarios with sensitive or widely distributed data. However, in these fields, it is well known that the labeling costs can be a significant expense, and models relying on labels are not suitable for these rapidly evolving fields especially in vehicular networks, or mobile internet of things (MIoT), where new data emerges constantly. To handle this issue, the self-supervised learning paves the way for training without labels. Additionally, for vehicles with high velocity, owing to blurred images, simple aggregation not only impacts the accuracy of the aggregated model but also reduces the convergence speed of FL. This paper proposes a FL algorithm based on image blur level to aggregation, called FLSimCo, which does not require labels and serves as a pre-training stage for self-supervised learning in the vehicular environment. Simulation results demonstrate that the proposed algorithm exhibits fast and stable convergence.
Authors: Xueying Gu, Qiong Wu, Pingyi Fan, Nan Cheng, Wen Chen, Khaled B. Letaief
Abstract: Intelligent Transportation Systems (ITS) leverage Integrated Sensing and Communications (ISAC) to enhance data exchange between vehicles and infrastructure in the Internet of Vehicles (IoV). This integration inevitably increases computing demands, risking real-time system stability. Vehicle Edge Computing (VEC) addresses this by offloading tasks to Road Side Unit (RSU), ensuring timely services. Our previous work FLSimCo algorithm, which uses local resources for Federated Self-Supervised Learning (SSL), though vehicles often can't complete all iterations task. Our improved algorithm offloads partial task to RSU and optimizes energy consumption by adjusting transmission power, CPU frequency, and task assignment ratios, balancing local and RSU-based training. Meanwhile, setting an offloading threshold further prevents inefficiencies. Simulation results show that the enhanced algorithm reduces energy consumption, improves offloading efficiency and the accuracy of Federated SSL.
Authors: Shuo Lu, Yingsheng Wang, Lijun Sheng, Lingxiao He, Aihua Zheng, Jian Liang
Abstract: Out-of-distribution (OOD) detection aims to detect test samples outside the training category space, which is an essential component in building reliable machine learning systems. Existing reviews on OOD detection primarily focus on method taxonomy, surveying the field by categorizing various approaches. However, many recent works concentrate on non-traditional OOD detection scenarios, such as test-time adaptation, multi-modal data sources and other novel contexts. In this survey, we uniquely review recent advances in OOD detection from the task-oriented perspective for the first time. According to the user's access to the model, that is, whether the OOD detection method is allowed to modify or retrain the model, we classify the methods as training-driven or training-agnostic. Besides, considering the rapid development of pre-trained models, large pre-trained model-based OOD detection is also regarded as an important category and discussed separately. Furthermore, we provide a discussion of the evaluation scenarios, a variety of applications, and several future research directions. We believe this survey with new taxonomy will benefit the proposal of new methods and the expansion of more practical scenarios. A curated list of related papers is provided in the Github repository: https://github.com/shuolucs/Awesome-Out-Of-Distribution-Detection.
URLs: https://github.com/shuolucs/Awesome-Out-Of-Distribution-Detection.
Authors: Amey Agrawal, Haoran Qiu, Junda Chen, \'I\~nigo Goiri, Chaojie Zhang, Rayyan Shahid, Ramachandran Ramjee, Alexey Tumanov, Esha Choukse
Abstract: As large language models (LLMs) handle increasingly longer contexts, serving long inference requests of millions of tokens presents unique challenges. We show that existing work for long context inference is largely based on techniques from long context training, and does not handle the high variability in input lengths during inference. This leads to inefficient resource utilization, server fragmentation, and head-of-line (HOL) blocking. We present Medha, an end-to-end system for efficient long-context LLM inference that addresses these challenges through fine-grained time sharing. Medha introduces three key innovations: (1) the mechanism of adaptive prefill chunking to help mitigate HOL blocking with preemption; (2) two new parallelism strategies: Sequence Pipeline Parallelism (SPP) to reduce time-to-first-token by pipelining prefill chunks, and KV-Cache Parallelism (KVP) to lower time-peroutput-token by distributing decoding across servers; and (3) a novel input-length aware least remaining slack scheduling to meet Service Level Objectives (SLOs). Medha enables exact inference scaling beyond 10 million tokens, maintaining high throughput and low latency across mixed-length workloads. Compared to state-of-the-art systems, Medha reduces server fragmentation, cuts median latency by up to 30x, and improves throughput by over 5x, delivering production-scale long-context inference without compromising performance on shorter requests.
Authors: Sergei Kholkin, Grigoriy Ksenofontov, David Li, Nikita Kornilov, Nikita Gushchin, Alexandra Suvorikova, Alexey Kroshnin, Evgeny Burnaev, Alexander Korotin
Abstract: The Iterative Markovian Fitting (IMF) procedure, which iteratively projects onto the space of Markov processes and the reciprocal class, successfully solves the Schr\"odinger Bridge (SB) problem. However, an efficient practical implementation requires a heuristic modification - alternating between fitting forward and backward time diffusion at each iteration. This modification is crucial for stabilizing training and achieving reliable results in applications such as unpaired domain translation. Our work reveals a close connection between the modified version of IMF and the Iterative Proportional Fitting (IPF) procedure - a foundational method for the SB problem, also known as Sinkhorn's algorithm. Specifically, we demonstrate that the heuristic modification of the IMF effectively integrates both IMF and IPF procedures. We refer to this combined approach as the Iterative Proportional Markovian Fitting (IPMF) procedure. Through theoretical and empirical analysis, we establish the convergence of IPMF procedure under various settings, contributing to developing a unified framework for solving SB problems. Moreover, from a practical standpoint, the IPMF procedure enables a flexible trade-off between image similarity and generation quality, offering a new mechanism for tailoring models to specific tasks.
Authors: Yinuo Ren, Tesi Xiao, Michael Shavlovsky, Lexing Ying, Holakou Rahmanian
Abstract: In LLM alignment and many other ML applications, one often faces the Multi-Objective Fine-Tuning (MOFT) problem, i.e., fine-tuning an existing model with datasets labeled w.r.t. different objectives simultaneously. To address the challenge, we propose a Conditioned One-Shot fine-tuning framework (COS-DPO) that extends the Direct Preference Optimization technique, originally developed for efficient LLM alignment with preference data, to accommodate the MOFT settings. By direct conditioning on the weight across auxiliary objectives, our Weight-COS-DPO method enjoys an efficient one-shot training process for profiling the Pareto front and is capable of achieving comprehensive trade-off solutions even in the post-training stage. Based on our theoretical findings on the linear transformation properties of the loss function, we further propose the Temperature-COS-DPO method that augments the temperature parameter to the model input, enhancing the flexibility of post-training control over the trade-offs between the main and auxiliary objectives. We demonstrate the effectiveness and efficiency of the COS-DPO framework through its applications to various tasks, including the Learning-to-Rank (LTR) and LLM alignment tasks, highlighting its viability for large-scale ML deployments.
Authors: Peter Shaw, James Cohan, Jacob Eisenstein, Kenton Lee, Jonathan Berant, Kristina Toutanova
Abstract: We propose a new programming language called ALTA and a compiler that can map ALTA programs to Transformer weights. ALTA is inspired by RASP, a language proposed by Weiss et al. (2021), and Tracr (Lindner et al., 2023), a compiler from RASP programs to Transformer weights. ALTA complements and extends this prior work, offering the ability to express loops and to compile programs to Universal Transformers, among other advantages. ALTA allows us to constructively show how Transformers can represent length-invariant algorithms for computing parity and addition, as well as a solution to the SCAN benchmark of compositional generalization tasks, without requiring intermediate scratchpad decoding steps. We also propose tools to analyze cases where the expressibility of an algorithm is established, but end-to-end training on a given training set fails to induce behavior consistent with the desired algorithm. To this end, we explore training from ALTA execution traces as a more fine-grained supervision signal. This enables additional experiments and theoretical analyses relating the learnability of various algorithms to data availability and modeling decisions, such as positional encodings. We make the ALTA framework -- language specification, symbolic interpreter, and weight compiler -- available to the community to enable further applications and insights.
Authors: Zeyuan Li, Yangfan He, Lewei He, Jianhui Wang, Tianyu Shi, Bin Lei, Yuchen Li, Qiuwu Chen
Abstract: Recently, large language models (LLMs) have achieved significant progress in automated code generation. Despite their strong instruction-following capabilities, these models frequently struggled to align with user intent in coding scenarios. In particular, they were hampered by datasets that lacked diversity and failed to address specialized tasks or edge cases. Furthermore, challenges in supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) led to failures in generating precise, human-intent-aligned code. To tackle these challenges and improve the code generation performance for automated programming systems, we propose Feedback-driven Adaptive Long/short-term memory reinforced Coding Optimization (i.e., FALCON). FALCON is structured into two hierarchical levels. From the global level, long-term memory improves code quality by retaining and applying learned knowledge. At the local level, short-term memory allows for the incorporation of immediate feedback from compilers and AI systems. Additionally, we introduce meta-reinforcement learning with feedback rewards to solve the global-local bi-level optimization problem and enhance the model's adaptability across diverse code generation tasks. Extensive experiments demonstrate that our technique achieves state-of-the-art performance, leading other reinforcement learning methods by more than 4.5 percentage points on the MBPP benchmark and 6.1 percentage points on the Humaneval benchmark. The open-sourced code is publicly available at https://github.com/titurte/FALCON.
Authors: Wenhan Chang, Tianqing Zhu, Ping Xiong, Yufeng Wu, Faqian Guan, Wanlei Zhou
Abstract: In the rapid advancement of artificial intelligence, privacy protection has become crucial, giving rise to machine unlearning. Machine unlearning is a technique that removes specific data influences from trained models without the need for extensive retraining. However, it faces several key challenges, including accurately implementing unlearning, ensuring privacy protection during the unlearning process, and achieving effective unlearning without significantly compromising model performance. This paper presents a novel approach to machine unlearning by employing Layer-wise Relevance Analysis and Neuronal Path Perturbation. We address three primary challenges: the lack of detailed unlearning principles, privacy guarantees in zero-shot unlearning scenario, and the balance between unlearning effectiveness and model utility. Our method balances machine unlearning performance and model utility by identifying and perturbing highly relevant neurons, thereby achieving effective unlearning. By using data not present in the original training set during the unlearning process, we satisfy the zero-shot unlearning scenario and ensure robust privacy protection. Experimental results demonstrate that our approach effectively removes targeted data from the target unlearning model while maintaining the model's utility, offering a practical solution for privacy-preserving machine learning.
Authors: Bohan Lyu, Yadi Cao, Duncan Watson-Parris, Leon Bergen, Taylor Berg-Kirkpatrick, Rose Yu
Abstract: Large Language Models (LLMs) demonstrate promising capabilities in solving scientific problems but often suffer from the issue of hallucination. While integrating LLMs with tools can mitigate this issue, models fine-tuned on tool usage become overreliant on them and incur unnecessary costs. Inspired by how human experts assess problem complexity before selecting solutions, we propose a novel two-component fine-tuning method, Adapting While Learning (AWL). In the first component, World Knowledge Learning (WKL), LLMs internalize scientific knowledge by learning from tool-generated solutions. In the second component, Tool Usage Adaptation (TUA), we categorize problems as easy or hard based on the model's accuracy, and train it to maintain direct reasoning for easy problems while switching to tools for hard ones. We validate our method on six scientific benchmark datasets across climate science, epidemiology, physics, and other domains. Compared to the original instruct model (8B), models post-trained with AWL achieve 29.11% higher answer accuracy and 12.72% better tool usage accuracy, even surpassing state-of-the-art models including GPT-4o and Claude-3.5 on four custom-created datasets. Our code is open-source at https://github.com/Rose-STL-Lab/Adapting-While-Learning.
URLs: https://github.com/Rose-STL-Lab/Adapting-While-Learning.
Authors: Guan Zhe Hong, Nishanth Dikkala, Enming Luo, Cyrus Rashtchian, Xin Wang, Rina Panigrahy
Abstract: Due to the size and complexity of modern large language models (LLMs), it has proven challenging to uncover the underlying mechanisms that models use to solve reasoning problems. For instance, is their reasoning for a specific problem localized to certain parts of the network? Do they break down the reasoning problem into modular components that are then executed as sequential steps as we go deeper in the model? To better understand the reasoning capability of LLMs, we study a minimal propositional logic problem that requires combining multiple facts to arrive at a solution. By studying this problem on Mistral and Gemma models, up to 27B parameters, we illuminate the core components the models use to solve such logic problems. From a mechanistic interpretability point of view, we use causal mediation analysis to uncover the pathways and components of the LLMs' reasoning processes. Then, we offer fine-grained insights into the functions of attention heads in different layers. We not only find a sparse circuit that computes the answer, but we decompose it into sub-circuits that have four distinct and modular uses. Finally, we reveal that three distinct models -- Mistral-7B, Gemma-2-9B and Gemma-2-27B -- contain analogous but not identical mechanisms.
Authors: Agnibh Dasgupta, Abdullah Tanvir, Xin Zhong
Abstract: Watermarking the outputs of large language models (LLMs) is critical for provenance tracing, content regulation, and model accountability. Existing approaches often rely on access to model internals or are constrained by static rules and token-level perturbations. Moreover, the idea of steering generative behavior via prompt-based instruction control remains largely underexplored. We introduce a prompt-guided watermarking framework that operates entirely at the input level and requires no access to model parameters or decoding logits. The framework comprises three cooperating components: a Prompting LM that synthesizes watermarking instructions from user prompts, a Marking LM that generates watermarked outputs conditioned on these instructions, and a Detecting LM trained to classify whether a response carries an embedded watermark. This modular design enables dynamic watermarking that adapts to individual prompts while remaining compatible with diverse LLM architectures, including both proprietary and open-weight models. We evaluate the framework over 25 combinations of Prompting and Marking LMs, such as GPT-4o, Mistral, LLaMA3, and DeepSeek. Experimental results show that watermark signals generalize across architectures and remain robust under fine-tuning, model distillation, and prompt-based adversarial attacks, demonstrating the effectiveness and robustness of the proposed approach.
Authors: Saul Santos, Vlad Niculae, Daniel McNamee, Andr\'e F. T. Martins
Abstract: Associative memory models, such as Hopfield networks and their modern variants, have garnered renewed interest due to advancements in memory capacity and connections with self-attention in transformers. In this work, we introduce a unified framework-Hopfield-Fenchel-Young networks-which generalizes these models to a broader family of energy functions. Our energies are formulated as the difference between two Fenchel-Young losses: one, parameterized by a generalized entropy, defines the Hopfield scoring mechanism, while the other applies a post-transformation to the Hopfield output. By utilizing Tsallis and norm entropies, we derive end-to-end differentiable update rules that enable sparse transformations, uncovering new connections between loss margins, sparsity, and exact retrieval of single memory patterns. We further extend this framework to structured Hopfield networks using the SparseMAP transformation, allowing the retrieval of pattern associations rather than a single pattern. Our framework unifies and extends traditional and modern Hopfield networks and provides an energy minimization perspective for widely used post-transformations like $\ell_2$-normalization and layer normalization-all through suitable choices of Fenchel-Young losses and by using convex analysis as a building block. Finally, we validate our Hopfield-Fenchel-Young networks on diverse memory recall tasks, including free and sequential recall. Experiments on simulated data, image retrieval, multiple instance learning, and text rationalization demonstrate the effectiveness of our approach.
Authors: Alkis Kalavasis, Anay Mehrotra, Grigoris Velegkas
Abstract: Specifying all desirable properties of a language model is challenging, but certain requirements seem essential. Given samples from an unknown language, the trained model should produce valid strings not seen in training and be expressive enough to capture the language's full richness. Otherwise, outputting invalid strings constitutes "hallucination," and failing to capture the full range leads to "mode collapse." We ask if a language model can meet both requirements. We investigate this within a statistical language generation setting building on Gold and Angluin. Here, the model receives random samples from a distribution over an unknown language K, which belongs to a possibly infinite collection of languages. The goal is to generate unseen strings from K. We say the model generates from K with consistency and breadth if, as training size increases, its output converges to all unseen strings in K. Kleinberg and Mullainathan [KM24] asked if consistency and breadth in language generation are possible. We answer this negatively: for a large class of language models, including next-token prediction models, this is impossible for most collections of candidate languages. This contrasts with [KM24]'s result, showing consistent generation without breadth is possible for any countable collection of languages. Our finding highlights that generation with breadth fundamentally differs from generation without breadth. As a byproduct, we establish near-tight bounds on the number of samples needed for generation with or without breadth. Finally, our results offer hope: consistent generation with breadth is achievable for any countable collection of languages when negative examples (strings outside K) are available alongside positive ones. This suggests that post-training feedback, which encodes negative examples, can be crucial in reducing hallucinations while limiting mode collapse.
Authors: Satush Parikh, Soumya Basu, Avishek Ghosh, Abishek Sankararaman
Abstract: Sequential learning in a multi-agent resource constrained matching market has received significant interest in the past few years. We study decentralized learning in two-sided matching markets where the demand side (aka players or agents) competes for the supply side (aka arms) with potentially time-varying preferences to obtain a stable match. Motivated by the linear contextual bandit framework, we assume that for each agent, an arm-mean may be represented by a linear function of a known feature vector and an unknown (agent-specific) parameter. Moreover, the preferences over arms depend on a latent environment in each round, where the latent environment varies across rounds in a non-stationary manner. We propose learning algorithms to identify the latent environment and obtain stable matchings simultaneously. Our proposed algorithms achieve instance-dependent logarithmic regret, scaling independently of the number of arms, and hence applicable for a large market.
Authors: Luca Colombo, Fabrizio Pittorino, Manuel Roveri
Abstract: Binary Neural Networks (BNNs) significantly reduce computational complexity and memory usage in machine and deep learning by representing weights and activations with just one bit. However, most existing training algorithms for BNNs rely on quantization-aware floating-point Stochastic Gradient Descent (SGD), limiting the full exploitation of binary operations to the inference phase only. In this work, we propose, for the first time, a fully binary and gradient-free training algorithm for multi-layer BNNs, eliminating the need for back-propagated floating-point gradients. Specifically, the proposed algorithm relies on local binary error signals and binary weight updates, employing integer-valued hidden weights that serve as a synaptic metaplasticity mechanism, thereby enhancing its neurobiological plausibility. Our proposed solution enables the training of binary multi-layer perceptrons by using exclusively XNOR, Popcount, and increment/decrement operations. Experimental results on multi-class classification benchmarks show test accuracy improvements of up to +35.47% over the only existing fully binary single-layer state-of-the-art solution. Compared to full-precision SGD, our solution improves test accuracy by up to +35.30% under the same total memory demand, while also reducing computational cost by two to three orders of magnitude in terms of the total number of Boolean gates. The proposed algorithm is made available to the scientific community as a public repository.
Authors: Taneesh Gupta, Rahul Madhavan, Xuchao Zhang, Nagarajan Natarajan, Chetan Bansal, Saravan Rajmohan
Abstract: Direct Preference Optimization (DPO) has become a popular approach for aligning language models using pairwise preferences. However, in practical post-training pipelines, on-policy generation typically yields multiple candidate responses per prompt, which are scored by a reward model to guide learning. In this setting, we propose $\textbf{Multi-Preference Optimization (MPO)}$, a generalization of DPO that optimizes over entire sets of responses by extending the Bradley-Terry model to groupwise comparisons between chosen and rejected sets. To further enhance learning, MPO employs deviation-based weighting, which emphasizes outlier responses that deviate most from the mean reward, effectively inducing a self-paced curriculum. We theoretically prove that MPO reduces alignment bias at a rate of $\mathcal{O}\left(\frac{1}{\sqrt{n}}\right)$ with respect to the number of responses per query. Empirically, MPO achieves state-of-the-art performance on the UltraFeedback benchmark and yields up to $\sim 17.5\%$ improvement over the state-of-the-art baseline in length-controlled win rate on AlpacaEval2, establishing a new baseline for preference-based alignment
Authors: Muru Zhang, Mayank Mishra, Zhongzhu Zhou, William Brandon, Jue Wang, Yoon Kim, Jonathan Ragan-Kelley, Shuaiwen Leon Song, Ben Athiwaratkun, Tri Dao
Abstract: Large language model inference is both memory-intensive and time-consuming, often requiring distributed algorithms to efficiently scale. Various model parallelism strategies are used in multi-gpu training and inference to partition computation across multiple devices, reducing memory load and computation time. However, using model parallelism necessitates communication of information between GPUs, which has been a major bottleneck and limits the gains obtained by scaling up the number of devices. We introduce Ladder Residual, a simple architectural modification applicable to all residual-based models that enables straightforward overlapping that effectively hides the latency of communication. Our insight is that in addition to systems optimization, one can also redesign the model architecture to decouple communication from computation. While Ladder Residual can allow communication-computation decoupling in conventional parallelism patterns, we focus on Tensor Parallelism in this paper, which is particularly bottlenecked by its heavy communication. For a Transformer model with 70B parameters, applying Ladder Residual to all its layers can achieve 29% end-to-end wall clock speed up at inference time with TP sharding over 8 devices. We refer the resulting Transformer model as the Ladder Transformer. We train a 1B and 3B Ladder Transformer from scratch and observe comparable performance to a standard dense transformer baseline. We also show that it is possible to convert parts of the Llama-3.1 8B model to our Ladder Residual architecture with minimal accuracy degradation by only retraining for 3B tokens. We release our code for training and inference for easier replication of experiments.
Authors: Abhinav Moudgil, Boris Knyazev, Guillaume Lajoie, Eugene Belilovsky
Abstract: Learned optimization has emerged as a promising alternative to hand-crafted optimizers, with the potential to discover stronger learned update rules that enable faster, hyperparameter-free training of neural networks. A critical element for practically useful learned optimizers, that can be used off-the-shelf after meta-training, is strong meta-generalization: the ability to apply the optimizers to new tasks. Recent state-of-the-art work in learned optimizers, VeLO (Metz et al., 2022), requires a large number of highly diverse meta-training tasks along with massive computational resources, 4000 TPU months, to achieve meta-generalization. This makes further improvements to such learned optimizers impractical. In this work, we identify several key elements in learned optimizer architectures and meta-training procedures that can lead to strong meta-generalization. We also propose evaluation metrics to reliably assess quantitative performance of an optimizer at scale on a set of evaluation tasks. Our proposed approach, Celo, makes a significant leap in improving the meta-generalization performance of learned optimizers and also outperforms tuned state-of-the-art optimizers on a diverse set of out-of-distribution tasks, despite being meta-trained for just 24 GPU hours.
