Authors: Xueqiang Ouyang, Jia Wei
Abstract: As a global disease, infertility has always affected human beings. The development of assisted reproductive technology can effectively solve this disease. However, the traditional in vitro fertilization-embryo transfer technology still faces many challenges in improving the success rate of pregnancy, such as the subjectivity of embryo grading and the inefficiency of integrating multi-modal data. Therefore, the introduction of artificial intelligence-based technologies is particularly crucial. This article reviews the application progress of multi-modal artificial intelligence in embryo grading and pregnancy prediction based on different data modalities (including static images, time-lapse videos and structured table data) from a new perspective, and discusses the main challenges in current research, such as the complexity of multi-modal information fusion and data scarcity.
Authors: Wanghan Xu, Wenlong Zhang, Fenghua Ling, Ben Fei, Yusong Hu, Fangxuan Ren, Jintai Lin, Wanli Ouyang, Lei Bai
Abstract: Meta-analysis is a systematic research methodology that synthesizes data from multiple existing studies to derive comprehensive conclusions. This approach not only mitigates limitations inherent in individual studies but also facilitates novel discoveries through integrated data analysis. Traditional meta-analysis involves a complex multi-stage pipeline including literature retrieval, paper screening, and data extraction, which demands substantial human effort and time. However, while LLM-based methods can accelerate certain stages, they still face significant challenges, such as hallucinations in paper screening and data extraction. In this paper, we propose a multi-agent system, Manalyzer, which achieves end-to-end automated meta-analysis through tool calls. The hybrid review, hierarchical extraction, self-proving, and feedback checking strategies implemented in Manalyzer significantly alleviate these two hallucinations. To comprehensively evaluate the performance of meta-analysis, we construct a new benchmark comprising 729 papers across 3 domains, encompassing text, image, and table modalities, with over 10,000 data points. Extensive experiments demonstrate that Manalyzer achieves significant performance improvements over the LLM baseline in multi meta-analysis tasks. Project page: https://black-yt.github.io/meta-analysis-page/ .
Authors: Son Tran, Edjard Mota, Artur d'Avila Garcez
Abstract: Knowledge representation and reasoning in neural networks have been a long-standing endeavor which has attracted much attention recently. The principled integration of reasoning and learning in neural networks is a main objective of the area of neurosymbolic Artificial Intelligence (AI). In this chapter, a simple energy-based neurosymbolic AI system is described that can represent and reason formally about any propositional logic formula. This creates a powerful combination of learning from data and knowledge and logical reasoning. We start by positioning neurosymbolic AI in the context of the current AI landscape that is unsurprisingly dominated by Large Language Models (LLMs). We identify important challenges of data efficiency, fairness and safety of LLMs that might be addressed by neurosymbolic reasoning systems with formal reasoning capabilities. We then discuss the representation of logic by the specific energy-based system, including illustrative examples and empirical evaluation of the correspondence between logical reasoning and energy minimization using Restricted Boltzmann Machines (RBM). Learning from data and knowledge is also evaluated empirically and compared with a symbolic, neural and a neurosymbolic system. Results reported in this chapter in an accessible way are expected to reignite the research on the use of neural networks as massively-parallel models for logical reasoning and promote the principled integration of reasoning and learning in deep networks. We conclude the chapter with a discussion of the importance of positioning neurosymbolic AI within a broader framework of formal reasoning and accountability in AI, discussing the challenges for neurosynbolic AI to tackle the various known problems of reliability of deep learning.