Authors: Yuta Suzuki, Tatsunori Taniai, Ryo Igarashi, Kotaro Saito, Naoya Chiba, Yoshitaka Ushiku, Kanta Ono
Abstract: Understanding structure-property relationships is an essential yet challenging aspect of materials discovery and development. To facilitate this process, recent studies in materials informatics have sought latent embedding spaces of crystal structures to capture their similarities based on properties and functionalities. However, abstract feature-based embedding spaces are human-unfriendly and prevent intuitive and efficient exploration of the vast materials space. Here we introduce Contrastive Language--Structure Pre-training (CLaSP), a learning paradigm for constructing crossmodal embedding spaces between crystal structures and texts. CLaSP aims to achieve material embeddings that 1) capture property- and functionality-related similarities between crystal structures and 2) allow intuitive retrieval of materials via user-provided description texts as queries. To compensate for the lack of sufficient datasets linking crystal structures with textual descriptions, CLaSP leverages a dataset of over 400,000 published crystal structures and corresponding publication records, including paper titles and abstracts, for training. We demonstrate the effectiveness of CLaSP through text-based crystal structure screening and embedding space visualization.
Authors: Ziyu Zhao, Yixiao Zhou, Zhi Zhang, Didi Zhu, Tao Shen, Zexi Li, Jinluan Yang, Xuwu Wang, Jing Su, Kun Kuang, Zhongyu Wei, Fei Wu, Yu Cheng
Abstract: Low-Rank Adaptation (LoRA) is widely used for adapting large language models (LLMs) to specific domains due to its efficiency and modularity. Meanwhile, vanilla LoRA struggles with task conflicts in multi-task scenarios. Recent works adopt Mixture of Experts (MoE) by treating each LoRA module as an expert, thereby mitigating task interference through multiple specialized LoRA modules. While effective, these methods often isolate knowledge within individual tasks, failing to fully exploit the shared knowledge across related tasks. In this paper, we establish a connection between single LoRA and multi-LoRA MoE, integrating them into a unified framework. We demonstrate that the dynamic routing of multiple LoRAs is functionally equivalent to rank partitioning and block-level activation within a single LoRA. We further empirically demonstrate that finer-grained LoRA partitioning, within the same total and activated parameter constraints, leads to better performance gains across heterogeneous tasks. Building on these findings, we propose Single-ranked Mixture of Experts LoRA (\textbf{SMoRA}), which embeds MoE into LoRA by \textit{treating each rank as an independent expert}. With a \textit{dynamic rank-wise activation} mechanism, SMoRA promotes finer-grained knowledge sharing while mitigating task conflicts. Experiments demonstrate that SMoRA activates fewer parameters yet achieves better performance in multi-task scenarios.
Authors: Haixia Liu, Boxiao Li, Can Yang, Yang Wang
Abstract: Numerous studies have shown that label noise can lead to poor generalization performance, negatively affecting classification accuracy. Therefore, understanding the effectiveness of classifiers trained using deep neural networks in the presence of noisy labels is of considerable practical significance. In this paper, we focus on the error bounds of excess risks for classification problems with noisy labels within deep learning frameworks. We derive error bounds for the excess risk, decomposing it into statistical error and approximation error. To handle statistical dependencies (e.g., mixing sequences), we employ an independent block construction to bound the error, leveraging techniques for dependent processes. For the approximation error, we establish these theoretical results to the vector-valued setting, where the output space consists of $K$-dimensional unit vectors. Finally, under the low-dimensional manifold hypothesis, we further refine the approximation error to mitigate the impact of high-dimensional input spaces.
Authors: Christian Wald, Gabriele Steidl
Abstract: Among generative neural models, flow matching techniques stand out for their simple applicability and good scaling properties. Here, velocity fields of curves connecting a simple latent and a target distribution are learned. Then the corresponding ordinary differential equation can be used to sample from a target distribution, starting in samples from the latent one. This paper reviews from a mathematical point of view different techniques to learn the velocity fields of absolutely continuous curves in the Wasserstein geometry. We show how the velocity fields can be characterized and learned via i) transport plans (couplings) between latent and target distributions, ii) Markov kernels and iii) stochastic processes, where the latter two include the coupling approach, but are in general broader. Besides this main goal, we show how flow matching can be used for solving Bayesian inverse problems, where the definition of conditional Wasserstein distances plays a central role. Finally, we briefly address continuous normalizing flows and score matching techniques, which approach the learning of velocity fields of curves from other directions.
Authors: Ander Gray, Vignesh Gopakumar, Sylvain Rousseau, S\'ebastien Destercke
Abstract: We propose a method for obtaining statistically guaranteed prediction sets for functional machine learning methods: surrogate models which map between function spaces, motivated by the need to build reliable PDE emulators. The method constructs nested prediction sets on a low-dimensional representation (an SVD) of the surrogate model's error, and then maps these sets to the prediction space using set-propagation techniques. This results in prediction sets for functional surrogate models with conformal prediction coverage guarantees. We use zonotopes as basis of the set construction, which allow an exact linear propagation and are closed under Cartesian products, making them well-suited to this high-dimensional problem. The method is model agnostic and can thus be applied to complex Sci-ML models, including Neural Operators, but also in simpler settings. We also introduce a technique to capture the truncation error of the SVD, preserving the guarantees of the method.
Authors: Argyrios Gerogiannis, Yu-Han Huang, Subhonmesh Bose, Venugopal V. Veeravalli
Abstract: We introduce a practical, black-box framework termed Detection Augmenting Learning (DAL) for the problem of non-stationary bandits without prior knowledge of the underlying non-stationarity. DAL is modular, accepting any stationary bandit algorithm as input and augmenting it with a change detector. Our approach is applicable to all common parametric and non-parametric bandit variants. Extensive experimentation demonstrates that DAL consistently surpasses current state-of-the-art methods across diverse non-stationary scenarios, including synthetic benchmarks and real-world datasets, underscoring its versatility and scalability. We provide theoretical insights into DAL's strong empirical performance on piecewise stationary and drift settings, complemented by thorough experimental validation.
Authors: Yingshan Chang, Yonatan Bisk
Abstract: The notion of generalization has moved away from the classical one defined in statistical learning theory towards an emphasis on out-of-domain generalization (OODG). There has been a growing focus on generalization from easy to hard, where a progression of difficulty implicitly governs the direction of domain shifts. This emerging regime has appeared in the literature under different names, such as length/logical/algorithmic extrapolation, but a formal definition is lacking. We argue that the unifying theme is induction -- based on finite samples observed in training, a learner should infer an inductive principle that applies in an unbounded manner. This work formalizes the notion of inductive generalization along a difficulty progression and argues that our path ahead lies in transforming the learning paradigm. We attempt to make inroads by proposing a novel learning paradigm, Inductive Learning, which involves a central concept called model successors. We outline practical steps to adapt well-established techniques towards learning model successors. This work calls for restructuring of the research discussion around induction and generalization from fragmented task-centric communities to a more unified effort, focused on universal properties of learning and computation.
Authors: Xiaotong Ji, Shyam Sundhar Ramesh, Matthieu Zimmer, Ilija Bogunovic, Jun Wang, Haitham Bou Ammar
Abstract: We introduce a novel inference-time alignment approach for LLMs that aims to generate safe responses almost surely, i.e., with probability approaching one. Our approach models the generation of safe responses as a constrained Markov Decision Process (MDP) within the LLM's latent space. We augment a safety state that tracks the evolution of safety constraints and dynamically penalize unsafe generations to ensure the generation of safe responses. Consequently, we demonstrate formal safety guarantees w.r.t. the given cost model upon solving the MDP in the latent space with sufficiently large penalties. Building on this foundation, we propose InferenceGuard, a practical implementation that safely aligns LLMs without modifying the model weights. Empirically, we demonstrate that InferenceGuard effectively balances safety and task performance, outperforming existing inference-time alignment methods in generating safe and aligned responses. Our findings contribute to the advancement of safer LLM deployment through alignment at inference-time, thus presenting a promising alternative to resource-intensive, overfitting-prone alignment techniques like RLHF.
Authors: Christopher E. Mower, Haitham Bou-Ammar
Abstract: Inferring physical laws from data is a central challenge in science and engineering, including but not limited to healthcare, physical sciences, biosciences, social sciences, sustainability, climate, and robotics. Deep networks offer high-accuracy results but lack interpretability, prompting interest in models built from simple components. The Sparse Identification of Nonlinear Dynamics (SINDy) method has become the go-to approach for building such modular and interpretable models. SINDy leverages sparse regression with L1 regularization to identify key terms from a library of candidate functions. However, SINDy's choice of candidate library and optimization method requires significant technical expertise, limiting its widespread applicability. This work introduces Al-Khwarizmi, a novel agentic framework for physical law discovery from data, which integrates foundational models with SINDy. Leveraging LLMs, VLMs, and Retrieval-Augmented Generation (RAG), our approach automates physical law discovery, incorporating prior knowledge and iteratively refining candidate solutions via reflection. Al-Khwarizmi operates in two steps: it summarizes system observations-comprising textual descriptions, raw data, and plots-followed by a secondary step that generates candidate feature libraries and optimizer configurations to identify hidden physics laws correctly. Evaluating our algorithm on over 198 models, we demonstrate state-of-the-art performance compared to alternatives, reaching a 20 percent increase against the best-performing alternative.
Authors: Muxing Li, Zesheng Ye, Yixuan Li, Andy Song, Guangquan Zhang, Feng Liu
Abstract: To detect unauthorized data usage in training large-scale generative models (e.g., ChatGPT or Midjourney), membership inference attacks (MIA) have proven effective in distinguishing a single training instance (a member) from a single non-training instance (a non-member). This success is mainly credited to a memorization effect: models tend to perform better on a member than a non-member. However, we find that standard MIAs fail against distilled generative models (i.e., student models) that are increasingly deployed in practice for efficiency (e.g., ChatGPT 4o-mini). Trained exclusively on data generated from a large-scale model (a teacher model), the student model lacks direct exposure to any members (teacher's training data), nullifying the memorization effect that standard MIAs rely on. This finding reveals a serious privacy loophole, where generation-service providers could deploy a student model whose teacher was potentially trained on unauthorized data, yet claim the deployed model is clean because it was not directly trained on such data. Hence, are distilled models inherently unauditable for upstream privacy violations, and should we discard them when we care about privacy? We contend no, as we uncover a memory chain connecting the student and teacher's member data: the distribution of student-generated data aligns more closely with the distribution of the teacher's members than with non-members, thus we can detect unauthorized data usage even when direct instance-level memorization is absent. This leads us to posit that MIAs on distilled generative models should shift from instance-level scores to distribution-level statistics. We further propose three principles of distribution-based MIAs for detecting unauthorized training data through distilled generative models, and validate our position through an exemplar framework. We lastly discuss the implications of our position.
Authors: Thomas Geert de Jong, Hirofumi Notsu, Kohei Nakajima
Abstract: Nature is pervaded with oscillatory dynamics. In networks of coupled oscillators patterns can arise when the system synchronizes to an external input. Hence, these networks provide processing and memory of input. We present a universal framework for harnessing oscillator networks as computational resource. This computing framework is introduced by the ubiquitous model for phase-locking, the Kuramoto model. We force the Kuramoto model by a nonlinear target-system, then after substituting the target-system with a trained feedback-loop it emulates the target-system. Our results are two-fold. Firstly, the trained network inherits performance properties of the Kuramoto model, where all-to-all coupling is performed in linear time with respect to the number of nodes and parameters for synchronization are abundant. The latter implies that the network is generically successful since the system learns via sychronization. Secondly, the learning capabilities of the oscillator network, which describe a type of collective intelligence, can be explained using Kuramoto model's order parameter. In summary, this work provides the foundation for utilizing nature's oscillator networks as a new class of information processing systems.
Authors: Yijun Dong, Yicheng Li, Yunai Li, Jason D. Lee, Qi Lei
Abstract: Weak-to-strong (W2S) generalization is a type of finetuning (FT) where a strong (large) student model is trained on pseudo-labels generated by a weak teacher. Surprisingly, W2S FT often outperforms the weak teacher. We seek to understand this phenomenon through the observation that FT often occurs in intrinsically low-dimensional spaces. Leveraging the low intrinsic dimensionality of FT, we analyze W2S in the ridgeless regression setting from a variance reduction perspective. For a strong student-weak teacher pair with sufficiently expressive low-dimensional feature subspaces $\mathcal{V}_s, \mathcal{V}_w$, we provide an exact characterization of the variance that dominates the generalization error of W2S. This unveils a virtue of discrepancy between the strong and weak models in W2S: the variance of the weak teacher is inherited by the strong student in $\mathcal{V}_s \cap \mathcal{V}_w$, while reduced by a factor of $\mathrm{dim}(\mathcal{V}_s)/N$ in the subspace of discrepancy $\mathcal{V}_w \setminus \mathcal{V}_s$ with $N$ pseudo-labels for W2S. Our analysis further casts light on the sample complexities and the scaling of performance gap recovery in W2S. The analysis is supported by experiments on synthetic regression problems, as well as real vision and NLP tasks.
Authors: Yunchu Han, Zhaojun Nan, Sheng Zhou, Zhisheng Niu
Abstract: The rapid development of deep neural networks (DNNs) is inherently accompanied by the problem of high computational costs. To tackle this challenge, dynamic voltage frequency scaling (DVFS) is emerging as a promising technology for balancing the latency and energy consumption of DNN inference by adjusting the computing frequency of processors. However, most existing models of DNN inference time are based on the CPU-DVFS technique, and directly applying the CPU-DVFS model to DNN inference on GPUs will lead to significant errors in optimizing latency and energy consumption. In this paper, we propose a DVFS-aware latency model to precisely characterize DNN inference time on GPUs. We first formulate the DNN inference time based on extensive experiment results for different devices and analyze the impact of fitting parameters. Then by dividing DNNs into multiple blocks and obtaining the actual inference time, the proposed model is further verified. Finally, we compare our proposed model with the CPU-DVFS model in two specific cases. Evaluation results demonstrate that local inference optimization with our proposed model achieves a reduction of no less than 66% and 69% in inference time and energy consumption respectively. In addition, cooperative inference with our proposed model can improve the partition policy and reduce the energy consumption compared to the CPU-DVFS model.
Authors: Feng Chen, Allan Raventos, Nan Cheng, Surya Ganguli, Shaul Druckmann
Abstract: Recent progress in large language models (LLMs) highlights the power of scaling test-time compute to achieve strong performance on complex tasks, such as mathematical reasoning and code generation. This raises a critical question: how should model training be modified to optimize performance under a subsequent test-time compute strategy and budget? To explore this, we focus on pass@N, a simple test-time strategy that searches for a correct answer in $N$ independent samples. We show, surprisingly, that training with cross-entropy (CE) loss can be ${\it misaligned}$ with pass@N in that pass@N accuracy ${\it decreases}$ with longer training. We explain the origins of this misalignment in terms of model overconfidence induced by CE, and experimentally verify our prediction of overconfidence as an impediment to scaling test-time compute via pass@N. Furthermore we suggest a principled, modified training loss that is better aligned to pass@N by limiting model confidence and rescuing pass@N test performance. Our algorithm demonstrates improved mathematical reasoning on MATH and MiniF2F benchmarks under several scenarios: (1) providing answers to math questions; and (2) proving theorems by searching over proof trees of varying shapes. Overall our work underscores the importance of co-designing two traditionally separate phases of LLM development: training-time protocols and test-time search and reasoning strategies.
Authors: Long-Fei Li, Yu-Yang Qian, Peng Zhao, Zhi-Hua Zhou
Abstract: Reinforcement Learning from Human Feedback (RLHF) has shown remarkable success in aligning Large Language Models (LLMs) with human preferences. Traditional RLHF approaches rely on a fixed dataset, which often suffers from limited coverage. To this end, online RLHF has emerged as a promising direction, enabling iterative data collection and model improvement. Despite its potential, this paradigm faces a key bottleneck: the requirement to continuously integrate new data into the historical dataset and re-optimize the model from scratch at each iteration, resulting in computational and storage costs that grow linearly with the number of iterations. In this work, we address this challenge by proposing a one-pass reward modeling method that does not require storing the historical data and can be computed in constant time. Specifically, we first formalize RLHF as a contextual preference bandit problem and design an online mirror descent algorithm with a tailored local norm to replace the standard maximum likelihood estimation for reward modeling. We then apply our method to various online RLHF settings, including passive data collection, active data collection, and deployment-time adaptation. We provide theoretical guarantees showing that our method improves both statistical and computational efficiency. Finally, we provide practical algorithms and conduct experiments using Llama-3-8B-Instruct and Qwen2.5-7B-Instruct models on the Ultrafeedback-binarized and Mixture2 datasets, validating the effectiveness of our proposed method.
Authors: Elias Kempf, Simon Schrodi, Max Argus, Thomas Brox
Abstract: The remarkable generalization performance of contrastive vision-language models like CLIP is often attributed to the diversity of their training distributions. However, key questions remain unanswered: Can CLIP generalize to an entirely unseen domain when trained on a diverse mixture of domains (domain generalization)? Can it generalize to unseen classes within partially seen domains (compositional generalization)? What factors affect such generalization? To answer these questions, we trained CLIP models on systematically constructed training distributions with controlled domain diversity and object class exposure. Our experiments show that domain diversity is essential for both domain and compositional generalization, yet compositional generalization can be surprisingly weaker than domain generalization when the training distribution contains a suboptimal subset of the test domain. Through data-centric and mechanistic analyses, we find that successful generalization requires the learning of sufficiently shared representations in intermediate layers and circuits.
Authors: Antoine Moulin, Gergely Neu, Luca Viano
Abstract: We study the problem of reinforcement learning in infinite-horizon discounted linear Markov decision processes (MDPs), and propose the first computationally efficient algorithm achieving rate-optimal regret guarantees in this setting. Our main idea is to combine two classic techniques for optimistic exploration: additive exploration bonuses applied to the reward function, and artificial transitions made to an absorbing state with maximal return. We show that, combined with a regularized approximate dynamic-programming scheme, the resulting algorithm achieves a regret of order $\tilde{\mathcal{O}} (\sqrt{d^3 (1 - \gamma)^{- 7 / 2} T})$, where $T$ is the total number of sample transitions, $\gamma \in (0,1)$ is the discount factor, and $d$ is the feature dimensionality. The results continue to hold against adversarial reward sequences, enabling application of our method to the problem of imitation learning in linear MDPs, where we achieve state-of-the-art results.
Authors: Sujeong Im, Jungwoo Oh, Edward Choi
Abstract: Lab tests are fundamental for diagnosing diseases and monitoring patient conditions. However, frequent testing can be burdensome for patients, and test results may not always be immediately available. To address these challenges, we propose LabTOP, a unified model that predicts lab test outcomes by leveraging a language modeling approach on EHR data. Unlike conventional methods that estimate only a subset of lab tests or classify discrete value ranges, LabTOP performs continuous numerical predictions for a diverse range of lab items. We evaluate LabTOP on three publicly available EHR datasets and demonstrate that it outperforms existing methods, including traditional machine learning models and state-of-the-art large language models. We also conduct extensive ablation studies to confirm the effectiveness of our design choices. We believe that LabTOP will serve as an accurate and generalizable framework for lab test outcome prediction, with potential applications in clinical decision support and early detection of critical conditions.
Authors: Piyushi Manupriya, Himanshu, SakethaNath Jagarlapudi, Ganesh Ghalme
Abstract: We investigate the problem of maximizing social welfare while ensuring fairness in a multi-agent multi-armed bandit (MA-MAB) setting. In this problem, a centralized decision-maker takes actions over time, generating random rewards for various agents. Our goal is to maximize the sum of expected cumulative rewards, a.k.a. social welfare, while ensuring that each agent receives an expected reward that is at least a constant fraction of the maximum possible expected reward. Our proposed algorithm, RewardFairUCB, leverages the Upper Confidence Bound (UCB) technique to achieve sublinear regret bounds for both fairness and social welfare. The fairness regret measures the positive difference between the minimum reward guarantee and the expected reward of a given policy, whereas the social welfare regret measures the difference between the social welfare of the optimal fair policy and that of the given policy. We show that RewardFairUCB algorithm achieves instance-independent social welfare regret guarantees of $\tilde{O}(T^{1/2})$ and a fairness regret upper bound of $\tilde{O}(T^{3/4})$. We also give the lower bound of $\Omega(\sqrt{T})$ for both social welfare and fairness regret. We evaluate RewardFairUCB's performance against various baseline and heuristic algorithms using simulated data and real world data, highlighting trade-offs between fairness and social welfare regrets.
Authors: Varun Babbar, Hayden McTavish, Cynthia Rudin, Margo Seltzer
Abstract: Decision tree optimization is fundamental to interpretable machine learning. The most popular approach is to greedily search for the best feature at every decision point, which is fast but provably suboptimal. Recent approaches find the global optimum using branch and bound with dynamic programming, showing substantial improvements in accuracy and sparsity at great cost to scalability. An ideal solution would have the accuracy of an optimal method and the scalability of a greedy method. We introduce a family of algorithms called SPLIT (SParse Lookahead for Interpretable Trees) that moves us significantly forward in achieving this ideal balance. We demonstrate that not all sub-problems need to be solved to optimality to find high quality trees; greediness suffices near the leaves. Since each depth adds an exponential number of possible trees, this change makes our algorithms orders of magnitude faster than existing optimal methods, with negligible loss in performance. We extend this algorithm to allow scalable computation of sets of near-optimal trees (i.e., the Rashomon set).