Authors: Yingpeng Du, Tianjun Wei, Zhu Sun, Jie Zhang
Abstract: Large Language Models (LLMs) have been widely adopted in ranking systems such as information retrieval (IR) systems and recommender systems (RSs). To alleviate the latency of auto-regressive decoding, some studies explore the single (first) token decoding for ranking approximation, but they suffer from severe degradation in tail positions. Although speculative decoding (SD) methods can be a remedy with verification at different positions, they face challenges in ranking systems due to their left-to-right decoding paradigm. Firstly, ranking systems require strict latency constraints, but verification rounds in SD methods remain agnostic; Secondly, SD methods usually discard listwise ranking knowledge about unaccepted items in previous rounds, hindering future multi-token prediction, especially when candidate tokens are the unaccepted items. In this paper, we propose a Reinforcement Speculative Decoding method for fast ranking inference of LLMs. To meet the ranking systems' latency requirement, we propose an up-to-down decoding paradigm that employs an agent to iteratively modify the ranking sequence under a constrained budget. Specifically, we design a ranking-tailored policy optimization, actively exploring optimal multi-round ranking modification policy verified by LLMs via reinforcement learning (RL). To better approximate the target LLM under the constrained budget, we trigger the agent fully utilizing the listwise ranking knowledge about all items verified by LLMs across different rounds in RL, enhancing the modification policy of the agent. More importantly, we demonstrate the theoretical robustness and advantages of our paradigm and implementation. Experiments on both IR and RS tasks show the effectiveness of our proposed method.
Authors: Antoni Gomila, Vincent C. M\"uller
Abstract: The declared goal of this paper is to fill this gap: "... cognitive systems research needs questions or challenges that define progress. The challenges are not (yet more) predictions of the future, but a guideline to what are the aims and what would constitute progress." -- the quotation being from the project description of EUCogII, the project for the European Network for Cognitive Systems within which this formulation of the 'challenges' was originally developed (http://www.eucognition.org). So, we stick out our neck and formulate the challenges for artificial cognitive systems. These challenges are articulated in terms of a definition of what a cognitive system is: a system that learns from experience and uses its acquired knowledge (both declarative and practical) in a flexible manner to achieve its own goals.
Authors: Paul de Font-Reaulx
Abstract: Value learning is a crucial aspect of safe and ethical AI. This is primarily pursued by methods inferring human values from behaviour. However, humans care about much more than we are able to demonstrate through our actions. Consequently, an AI must predict the rest of our seemingly complex values from a limited sample. I call this the value generalization problem. In this paper, I argue that human values have a generative rational structure and that this allows us to solve the value generalization problem. In particular, we can use Bayesian Theory of Mind models to infer human values not only from behaviour, but also from other values. This has been obscured by the widespread use of simple utility functions to represent human values. I conclude that developing generative value-to-value inference is a crucial component of achieving a scalable machine theory of mind.
Authors: Meng Cao, Shuyuan Zhang, Xiao-Wen Chang, Doina Precup
Abstract: Reinforcement Learning from Human Feedback (RLHF) is a widely used technique for aligning Large Language Models (LLMs) with human preferences, yet it often suffers from sparse reward signals, making effective credit assignment challenging. In typical setups, the reward model provides a single scalar score for an entire generated sequence, offering little insight into which token or span-level decisions were responsible for the outcome. To address this, we propose Shapley Credit Assignment Rewards (SCAR), a novel method that leverages Shapley values in cooperative game theory. SCAR distributes the total sequence-level reward among constituent tokens or text spans based on their principled marginal contributions. This creates dense reward signals, crucially, without necessitating the training of auxiliary critique models or recourse to fine-grained human annotations at intermediate generation stages. Unlike prior dense reward methods, SCAR offers a game-theoretic foundation for fair credit attribution. Theoretically, we demonstrate that SCAR preserves the original optimal policy, and empirically, across diverse tasks including sentiment control, text summarization, and instruction tuning, we show that SCAR converges significantly faster and achieves higher final reward scores compared to standard RLHF and attention-based dense reward baselines. Our findings suggest that SCAR provides a more effective and theoretically sound method for credit assignment in RLHF, leading to more efficient alignment of LLMs.