Authors: Emiliano Seri, Marcello Petitta, Chryssoula Papaioannou, Nikolaos Katsoulas, Cristina Cornaro
Abstract: The integration of photovoltaic (PV) systems into greenhouses not only optimizes land use but also enhances sustainable agricultural practices by enabling dual benefits of food production and renewable energy generation. However, accurate prediction of internal environmental conditions is crucial to ensure optimal crop growth while maximizing energy production. This study introduces a novel application of Spatio-Temporal Graph Neural Networks (STGNNs) to greenhouse microclimate modeling, comparing their performance with traditional Recurrent Neural Networks (RNNs). While RNNs excel at temporal pattern recognition, they cannot explicitly model the directional relationships between environmental variables. Our STGNN approach addresses this limitation by representing these relationships as directed graphs, enabling the model to capture both environmental dependencies and their directionality. We benchmark RNNs against directed STGNNs on two 15-min-resolution datasets from Volos (Greece): a six-variable 2020 installation and a more complex eight-variable greenhouse monitored in autumn 2024. In the simpler 2020 case the RNN attains near-perfect accuracy, outperforming the STGNN. When additional drivers are available in 2024, the STGNN overtakes the RNN ($R^{2}=0.905$ vs $0.740$), demonstrating that explicitly modelling directional dependencies becomes critical as interaction complexity grows. These findings indicate when graph-based models are warranted and provide a stepping-stone toward digital twins that jointly optimise crop yield and PV power in agrivoltaic greenhouses.
Authors: Th\'eo Vincent, Tim Faust, Yogesh Tripathi, Jan Peters, Carlo D'Eramo
Abstract: Recent works have successfully demonstrated that sparse deep reinforcement learning agents can be competitive against their dense counterparts. This opens up opportunities for reinforcement learning applications in fields where inference time and memory requirements are cost-sensitive or limited by hardware. Until now, dense-to-sparse methods have relied on hand-designed sparsity schedules that are not synchronized with the agent's learning pace. Crucially, the final sparsity level is chosen as a hyperparameter, which requires careful tuning as setting it too high might lead to poor performances. In this work, we address these shortcomings by crafting a dense-to-sparse algorithm that we name Eau De $Q$-Network (EauDeQN). To increase sparsity at the agent's learning pace, we consider multiple online networks with different sparsity levels, where each online network is trained from a shared target network. At each target update, the online network with the smallest loss is chosen as the next target network, while the other networks are replaced by a pruned version of the chosen network. We evaluate the proposed approach on the Atari $2600$ benchmark and the MuJoCo physics simulator, showing that EauDeQN reaches high sparsity levels while keeping performances high.
Authors: Mohamed Hassouna, Clara Holzh\"uter, Malte Lehna, Matthijs de Jong, Jan Viebahn, Bernhard Sick, Christoph Scholz
Abstract: The rising proportion of renewable energy in the electricity mix introduces significant operational challenges for power grid operators. Effective power grid management demands adaptive decision-making strategies capable of handling dynamic conditions. With the increase in complexity, more and more Deep Learning (DL) approaches have been proposed to find suitable grid topologies for congestion management. In this work, we contribute to this research by introducing a novel Imitation Learning (IL) approach that leverages soft labels derived from simulated topological action outcomes, thereby capturing multiple viable actions per state. Unlike traditional IL methods that rely on hard labels to enforce a single optimal action, our method constructs soft labels that capture the effectiveness of actions that prove suitable in resolving grid congestion. To further enhance decision-making, we integrate Graph Neural Networks (GNNs) to encode the structural properties of power grids, ensuring that the topology-aware representations contribute to better agent performance. Our approach significantly outperforms its hard-label counterparts as well as state-of-the-art Deep Reinforcement Learning (DRL) baseline agents. Most notably, it achieves a 17% better performance compared to the greedy expert agent from which the imitation targets were derived.
Authors: Zubair Shaban, Nazreen Shah, Ranjitha Prasad
Abstract: In 6G wireless networks, Artificial Intelligence (AI)-driven applications demand the adoption of Federated Learning (FL) to enable efficient and privacy-preserving model training across distributed devices. Over-The-Air Federated Learning (OTA-FL) exploits the superposition property of multiple access channels, allowing edge users in 6G networks to efficiently share spectral resources and perform low-latency global model aggregation. However, these advantages come with challenges, as traditional OTA-FL techniques suffer due to the joint effects of Additive White Gaussian Noise (AWGN) at the server, fading, and both data and system heterogeneity at the participating edge devices. In this work, we propose the novel Noise Resilient Over-the-Air Federated Learning (NoROTA-FL) framework to jointly tackle these challenges in federated wireless networks. In NoROTA-FL, the local optimization problems find controlled inexact solutions, which manifests as an additional proximal constraint at the clients. This approach provides robustness against straggler-induced partial work, heterogeneity, noise, and fading. From a theoretical perspective, we leverage the zeroth- and first-order inexactness and establish convergence guarantees for non-convex optimization problems in the presence of heterogeneous data and varying system capabilities. Experimentally, we validate NoROTA-FL on real-world datasets, including FEMNIST, CIFAR10, and CIFAR100, demonstrating its robustness in noisy and heterogeneous environments. Compared to state-of-the-art baselines such as COTAF and FedProx, NoROTA-FL achieves significantly more stable convergence and higher accuracy, particularly in the presence of stragglers.
Authors: Enrico Marchesini, Benjamin Donnot, Constance Crozier, Ian Dytham, Christian Merz, Lars Schewe, Nico Westerbeck, Cathy Wu, Antoine Marot, Priya L. Donti
Abstract: Reinforcement learning (RL) can provide adaptive and scalable controllers essential for power grid decarbonization. However, RL methods struggle with power grids' complex dynamics, long-horizon goals, and hard physical constraints. For these reasons, we present RL2Grid, a benchmark designed in collaboration with power system operators to accelerate progress in grid control and foster RL maturity. Built on RTE France's power simulation framework, RL2Grid standardizes tasks, state and action spaces, and reward structures for a systematic evaluation and comparison of RL algorithms. Moreover, we integrate operational heuristics and design safety constraints based on human expertise to ensure alignment with physical requirements. By establishing reference performance metrics for classic RL baselines on RL2Grid's tasks, we highlight the need for novel methods capable of handling real systems and discuss future directions for RL-based grid control.
Authors: Mengjia Niu, Hamed Haddadi, Guansong Pang
Abstract: Hallucinations in large language models (LLMs) pose significant safety concerns that impede their broader deployment. Recent research in hallucination detection has demonstrated that LLMs' internal representations contain truthfulness hints, which can be harnessed for detector training. However, the performance of these detectors is heavily dependent on the internal representations of predetermined tokens, fluctuating considerably when working on free-form generations with varying lengths and sparse distributions of hallucinated entities. To address this, we propose HaMI, a novel approach that enables robust detection of hallucinations through adaptive selection and learning of critical tokens that are most indicative of hallucinations. We achieve this robustness by an innovative formulation of the Hallucination detection task as Multiple Instance (HaMI) learning over token-level representations within a sequence, thereby facilitating a joint optimisation of token selection and hallucination detection on generation sequences of diverse forms. Comprehensive experimental results on four hallucination benchmarks show that HaMI significantly outperforms existing state-of-the-art approaches.
Authors: Ruochi Zhang, Qian Yang, Xiaoyang Wang, Tian Wang, Qiong Zhou, Ziqi Deng, Kewei Li, Yueying Wang, Yusi Fan, Jiale Zhang, Lan Huang, Chang Liu, Fengfeng Zhou
Abstract: The rapid accumulation of Electronic Health Records (EHRs) has transformed healthcare by providing valuable data that enhance clinical predictions and diagnoses. While conventional machine learning models have proven effective, they often lack robust representation learning and depend heavily on expert-crafted features. Although deep learning offers powerful solutions, it is often criticized for its lack of interpretability. To address these challenges, we propose DeepSelective, a novel end to end deep learning framework for predicting patient prognosis using EHR data, with a strong emphasis on enhancing model interpretability. DeepSelective combines data compression techniques with an innovative feature selection approach, integrating custom-designed modules that work together to improve both accuracy and interpretability. Our experiments demonstrate that DeepSelective not only enhances predictive accuracy but also significantly improves interpretability, making it a valuable tool for clinical decision-making. The source code is freely available at http://www.healthinformaticslab.org/supp/resources.php .
URLs: http://www.healthinformaticslab.org/supp/resources.php
Authors: Xiang Li, Jianpeng Qi, Haobing Liu, Yuan Cao, Guoqing Chao, Zhongying Zhao, Junyu Dong, Yanwei Yu
Abstract: Graph Neural Networks (GNNs) have demonstrated impressive performance across diverse graph-based tasks by leveraging message passing to capture complex node relationships. However, when applied to large-scale real-world graphs, GNNs face two major challenges: First, it becomes increasingly difficult to ensure both scalability and efficiency, as the repeated aggregation of large neighborhoods leads to significant computational overhead; Second, the over-smoothing problem arises, where excessive or deep propagation makes node representations indistinguishable, severely hindering model expressiveness. To tackle these issues, we propose ScaleGNN, a novel framework that adaptively fuses multi-level graph features for both scalable and effective graph learning. ScaleGNN first constructs per-order neighbor matrices that capture only the exclusive structural information at each hop, avoiding the redundancy of conventional aggregation. A learnable fusion module then selectively integrates these features, emphasizing the most informative high-order neighbors. To further reduce redundancy and over-smoothing, we introduce a Local Contribution Score (LCS)-based masking mechanism to filter out less relevant high-order neighbors, ensuring that only the most meaningful information is aggregated. In addition, a task-aware feature fusion strategy dynamically balances low- and high-order information, preserving both local detail and global context without incurring excessive complexity. Extensive experiments on real-world datasets demonstrate that ScaleGNN consistently outperforms state-of-the-art GNNs in both predictive accuracy and computational efficiency, highlighting its practical value for large-scale graph learning.
Authors: Wenqing Peng, Zhi-Song Liu, Michael Boy
Abstract: Estimating rate coefficients from complex chemical reactions is essential for advancing detailed chemistry. However, the stiffness inherent in real-world atmospheric chemistry systems poses severe challenges, leading to training instability and poor convergence that hinder effective rate coefficient estimation using learning-based approaches. To address this, we propose a Stiff Physics-Informed Neural ODE framework (SPIN-ODE) for chemical reaction modelling. Our method introduces a three-stage optimisation process: first, a latent neural ODE learns the continuous and differentiable trajectory between chemical concentrations and their time derivatives; second, an explicit Chemical Reaction Neural Network (CRNN) extracts the underlying rate coefficients based on the learned dynamics; and third, fine-tune CRNN using a neural ODE solver to further improve rate coefficient estimation. Extensive experiments on both synthetic and newly proposed real-world datasets validate the effectiveness and robustness of our approach. As the first work on stiff Neural ODEs for chemical rate coefficient discovery, our study opens promising directions for integrating neural networks with detailed chemistry.
Authors: Winston Chen, Trenton Chang, Jenna Wiens
Abstract: Estimates of heterogeneous treatment assignment effects can inform treatment decisions. Under the presence of non-adherence (e.g., patients do not adhere to their assigned treatment), both the standard backdoor adjustment (SBD) and the conditional front-door adjustment (CFD) can recover unbiased estimates of the treatment assignment effects. However, the estimation variance of these approaches may vary widely across settings, which remains underexplored in the literature. In this work, we demonstrate theoretically and empirically that CFD yields lower-variance estimates than SBD when the true effect of treatment assignment is small (i.e., assigning an intervention leads to small changes in patients' future outcome). Additionally, since CFD requires estimating multiple nuisance parameters, we introduce LobsterNet, a multi-task neural network that implements CFD with joint modeling of the nuisance parameters. Empirically, LobsterNet reduces estimation error across several semi-synthetic and real-world datasets compared to baselines. Our findings suggest CFD with shared nuisance parameter modeling can improve treatment assignment effect estimation under non-adherence.
Authors: Feilong Jiang, Xiaonan Hou, Jianqiao Ye, Min Xia
Abstract: Physics-Informed Neural Networks (PINNs) have emerged as a powerful framework for solving partial differential equations (PDEs) by embedding physical laws directly into the loss function. However, effective training of PINNs remains challenging due to internal covariate shift, which destabilizes feature distributions and impairs model expressiveness. While normalization techniques like Batch Normalization and Layer Normalization are standard remedies in deep learning, they disrupt the pointwise input-output mappings critical to preserving the physical consistency in PINNs. In this work, we introduce Mask-PINNs, a novel architecture that regulates internal feature distributions through a smooth, learnable mask function applied pointwise across hidden layers. Unlike conventional normalization methods, the proposed mask function preserves the deterministic nature of input-output relationships while suppressing activation drift and saturation. Theoretically, we demonstrate that Mask-PINNs control feature spread near initialization by attenuating gradient variance growth through a tailored modulation mechanism. Empirically, we validate the method on multiple PDE benchmarks across diverse activation functions. Our results show consistent improvements in prediction accuracy, convergence stability, and robustness, with relative L2 errors reduced by up to two orders of magnitude over baseline models. Furthermore, we demonstrate that Mask-PINNs enable the effective use of wider networks, overcoming a key limitation in existing PINN frameworks.
Authors: Muhamed Amin, Bernard R. Brooks
Abstract: We present the Boltzmann classifier, a novel distance based probabilistic classification algorithm inspired by the Boltzmann distribution. Unlike traditional classifiers that produce hard decisions or uncalibrated probabilities, the Boltzmann classifier assigns class probabilities based on the average distance to the nearest neighbors within each class, providing interpretable, physically meaningful outputs. We evaluate the performance of the method across three application domains: molecular activity prediction, oxidation state classification of transition metal complexes, and breast cancer diagnosis. In the molecular activity task, the classifier achieved the highest accuracy in predicting active compounds against two protein targets, with strong correlations observed between the predicted probabilities and experimental pIC50 values. For metal complexes, the classifier accurately distinguished between oxidation states II and III for Fe, Mn, and Co, using only metal-ligand bond lengths extracted from crystallographic data, and demonstrated high consistency with known chemical trends. In the breast cancer dataset, the classifier achieved 97% accuracy, with low confidence predictions concentrated in inherently ambiguous cases. Across all tasks, the Boltzmann classifier performed competitively or better than standard models such as logistic regression, support vector machines, random forests, and k-nearest neighbors. Its probabilistic outputs were found to correlate with continuous physical or biological properties, highlighting its potential utility in both classification and regression contexts. The results suggest that the Boltzmann classifier is a robust and interpretable alternative to conventional machine learning approaches, particularly in scientific domains where underlying structure property relationships are important.
Authors: Jiafan Li, Jiaqi Zhu, Liang Chang, Yilin Li, Miaomiao Li, Yang Wang, Hongan Wang
Abstract: Nowadays, numerous online platforms can be described as multi-modal heterogeneous networks (MMHNs), such as Douban's movie networks and Amazon's product review networks. Accurately categorizing nodes within these networks is crucial for analyzing the corresponding entities, which requires effective representation learning on nodes. However, existing multi-modal fusion methods often adopt either early fusion strategies which may lose the unique characteristics of individual modalities, or late fusion approaches overlooking the cross-modal guidance in GNN-based information propagation. In this paper, we propose a novel model for node classification in MMHNs, named Heterogeneous Graph Neural Network with Inter-Modal Attention (HGNN-IMA). It learns node representations by capturing the mutual influence of multiple modalities during the information propagation process, within the framework of heterogeneous graph transformer. Specifically, a nested inter-modal attention mechanism is integrated into the inter-node attention to achieve adaptive multi-modal fusion, and modality alignment is also taken into account to encourage the propagation among nodes with consistent similarities across all modalities. Moreover, an attention loss is augmented to mitigate the impact of missing modalities. Extensive experiments validate the superiority of the model in the node classification task, providing an innovative view to handle multi-modal data, especially when accompanied with network structures.
Authors: Xiaohui Wang, Peng Ye, Chenyu Huang, Shenghe Zheng, Bo Zhang, Lei Bai, Wanli Ouyang, Tao Chen
Abstract: With the rise of the fine-tuned--pretrained paradigm, storing numerous fine-tuned models for multi-tasking creates significant storage overhead. Delta compression alleviates this by storing only the pretrained model and the highly compressed delta weights (the differences between fine-tuned and pretrained model weights). However, existing methods fail to maintain both high compression and performance, and often rely on data. To address these challenges, we propose UltraDelta, the first data-free delta compression pipeline that achieves both ultra-high compression and strong performance. UltraDelta is designed to minimize redundancy, maximize information, and stabilize performance across inter-layer, intra-layer, and global dimensions, using three key components: (1) Variance-Based Mixed Sparsity Allocation assigns sparsity based on variance, giving lower sparsity to high-variance layers to preserve inter-layer information. (2) Distribution-Aware Compression applies uniform quantization and then groups parameters by value, followed by group-wise pruning, to better preserve intra-layer distribution. (3) Trace-Norm-Guided Rescaling uses the trace norm of delta weights to estimate a global rescaling factor, improving model stability under higher compression. Extensive experiments across (a) large language models (fine-tuned on LLaMA-2 7B and 13B) with up to 133x, (b) general NLP models (RoBERTa-base, T5-base) with up to 800x, (c) vision models (ViT-B/32, ViT-L/14) with up to 400x, and (d) multi-modal models (BEiT-3) with 40x compression ratio, demonstrate that UltraDelta consistently outperforms existing methods, especially under ultra-high compression.
Authors: Jihwan Oh, Murad Aghazada, Se-Young Yun, Taehyeon Kim
Abstract: Bargaining, a critical aspect of real-world interactions, presents challenges for large language models (LLMs) due to limitations in strategic depth and adaptation to complex human factors. Existing benchmarks often fail to capture this real-world complexity. To address this and enhance LLM capabilities in realistic bargaining, we introduce a comprehensive framework centered on utility-based feedback. Our contributions are threefold: (1) BargainArena, a novel benchmark dataset with six intricate scenarios (e.g., deceptive practices, monopolies) to facilitate diverse strategy modeling; (2) human-aligned, economically-grounded evaluation metrics inspired by utility theory, incorporating agent utility and negotiation power, which implicitly reflect and promote opponent-aware reasoning (OAR); and (3) a structured feedback mechanism enabling LLMs to iteratively refine their bargaining strategies. This mechanism can positively collaborate with in-context learning (ICL) prompts, including those explicitly designed to foster OAR. Experimental results show that LLMs often exhibit negotiation strategies misaligned with human preferences, and that our structured feedback mechanism significantly improves their performance, yielding deeper strategic and opponent-aware reasoning.
Authors: Yaoyu Zhu, Di Huang, Hanqi Lyu, Xiaoyun Zhang, Chongxiao Li, Wenxuan Shi, Yutong Wu, Jianan Mu, Jinghua Wang, Yang Zhao, Pengwei Jin, Shuyao Cheng, Shengwen Liang, Xishan Zhang, Rui Zhang, Zidong Du, Qi Guo, Xing Hu, Yunji Chen
Abstract: Large language models (LLMs) trained via reinforcement learning with verifiable reward (RLVR) have achieved breakthroughs on tasks with explicit, automatable verification, such as software programming and mathematical problems. Extending RLVR to electronic design automation (EDA), especially automatically generating hardware description languages (HDLs) like Verilog from natural-language (NL) specifications, however, poses three key challenges: the lack of automated and accurate verification environments, the scarcity of high-quality NL-code pairs, and the prohibitive computation cost of RLVR. To this end, we introduce CodeV-R1, an RLVR framework for training Verilog generation LLMs. First, we develop a rule-based testbench generator that performs robust equivalence checking against golden references. Second, we propose a round-trip data synthesis method that pairs open-source Verilog snippets with LLM-generated NL descriptions, verifies code-NL-code consistency via the generated testbench, and filters out inequivalent examples to yield a high-quality dataset. Third, we employ a two-stage "distill-then-RL" training pipeline: distillation for the cold start of reasoning abilities, followed by adaptive DAPO, our novel RLVR algorithm that can reduce training cost by adaptively adjusting sampling rate. The resulting model, CodeV-R1-7B, achieves 68.6% and 72.9% pass@1 on VerilogEval v2 and RTLLM v1.1, respectively, surpassing prior state-of-the-art by 12~20%, while matching or even exceeding the performance of 671B DeepSeek-R1. We will release our model, training pipeline, and dataset to facilitate research in EDA and LLM communities.
Authors: Junaid Muzaffar, Khubaib Ahmed, Ingo Frommholz, Zeeshan Pervez, Ahsan ul Haq
Abstract: Training reinforcement learning (RL) agents often requires significant computational resources and prolonged training durations. To address this challenge, we build upon prior work that introduced a neural architecture with permutation-invariant sensory processing. We propose a modified attention mechanism that applies a non-linear transformation to the key vectors (K), producing enriched representations (K') through a custom mapping function. This Nonlinear Attention (NLA) mechanism enhances the representational capacity of the attention layer, enabling the agent to learn more expressive feature interactions. As a result, our model achieves significantly faster convergence and improved training efficiency, while maintaining performance on par with the baseline. These results highlight the potential of nonlinear attention mechanisms to accelerate reinforcement learning without sacrificing effectiveness.
Authors: Wenhao Song, Xuan Wu, Bo Yang, You Zhou, Yubin Xiao, Yanchun Liang, Hongwei Ge, Heow Pueh Lee, Chunguo Wu
Abstract: To address the weight coupling problem, certain studies introduced few-shot Neural Architecture Search (NAS) methods, which partition the supernet into multiple sub-supernets. However, these methods often suffer from computational inefficiency and tend to provide suboptimal partitioning schemes. To address this problem more effectively, we analyze the weight coupling problem from a novel perspective, which primarily stems from distinct modules in succeeding layers imposing conflicting gradient directions on the preceding layer modules. Based on this perspective, we propose the Gradient Contribution (GC) method that efficiently computes the cosine similarity of gradient directions among modules by decomposing the Vector-Jacobian Product during supernet backpropagation. Subsequently, the modules with conflicting gradient directions are allocated to distinct sub-supernets while similar ones are grouped together. To assess the advantages of GC and address the limitations of existing Graph Neural Architecture Search methods, which are limited to searching a single type of Graph Neural Networks (Message Passing Neural Networks (MPNNs) or Graph Transformers (GTs)), we propose the Unified Graph Neural Architecture Search (UGAS) framework, which explores optimal combinations of MPNNs and GTs. The experimental results demonstrate that GC achieves state-of-the-art (SOTA) performance in supernet partitioning quality and time efficiency. In addition, the architectures searched by UGAS+GC outperform both the manually designed GNNs and those obtained by existing NAS methods. Finally, ablation studies further demonstrate the effectiveness of all proposed methods.
Authors: Sujeevan Aseervatham, Achraf Kerzazi, Youn\`es Bennani
Abstract: Backdoor attacks represent a subtle yet effective class of cyberattacks targeting AI models, primarily due to their stealthy nature. The model behaves normally on clean data but exhibits malicious behavior only when the attacker embeds a specific trigger into the input. This attack is performed during the training phase, where the adversary corrupts a small subset of the training data by embedding a pattern and modifying the labels to a chosen target. The objective is to make the model associate the pattern with the target label while maintaining normal performance on unaltered data. Several defense mechanisms have been proposed to sanitize training data-sets. However, these methods often rely on the availability of a clean dataset to compute statistical anomalies, which may not always be feasible in real-world scenarios where datasets can be unavailable or compromised. To address this limitation, we propose a novel defense method that trains a model on the given dataset, detects poisoned classes, and extracts the critical part of the attack trigger before identifying the poisoned instances. This approach enhances explainability by explicitly revealing the harmful part of the trigger. The effectiveness of our method is demonstrated through experimental evaluations on well-known image datasets and comparative analysis against three state-of-the-art algorithms: SCAn, ABL, and AGPD.
Authors: Andre He, Daniel Fried, Sean Welleck
Abstract: Reinforcement learning is emerging as a primary driver for improving language model reasoning capabilities. A fundamental question is whether current reinforcement learning algorithms -- such as Group Relative Policy Optimization (GRPO), the de facto standard algorithm used to improve language model reasoning -- merely sharpen the base model's distribution around problems it can already solve. We investigate this question in the context of formal theorem proving, which has access to a perfect verifier. We identify a degenerate rank bias in GRPO in which highly probable trajectories are reinforced and rare ones are neglected. This results in distribution sharpening: the model can solve some problems with fewer samples, but underperforms simply sampling more solutions from the original model. To overcome GRPO's rank bias we introduce unlikeliness reward, a simple method for explicitly up-weighting rare but correct solutions. We show that unlikeliness reward mitigates rank bias and improves pass@$N$ across a large range of $N$ in both synthetic and real theorem proving settings. We also uncover an unexpected link between rank bias and a seemingly mundane hyperparameter -- the number of updates per batch -- that leads to a second, complementary mitigation. We combine our insights into a revised GRPO training recipe for formal theorem proving, yielding an open pipeline that achieves competitive performance to DeepSeek-Prover-V1.5-RL on the miniF2F-test benchmark. We release our implementation at https://github.com/AndreHe02/rewarding-unlikely-release
URLs: https://github.com/AndreHe02/rewarding-unlikely-release
Authors: Pau Vilimelis Aceituno, Jack William Miller, Noah Marti, Youssef Farag, Victor Boussange
Abstract: When training autoregressive models to forecast dynamical systems, a critical question arises: how far into the future should the model be trained to predict? Too short a horizon may miss long-term trends, while too long a horizon can impede convergence due to accumulating prediction errors. In this work, we formalize this trade-off by analyzing how the geometry of the loss landscape depends on the training horizon. We prove that for chaotic systems, the loss landscape's roughness grows exponentially with the training horizon, while for limit cycles, it grows linearly, making long-horizon training inherently challenging. However, we also show that models trained on long horizons generalize well to short-term forecasts, whereas those trained on short horizons suffer exponentially (resp. linearly) worse long-term predictions in chaotic (resp. periodic) systems. We validate our theory through numerical experiments and discuss practical implications for selecting training horizons. Our results provide a principled foundation for hyperparameter optimization in autoregressive forecasting models.
Authors: Ranajoy Sadhukhan, Zhuoming Chen, Haizhong Zheng, Yang Zhou, Emma Strubell, Beidi Chen
Abstract: We rethink test-time scaling laws from a practical efficiency perspective, revealing that the effectiveness of smaller models is significantly overestimated. Prior work, grounded in compute-optimality, overlooks critical memory access bottlenecks introduced by inference-time strategies (e.g., Best-of-$N$, long CoTs). Our holistic analysis, spanning models from 0.6B to 32B parameters, reveals a new Kinetics Scaling Law that better guides resource allocation by incorporating both computation and memory access costs. Kinetics Scaling Law suggests that test-time compute is more effective when used on models above a threshold than smaller ones. A key reason is that in TTS, attention, rather than parameter count, emerges as the dominant cost factor. Motivated by this, we propose a new scaling paradigm centered on sparse attention, which lowers per-token cost and enables longer generations and more parallel samples within the same resource budget. Empirically, we show that sparse attention models consistently outperform dense counterparts, achieving over 60 points gains in low-cost regimes and over 5 points gains in high-cost regimes for problem-solving accuracy on AIME, encompassing evaluations on state-of-the-art MoEs. These results suggest that sparse attention is essential and increasingly important with more computing invested, for realizing the full potential of test-time scaling where, unlike training, accuracy has yet to saturate as a function of computation, and continues to improve through increased generation. The code is available at https://github.com/Infini-AI-Lab/Kinetics.
Authors: Ziheng Qin, Hailun Xu, Wei Chee Yew, Qi Jia, Yang Luo, Kanchan Sarkar, Danhui Guan, Kai Wang, Yang You
Abstract: Machine learning relies heavily on data, yet the continuous growth of real-world data poses challenges for efficient dataset construction and training. A fundamental yet unsolved question is: given our current model and data, does a new data (sample/batch) need annotation/learning? Conventional approaches retain all available data, leading to non-optimal data and training efficiency. Active learning aims to reduce data redundancy by selecting a subset of samples to annotate, while it increases pipeline complexity and introduces bias. In this work, we propose Info-Coevolution, a novel framework that efficiently enables models and data to coevolve through online selective annotation with no bias. Leveraging task-specific models (and open-source models), it selectively annotates and integrates online and web data to improve datasets efficiently. For real-world datasets like ImageNet-1K, Info-Coevolution reduces annotation and training costs by 32\% without performance loss. It is able to automatically give the saving ratio without tuning the ratio. It can further reduce the annotation ratio to 50\% with semi-supervised learning. We also explore retrieval-based dataset enhancement using unlabeled open-source data. Code is available at https://github.com/NUS-HPC-AI-Lab/Info-Coevolution/.
Authors: Sebastian Schmidt, Prasanga Dhungel, Christoffer L\"offler, Bj\"orn Nieth, Stephan G\"unnemann, Leo Schwinn
Abstract: Training advanced machine learning models demands massive datasets, resulting in prohibitive computational costs. To address this challenge, data pruning techniques identify and remove redundant training samples while preserving model performance. Yet, existing pruning techniques predominantly require a full initial training pass to identify removable samples, negating any efficiency benefits for single training runs. To overcome this limitation, we introduce a novel importance score extrapolation framework that requires training on only a small subset of data. We present two initial approaches in this framework - k-nearest neighbors and graph neural networks - to accurately predict sample importance for the entire dataset using patterns learned from this minimal subset. We demonstrate the effectiveness of our approach for 2 state-of-the-art pruning methods (Dynamic Uncertainty and TDDS), 4 different datasets (CIFAR-10, CIFAR-100, Places-365, and ImageNet), and 3 training paradigms (supervised, unsupervised, and adversarial). Our results indicate that score extrapolation is a promising direction to scale expensive score calculation methods, such as pruning, data attribution, or other tasks.
Authors: Shikun Li, Shipeng Li, Zhiqin Yang, Xinghua Zhang, Gaode Chen, Xiaobo Xia, Hengyu Liu, Zhe Peng
Abstract: Reinforcement learning (RL) has become a key technique for enhancing LLMs' reasoning abilities, yet its data inefficiency remains a major bottleneck. To address this critical yet challenging issue, we present a novel gradient-alignment-based method, named LearnAlign, which intelligently selects the learnable and representative training reasoning data for RL post-training. To overcome the issue of response-length bias in gradient norms, we introduce the data learnability based on the success rate, which can indicate the learning potential of each data point. Experiments across three mathematical reasoning benchmarks demonstrate that our method significantly reduces training data requirements while achieving minor performance degradation or even improving performance compared to full-data training. For example, it reduces data requirements by up to 1,000 data points with better performance (77.53%) than that on the full dataset on GSM8K benchmark (77.04%). Furthermore, we show its effectiveness in the staged RL setting. This work provides valuable insights into data-efficient RL post-training and establishes a foundation for future research in optimizing reasoning data selection. To facilitate future work, we will release code.
Authors: Yaning Jia, Shenyang Deng, Chiyu Ma, Yaoqing Yang, Soroush Vosoughi
Abstract: Graph Neural Networks (GNNs) have achieved impressive success across a wide range of graph-based tasks, yet they remain highly vulnerable to small, imperceptible perturbations and adversarial attacks. Although numerous defense methods have been proposed to address these vulnerabilities, many rely on heuristic metrics, overfit to specific attack patterns, and suffer from high computational complexity. In this paper, we propose Kernel Complexity-Based Edge Sanitization (KCES), a training-free, model-agnostic defense framework. KCES leverages Graph Kernel Complexity (GKC), a novel metric derived from the graph's Gram matrix that characterizes GNN generalization via its test error bound. Building on GKC, we define a KC score for each edge, measuring the change in GKC when the edge is removed. Edges with high KC scores, typically introduced by adversarial perturbations, are pruned to mitigate their harmful effects, thereby enhancing GNNs' robustness. KCES can also be seamlessly integrated with existing defense strategies as a plug-and-play module without requiring training. Theoretical analysis and extensive experiments demonstrate that KCES consistently enhances GNN robustness, outperforms state-of-the-art baselines, and amplifies the effectiveness of existing defenses, offering a principled and efficient solution for securing GNNs.
Authors: Anna Soligo, Edward Turner, Senthooran Rajamanoharan, Neel Nanda
Abstract: Fine-tuning large language models on narrow datasets can cause them to develop broadly misaligned behaviours: a phenomena known as emergent misalignment. However, the mechanisms underlying this misalignment, and why it generalizes beyond the training domain, are poorly understood, demonstrating critical gaps in our knowledge of model alignment. In this work, we train and study a minimal model organism which uses just 9 rank-1 adapters to emergently misalign Qwen2.5-14B-Instruct. Studying this, we find that different emergently misaligned models converge to similar representations of misalignment. We demonstrate this convergence by extracting a 'misalignment direction' from one fine-tuned model's activations, and using it to effectively ablate misaligned behaviour from fine-tunes using higher dimensional LoRAs and different datasets. Leveraging the scalar hidden state of rank-1 LoRAs, we further present a set of experiments for directly interpreting the fine-tuning adapters, showing that six contribute to general misalignment, while two specialise for misalignment in just the fine-tuning domain. Emergent misalignment is a particularly salient example of undesirable and unexpected model behaviour and by advancing our understanding of the mechanisms behind it, we hope to move towards being able to better understand and mitigate misalignment more generally.
Authors: Dylan Kline
Abstract: This study bridges cognitive science and neural network design by examining whether artificial models exhibit human-like forgetting curves. Drawing upon Ebbinghaus' seminal work on memory decay and principles of spaced repetition, we propose a quantitative framework to measure information retention in neural networks. Our approach computes the recall probability by evaluating the similarity between a network's current hidden state and previously stored prototype representations. This retention metric facilitates the scheduling of review sessions, thereby mitigating catastrophic forgetting during deployment and enhancing training efficiency by prompting targeted reviews. Our experiments with Multi-Layer Perceptrons reveal human-like forgetting curves, with knowledge becoming increasingly robust through scheduled reviews. This alignment between neural network forgetting curves and established human memory models identifies neural networks as an architecture that naturally emulates human memory decay and can inform state-of-the-art continual learning algorithms.
Authors: Yanting Miao, William Loh, Suraj Kothawade, Pacal Poupart
Abstract: Recent work uses reinforcement learning (RL) to fine-tune text-to-image diffusion models, improving text-image alignment and sample quality. However, existing approaches introduce unnecessary complexity: they cache the full sampling trajectory, depend on differentiable reward models or large preference datasets, or require specialized guidance techniques. Motivated by the "golden noise" hypothesis -- that certain initial noise samples can consistently yield superior alignment -- we introduce Noise PPO, a minimalist RL algorithm that leaves the pre-trained diffusion model entirely frozen and learns a prompt-conditioned initial noise generator. Our approach requires no trajectory storage, reward backpropagation, or complex guidance tricks. Extensive experiments show that optimizing the initial noise distribution consistently improves alignment and sample quality over the original model, with the most significant gains at low inference steps. As the number of inference steps increases, the benefit of noise optimization diminishes but remains present. These findings clarify the scope and limitations of the golden noise hypothesis and reinforce the practical value of minimalist RL fine-tuning for diffusion models.
Authors: Hantao Yu, Josh Alman
Abstract: The quadratic complexity of self-attention prevents transformers from scaling effectively to long input sequences. On the other hand, modern GPUs and other specialized hardware accelerators are well-optimized for processing small input sequences in transformers during both training and inference. A natural question arises: can we take advantage of the efficiency of small transformers to deal with long input sequences? In this paper, we show that transformers with long input sequences (large transformers) can be efficiently simulated by transformers that can only take short input sequences (small transformers). Specifically, we prove that any transformer with input length $N$ can be efficiently simulated by only $O((N/M)^2)$ transformers with input length $M \ll N$, and that this cannot be improved in the worst case. However, we then prove that in various natural scenarios including average-case inputs, sliding window masking and attention sinks, the optimal number $O(N/M)$ of small transformers suffice.
Authors: Johnny Peng, Thanh Tung Khuat, Katarzyna Musial, Bogdan Gabrys
Abstract: Data is crucial for machine learning (ML) applications, yet acquiring large datasets can be costly and time-consuming, especially in complex, resource-intensive fields like biopharmaceuticals. A key process in this industry is upstream bioprocessing, where living cells are cultivated and optimised to produce therapeutic proteins and biologics. The intricate nature of these processes, combined with high resource demands, often limits data collection, resulting in smaller datasets. This comprehensive review explores ML methods designed to address the challenges posed by small data and classifies them into a taxonomy to guide practical applications. Furthermore, each method in the taxonomy was thoroughly analysed, with a detailed discussion of its core concepts and an evaluation of its effectiveness in tackling small data challenges, as demonstrated by application results in the upstream bioprocessing and other related domains. By analysing how these methods tackle small data challenges from different perspectives, this review provides actionable insights, identifies current research gaps, and offers guidance for leveraging ML in data-constrained environments.
Authors: Khizar Anjum, Muhammad Arbab Arshad, Kadhim Hayawi, Efstathios Polyzos, Asadullah Tariq, Mohamed Adel Serhani, Laiba Batool, Brady Lund, Nishith Reddy Mannuru, Ravi Varma Kumar Bevara, Taslim Mahbub, Muhammad Zeeshan Akram, Sakib Shahriar
Abstract: Large language models (LLMs) are increasingly being deployed across disciplines due to their advanced reasoning and problem solving capabilities. To measure their effectiveness, various benchmarks have been developed that measure aspects of LLM reasoning, comprehension, and problem-solving. While several surveys address LLM evaluation and benchmarks, a domain-specific analysis remains underexplored in the literature. This paper introduces a taxonomy of seven key disciplines, encompassing various domains and application areas where LLMs are extensively utilized. Additionally, we provide a comprehensive review of LLM benchmarks and survey papers within each domain, highlighting the unique capabilities of LLMs and the challenges faced in their application. Finally, we compile and categorize these benchmarks by domain to create an accessible resource for researchers, aiming to pave the way for advancements toward artificial general intelligence (AGI)
Authors: Haiyang Guo, Fanhu Zeng, Fei Zhu, Jiayi Wang, Xukai Wang, Jingang Zhou, Hongbo Zhao, Wenzhuo Liu, Shijie Ma, Da-Han Wang, Xu-Yao Zhang, Cheng-Lin Liu
Abstract: The rapid advancement of generative models has enabled modern AI systems to comprehend and produce highly sophisticated content, even achieving human-level performance in specific domains. However, these models remain fundamentally constrained by catastrophic forgetting - a persistent challenge where adapting to new tasks typically leads to significant degradation in performance on previously learned tasks. To address this practical limitation, numerous approaches have been proposed to enhance the adaptability and scalability of generative models in real-world applications. In this work, we present a comprehensive survey of continual learning methods for mainstream generative models, including large language models, multimodal large language models, vision language action models, and diffusion models. Drawing inspiration from the memory mechanisms of the human brain, we systematically categorize these approaches into three paradigms: architecture-based, regularization-based, and replay-based methods, while elucidating their underlying methodologies and motivations. We further analyze continual learning setups for different generative models, including training objectives, benchmarks, and core backbones, offering deeper insights into the field. The project page of this paper is available at https://github.com/Ghy0501/Awesome-Continual-Learning-in-Generative-Models.
URLs: https://github.com/Ghy0501/Awesome-Continual-Learning-in-Generative-Models.
Authors: Kai Tang, Ji Zhang, Hua Meng, Minbo Ma, Qi Xiong, Fengmao Lv, Jie Xu, Tianrui Li
Abstract: Multivariate time series forecasting (MTSF) is a critical task with broad applications in domains such as meteorology, transportation, and economics. Nevertheless, pervasive missing values caused by sensor failures or human errors significantly degrade forecasting accuracy. Prior efforts usually employ an impute-then-forecast paradigm, leading to suboptimal predictions due to error accumulation and misaligned objectives between the two stages. To address this challenge, we propose the Collaborative Imputation-Forecasting Network (CoIFNet), a novel framework that unifies imputation and forecasting to achieve robust MTSF in the presence of missing values. Specifically, CoIFNet takes the observed values, mask matrix and timestamp embeddings as input, processing them sequentially through the Cross-Timestep Fusion (CTF) and Cross-Variate Fusion (CVF) modules to capture temporal dependencies that are robust to missing values. We provide theoretical justifications on how our CoIFNet learning objective improves the performance bound of MTSF with missing values. Through extensive experiments on challenging MSTF benchmarks, we demonstrate the effectiveness and computational efficiency of our proposed approach across diverse missing-data scenarios, e.g., CoIFNet outperforms the state-of-the-art method by $\underline{\textbf{24.40}}$% ($\underline{\textbf{23.81}}$%) at a point (block) missing rate of 0.6, while improving memory and time efficiency by $\underline{\boldsymbol{4.3\times}}$ and $\underline{\boldsymbol{2.1\times}}$, respectively. Our code is available at: https://github.com/KaiTang-eng/CoIFNet.
Authors: Hen Davidov, Gilad Freidkin, Shai Feldman, Yaniv Romano
Abstract: We develop a framework to quantify the time-to-unsafe-sampling - the number of large language model (LLM) generations required to trigger an unsafe (e.g., toxic) response. Estimating this quantity is challenging, since unsafe responses are exceedingly rare in well-aligned LLMs, potentially occurring only once in thousands of generations. As a result, directly estimating time-to-unsafe-sampling would require collecting training data with a prohibitively large number of generations per prompt. However, with realistic sampling budgets, we often cannot generate enough responses to observe an unsafe outcome for every prompt, leaving the time-to-unsafe-sampling unobserved in many cases, making the estimation and evaluation tasks particularly challenging. To address this, we frame this estimation problem as one of survival analysis and develop a provably calibrated lower predictive bound (LPB) on the time-to-unsafe-sampling of a given prompt, leveraging recent advances in conformal prediction. Our key innovation is designing an adaptive, per-prompt sampling strategy, formulated as a convex optimization problem. The objective function guiding this optimized sampling allocation is designed to reduce the variance of the estimators used to construct the LPB, leading to improved statistical efficiency over naive methods that use a fixed sampling budget per prompt. Experiments on both synthetic and real data support our theoretical results and demonstrate the practical utility of our method for safety risk assessment in generative AI models.
Authors: Vaskar Nath, Elaine Lau, Anisha Gunjal, Manasi Sharma, Nikhil Baharte, Sean Hendryx
Abstract: We study the process through which reasoning models trained with reinforcement learning on verifiable rewards (RLVR) can learn to solve new problems. We find that RLVR drives performance in two main ways: (1) by compressing pass@$k$ into pass@1 and (2) via "capability gain" in which models learn to solve new problems that they previously could not solve even at high $k$. We find that while capability gain exists across model scales, learning to solve new problems is primarily driven through self-distillation. We demonstrate these findings across model scales ranging from 0.5B to 72B parameters on >500,000 reasoning problems with prompts and verifiable final answers across math, science, and code domains. We further show that we can significantly improve pass@$k$ rates by leveraging natural language guidance for the model to consider within context while still requiring the model to derive a solution chain from scratch. Based of these insights, we derive $\text{Guide}$ -- a new class of online training algorithms. $\text{Guide}$ adaptively incorporates hints into the model's context on problems for which all rollouts were initially incorrect and adjusts the importance sampling ratio for the "off-policy" trajectories in order to optimize the policy for contexts in which the hints are no longer present. We describe variants of $\text{Guide}$ for GRPO and PPO and empirically show that Guide-GRPO on 7B and 32B parameter models improves generalization over its vanilla counterpart with up to 4$\%$ macro-average improvement across math benchmarks. We include careful ablations to analyze $\text{Guide}$'s components and theoretically analyze Guide's learning efficiency.
Authors: Md Abrar Jahin, Adiba Abid, M. F. Mridha
Abstract: Expert systems often operate in domains characterized by class-imbalanced tabular data, where detecting rare but critical instances is essential for safety and reliability. While conventional approaches, such as cost-sensitive learning, oversampling, and graph neural networks, provide partial solutions, they suffer from drawbacks like overfitting, label noise, and poor generalization in low-density regions. To address these challenges, we propose QCL-MixNet, a novel Quantum-Informed Contrastive Learning framework augmented with k-nearest neighbor (kNN) guided dynamic mixup for robust classification under imbalance. QCL-MixNet integrates three core innovations: (i) a Quantum Entanglement-inspired layer that models complex feature interactions through sinusoidal transformations and gated attention, (ii) a sample-aware mixup strategy that adaptively interpolates feature representations of semantically similar instances to enhance minority class representation, and (iii) a hybrid loss function that unifies focal reweighting, supervised contrastive learning, triplet margin loss, and variance regularization to improve both intra-class compactness and inter-class separability. Extensive experiments on 18 real-world imbalanced datasets (binary and multi-class) demonstrate that QCL-MixNet consistently outperforms 20 state-of-the-art machine learning, deep learning, and GNN-based baselines in macro-F1 and recall, often by substantial margins. Ablation studies further validate the critical role of each architectural component. Our results establish QCL-MixNet as a new benchmark for tabular imbalance handling in expert systems. Theoretical analyses reinforce its expressiveness, generalization, and optimization robustness.