Authors: P. S. Kesavan, Pontus Nordenfelt
Abstract: Smart microscopy represents a paradigm shift in biological imaging, moving from passive observation tools to active collaborators in scientific inquiry. Enabled by advances in automation, computational power, and artificial intelligence, these systems are now capable of adaptive decision-making and real-time experimental control. Here, we introduce a theoretical framework that reconceptualizes smart microscopy as a partner in scientific investigation. Central to our framework is the concept of the 'epistemic-empirical divide' in cellular investigation-the gap between what is observable (empirical domain) and what must be understood (epistemic domain). We propose six core design principles: epistemic-empirical awareness, hierarchical context integration, an evolution from detection to perception, adaptive measurement frameworks, narrative synthesis capabilities, and cross-contextual reasoning. Together, these principles guide a multi-agent architecture designed to align empirical observation with the goals of scientific understanding. Our framework provides a roadmap for building microscopy systems that go beyond automation to actively support hypothesis generation, insight discovery, and theory development, redefining the role of scientific instruments in the process of knowledge creation.
Authors: Ana Rita Ortigoso, Gabriel Vieira, Daniel Fuentes, Luis Fraz\~ao, Nuno Costa, Ant\'onio Pereira
Abstract: This paper presents Project Riley, a novel multimodal and multi-model conversational AI architecture oriented towards the simulation of reasoning influenced by emotional states. Drawing inspiration from Pixar's Inside Out, the system comprises five distinct emotional agents - Joy, Sadness, Fear, Anger, and Disgust - that engage in structured multi-round dialogues to generate, criticise, and iteratively refine responses. A final reasoning mechanism synthesises the contributions of these agents into a coherent output that either reflects the dominant emotion or integrates multiple perspectives. The architecture incorporates both textual and visual large language models (LLMs), alongside advanced reasoning and self-refinement processes. A functional prototype was deployed locally in an offline environment, optimised for emotional expressiveness and computational efficiency. From this initial prototype, another one emerged, called Armando, which was developed for use in emergency contexts, delivering emotionally calibrated and factually accurate information through the integration of Retrieval-Augmented Generation (RAG) and cumulative context tracking. The Project Riley prototype was evaluated through user testing, in which participants interacted with the chatbot and completed a structured questionnaire assessing three dimensions: Emotional Appropriateness, Clarity and Utility, and Naturalness and Human-likeness. The results indicate strong performance in structured scenarios, particularly with respect to emotional alignment and communicative clarity.
Authors: Jian Wang, Boyan Zhu, Chak Tou Leong, Yongqi Li, Wenjie Li
Abstract: Large reasoning models (LRMs) have exhibited the capacity of enhancing reasoning performance via internal test-time scaling. Building upon this, a promising direction is to further scale test-time compute to unlock even greater reasoning capabilities. However, as we push these scaling boundaries, systematically understanding the practical limits and achieving optimal resource allocation becomes a critical challenge. In this paper, we investigate the scaling Pareto of test-time scaling and introduce the Test-Time Scaling Performance Model (TTSPM). We theoretically analyze two fundamental paradigms for such extended scaling, parallel scaling and sequential scaling, from a probabilistic modeling perspective. Our primary contribution is the derivation of the saturation point on the scaling budget for both strategies, identifying thresholds beyond which additional computation yields diminishing returns. Remarkably, despite their distinct mechanisms, both paradigms converge to a unified mathematical structure in their upper bounds. We empirically validate our theoretical findings on challenging reasoning benchmarks, including AIME, MATH-500, and GPQA, demonstrating the practical utility of these bounds for test-time resource allocation. We hope that this work provides insights into the cost-benefit trade-offs of test-time scaling, guiding the development of more resource-efficient inference strategies for large reasoning models.