Authors: Joseph Geraci, Bessi Qorri, Christian Cumbaa, Mike Tsay, Paul Leonczyk, Luca Pani
Abstract: Artificial intelligence (AI) has evolved into an ecosystem of specialized "species," each with unique strengths. We analyze two: DeepSeek-V3, a 671-billion-parameter Mixture of Experts large language model (LLM) exemplifying scale-driven generality, and NetraAI, a dynamical system-based framework engineered for stability and interpretability on small clinical trial datasets. We formalize NetraAI's foundations, combining contraction mappings, information geometry, and evolutionary algorithms to identify predictive patient cohorts. Features are embedded in a metric space and iteratively contracted toward stable attractors that define latent subgroups. A pseudo-temporal embedding and long-range memory enable exploration of higher-order feature interactions, while an internal evolutionary loop selects compact, explainable 2-4-variable bundles ("Personas"). To guide discovery, we introduce an LLM Strategist as a meta-evolutionary layer that observes Persona outputs, prioritizes promising variables, injects domain knowledge, and assesses robustness. This two-tier architecture mirrors the human scientific process: NetraAI as experimentalist, the LLM as theorist, forming a self-improving loop. In case studies (schizophrenia, depression, pancreatic cancer), NetraAI uncovered small, high-effect-size subpopulations that transformed weak baseline models (AUC ~0.50-0.68) into near-perfect classifiers using only a few features. We position NetraAI at the intersection of dynamical systems, information geometry, and evolutionary learning, aligned with emerging concept-level reasoning paradigms such as LeCun's Joint Embedding Predictive Architecture (JEPA). By prioritizing reliable, explainable knowledge, NetraAI offers a new generation of adaptive, self-reflective AI to accelerate clinical discovery.
Authors: Naoki Matsumura, Yuta Yoshimoto, Yuto Iwasaki, Meguru Yamazaki, Yasufumi Sakai
Abstract: Neural network potentials (NNPs) offer a powerful alternative to traditional force fields for molecular dynamics (MD) simulations. Accurate and stable MD simulations, crucial for evaluating material properties, require training data encompassing both low-energy stable structures and high-energy structures. Conventional knowledge distillation (KD) methods fine-tune a pre-trained NNP as a teacher model to generate training data for a student model. However, in material-specific models, this fine-tuning process increases energy barriers, making it difficult to create training data containing high-energy structures. To address this, we propose a novel KD framework that leverages a non-fine-tuned, off-the-shelf pre-trained NNP as a teacher. Its gentler energy landscape facilitates the exploration of a wider range of structures, including the high-energy structures crucial for stable MD simulations. Our framework employs a two-stage training process: first, the student NNP is trained with a dataset generated by the off-the-shelf teacher; then, it is fine-tuned with a smaller, high-accuracy density functional theory (DFT) dataset. We demonstrate the effectiveness of our framework by applying it to both organic (polyethylene glycol) and inorganic (L$_{10}$GeP$_{2}$S$_{12}$) materials, achieving comparable or superior accuracy in reproducing physical properties compared to existing methods. Importantly, our method reduces the number of expensive DFT calculations by 10x compared to existing NNP generation methods, without sacrificing accuracy. Furthermore, the resulting student NNP achieves up to 106x speedup in inference compared to the teacher NNP, enabling significantly faster and more efficient MD simulations.
Authors: Laura Kopf, Nils Feldhus, Kirill Bykov, Philine Lou Bommer, Anna Hedstr\"om, Marina M. -C. H\"ohne, Oliver Eberle
Abstract: Automated interpretability research aims to identify concepts encoded in neural network features to enhance human understanding of model behavior. Current feature description methods face two critical challenges: limited robustness and the flawed assumption that each neuron encodes only a single concept (monosemanticity), despite growing evidence that neurons are often polysemantic. This assumption restricts the expressiveness of feature descriptions and limits their ability to capture the full range of behaviors encoded in model internals. To address this, we introduce Polysemantic FeatuRe Identification and Scoring Method (PRISM), a novel framework that captures the inherent complexity of neural network features. Unlike prior approaches that assign a single description per feature, PRISM provides more nuanced descriptions for both polysemantic and monosemantic features. We apply PRISM to language models and, through extensive benchmarking against existing methods, demonstrate that our approach produces more accurate and faithful feature descriptions, improving both overall description quality (via a description score) and the ability to capture distinct concepts when polysemanticity is present (via a polysemanticity score).
Authors: Vincent Roca, Marc Tommasi, Paul Andrey, Aur\'elien Bellet, Markus D. Schirmer, Hilde Henon, Laurent Puy, Julien Ramon, Gr\'egory Kuchcinski, Martin Bretzner, Renaud Lopes
Abstract: $\textbf{Objective:}$ Brain-predicted age difference (BrainAGE) is a neuroimaging biomarker reflecting brain health. However, training robust BrainAGE models requires large datasets, often restricted by privacy concerns. This study evaluates the performance of federated learning (FL) for BrainAGE estimation in ischemic stroke patients treated with mechanical thrombectomy, and investigates its association with clinical phenotypes and functional outcomes. $\textbf{Methods:}$ We used FLAIR brain images from 1674 stroke patients across 16 hospital centers. We implemented standard machine learning and deep learning models for BrainAGE estimates under three data management strategies: centralized learning (pooled data), FL (local training at each site), and single-site learning. We reported prediction errors and examined associations between BrainAGE and vascular risk factors (e.g., diabetes mellitus, hypertension, smoking), as well as functional outcomes at three months post-stroke. Logistic regression evaluated BrainAGE's predictive value for these outcomes, adjusting for age, sex, vascular risk factors, stroke severity, time between MRI and arterial puncture, prior intravenous thrombolysis, and recanalisation outcome. $\textbf{Results:}$ While centralized learning yielded the most accurate predictions, FL consistently outperformed single-site models. BrainAGE was significantly higher in patients with diabetes mellitus across all models. Comparisons between patients with good and poor functional outcomes, and multivariate predictions of these outcomes showed the significance of the association between BrainAGE and post-stroke recovery. $\textbf{Conclusion:}$ FL enables accurate age predictions without data centralization. The strong association between BrainAGE, vascular risk factors, and post-stroke recovery highlights its potential for prognostic modeling in stroke care.
Authors: Yiwen Ding, Jiarui Liu, Zhiheng Lyu, Kun Zhang, Bernhard Schoelkopf, Zhijing Jin, Rada Mihalcea
Abstract: While several previous studies have analyzed gender bias in research, we are still missing a comprehensive analysis of gender differences in the AI community, covering diverse topics and different development trends. Using the AI Scholar dataset of 78K researchers in the field of AI, we identify several gender differences: (1) Although female researchers tend to have fewer overall citations than males, this citation difference does not hold for all academic-age groups; (2) There exist large gender homophily in co-authorship on AI papers; (3) Female first-authored papers show distinct linguistic styles, such as longer text, more positive emotion words, and more catchy titles than male first-authored papers. Our analysis provides a window into the current demographic trends in our AI community, and encourages more gender equality and diversity in the future. Our code and data are at https://github.com/causalNLP/ai-scholar-gender.
Authors: Helton Graziadei, Paulo C. Marques F., Eduardo F. L. de Melo, Rodrigo S. Targino
Abstract: We present a model-agnostic framework for the construction of prediction intervals of insurance claims, with finite sample statistical guarantees, extending the technique of split conformal prediction to the domain of two-stage frequency-severity modeling. The framework effectiveness is showcased with simulated and real datasets using classical parametric models and contemporary machine learning methods. When the underlying severity model is a random forest, we extend the two-stage split conformal prediction algorithm, showing how the out-of-bag mechanism can be leveraged to eliminate the need for a calibration set in the conformal procedure.
Authors: John Y. Zhu
Abstract: I study statistical discrimination driven by verifiable beliefs, such as those generated by machine learning, rather than by humans. When beliefs are verifiable, interventions against statistical discrimination can move beyond simple, belief-free designs like affirmative action, to more sophisticated ones, that constrain decision makers based on what they are thinking. I design a belief-contingent intervention I call common identity. I show that it is effective at eliminating equilibrium statistical discrimination, even when training data exhibit the various statistical biases that often plague algorithmic decision problems.
Authors: Shaobo Xia, Jun Yue, Kacper Kania, Leyuan Fang, Andrea Tagliasacchi, Kwang Moo Yi, Weiwei Sun
Abstract: We propose a weakly supervised semantic segmentation method for point clouds that predicts "per-point" labels from just "whole-scene" annotations. The key challenge here is the discrepancy between the target of dense per-point semantic prediction and training losses derived from only scene-level labels. To address this, in addition to the typical weakly-supervised setup that supervises all points with the scene label, we propose to conservatively propagate the scene-level labels to points selectively. Specifically, we over-segment point cloud features via unsupervised clustering in the entire dataset and form primitives. We then associate scene-level labels with primitives through bipartite matching. Then, we allow labels to pass through this primitive-label relationship, while further encouraging features to form narrow clusters around the primitives. Importantly, through bipartite matching, this additional pathway through which labels flow, only propagates scene labels to the most relevant points, reducing the potential negative impact caused by the global approach that existing methods take. We evaluate our method on ScanNet and S3DIS datasets, outperforming the state of the art by a large margin.
Authors: Manish Prajapat, Johannes K\"ohler, Matteo Turchetta, Andreas Krause, Melanie N. Zeilinger
Abstract: Safely exploring environments with a-priori unknown constraints is a fundamental challenge that restricts the autonomy of robots. While safety is paramount, guarantees on sufficient exploration are also crucial for ensuring autonomous task completion. To address these challenges, we propose a novel safe guaranteed exploration framework using optimal control, which achieves first-of-its-kind results: guaranteed exploration for non-linear systems with finite time sample complexity bounds, while being provably safe with arbitrarily high probability. The framework is general and applicable to many real-world scenarios with complex non-linear dynamics and unknown domains. We improve the efficiency of this general framework by proposing an algorithm, SageMPC, SAfe Guaranteed Exploration using Model Predictive Control. SageMPC leverages three key techniques: i) exploiting a Lipschitz bound, ii) goal-directed exploration, and iii) receding horizon style re-planning, all while maintaining the desired sample complexity, safety and exploration guarantees of the framework. Lastly, we demonstrate safe efficient exploration in challenging unknown environments using SageMPC with a car model.
Authors: Siwei Yang, Bingchen Zhao, Cihang Xie
Abstract: This paper introduces AQA-Bench, a novel benchmark to assess the sequential reasoning capabilities of large language models (LLMs) in algorithmic contexts, such as depth-first search (DFS). The key feature of our evaluation benchmark lies in its interactive evaluation protocol - for example, in DFS, the availability of each node's connected edge is contingent upon the model's traversal to that node, thereby necessitating the LLM's ability to effectively remember visited nodes and strategize subsequent moves considering the possible environmental feedback in the future steps. We comprehensively build AQA-Bench with three different algorithms, namely binary search, depth-first search, and breadth-first search, and to evaluate the sequential reasoning ability of 14 different LLMs. Our investigations reveal several interesting findings: (1) Closed-source models like GPT-4 and Gemini generally show much stronger sequential reasoning ability, significantly outperforming open-source LLMs. (2) Naively providing in-context examples may inadvertently hurt few-shot performance in an interactive environment due to over-fitting to examples. (3) Instead of using optimal steps from another test case as the in-context example, a very limited number of predecessor steps in the current test case following the optimal policy can substantially boost small models' performance. (4) The performance gap between weak models and strong models is greatly due to the incapability of weak models to start well. (5) The scaling correlation between performance and model size is not always significant, sometimes even showcasing an inverse trend. We hope our study can catalyze future work on advancing the understanding and enhancement of LLMs' capabilities in sequential reasoning. The code is available at https://github.com/UCSC-VLAA/AQA-Bench.
Authors: Zhe Jiao, Wantao Jia, Weiqiu Zhu
Abstract: The aim of this work is to develop a deep learning method for solving high-dimensional stochastic control problems based on the Hamilton--Jacobi--Bellman (HJB) equation and physics-informed learning. Our approach is to parameterize the feedback control and the value function using a decoupled neural network with multiple outputs. We train this network by using a loss function with penalty terms that enforce the HJB equation along the sampled trajectories generated by the controlled system. More significantly, numerical results on various applications are carried out to demonstrate that the proposed approach is efficient and applicable.
Authors: Kyla H. Levin, Nicolas van Kempen, Emery D. Berger, Stephen N. Freund
Abstract: Debugging is a critical but challenging task for programmers. This paper proposes ChatDBG, an AI-powered debugging assistant. ChatDBG integrates large language models (LLMs) to significantly enhance the capabilities and user-friendliness of conventional debuggers. ChatDBG lets programmers engage in a collaborative dialogue with the debugger, allowing them to pose complex questions about program state, perform root cause analysis for crashes or assertion failures, and explore open-ended queries like "why is x null?". To handle these queries, ChatDBG grants the LLM autonomy to "take the wheel": it can act as an independent agent capable of querying and controlling the debugger to navigate through stacks and inspect program state. It then reports its findings and yields back control to the programmer. By leveraging the real-world knowledge embedded in LLMs, ChatDBG can diagnose issues identifiable only through the use of domain-specific reasoning. Our ChatDBG prototype integrates with standard debuggers including LLDB and GDB for native code and Pdb for Python. Our evaluation across a diverse set of code, including C/C++ code with known bugs and a suite of Python code including standalone scripts and Jupyter notebooks, demonstrates that ChatDBG can successfully analyze root causes, explain bugs, and generate accurate fixes for a wide range of real-world errors. For the Python programs, a single query led to an actionable bug fix 67% of the time; one additional follow-up query increased the success rate to 85%. ChatDBG has seen rapid uptake; it has already been downloaded more than 75,000 times.
Authors: Qianglin Wen, Chengchun Shi, Ying Yang, Niansheng Tang, Hongtu Zhu
Abstract: A/B testing has become the gold standard for policy evaluation in modern technological industries. Motivated by the widespread use of switchback experiments in A/B testing, this paper conducts a comprehensive comparative analysis of various switchback designs in Markovian environments. Unlike many existing works which derive the optimal design based on specific and relatively simple estimators, our analysis covers a range of state-of-the-art estimators developed in the reinforcement learning (RL) literature. It reveals that the effectiveness of different switchback designs depends crucially on (i) the size of the carryover effect and (ii) the auto-correlations among reward errors over time. Meanwhile, these findings are estimator-agnostic, i.e., they apply to most RL estimators. Based on these insights, we provide a workflow to offer guidelines for practitioners on designing switchback experiments in A/B testing.
Authors: A. Martina Neuman, Dominik Dold, Philipp Christian Petersen
Abstract: We study the learning problem associated with spiking neural networks. Specifically, we focus on spiking neural networks composed of simple spiking neurons having only positive synaptic weights, equipped with an affine encoder and decoder; we refer to these as affine spiking neural networks. These neural networks are shown to depend continuously on their parameters, which facilitates classical covering number-based generalization statements and supports stable gradient-based training. We demonstrate that the positivity of the weights enables a wide range of expressivity results, including rate-optimal approximation of smooth functions and dimension-independent approximation of Barron regular functions. In particular, we show in theory and simulations that affine spiking neural networks are capable of approximating shallow ReLU neural networks. Furthermore, we apply these affine spiking neural networks to standard machine learning benchmarks and reach competitive results. Finally, we observe that from a generalization perspective, contrary to feedforward neural networks or previous results for general spiking neural networks, the depth has little to no adverse effect on the generalization capabilities.
Authors: Jun Huang, Yan Liu
Abstract: This paper proposes a new gradient-based XAI method called Guided AbsoluteGrad for saliency map explanations. We utilize both positive and negative gradient magnitudes and employ gradient variance to distinguish the important areas for noise deduction. We also introduce a novel evaluation metric named ReCover And Predict (RCAP), which considers the Localization and Visual Noise Level objectives of the explanations. We propose two propositions for these two objectives and prove the necessity of evaluating them. We evaluate Guided AbsoluteGrad with seven gradient-based XAI methods using the RCAP metric and other SOTA metrics in three case studies: (1) ImageNet dataset with ResNet50 model; (2) International Skin Imaging Collaboration (ISIC) dataset with EfficientNet model; (3) the Places365 dataset with DenseNet161 model. Our method surpasses other gradient-based approaches, showcasing the quality of enhanced saliency map explanations through gradient magnitude.
Authors: Tugba Torun, Ameer Taweel, Didem Unat
Abstract: Sparse tensor operations are increasingly important in diverse applications such as social networks, deep learning, diagnosis, crime, and review analysis. However, a major obstacle in sparse tensor research is the lack of large-scale sparse tensor datasets. Another challenge lies in analyzing sparse tensor features, which are essential not only for understanding the nonzero pattern but also for selecting the most suitable storage format, decomposition algorithm, and reordering methods. However, due to the large size of real-world tensors, even extracting these features can be computationally expensive without careful optimization. To address these limitations, we have developed a smart sparse tensor generator that replicates key characteristics of real sparse tensors. Additionally, we propose efficient methods for extracting a comprehensive set of sparse tensor features. The effectiveness of our generator is validated through the quality of extracted features and the performance of decomposition on the generated tensors. Both the sparse tensor feature extractor and the tensor generator are open source with all the artifacts available at https://github.com/sparcityeu/FeaTensor and https://github.com/sparcityeu/GenTensor, respectively.
URLs: https://github.com/sparcityeu/FeaTensor, https://github.com/sparcityeu/GenTensor,
Authors: Vincent Froese, Moritz Grillo, Martin Skutella
Abstract: Neural networks with ReLU activation play a key role in modern machine learning. Understanding the functions represented by ReLU networks is a major topic in current research as this enables a better interpretability of learning processes. Injectivity of a function computed by a ReLU network, that is, the question if different inputs to the network always lead to different outputs, plays a crucial role whenever invertibility of the function is required, such as, e.g., for inverse problems or generative models. The exact computational complexity of deciding injectivity was recently posed as an open problem (Puthawala et al. [JMLR 2022]). We answer this question by proving coNP-completeness. On the positive side, we show that the problem for a single ReLU-layer is still tractable for small input dimension; more precisely, we present a parameterized algorithm which yields fixed-parameter tractability with respect to the input dimension. In addition, we study the network verification problem which is to verify that certain inputs only yield specific outputs. This is of great importance since neural networks are increasingly used in safety-critical systems. We prove that network verification is coNP-hard for a general class of input domains. Our results also exclude constant-factor polynomial-time approximations for the maximum of a function computed by a ReLU network. In this context, we also characterize surjectivity of functions computed by ReLU networks with one-dimensional output which turns out to be the complement of a basic network verification task. We reveal interesting connections to computational convexity by formulating the surjectivity problem as a zonotope containment problem
Authors: Kangwei Qi, Qiong Wu, Pingyi Fan, Nan Cheng, Qiang Fan, Jiangzhou Wang
Abstract: Vehicular edge computing (VEC) is an emerging technology that enables vehicles to perform high-intensity tasks by executing tasks locally or offloading them to nearby edge devices. However, obstacles such as buildings may degrade the communications and incur communication interruptions, and thus the vehicle may not meet the requirement for task offloading. Reconfigurable intelligent surfaces (RIS) is introduced to support vehicle communication and provide an alternative communication path. The system performance can be improved by flexibly adjusting the phase-shift of the RIS. For RIS-assisted VEC system where tasks arrive randomly, we design a control scheme that considers offloading power, local power allocation and phase-shift optimization. To solve this non-convex problem, we propose a new deep reinforcement learning (DRL) framework that employs modified multi-agent deep deterministic policy gradient (MADDPG) approach to optimize the power allocation for vehicle users (VUs) and block coordinate descent (BCD) algorithm to optimize the phase-shift of the RIS. Simulation results show that our proposed scheme outperforms the centralized deep deterministic policy gradient (DDPG) scheme and random scheme.
Authors: Tuan-Luc Huynh, Thuy-Trang Vu, Weiqing Wang, Yinwei Wei, Trung Le, Dragan Gasevic, Yuan-Fang Li, Thanh-Toan Do
Abstract: Differentiable Search Index (DSI) utilizes pre-trained language models to perform indexing and document retrieval via end-to-end learning without relying on external indexes. However, DSI requires full re-training to index new documents, causing significant computational inefficiencies. Continual learning (CL) offers a solution by enabling the model to incrementally update without full re-training. Existing CL solutions in document retrieval rely on memory buffers or generative models for rehearsal, which is infeasible when accessing previous training data is restricted due to privacy concerns. To this end, we introduce PromptDSI, a prompt-based, rehearsal-free continual learning approach for document retrieval. PromptDSI follows the Prompt-based Continual Learning (PCL) framework, using learnable prompts to efficiently index new documents without accessing previous documents or queries. To improve retrieval latency, we remove the initial forward pass of PCL, which otherwise greatly increases training and inference time, with a negligible trade-off in performance. Additionally, we introduce a novel topic-aware prompt pool that employs neural topic embeddings as fixed keys, eliminating the instability of prompt key optimization while maintaining competitive performance with existing PCL prompt pools. In a challenging rehearsal-free continual learning setup, we demonstrate that PromptDSI variants outperform rehearsal-based baselines, match the strong cache-based baseline in mitigating forgetting, and significantly improving retrieval performance on new corpora.
Authors: Yunpeng Jiang, Yutong Ban, Paul Weng
Abstract: Data augmentation is widely applied and has shown its benefits in different machine learning tasks. However, as recently observed, it may have an unfair effect in multi-class classification. While data augmentation generally improves the overall performance (and therefore is beneficial for many classes), it can actually be detrimental for other classes, which can be problematic in some application domains. In this paper, to counteract this phenomenon, we propose CLAM, a CLAss-dependent Multiplicative-weights method. To derive it, we first formulate the training of a classifier as a non-linear optimization problem that aims at simultaneously maximizing the individual class performances and balancing them. By rewriting this optimization problem as an adversarial two-player game, we propose a novel multiplicative weight algorithm, for which we prove the convergence. Interestingly, our formulation also reveals that the class-dependent effects of data augmentation is not due to data augmentation only, but is in fact a general phenomenon. Our empirical results over six datasets demonstrate that the performance of learned classifiers is indeed more fairly distributed over classes, with only limited impact on the average accuracy.