Authors: Hyungjun Park (Department of Pulmonology, Shihwa Medical Center, Siheung, Republic of Korea, Helpmedoc Inc., Republic of Korea), Chang-Yun Woo (Department of Internal Medicine, Asan Medical Center, Seoul, Republic of Korea), Seungjo Lim (Helpmedoc Inc., Republic of Korea), Seunghwan Lim (Helpmedoc Inc., Republic of Korea), Keunho Kwak (Helpmedoc Inc., Republic of Korea), Ju Young Jeong (Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea), Chong Hyun Suh (Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea)
Abstract: Objective To develop an LLM based realtime compound diagnostic medical AI interface and performed a clinical trial comparing this interface and physicians for common internal medicine cases based on the United States Medical License Exam (USMLE) Step 2 Clinical Skill (CS) style exams. Methods A nonrandomized clinical trial was conducted on August 20, 2024. We recruited one general physician, two internal medicine residents (2nd and 3rd year), and five simulated patients. The clinical vignettes were adapted from the USMLE Step 2 CS style exams. We developed 10 representative internal medicine cases based on actual patients and included information available on initial diagnostic evaluation. Primary outcome was the accuracy of the first differential diagnosis. Repeatability was evaluated based on the proportion of agreement. Results The accuracy of the physicians' first differential diagnosis ranged from 50% to 70%, whereas the realtime compound diagnostic medical AI interface achieved an accuracy of 80%. The proportion of agreement for the first differential diagnosis was 0.7. The accuracy of the first and second differential diagnoses ranged from 70% to 90% for physicians, whereas the AI interface achieved an accuracy rate of 100%. The average time for the AI interface (557 sec) was 44.6% shorter than that of the physicians (1006 sec). The AI interface ($0.08) also reduced costs by 98.1% compared to the physicians' average ($4.2). Patient satisfaction scores ranged from 4.2 to 4.3 for care by physicians and were 3.9 for the AI interface Conclusion An LLM based realtime compound diagnostic medical AI interface demonstrated diagnostic accuracy and patient satisfaction comparable to those of a physician, while requiring less time and lower costs. These findings suggest that AI interfaces may have the potential to assist primary care consultations for common internal medicine cases.
Authors: Yi Zhan, Qi Liu, Weibo Gao, Zheng Zhang, Tianfu Wang, Shuanghong Shen, Junyu Lu, Zhenya Huang
Abstract: Personalized programming tutoring, such as exercise recommendation, can enhance learners' efficiency, motivation, and outcomes, which is increasingly important in modern digital education. However, the lack of sufficient and high-quality programming data, combined with the mismatch between offline evaluation and real-world learning, hinders the practical deployment of such systems. To address this challenge, many approaches attempt to simulate learner practice data, yet they often overlook the fine-grained, iterative nature of programming learning, resulting in a lack of interpretability and granularity. To fill this gap, we propose a LLM-based agent, CoderAgent, to simulate students' programming processes in a fine-grained manner without relying on real data. Specifically, we equip each human learner with an intelligent agent, the core of which lies in capturing the cognitive states of the human programming practice process. Inspired by ACT-R, a cognitive architecture framework, we design the structure of CoderAgent to align with human cognitive architecture by focusing on the mastery of programming knowledge and the application of coding ability. Recognizing the inherent patterns in multi-layered cognitive reasoning, we introduce the Programming Tree of Thought (PTOT), which breaks down the process into four steps: why, how, where, and what. This approach enables a detailed analysis of iterative problem-solving strategies. Finally, experimental evaluations on real-world datasets demonstrate that CoderAgent provides interpretable insights into learning trajectories and achieves accurate simulations, paving the way for personalized programming education.
Authors: Xuanle Zhao, Zilin Sang, Yuxuan Li, Qi Shi, Shuo Wang, Duzhen Zhang, Xu Han, Zhiyuan Liu, Maosong Sun
Abstract: Efficient experiment reproduction is critical to accelerating progress in artificial intelligence. However, the inherent complexity of method design and training procedures presents substantial challenges for automation. Notably, reproducing experiments often requires implicit domain-specific knowledge not explicitly documented in the original papers. To address this, we introduce the paper lineage algorithm, which identifies and extracts implicit knowledge from the relevant references cited by the target paper. Building on this idea, we propose AutoReproduce, a multi-agent framework capable of automatically reproducing experiments described in research papers in an end-to-end manner. AutoReproduce enhances code executability by generating unit tests alongside the reproduction process. To evaluate the reproduction capability, we construct ReproduceBench, a benchmark annotated with verified implementations, and introduce novel evaluation metrics to assess both the reproduction and execution fidelity. Experimental results demonstrate that AutoReproduce outperforms the existing strong agent baselines on all five evaluation metrics by a peak margin of over $70\%$. In particular, compared to the official implementations, AutoReproduce achieves an average performance gap of $22.1\%$ on $89.74\%$ of the executable experiment runs. The code will be available at https://github.com/AI9Stars/AutoReproduce.