Authors: Amit Attia, Ofir Gaash, Tomer Koren
Abstract: We consider the problem of asynchronous stochastic optimization, where an optimization algorithm makes updates based on stale stochastic gradients of the objective that are subject to an arbitrary (possibly adversarial) sequence of delays. We present a procedure which, for any given $q \in (0,1]$, transforms any standard stochastic first-order method to an asynchronous method with convergence guarantee depending on the $q$-quantile delay of the sequence. This approach leads to convergence rates of the form $O(\tau_q/qT+\sigma/\sqrt{qT})$ for non-convex and $O(\tau_q^2/(q T)^2+\sigma/\sqrt{qT})$ for convex smooth problems, where $\tau_q$ is the $q$-quantile delay, generalizing and improving on existing results that depend on the average delay. We further show a method that automatically adapts to all quantiles simultaneously, without any prior knowledge of the delays, achieving convergence rates of the form $O(\inf_{q} \tau_q/qT+\sigma/\sqrt{qT})$ for non-convex and $O(\inf_{q} \tau_q^2/(q T)^2+\sigma/\sqrt{qT})$ for convex smooth problems. Our technique is based on asynchronous mini-batching with a careful batch-size selection and filtering of stale gradients.
Authors: Junwei You, Haotian Shi, Zhuoyu Jiang, Zilin Huang, Rui Gan, Keshu Wu, Xi Cheng, Xiaopeng Li, Bin Ran
Abstract: Vehicle-to-everything (V2X) cooperation has emerged as a promising paradigm to overcome the perception limitations of classical autonomous driving by leveraging information from both ego-vehicle and infrastructure sensors. However, effectively fusing heterogeneous visual and semantic information while ensuring robust trajectory planning remains a significant challenge. This paper introduces V2X-VLM, a novel end-to-end (E2E) cooperative autonomous driving framework based on vision-language models (VLMs). V2X-VLM integrates multiperspective camera views from vehicles and infrastructure with text-based scene descriptions to enable a more comprehensive understanding of driving environments. Specifically, we propose a contrastive learning-based mechanism to reinforce the alignment of heterogeneous visual and textual characteristics, which enhances the semantic understanding of complex driving scenarios, and employ a knowledge distillation strategy to stabilize training. Experiments on a large real-world dataset demonstrate that V2X-VLM achieves state-of-the-art trajectory planning accuracy, significantly reducing L2 error and collision rate compared to existing cooperative autonomous driving baselines. Ablation studies validate the contributions of each component. Moreover, the evaluation of robustness and efficiency highlights the practicality of V2X-VLM for real-world deployment to enhance overall autonomous driving safety and decision-making.
Authors: Nicolas Yax, Pierre-Yves Oudeyer, Stefano Palminteri
Abstract: In machine learning, contamination refers to situations where testing data leak into the training set. The issue is particularly relevant for the evaluation of the performance of Large Language Models (LLMs), which are generally trained on gargantuan, and generally opaque, corpora of text scraped from the world wide web. Developing tools to detect contamination is therefore crucial to be able to fairly and properly track the evolution of the performance of LLMs. To date, only a few recent studies have attempted to address the issue of quantifying and detecting contamination in short text sequences, such as those commonly found in benchmarks. However, these methods have limitations that can sometimes render them impractical. In the present paper, we introduce LogProber, a novel, efficient algorithm that we show to be able to detect contamination in a black box setting that tries to tackle some of these drawbacks by focusing on the familiarity with the question rather than the answer. Here, we explore the properties of the proposed method in comparison with concurrent approaches, identify its advantages and limitations, and illustrate how different forms of contamination can go undetected depending on the design of the detection algorithm.
Authors: Syed Abdul Gaffar Shakhadri, Kruthika KR, Rakshit Aralimatti
Abstract: We introduce Shakti, a 2.5 billion parameter language model specifically optimized for resource-constrained environments such as edge devices, including smartphones, wearables, and IoT systems. Shakti combines high-performance NLP with optimized efficiency and precision, making it ideal for real-time AI applications where computational resources and memory are limited. With support for vernacular languages and domain-specific tasks, Shakti excels in industries such as healthcare, finance, and customer service. Benchmark evaluations demonstrate that Shakti performs competitively against larger models while maintaining low latency and on-device efficiency, positioning it as a leading solution for edge AI.
Authors: Yu-Neng Chuang, Prathusha Kameswara Sarma, Parikshit Gopalan, John Boccio, Sara Bolouki, Xia Hu, Helen Zhou
Abstract: Large language models (LLMs) have demonstrated impressive performance on several tasks and are increasingly deployed in real-world applications. However, especially in high-stakes settings, it becomes vital to know when the output of an LLM may be unreliable. Depending on whether an answer is trustworthy, a system can then choose to route the question to another expert, or otherwise fall back on a safe default behavior. In this work, we study the extent to which LLMs can reliably indicate confidence in their answers, and how this notion of confidence can translate into downstream accuracy gains. We propose Self-Reflection with Error-based Feedback (Self-REF), a lightweight training strategy to teach LLMs to express confidence in whether their answers are correct in a reliable manner. Self-REF introduces confidence tokens into the LLM, from which a confidence score can be extracted. Compared to conventional approaches such as verbalizing confidence and examining token probabilities, we demonstrate empirically that confidence tokens show significant improvements in downstream routing and rejection learning tasks.
Authors: Naitong Chen, Jonathan H. Huggins, Trevor Campbell
Abstract: A Bayesian coreset is a small, weighted subset of a data set that replaces the full data during inference to reduce computational cost. The state-of-the-art coreset construction algorithm, Coreset Markov chain Monte Carlo (Coreset MCMC), uses draws from an adaptive Markov chain targeting the coreset posterior to train the coreset weights via stochastic gradient optimization. However, the quality of the constructed coreset, and thus the quality of its posterior approximation, is sensitive to the stochastic optimization learning rate. In this work, we propose a learning-rate-free stochastic gradient optimization procedure, Hot-start Distance over Gradient (Hot DoG), for training coreset weights in Coreset MCMC without user tuning effort. We provide a theoretical analysis of the convergence of the coreset weights produced by Hot DoG. We also provide empirical results demonstrate that Hot DoG provides higher quality posterior approximations than other learning-rate-free stochastic gradient methods, and performs competitively to optimally-tuned ADAM.
Authors: Ali Reza Ibrahimzada, Kaiyao Ke, Mrigank Pawagi, Muhammad Salman Abid, Rangeet Pan, Saurabh Sinha, Reyhaneh Jabbarvand
Abstract: Code translation transforms programs from one programming language (PL) to another. Several rule-based transpilers have been designed to automate code translation between different pairs of PLs. However, the rules can become obsolete as the PLs evolve and cannot generalize to other PLs. Recent studies have explored the automation of code translation using Large Language Models (LLMs). One key observation is that such techniques may work well for crafted benchmarks but fail to generalize to the scale and complexity of real-world projects with dependencies, custom types, PL-specific features, etc. We propose AlphaTrans, a neuro-symbolic approach to automate repository-level code translation. AlphaTrans translates both source and test code, and employs multiple levels of validation to ensure the translation preserves the functionality of the source program. To break down the problem for LLMs, AlphaTrans leverages program analysis to decompose the program into fragments and translates them in the reverse call order. We leveraged AlphaTrans to translate ten real-world open-source projects consisting of <836, 8575, 2719> classes, methods, and tests. AlphaTrans breaks down these projects into 17874 fragments and translates the entire repository. 96.40% of the translated fragments are syntactically correct, and AlphaTrans validates the translations' runtime behavior and functional correctness for 27.03% and 25.14% of fragments. On average, the integrated translation and validation take 34 hours to translate a project, showing its scalability in practice. For the incorrect translations, AlphaTrans generates a report including existing translation, stack trace, test errors, or assertion failures. We provided these artifacts to two developers to fix the translation bugs in four projects. They were able to fix the issues in 20.1 hours on average and achieve all passing tests.
Authors: Shiwei Sun, Giovanni S. Alberti
Abstract: We consider the inverse problem consisting of the reconstruction of an inclusion $B$ contained in a bounded domain $\Omega\subset\mathbb{R}^d$ from a single pair of Cauchy data $(u|_{\partial\Omega},\partial_\nu u|_{\partial\Omega})$, where $\Delta u=0$ in $\Omega\setminus\overline B$ and $u=0$ on $\partial B$. We show that the reconstruction algorithm based on the range test, a domain sampling method, can be written as a neural network with a specific architecture. We propose to learn the weights of this network in the framework of supervised learning, and to combine it with a pre-trained classifier, with the purpose of distinguishing the inclusions based on their distance from the boundary. The numerical simulations show that this learned range test method provides accurate and stable reconstructions of polygonal inclusions. Furthermore, the results are superior to those obtained with the standard range test method (without learning) and with an end-to-end fully connected deep neural network, a purely data-driven method.
Authors: Kiran Doshi, Marco Bagatella, Stelian Coros
Abstract: The combination of behavioural cloning and neural networks has driven significant progress in robotic manipulation. As these algorithms may require a large number of demonstrations for each task of interest, they remain fundamentally inefficient in complex scenarios, in which finite datasets can hardly cover the state space. One of the remaining challenges is thus out-of-distribution (OOD) generalisation, i.e. the ability to predict correct actions for states with a low likelihood with respect to the state occupancy induced by the dataset. This issue is aggravated when the system to control is treated as a black-box, ignoring its physical properties. This work characterises widespread properties of robotic manipulation, specifically pose equivariance and locality. We investigate the effect of the choice of problem space on OOD performance of BC policies and how transformations arising from characteristic properties of manipulation could be employed for its improvement. We empirically demonstrate that these transformations allow behaviour cloning policies, using either standard MLP-based one-step action prediction or diffusion-based action-sequence prediction, to generalise better to OOD problem instances.
Authors: Yingying Fang, Zihao Jin, Shaojie Guo, Jinda Liu, Zhiling Yue, Yijian Gao, Junzhi Ning, Zhi Li, Simon Walsh, Guang Yang
Abstract: Despite significant advancements in automated report generation, the opaqueness of text interpretability continues to cast doubt on the reliability of the content produced. This paper introduces a novel approach to identify specific image features in X-ray images that influence the outputs of report generation models. Specifically, we propose Cyclic Vision-Language Manipulator CVLM, a module to generate a manipulated X-ray from an original X-ray and its report from a designated report generator. The essence of CVLM is that cycling manipulated X-rays to the report generator produces altered reports aligned with the alterations pre-injected into the reports for X-ray generation, achieving the term "cyclic manipulation". This process allows direct comparison between original and manipulated X-rays, clarifying the critical image features driving changes in reports and enabling model users to assess the reliability of the generated texts. Empirical evaluations demonstrate that CVLM can identify more precise and reliable features compared to existing explanation methods, significantly enhancing the transparency and applicability of AI-generated reports.
Authors: Takuya Ito, Murray Campbell, Lior Horesh, Tim Klinger, Parikshit Ram
Abstract: The rapid development of artificial intelligence (AI) systems has created an urgent need for their scientific quantification. While their fluency across a variety of domains is impressive, AI systems fall short on tests requiring algorithmic reasoning -- a glaring limitation given the necessity for interpretable and reliable technology. Despite a surge of reasoning benchmarks emerging from the academic community, no theoretical framework exists to quantify algorithmic reasoning in AI systems. Here, we adopt a framework from computational complexity theory to quantify algorithmic generalization using algebraic expressions: algebraic circuit complexity. Algebraic circuit complexity theory -- the study of algebraic expressions as circuit models -- is a natural framework to study the complexity of algorithmic computation. Algebraic circuit complexity enables the study of generalization by defining benchmarks in terms of the computational requirements to solve a problem. Moreover, algebraic circuits are generic mathematical objects; an arbitrarily large number of samples can be generated for a specified circuit, making it an ideal experimental sandbox for the data-hungry models that are used today. In this Perspective, we adopt tools from algebraic circuit complexity, apply them to formalize a science of algorithmic generalization, and address key challenges for its successful application to AI science.
Authors: Daixuan Cheng, Shaohan Huang, Ziyu Zhu, Xintong Zhang, Wayne Xin Zhao, Zhongzhi Luan, Bo Dai, Zhenliang Zhang
Abstract: Adapting general multimodal large language models (MLLMs) to specific domains, such as scientific and industrial fields, is highly significant in promoting their practical applications. This paper systematically investigates domain adaptation of MLLMs via post-training, focusing on data synthesis, training pipeline, and task evaluation. (1) Data Synthesis: Using only open-source models, we develop a generate-then-filter pipeline that curates diverse visual instruction tasks based on domain-specific image-caption pairs. The resulting data surpass the data synthesized by manual rules or strong closed-source models in enhancing domain-specific performance. (2) Training Pipeline: Unlike general MLLMs that typically adopt a two-stage training paradigm, we find that a single-stage approach is more effective for domain adaptation. (3) Task Evaluation: We conduct extensive experiments in high-impact domains such as biomedicine, food, and remote sensing, by post-training a variety of MLLMs and then evaluating MLLM performance on various domain-specific tasks. Finally, we fully open-source our models, code, and data to encourage future research in this area.
Authors: Yin Tang, Bing Li
Abstract: We introduce a unified, flexible, and easy-to-implement framework of sufficient dimension reduction that can accommodate both linear and nonlinear dimension reduction, and both the conditional distribution and the conditional mean as the targets of estimation. This unified framework is achieved by a specially structured neural network -- the Belted and Ensembled Neural Network (BENN) -- that consists of a narrow latent layer, which we call the belt, and a family of transformations of the response, which we call the ensemble. By strategically placing the belt at different layers of the neural network, we can achieve linear or nonlinear sufficient dimension reduction, and by choosing the appropriate transformation families, we can achieve dimension reduction for the conditional distribution or the conditional mean. Moreover, thanks to the advantage of the neural network, the method is very fast to compute, overcoming a computation bottleneck of the traditional sufficient dimension reduction estimators, which involves the inversion of a matrix of dimension either p or n. We develop the algorithm and convergence rate of our method, compare it with existing sufficient dimension reduction methods, and apply it to two data examples.
Authors: Ed Mackay, Callum Murphy-Barltrop, Jordan Richards, Philip Jonathan
Abstract: This paper presents a novel deep learning framework for estimating multivariate joint extremes of metocean variables, based on the Semi-Parametric Angular-Radial (SPAR) model. When considered in polar coordinates, the problem of modelling multivariate extremes is transformed to one of modelling an angular density, and the tail of a univariate radial variable conditioned on angle. In the SPAR approach, the tail of the radial variable is modelled using a generalised Pareto (GP) distribution, providing a natural extension of univariate extreme value theory to the multivariate setting. In this work, we show how the method can be applied in higher dimensions, using a case study for five metocean variables: wind speed, wind direction, wave height, wave period, and wave direction. The angular variable is modelled using a kernel density method, while the parameters of the GP model are approximated using fully-connected deep neural networks. Our approach provides great flexibility in the dependence structures that can be represented, together with computationally efficient routines for training the model. Furthermore, the application of the method requires fewer assumptions about the underlying distribution(s) compared to existing approaches, and an asymptotically justified means for extrapolating outside the range of observations. Using various diagnostic plots, we show that the fitted models provide a good description of the joint extremes of the metocean variables considered.
Authors: Jan Skvrna, Lukas Neumann
Abstract: Inferring object 3D position and orientation from a single RGB camera is a foundational task in computer vision with many important applications. Traditionally, 3D object detection methods are trained in a fully-supervised setup, requiring LiDAR and vast amounts of human annotations, which are laborious, costly, and do not scale well with the ever-increasing amounts of data being captured. We present a novel method to train a 3D object detector from a single RGB camera without domain-specific human annotations, making orders of magnitude more data available for training. The method uses newly proposed Local Object Motion Model to disentangle object movement source between subsequent frames, is approximately 700 times faster than previous work and compensates camera focal length differences to aggregate multiple datasets. The method is evaluated on three public datasets, where despite using no human labels, it outperforms prior work by a significant margin. It also shows its versatility as a pre-training tool for fully-supervised training and shows that combining pseudo-labels from multiple datasets can achieve comparable accuracy to using human labels from a single dataset. The source code and model are available at https://github.com/jskvrna/MonoSOWA.
Authors: Kavitha Viswanathan, Shashwat Pathak, Piyush Bharambe, Harsh Choudhary, Amit Sethi
Abstract: The tradeoff between reconstruction quality and compute required for video super-resolution (VSR) remains a formidable challenge in its adoption for deployment on resource-constrained edge devices. While transformer-based VSR models have set new benchmarks for reconstruction quality in recent years, these require substantial computational resources. On the other hand, lightweight models that have been introduced even recently struggle to deliver state-of-the-art reconstruction. We propose a novel lightweight and parameter-efficient neural architecture for VSR that achieves state-of-the-art reconstruction accuracy with just 2.3 million parameters. Our model enhances information utilization based on several architectural attributes. Firstly, it uses 2D wavelet decompositions strategically interlayered with learnable convolutional layers to utilize the inductive prior of spatial sparsity of edges in visual data. Secondly, it uses a single memory tensor to capture inter-frame temporal information while avoiding the computational cost of previous memory-based schemes. Thirdly, it uses residual deformable convolutions for implicit inter-frame object alignment that improve upon deformable convolutions by enhancing spatial information in inter-frame feature differences. Architectural insights from our model can pave the way for real-time VSR on the edge, such as display devices for streaming data.
Authors: Yingce Xia, Peiran Jin, Shufang Xie, Liang He, Chuan Cao, Renqian Luo, Guoqing Liu, Yue Wang, Zequn Liu, Yuan-Jyue Chen, Zekun Guo, Yeqi Bai, Pan Deng, Yaosen Min, Ziheng Lu, Hongxia Hao, Han Yang, Jielan Li, Chang Liu, Jia Zhang, Jianwei Zhu, Ran Bi, Kehan Wu, Wei Zhang, Kaiyuan Gao, Qizhi Pei, Qian Wang, Xixian Liu, Yanting Li, Houtian Zhu, Yeqing Lu, Mingqian Ma, Zun Wang, Tian Xie, Krzysztof Maziarz, Marwin Segler, Zhao Yang, Zilong Chen, Yu Shi, Shuxin Zheng, Lijun Wu, Chen Hu, Peggy Dai, Tie-Yan Liu, Haiguang Liu, Tao Qin
Abstract: Foundation models have revolutionized natural language processing and artificial intelligence, significantly enhancing how machines comprehend and generate human languages. Inspired by the success of these foundation models, researchers have developed foundation models for individual scientific domains, including small molecules, materials, proteins, DNA, RNA and even cells. However, these models are typically trained in isolation, lacking the ability to integrate across different scientific domains. Recognizing that entities within these domains can all be represented as sequences, which together form the "language of nature", we introduce Nature Language Model (NatureLM), a sequence-based science foundation model designed for scientific discovery. Pre-trained with data from multiple scientific domains, NatureLM offers a unified, versatile model that enables various applications including: (i) generating and optimizing small molecules, proteins, RNA, and materials using text instructions; (ii) cross-domain generation/design, such as protein-to-molecule and protein-to-RNA generation; and (iii) top performance across different domains, matching or surpassing state-of-the-art specialist models. NatureLM offers a promising generalist approach for various scientific tasks, including drug discovery (hit generation/optimization, ADMET optimization, synthesis), novel material design, and the development of therapeutic proteins or nucleotides. We have developed NatureLM models in different sizes (1 billion, 8 billion, and 46.7 billion parameters) and observed a clear improvement in performance as the model size increases.
Authors: Yunsik Kim, Yonghun Song, Yoonyoung Chung
Abstract: In high-noise environments such as factories, subways, and busy streets, capturing clear speech is challenging. Throat microphones can offer a solution because of their inherent noise-suppression capabilities; however, the passage of sound waves through skin and tissue attenuates high-frequency information, reducing speech clarity. Recent deep learning approaches have shown promise in enhancing throat microphone recordings, but further progress is constrained by the lack of a standard dataset. Here, we introduce the Throat and Acoustic Paired Speech (TAPS) dataset, a collection of paired utterances recorded from 60 native Korean speakers using throat and acoustic microphones. Furthermore, an optimal alignment approach was developed and applied to address the inherent signal mismatch between the two microphones. We tested three baseline deep learning models on the TAPS dataset and found mapping-based approaches to be superior for improving speech quality and restoring content. These findings demonstrate the TAPS dataset's utility for speech enhancement tasks and support its potential as a standard resource for advancing research in throat microphone-based applications.
Authors: Gefan Yang, Frank van der Meulen, Stefan Sommer
Abstract: We propose a novel method for simulating conditioned diffusion processes (diffusion bridges) in Euclidean spaces. By training a neural network to approximate bridge dynamics, our approach eliminates the need for computationally intensive Markov Chain Monte Carlo (MCMC) methods or score modeling. Compared to existing methods, it offers greater robustness across various diffusion specifications and conditioning scenarios. This applies in particular to rare events and multimodal distributions, which pose challenges for score-learning- and MCMC-based approaches. We introduce a flexible variational family, partially specified by a neural network, for approximating the diffusion bridge path measure. Once trained, it enables efficient sampling of independent bridges at a cost comparable to sampling the unconditioned (forward) process.
Authors: Sunay Joshi, Shayan Kiyani, George Pappas, Edgar Dobriban, Hamed Hassani
Abstract: We consider the problem of conformal prediction under covariate shift. Given labeled data from a source domain and unlabeled data from a covariate shifted target domain, we seek to construct prediction sets with valid marginal coverage in the target domain. Most existing methods require estimating the unknown likelihood ratio function, which can be prohibitive for high-dimensional data such as images. To address this challenge, we introduce the likelihood ratio regularized quantile regression (LR-QR) algorithm, which combines the pinball loss with a novel choice of regularization in order to construct a threshold function without directly estimating the unknown likelihood ratio. We show that the LR-QR method has coverage at the desired level in the target domain, up to a small error term that we can control. Our proofs draw on a novel analysis of coverage via stability bounds from learning theory. Our experiments demonstrate that the LR-QR algorithm outperforms existing methods on high-dimensional prediction tasks, including a regression task for the Communities and Crime dataset, an image classification task from the WILDS repository, and an LLM question-answering task on the MMLU benchmark.
Authors: Zichun Yu, Fei Peng, Jie Lei, Arnold Overwijk, Wen-tau Yih, Chenyan Xiong
Abstract: In this paper, we introduce Group-MATES, an efficient group-level data selection approach to optimize the speed-quality frontier of language model pretraining. Specifically, Group-MATES parameterizes costly group-level selection with a relational data influence model. To train this model, we sample training trajectories of the language model and collect oracle data influences alongside. The relational data influence model approximates the oracle data influence by weighting individual influence with relationships among training data. To enable efficient selection with our relational data influence model, we partition the dataset into small clusters using relationship weights and select data within each cluster independently. Experiments on DCLM 400M-4x, 1B-1x, and 3B-1x show that Group-MATES achieves 3.5%-9.4% relative performance gains over random selection across 22 downstream tasks, nearly doubling the improvements achieved by state-of-the-art individual data selection baselines. Furthermore, Group-MATES reduces the number of tokens required to reach a certain downstream performance by up to 1.75x, substantially elevating the speed-quality frontier. Further analyses highlight the critical role of relationship weights in the relational data influence model and the effectiveness of our cluster-based inference. Our code is open-sourced at https://github.com/facebookresearch/Group-MATES.
Authors: Yoonyoung Cho, Junhyek Han, Jisu Han, Beomjoon Kim
Abstract: For robots to operate in general environments like households, they must be able to perform non-prehensile manipulation actions such as toppling and rolling to manipulate ungraspable objects. However, prior works on non-prehensile manipulation cannot yet generalize across environments with diverse geometries. The main challenge lies in adapting to varying environmental constraints: within a cabinet, the robot must avoid walls and ceilings; to lift objects to the top of a step, the robot must account for the step's pose and extent. While deep reinforcement learning (RL) has demonstrated impressive success in non-prehensile manipulation, accounting for such variability presents a challenge for the generalist policy, as it must learn diverse strategies for each new combination of constraints. To address this, we propose a modular and reconfigurable architecture that adaptively reconfigures network modules based on task requirements. To capture the geometric variability in environments, we extend the contact-based object representation (CORN) to environment geometries, and propose a procedural algorithm for generating diverse environments to train our agent. Taken together, the resulting policy can zero-shot transfer to novel real-world environments and objects despite training entirely within a simulator. We additionally release a simulation-based benchmark featuring nine digital twins of real-world scenes with 353 objects to facilitate non-prehensile manipulation research in realistic domains.
Authors: Songming Zhang, Xue Zhang, Tong Zhang, Bojie Hu, Yufeng Chen, Jinan Xu
Abstract: In modern large language models (LLMs), LLM alignment is of crucial importance and is typically achieved through methods such as reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO). However, in most existing methods for LLM alignment, all tokens in the response are optimized using a sparse, response-level reward or preference annotation. The ignorance of token-level rewards may erroneously punish high-quality tokens or encourage low-quality tokens, resulting in suboptimal performance and slow convergence speed. To address this issue, we propose AlignDistil, an RLHF-equivalent distillation method for token-level reward optimization. Specifically, we introduce the reward learned by DPO into the RLHF objective and theoretically prove the equivalence between this objective and a token-level distillation process, where the teacher distribution linearly combines the logits from the DPO model and a reference model. On this basis, we further bridge the accuracy gap between the reward from the DPO model and the pure reward model, by building a contrastive DPO reward with a normal and a reverse DPO model. Moreover, to avoid under- and over-optimization on different tokens, we design a token adaptive logit extrapolation mechanism to construct an appropriate teacher distribution for each token. Experimental results demonstrate the superiority of our AlignDistil over existing methods and showcase fast convergence due to its token-level distributional reward optimization.
Authors: Bang Nguyen, Tingting Du, Mengxia Yu, Lawrence Angrave, Meng Jiang
Abstract: While the Question Generation (QG) task has been increasingly adopted in educational assessments, its evaluation remains limited by approaches that lack a clear connection to the educational values of test items. In this work, we introduce test item analysis, a method frequently used by educators to assess test question quality, into QG evaluation. Specifically, we construct pairs of candidate questions that differ in quality across dimensions such as topic coverage, item difficulty, item discrimination, and distractor efficiency. We then examine whether existing QG evaluation approaches can effectively distinguish these differences. Our findings reveal significant shortcomings in these approaches with respect to accurately assessing test item quality in relation to student performance. To address this gap, we propose a novel QG evaluation framework, QG-SMS, which leverages Large Language Model for Student Modeling and Simulation to perform test item analysis. As demonstrated in our extensive experiments and human evaluation study, the additional perspectives introduced by the simulated student profiles lead to a more effective and robust assessment of test items.
Authors: Chen Xu, Tony Khuong Nguyen, Emma Dixon, Christopher Rodriguez, Patrick Miller, Robert Lee, Paarth Shah, Rares Ambrus, Haruki Nishimura, Masha Itkina
Abstract: Recent years have witnessed impressive robotic manipulation systems driven by advances in imitation learning and generative modeling, such as diffusion- and flow-based approaches. As robot policy performance increases, so does the complexity and time horizon of achievable tasks, inducing unexpected and diverse failure modes that are difficult to predict a priori. To enable trustworthy policy deployment in safety-critical human environments, reliable runtime failure detection becomes important during policy inference. However, most existing failure detection approaches rely on prior knowledge of failure modes and require failure data during training, which imposes a significant challenge in practicality and scalability. In response to these limitations, we present FAIL-Detect, a modular two-stage approach for failure detection in imitation learning-based robotic manipulation. To accurately identify failures from successful training data alone, we frame the problem as sequential out-of-distribution (OOD) detection. We first distill policy inputs and outputs into scalar signals that correlate with policy failures and capture epistemic uncertainty. FAIL-Detect then employs conformal prediction (CP) as a versatile framework for uncertainty quantification with statistical guarantees. Empirically, we thoroughly investigate both learned and post-hoc scalar signal candidates on diverse robotic manipulation tasks. Our experiments show learned signals to be mostly consistently effective, particularly when using our novel flow-based density estimator. Furthermore, our method detects failures more accurately and faster than state-of-the-art (SOTA) failure detection baselines. These results highlight the potential of FAIL-Detect to enhance the safety and reliability of imitation learning-based robotic systems as they progress toward real-world deployment.
Authors: Alberto Coffrini, Paolo Barsocchi, Francesco Furfari, Antonino Crivello, Alessio Ferrari
Abstract: Indoor navigation presents unique challenges due to complex layouts and the unavailability of GNSS signals. Existing solutions often struggle with contextual adaptation, and typically require dedicated hardware. In this work, we explore the potential of a Large Language Model (LLM), i.e., ChatGPT, to generate natural, context-aware navigation instructions from indoor map images. We design and evaluate test cases across different real-world environments, analyzing the effectiveness of LLMs in interpreting spatial layouts, handling user constraints, and planning efficient routes. Our findings demonstrate the potential of LLMs for supporting personalized indoor navigation, with an average of 86.59% correct indications and a maximum of 97.14%. The proposed system achieves high accuracy and reasoning performance. These results have key implications for AI-driven navigation and assistive technologies.
Authors: Jakub Malinowski, Marcin Kostrzewa, Micha{\l} Balcerek, Weronika Tomczuk, Janusz Szwabi\'nski
Abstract: Change point detection has become an important part of the analysis of the single-particle tracking data, as it allows one to identify moments, in which the motion patterns of observed particles undergo significant changes. The segmentation of diffusive trajectories based on those moments may provide insight into various phenomena in soft condensed matter and biological physics. In this paper, we propose CINNAMON, a hybrid approach to classifying single-particle tracking trajectories, detecting change points within them, and estimating diffusion parameters in the segments between the change points. Our method is based on a combination of neural networks, feature-based machine learning, and statistical techniques. It has been benchmarked in the second Anomalous Diffusion Challenge. The method offers a high level of interpretability due to its analytical and feature-based components. A potential use of features from topological data analysis is also discussed.
Authors: Murong Yue, Ziyu Yao
Abstract: Batch prompting, which combines a batch of multiple queries sharing the same context in one inference, has emerged as a promising solution to reduce inference costs. However, our study reveals a significant security vulnerability in batch prompting: malicious users can inject attack instructions into a batch, leading to unwanted interference across all queries, which can result in the inclusion of harmful content, such as phishing links, or the disruption of logical reasoning. In this paper, we construct BATCHSAFEBENCH, a comprehensive benchmark comprising 150 attack instructions of two types and 8k batch instances, to study the batch prompting vulnerability systematically. Our evaluation of both closed-source and open-weight LLMs demonstrates that all LLMs are susceptible to batch-prompting attacks. We then explore multiple defending approaches. While the prompting-based defense shows limited effectiveness for smaller LLMs, the probing-based approach achieves about 95% accuracy in detecting attacks. Additionally, we perform a mechanistic analysis to understand the attack and identify attention heads that are responsible for it.
Authors: Claudio Fantasia, Luca Calatroni, Xavier Descombes, Rim Rekik
Abstract: We consider a patch-based learning approach defined in terms of neural networks to estimate spatially adaptive regularisation parameter maps for image denoising with weighted Total Variation (TV) and test it to situations when the noise distribution is unknown. As an example, we consider situations where noise could be either Gaussian or Poisson and perform preliminary model selection by a standard binary classification network. Then, we define a patch-based approach where at each image pixel an optimal weighting between TV regularisation and the corresponding data fidelity is learned in a supervised way using reference natural image patches upon optimisation of SSIM and in a sliding window fashion. Extensive numerical results are reported for both noise models, showing significant improvement w.r.t. results obtained by means of optimal scalar regularisation.
Authors: Ziwei Yang, Takeyuki Tamura
Abstract: In genome-scale constraint-based metabolic models, gene deletion strategies are crucial for achieving growth-coupled production, where cell growth and target metabolite production are simultaneously achieved. While computational methods for calculating gene deletions have been widely explored and contribute to developing gene deletion strategy databases, current approaches are limited in leveraging new data-driven paradigms, such as machine learning, for more efficient strain design. Therefore, it is necessary to propose a fundamental framework for this objective. In this study, we first formulate the problem of gene deletion strategy prediction and then propose a framework for predicting gene deletion strategies for growth-coupled production in genome-scale metabolic models. The proposed framework leverages deep learning algorithms to learn and integrate sequential gene and metabolite data representation, enabling the automatic gene deletion strategy prediction. Computational experiment results demonstrate the feasibility of the proposed framework, showing substantial improvements over baseline methods. Specifically, the proposed framework achieves a 14.69%, 22.52%, and 13.03% increase in overall accuracy across three metabolic models of different scales under study, while maintaining balanced precision and recall in predicting gene deletion statuses. The source code and examples for the framework are publicly available at https://github.com/MetNetComp/DeepGDel.
Authors: Hugo Lebeau
Abstract: We are interested in the estimation of a rank-one tensor signal when only a portion $\varepsilon$ of its noisy observation is available. We show that the study of this problem can be reduced to that of a random matrix model whose spectral analysis gives access to the reconstruction performance. These results shed light on and specify the loss of performance induced by an artificial reduction of the memory cost of a tensor via the deletion of a random part of its entries.
Authors: Mahshad Lotfinia, Arash Tayebiarasteh, Samaneh Samiei, Mehdi Joodaki, Soroosh Tayebi Arasteh
Abstract: Reliable artificial intelligence (AI) models for medical image analysis often depend on large and diverse labeled datasets. Federated learning (FL) offers a decentralized and privacy-preserving approach to training but struggles in highly non-independent and identically distributed (non-IID) settings, where institutions with more representative data may experience degraded performance. Moreover, existing large-scale FL studies have been limited to adult datasets, neglecting the unique challenges posed by pediatric data, which introduces additional non-IID variability. To address these limitations, we analyzed n=398,523 adult chest radiographs from diverse institutions across multiple countries and n=9,125 pediatric images, leveraging transfer learning from general-purpose self-supervised image representations to classify pneumonia and cases with no abnormality. Using state-of-the-art vision transformers, we found that FL improved performance only for smaller adult datasets (P<0.001) but degraded performance for larger datasets (P<0.064) and pediatric cases (P=0.242). However, equipping FL with self-supervised weights significantly enhanced outcomes across pediatric cases (P=0.031) and most adult datasets (P<0.008), except the largest dataset (P=0.052). These findings underscore the potential of easily deployable general-purpose self-supervised image representations to address non-IID challenges in clinical FL applications and highlight their promise for enhancing patient outcomes and advancing pediatric healthcare, where data scarcity and variability remain persistent obstacles.
Authors: Yijun Liu
Abstract: Recent advances in recommender systems have underscored the complementary strengths of generative modeling and pretrained language models. We propose HSTU-BLaIR, a hybrid framework that augments the Hierarchical Sequential Transduction Unit (HSTU)-based generative recommender with BLaIR, a lightweight contrastive text embedding model. This integration enriches item representations with semantic signals from textual metadata while preserving HSTU's powerful sequence modeling capabilities. We evaluate HSTU-BLaIR on two e-commerce datasets: three subsets from the Amazon Reviews 2023 dataset and the Steam dataset. We compare its performance against both the original HSTU-based recommender and a variant augmented with embeddings from OpenAI's state-of-the-art \texttt{text-embedding-3-large} model. Despite the latter being trained on a substantially larger corpus with significantly more parameters, our lightweight BLaIR-enhanced approach -- pretrained on domain-specific data -- achieves better performance in nearly all cases. Specifically, HSTU-BLaIR outperforms the OpenAI embedding-based variant on all but one metric, where it is marginally lower, and matches it on another. These findings highlight the effectiveness of contrastive text embeddings in compute-efficient recommendation settings.
Authors: Jang Hyun Cho, Andrea Madotto, Effrosyni Mavroudi, Triantafyllos Afouras, Tushar Nagarajan, Muhammad Maaz, Yale Song, Tengyu Ma, Shuming Hu, Suyog Jain, Miguel Martin, Huiyu Wang, Hanoona Rasheed, Peize Sun, Po-Yao Huang, Daniel Bolya, Nikhila Ravi, Shashank Jain, Tammy Stark, Shane Moon, Babak Damavandi, Vivian Lee, Andrew Westbury, Salman Khan, Philipp Kr\"ahenb\"uhl, Piotr Doll\'ar, Lorenzo Torresani, Kristen Grauman, Christoph Feichtenhofer
Abstract: Vision-language models are integral to computer vision research, yet many high-performing models remain closed-source, obscuring their data, design and training recipe. The research community has responded by using distillation from black-box models to label training data, achieving strong benchmark results, at the cost of measurable scientific progress. However, without knowing the details of the teacher model and its data sources, scientific progress remains difficult to measure. In this paper, we study building a Perception Language Model (PLM) in a fully open and reproducible framework for transparent research in image and video understanding. We analyze standard training pipelines without distillation from proprietary models and explore large-scale synthetic data to identify critical data gaps, particularly in detailed video understanding. To bridge these gaps, we release 2.8M human-labeled instances of fine-grained video question-answer pairs and spatio-temporally grounded video captions. Additionally, we introduce PLM-VideoBench, a suite for evaluating challenging video understanding tasks focusing on the ability to reason about "what", "where", "when", and "how" of a video. We make our work fully reproducible by providing data, training recipes, code & models. https://github.com/facebookresearch/perception_models
Authors: Nahshon Mokua Obiri, Kristof Van Laerhoven
Abstract: Modeling path loss in indoor LoRaWAN technology deployments is inherently challenging due to structural obstructions, occupant density and activities, and fluctuating environmental conditions. This study proposes a two-stage approach to capture and analyze these complexities using an extensive dataset of 1,328,334 field measurements collected over six months in a single-floor office at the University of Siegen's Hoelderlinstrasse Campus, Germany. First, we implement a multiple linear regression model that includes traditional propagation metrics (distance, structural walls) and an extension with proposed environmental variables (relative humidity, temperature, carbon dioxide, particulate matter, and barometric pressure). Using analysis of variance, we demonstrate that adding these environmental factors can reduce unexplained variance by 42.32 percent. Secondly, we examine residual distributions by fitting five candidate probability distributions: Normal, Skew-Normal, Cauchy, Student's t, and Gaussian Mixture Models (GMMs) with 2 to 5 components. Our results show that a four-component Gaussian Mixture Model captures the residual heterogeneity of indoor signal propagation most accurately, significantly outperforming single-distribution approaches. Given the push toward ultra-reliable, context-aware communications in 6G networks, our analysis shows that environment-aware modeling can substantially improve LoRaWAN network design in dynamic indoor IoT deployments.
Authors: Maximilian Schuessler, Erik Sverdrup, Robert Tibshirani
Abstract: Robust estimation of heterogeneous treatment effects is a fundamental challenge for optimal decision-making in domains ranging from personalized medicine to educational policy. In recent years, predictive machine learning has emerged as a valuable toolbox for causal estimation, enabling more flexible effect estimation. However, accurately estimating conditional average treatment effects (CATE) remains a major challenge, particularly in the presence of many covariates. In this article, we propose pretraining strategies that leverage a phenomenon in real-world applications: factors that are prognostic of the outcome are frequently also predictive of treatment effect heterogeneity. In medicine, for example, components of the same biological signaling pathways frequently influence both baseline risk and treatment response. Specifically, we demonstrate our approach within the R-learner framework, which estimates the CATE by solving individual prediction problems based on a residualized loss. We use this structure to incorporate side information and develop models that can exploit synergies between risk prediction and causal effect estimation. In settings where these synergies are present, this cross-task learning enables more accurate signal detection, yields lower estimation error, reduced false discovery rates, and higher power for detecting heterogeneity.
Authors: Xingjin Wang, Howe Tissue, Lu Wang, Linjing Li, Daniel Dajun Zeng
Abstract: Continual Pre-Training (CPT) has become a popular and effective method to apply strong foundation models to specific downstream tasks. In this work, we explore the learning dynamics throughout the CPT process for large language models. We specifically focus on how general and downstream domain performance evolves at each training step, with domain performance measured via validation losses. We have observed that the CPT loss curve fundamentally characterizes the transition from one curve to another hidden curve, and could be described by decoupling the effects of distribution shift and learning rate annealing. We derive a CPT scaling law that combines the two factors, enabling the prediction of loss at any (continual) training steps and across learning rate schedules (LRS) in CPT. Our formulation presents a comprehensive understanding of several critical factors in CPT, including loss potential, peak learning rate, training steps, replay ratio, etc. Moreover, our approach can be adapted to customize training hyper-parameters to different CPT goals such as balancing general and domain-specific performance. Extensive experiments demonstrate that our scaling law holds across various CPT datasets and training hyper-parameters.
Authors: Maris F. L. Galesloot, Roman Andriushchenko, Milan \v{C}e\v{s}ka, Sebastian Junges, Nils Jansen
Abstract: Partially observable Markov decision processes (POMDPs) model specific environments in sequential decision-making under uncertainty. Critically, optimal policies for POMDPs may not be robust against perturbations in the environment. Hidden-model POMDPs (HM-POMDPs) capture sets of different environment models, that is, POMDPs with a shared action and observation space. The intuition is that the true model is hidden among a set of potential models, and it is unknown which model will be the environment at execution time. A policy is robust for a given HM-POMDP if it achieves sufficient performance for each of its POMDPs.We compute such robust policies by combining two orthogonal techniques: (1) a deductive formal verification technique that supports tractable robust policy evaluation by computing a worst-case POMDP within the HM-POMDP, and (2) subgradient ascent to optimize the candidate policy for a worst-case POMDP. The empirical evaluation shows that, compared to various baselines, our approach (1) produces policies that are more robust and generalize better to unseen POMDPs, and (2) scales to HM-POMDPs that consist of over a hundred thousand environments.
Authors: Drashthi Doshi, Aditya Vema Reddy Kesari, Swaprava Nath, Avishek Ghosh, Suhas S Kowshik
Abstract: Classical federated learning (FL) assumes that the clients have a limited amount of noisy data with which they voluntarily participate and contribute towards learning a global, more accurate model in a principled manner. The learning happens in a distributed fashion without sharing the data with the center. However, these methods do not consider the incentive of an agent for participating and contributing to the process, given that data collection and running a distributed algorithm is costly for the clients. The question of rationality of contribution has been asked recently in the literature and some results exist that consider this problem. This paper addresses the question of simultaneous parameter learning and incentivizing contribution, which distinguishes it from the extant literature. Our first mechanism incentivizes each client to contribute to the FL process at a Nash equilibrium and simultaneously learn the model parameters. However, this equilibrium outcome can be away from the optimal, where clients contribute with their full data and the algorithm learns the optimal parameters. We propose a second mechanism with monetary transfers that is budget balanced and enables the full data contribution along with optimal parameter learning. Large scale experiments with real (federated) datasets (CIFAR-10, FeMNIST, and Twitter) show that these algorithms converge quite fast in practice, yield good welfare guarantees, and better model performance for all agents.
Authors: Kaiyuan Chen, Shuangyu Xie, Zehan Ma, Pannag R Sanketi, Ken Goldberg
Abstract: Vision-Language Models (VLMs) acquire real-world knowledge and general reasoning ability through Internet-scale image-text corpora. They can augment robotic systems with scene understanding and task planning, and assist visuomotor policies that are trained on robot trajectory data. We explore the reverse paradigm - using rich, real, multi-modal robot trajectory data to enhance and evaluate VLMs. In this paper, we present Robo2VLM, a Visual Question Answering (VQA) dataset generation framework for VLMs. Given a human tele-operated robot trajectory, Robo2VLM derives ground-truth from non-visual and non-descriptive sensory modalities, such as end-effector pose, gripper aperture, and force sensing. Based on these modalities, it segments the robot trajectory into a sequence of manipulation phases. At each phase, Robo2VLM uses scene and interaction understanding to identify 3D properties of the robot, task goal, and the target object. The properties are used to generate representative VQA queries - images with textural multiple-choice questions - based on spatial, goal-conditioned, and interaction reasoning question templates. We curate Robo2VLM-1, a large-scale in-the-wild dataset with 684,710 questions covering 463 distinct scenes and 3,396 robotic manipulation tasks from 176k real robot trajectories. Results suggest that Robo2VLM-1 can benchmark and improve VLM capabilities in spatial and interaction reasoning.
Authors: Jiquan Shan, Junxiao Wang, Lifeng Zhao, Liang Cai, Hongyuan Zhang, Ioannis Liritzis
Abstract: Recently, vision transformers (ViTs) have achieved excellent performance on vision tasks by measuring the global self-attention among the image patches. Given $n$ patches, they will have quadratic complexity such as $\mathcal{O}(n^2)$ and the time cost is high when splitting the input image with a small granularity. Meanwhile, the pivotal information is often randomly gathered in a few regions of an input image, some tokens may not be helpful for the downstream tasks. To handle this problem, we introduce an anchor-based efficient vision transformer (AnchorFormer), which employs the anchor tokens to learn the pivotal information and accelerate the inference. Firstly, by estimating the bipartite attention between the anchors and tokens, the complexity will be reduced from $\mathcal{O}(n^2)$ to $\mathcal{O}(mn)$, where $m$ is an anchor number and $m < n$. Notably, by representing the anchors with the neurons in a neural layer, we can differentiably learn these anchors and approximate global self-attention through the Markov process. It avoids the burden caused by non-differentiable operations and further speeds up the approximate attention. Moreover, we extend the proposed model to three downstream tasks including classification, detection, and segmentation. Extensive experiments show the effectiveness of our AnchorFormer, e.g., achieving up to a 9.0% higher accuracy or 46.7% FLOPs reduction on ImageNet classification, 81.3% higher mAP on COCO detection under comparable FLOPs, as compared to the current baselines.
Authors: Wenjie Yang, Mao Zheng, Mingyang Song, Zheng Li, Sitong Wang
Abstract: Large language models (LLMs) have recently demonstrated remarkable capabilities in machine translation (MT). However, most advanced MT-specific LLMs heavily rely on external supervision signals during training, such as human-annotated reference data or trained reward models (RMs), which are often expensive to obtain and challenging to scale. To overcome this limitation, we propose a Simple Self-Rewarding (SSR) Reinforcement Learning (RL) framework for MT that is reference-free, fully online, and relies solely on self-judging rewards. Training with SSR using 13K monolingual examples and Qwen-2.5-7B as the backbone, our model SSR-Zero-7B outperforms existing MT-specific LLMs, e.g., TowerInstruct-13B and GemmaX-28-9B, as well as larger general LLMs like Qwen2.5-32B-Instruct in English $\leftrightarrow$ Chinese translation tasks from WMT23, WMT24, and Flores200 benchmarks. Furthermore, by augmenting SSR with external supervision from COMET, our strongest model, SSR-X-Zero-7B, achieves state-of-the-art performance in English $\leftrightarrow$ Chinese translation, surpassing all existing open-source models under 72B parameters and even outperforming closed-source models, e.g., GPT-4o and Gemini 1.5 Pro. Our analysis highlights the effectiveness of the self-rewarding mechanism compared to the external LLM-as-a-judge approach in MT and demonstrates its complementary benefits when combined with trained RMs. Our findings provide valuable insight into the potential of self-improving RL methods. We have publicly released our code, data and models.
Authors: Hongyuan Tao, Ying Zhang, Zhenhao Tang, Hongen Peng, Xukun Zhu, Bingchang Liu, Yingguang Yang, Ziyin Zhang, Zhaogui Xu, Haipeng Zhang, Linchao Zhu, Rui Wang, Hang Yu, Jianguo Li, Peng Di
Abstract: Recent advances in Large Language Models (LLMs) have shown promise in function-level code generation, yet repository-level software engineering tasks remain challenging. Current solutions predominantly rely on proprietary LLM agents, which introduce unpredictability and limit accessibility, raising concerns about data privacy and model customization. This paper investigates whether open-source LLMs can effectively address repository-level tasks without requiring agent-based approaches. We demonstrate this is possible by enabling LLMs to comprehend functions and files within codebases through their semantic information and structural dependencies. To this end, we introduce Code Graph Models (CGMs), which integrate repository code graph structures into the LLM's attention mechanism and map node attributes to the LLM's input space using a specialized adapter. When combined with an agentless graph RAG framework, our approach achieves a 43.00% resolution rate on the SWE-bench Lite benchmark using the open-source Qwen2.5-72B model. This performance ranks first among open weight models, second among methods with open-source systems, and eighth overall, surpassing the previous best open-source model-based method by 12.33%.
Authors: Xiang Li, Yunai Li, Huiying Zhong, Lihua Lei, Zhun Deng
Abstract: Performativity of predictions refers to the phenomena that prediction-informed decisions may influence the target they aim to predict, which is widely observed in policy-making in social sciences and economics. In this paper, we initiate the study of statistical inference under performativity. Our contribution is two-fold. First, we build a central limit theorem for estimation and inference under performativity, which enables inferential purposes in policy-making such as constructing confidence intervals or testing hypotheses. Second, we further leverage the derived central limit theorem to investigate prediction-powered inference (PPI) under performativity, which is based on a small labeled dataset and a much larger dataset of machine-learning predictions. This enables us to obtain more precise estimation and improved confidence regions for the model parameter (i.e., policy) of interest in performative prediction. We demonstrate the power of our framework by numerical experiments. To the best of our knowledge, this paper is the first one to establish statistical inference under performativity, which brings up new challenges and inference settings that we believe will add significant values to policy-making, statistics, and machine learning.
Authors: Tonghe Zhang, Chao Yu, Sichang Su, Yu Wang
Abstract: We propose ReinFlow, a simple yet effective online reinforcement learning (RL) framework that fine-tunes a family of flow matching policies for continuous robotic control. Derived from rigorous RL theory, ReinFlow injects learnable noise into a flow policy's deterministic path, converting the flow into a discrete-time Markov Process for exact and straightforward likelihood computation. This conversion facilitates exploration and ensures training stability, enabling ReinFlow to fine-tune diverse flow model variants, including Rectified Flow [35] and Shortcut Models [19], particularly at very few or even one denoising step. We benchmark ReinFlow in representative locomotion and manipulation tasks, including long-horizon planning with visual input and sparse reward. The episode reward of Rectified Flow policies obtained an average net growth of 135.36% after fine-tuning in challenging legged locomotion tasks while saving denoising steps and 82.63% of wall time compared to state-of-the-art diffusion RL fine-tuning method DPPO [43]. The success rate of the Shortcut Model policies in state and visual manipulation tasks achieved an average net increase of 40.34% after fine-tuning with ReinFlow at four or even one denoising step, whose performance is comparable to fine-tuned DDIM policies while saving computation time for an average of 23.20%. Project webpage: https://reinflow.github.io/
Authors: Yongjin Yang, Euiin Yi, Jongwoo Ko, Kimin Lee, Zhijing Jin, Se-Young Yun
Abstract: The remarkable growth in large language model (LLM) capabilities has spurred exploration into multi-agent systems, with debate frameworks emerging as a promising avenue for enhanced problem-solving. These multi-agent debate (MAD) approaches, where agents collaboratively present, critique, and refine arguments, potentially offer improved reasoning, robustness, and diverse perspectives over monolithic models. Despite prior studies leveraging MAD, a systematic understanding of its effectiveness compared to self-agent methods, particularly under varying conditions, remains elusive. This paper seeks to fill this gap by conceptualizing MAD as a test-time computational scaling technique, distinguished by collaborative refinement and diverse exploration capabilities. We conduct a comprehensive empirical investigation comparing MAD with strong self-agent test-time scaling baselines on mathematical reasoning and safety-related tasks. Our study systematically examines the influence of task difficulty, model scale, and agent diversity on MAD's performance. Key findings reveal that, for mathematical reasoning, MAD offers limited advantages over self-agent scaling but becomes more effective with increased problem difficulty and decreased model capability, while agent diversity shows little benefit. Conversely, for safety tasks, MAD's collaborative refinement can increase vulnerability, but incorporating diverse agent configurations facilitates a gradual reduction in attack success through the collaborative refinement process. We believe our findings provide critical guidance for the future development of more effective and strategically deployed MAD systems.
Authors: Yongjin Yang, Sinjae Kang, Juyong Lee, Dongjun Lee, Se-Young Yun, Kimin Lee
Abstract: Training large language model (LLM) agents to acquire necessary skills and perform diverse tasks within an environment is gaining interest as a means to enable open-endedness. However, creating the training dataset for their skill acquisition faces several challenges. Manual trajectory collection requires significant human effort. Another approach, where LLMs directly propose tasks to learn, is often invalid, as the LLMs lack knowledge of which tasks are actually feasible. Moreover, the generated data may not provide a meaningful learning signal, as agents often already perform well on the proposed tasks. To address this, we propose a novel automatic skill discovery framework EXIF for LLM-powered agents, designed to improve the feasibility of generated target behaviors while accounting for the agents' capabilities. Our method adopts an exploration-first strategy by employing an exploration agent (Alice) to train the target agent (Bob) to learn essential skills in the environment. Specifically, Alice first interacts with the environment to retrospectively generate a feasible, environment-grounded skill dataset, which is then used to train Bob. Crucially, we incorporate an iterative feedback loop, where Alice evaluates Bob's performance to identify areas for improvement. This feedback then guides Alice's next round of exploration, forming a closed-loop data generation process. Experiments on Webshop and Crafter demonstrate EXIF's ability to effectively discover meaningful skills and iteratively expand the capabilities of the trained agent without any human intervention, achieving substantial performance improvements. Interestingly, we observe that setting Alice to the same model as Bob also notably improves performance, demonstrating EXIF's potential for building a self-evolving system.
Authors: Benjamin Smith, Jianhui Gao, Jessica Gronsbell
Abstract: Fairness is a growing area of machine learning (ML) that focuses on ensuring models do not produce systematically biased outcomes for specific groups, particularly those defined by protected attributes such as race, gender, or age. Evaluating fairness is a critical aspect of ML model development, as biased models can perpetuate structural inequalities. The {fairmetrics} R package offers a user-friendly framework for rigorously evaluating numerous group-based fairness criteria, including metrics based on independence (e.g., statistical parity), separation (e.g., equalized odds), and sufficiency (e.g., predictive parity). Group-based fairness criteria assess whether a model is equally accurate or well-calibrated across a set of predefined groups so that appropriate bias mitigation strategies can be implemented. {fairmetrics} provides both point and interval estimates for multiple metrics through a convenient wrapper function and includes an example dataset derived from the Medical Information Mart for Intensive Care, version II (MIMIC-II) database (Goldberger et al., 2000; Raffa, 2016).
Authors: Pengfei Zuo, Huimin Lin, Junbo Deng, Nan Zou, Xingkun Yang, Yingyu Diao, Weifeng Gao, Ke Xu, Zhangyu Chen, Shirui Lu, Zhao Qiu, Peiyang Li, Xianyu Chang, Zhengzhong Yu, Fangzheng Miao, Jia Zheng, Ying Li, Yuan Feng, Bei Wang, Zaijian Zong, Mosong Zhou, Wenli Zhou, Houjiang Chen, Xingyu Liao, Yipeng Li, Wenxiao Zhang, Ping Zhu, Yinggang Wang, Chuanjie Xiao, Depeng Liang, Dong Cao, Juncheng Liu, Yongqiang Yang, Xiaolong Bai, Yi Li, Huaguo Xie, Huatao Wu, Zhibin Yu, Lv Chen, Hu Liu, Yujun Ding, Haipei Zhu, Jing Xia, Yi Xiong, Zhou Yu, Heng Liao
Abstract: The rapid evolution of large language models (LLMs), driven by growing parameter scales, adoption of mixture-of-experts (MoE) architectures, and expanding context lengths, imposes unprecedented demands on AI infrastructure. Traditional AI clusters face limitations in compute intensity, memory bandwidth, inter-chip communication, and latency, compounded by variable workloads and strict service-level objectives. Addressing these issues requires fundamentally redesigned hardware-software integration. This paper introduces Huawei CloudMatrix, a next-generation AI datacenter architecture, realized in the production-grade CloudMatrix384 supernode. It integrates 384 Ascend 910 NPUs and 192 Kunpeng CPUs interconnected via an ultra-high-bandwidth Unified Bus (UB) network, enabling direct all-to-all communication and dynamic pooling of resources. These features optimize performance for communication-intensive operations, such as large-scale MoE expert parallelism and distributed key-value cache access. To fully leverage CloudMatrix384, we propose CloudMatrix-Infer, an advanced LLM serving solution incorporating three core innovations: a peer-to-peer serving architecture that independently scales prefill, decode, and caching; a large-scale expert parallelism strategy supporting EP320 via efficient UB-based token dispatch; and hardware-aware optimizations including specialized operators, microbatch-based pipelining, and INT8 quantization. Evaluation with the DeepSeek-R1 model shows CloudMatrix-Infer achieves state-of-the-art efficiency: prefill throughput of 6,688 tokens/s per NPU and decode throughput of 1,943 tokens/s per NPU (<50 ms TPOT). It effectively balances throughput and latency, sustaining 538 tokens/s per NPU even under stringent 15 ms latency constraints, while INT8 quantization maintains model accuracy across benchmarks.
Authors: Yuxuan Wang, Ming Yang, Weishuai Zeng, Yu Zhang, Xinrun Xu, Haobin Jiang, Ziluo Ding, Zongqing Lu
Abstract: Achieving general agile whole-body control on humanoid robots remains a major challenge due to diverse motion demands and data conflicts. While existing frameworks excel in training single motion-specific policies, they struggle to generalize across highly varied behaviors due to conflicting control requirements and mismatched data distributions. In this work, we propose BumbleBee (BB), an expert-generalist learning framework that combines motion clustering and sim-to-real adaptation to overcome these challenges. BB first leverages an autoencoder-based clustering method to group behaviorally similar motions using motion features and motion descriptions. Expert policies are then trained within each cluster and refined with real-world data through iterative delta action modeling to bridge the sim-to-real gap. Finally, these experts are distilled into a unified generalist controller that preserves agility and robustness across all motion types. Experiments on two simulations and a real humanoid robot demonstrate that BB achieves state-of-the-art general whole-body control, setting a new benchmark for agile, robust, and generalizable humanoid performance in the real world.
Authors: Kshitij Kavimandan, Angelos Nalmpantis, Emma Beauxis-Aussalet, Robert-Jan Sips
Abstract: Patent images are technical drawings that convey information about a patent's innovation. Patent image retrieval systems aim to search in vast collections and retrieve the most relevant images. Despite recent advances in information retrieval, patent images still pose significant challenges due to their technical intricacies and complex semantic information, requiring efficient fine-tuning for domain adaptation. Current methods neglect patents' hierarchical relationships, such as those defined by the Locarno International Classification (LIC) system, which groups broad categories (e.g., "furnishing") into subclasses (e.g., "seats" and "beds") and further into specific patent designs. In this work, we introduce a hierarchical multi-positive contrastive loss that leverages the LIC's taxonomy to induce such relations in the retrieval process. Our approach assigns multiple positive pairs to each patent image within a batch, with varying similarity scores based on the hierarchical taxonomy. Our experimental analysis with various vision and multimodal models on the DeepPatent2 dataset shows that the proposed method enhances the retrieval results. Notably, our method is effective with low-parameter models, which require fewer computational resources and can be deployed on environments with limited hardware.
Authors: Rylan Schaeffer, Joshua Kazdan, Yegor Denisov-Blanch
Abstract: Sampling from language models impacts the quality and diversity of outputs, affecting both research and real-world applications. Recently, Nguyen et al. 2024's "Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputs" introduced a new sampler called min-p, claiming it achieves superior quality and diversity over established samplers such as basic, top-k, and top-p sampling. The significance of these claims was underscored by the paper's recognition as the 18th highest-scoring submission to ICLR 2025 and selection for an Oral presentation. This paper conducts a comprehensive re-examination of the evidence supporting min-p and reaches different conclusions from the original paper's four lines of evidence. First, the original paper's human evaluations omitted data, conducted statistical tests incorrectly, and described qualitative feedback inaccurately; our reanalysis demonstrates min-p did not outperform baselines in quality, diversity, or a trade-off between quality and diversity; in response to our findings, the authors of the original paper conducted a new human evaluation using a different implementation, task, and rubric that nevertheless provides further evidence min-p does not improve over baselines. Second, comprehensively sweeping the original paper's NLP benchmarks reveals min-p does not surpass baselines when controlling for the number of hyperparameters. Third, the original paper's LLM-as-a-Judge evaluations lack methodological clarity and appear inconsistently reported. Fourth, community adoption claims (49k GitHub repositories, 1.1M GitHub stars) were found to be unsubstantiated, leading to their removal; the revised adoption claim remains misleading. We conclude that evidence presented in the original paper fails to support claims that min-p improves quality, diversity, or a trade-off between quality and diversity.
Authors: Essential AI, :, Andrew Hojel, Michael Pust, Tim Romanski, Yash Vanjani, Ritvik Kapila, Mohit Parmar, Adarsh Chaluvaraju, Alok Tripathy, Anil Thomas, Ashish Tanwer, Darsh J Shah, Ishaan Shah, Karl Stratos, Khoi Nguyen, Kurt Smith, Michael Callahan, Peter Rushton, Philip Monk, Platon Mazarakis, Saad Jamal, Saurabh Srivastava, Somanshu Singla, Ashish Vaswani
Abstract: Data plays the most prominent role in how language models acquire skills and knowledge. The lack of massive, well-organized pre-training datasets results in costly and inaccessible data pipelines. We present Essential-Web v1.0, a 24-trillion-token dataset in which every document is annotated with a twelve-category taxonomy covering topic, format, content complexity, and quality. Taxonomy labels are produced by EAI-Distill-0.5b, a fine-tuned 0.5b-parameter model that achieves an annotator agreement within 3% of Qwen2.5-32B-Instruct. With nothing more than SQL-style filters, we obtain competitive web-curated datasets in math (-8.0% relative to SOTA), web code (+14.3%), STEM (+24.5%) and medical (+8.6%). Essential-Web v1.0 is available on HuggingFace: https://huggingface.co/datasets/EssentialAI/essential-web-v1.0
URLs: https://huggingface.co/datasets/EssentialAI/essential-web-v1.0
Authors: Mikael M{\o}ller H{\o}gsgaard, Andrea Paudice
Abstract: The Median of Means (MoM) is a mean estimator that has gained popularity in the context of heavy-tailed data. In this work, we analyze its performance in the task of simultaneously estimating the mean of each function in a class $\mathcal{F}$ when the data distribution possesses only the first $p$ moments for $p \in (1,2]$. We prove a new sample complexity bound using a novel symmetrization technique that may be of independent interest. Additionally, we present applications of our result to $k$-means clustering with unbounded inputs and linear regression with general losses, improving upon existing works.
Authors: Jakob De Moor, Hans Weytjens, Johannes De Smedt, Jochen De Weerdt
Abstract: Prescriptive Process Monitoring (PresPM) is an emerging area within Process Mining, focused on optimizing processes through real-time interventions for effective decision-making. PresPM holds significant promise for organizations seeking enhanced operational performance. However, the current literature faces two key limitations: a lack of extensive comparisons between techniques and insufficient evaluation approaches. To address these gaps, we introduce SimBank: a simulator designed for accurate benchmarking of PresPM methods. Modeled after a bank's loan application process, SimBank enables extensive comparisons of both online and offline PresPM methods. It incorporates a variety of intervention optimization problems with differing levels of complexity and supports experiments on key causal machine learning challenges, such as assessing a method's robustness to confounding in data. SimBank additionally offers a comprehensive evaluation capability: for each test case, it can generate the true outcome under each intervention action, which is not possible using recorded datasets. The simulator incorporates parallel activities and loops, drawing from common logs to generate cases that closely resemble real-life process instances. Our proof of concept demonstrates SimBank's benchmarking capabilities through experiments with various PresPM methods across different interventions, highlighting its value as a publicly available simulator for advancing research and practice in PresPM.
Authors: Jules Leguy, Pierre-Antoine Jean, Felipe Torres Figueroa, S\'ebastien Harispe
Abstract: This article introduces WebXAII, an open-source web framework designed to facilitate research on human interaction with eXplainable Artificial Intelligence (XAI) systems. The field of XAI is rapidly expanding, driven by the growing societal implications of the widespread adoption of AI (and in particular machine learning) across diverse applications. Researchers who study the interaction between humans and XAI techniques typically develop ad hoc interfaces in order to conduct their studies. These interfaces are usually not shared alongside the results of the studies, which limits their reusability and the reproducibility of experiments. In response, we design and implement WebXAII, a web-based platform that can embody full experimental protocols, meaning that it can present all aspects of the experiment to human participants and record their responses. The experimental protocols are translated into a composite architecture of generic views and modules, which offers a lot of flexibility. The architecture is defined in a structured configuration file, so that protocols can be implemented with minimal programming skills. We demonstrate that WebXAII can effectively embody relevant protocols, by reproducing the protocol of a state-of-the-art study of the literature.
Authors: Varun Mannam, Zhenyu Shi
Abstract: Accurate video annotation plays a vital role in modern retail applications, including customer behavior analysis, product interaction detection, and in-store activity recognition. However, conventional annotation methods heavily rely on time-consuming manual labeling by human annotators, introducing non-robust frame selection and increasing operational costs. To address these challenges in the retail domain, we propose a deep learning-based approach that automates key-frame identification in retail videos and provides automatic annotations of products and customers. Our method leverages deep neural networks to learn discriminative features by embedding video frames and incorporating object detection-based techniques tailored for retail environments. Experimental results showcase the superiority of our approach over traditional methods, achieving accuracy comparable to human annotator labeling while enhancing the overall efficiency of retail video annotation. Remarkably, our approach leads to an average of 2 times cost savings in video annotation. By allowing human annotators to verify/adjust less than 5% of detected frames in the video dataset, while automating the annotation process for the remaining frames without reducing annotation quality, retailers can significantly reduce operational costs. The automation of key-frame detection enables substantial time and effort savings in retail video labeling tasks, proving highly valuable for diverse retail applications such as shopper journey analysis, product interaction detection, and in-store security monitoring.