Authors: Zikang Guo, Benfeng Xu, Xiaorui Wang, Zhendong Mao
Abstract: Complex tasks involving tool integration pose significant challenges for Large Language Models (LLMs), leading to the emergence of multi-agent workflows as a promising solution. Reflection has emerged as an effective strategy for correcting erroneous trajectories in agentic workflows. However, existing approaches only exploit such capability in the post-action stage, where the agent observes the execution outcomes. We argue that, like humans, LLMs can also engage in reflection before action execution: the agent can anticipate undesirable outcomes from its own decisions, which not only provides a necessarily complementary perspective to evaluate the decision but also prevents the propagation of errors throughout the trajectory. In this paper, we propose MIRROR, a framework that consists of both intra-reflection, which critically assesses intended actions before execution, and inter-reflection, which further adjusts the trajectory based on observations. This design systematically leverages LLM reflection capabilities to eliminate and rectify erroneous actions on a more comprehensive scope. Evaluations on both the StableToolBench and TravelPlanner benchmarks demonstrate MIRROR's superior performance, achieving state-of-the-art results compared to existing approaches.
Authors: Heng Tan, Hua Yan, Yu Yang
Abstract: While reinforcement learning (RL) has achieved notable success in various domains, training effective policies for complex tasks remains challenging. Agents often converge to local optima and fail to maximize long-term rewards. Existing approaches to mitigate training bottlenecks typically fall into two categories: (i) Automated policy refinement, which identifies critical states from past trajectories to guide policy updates, but suffers from costly and uncertain model training; and (ii) Human-in-the-loop refinement, where human feedback is used to correct agent behavior, but this does not scale well to environments with large or continuous action spaces. In this work, we design a large language model-guided policy modulation framework that leverages LLMs to improve RL training without additional model training or human intervention. We first prompt an LLM to identify critical states from a sub-optimal agent's trajectories. Based on these states, the LLM then provides action suggestions and assigns implicit rewards to guide policy refinement. Experiments across standard RL benchmarks demonstrate that our method outperforms state-of-the-art baselines, highlighting the effectiveness of LLM-based explanations in addressing RL training bottlenecks.
Authors: Woochang Sim, Hyunseok Ryu, Kyungmin Choi, Sungwon Han, Sundong Kim
Abstract: The Abstraction and Reasoning Corpus (ARC) poses a stringent test of general AI capabilities, requiring solvers to infer abstract patterns from only a handful of examples. Despite substantial progress in deep learning, state-of-the-art models still achieve accuracy rates of merely 40-55% on 2024 ARC Competition, indicative of a significant gap between their performance and human-level reasoning. In this work, we seek to bridge that gap by introducing an analogy-inspired ARC dataset, GIFARC. Leveraging large language models (LLMs) and vision-language models (VLMs), we synthesize new ARC-style tasks from a variety of GIF images that include analogies. Each new task is paired with ground-truth analogy, providing an explicit mapping between visual transformations and everyday concepts. By embedding robust human-intuitive analogies into ARC-style tasks, GIFARC guides AI agents to evaluate the task analogically before engaging in brute-force pattern search, thus efficiently reducing problem complexity and build a more concise and human-understandable solution. We empirically validate that guiding LLM with analogic approach with GIFARC affects task-solving approaches of LLMs to align with analogic approach of human.
Authors: Zesen Lyu, Dandan Zhang, Wei Ye, Fangdi Li, Zhihang Jiang, Yao Yang
Abstract: Spatial reasoning is a core component of human cognition, enabling individuals to perceive, comprehend, and interact with the physical world. It relies on a nuanced understanding of spatial structures and inter-object relationships, serving as the foundation for complex reasoning and decision-making. To investigate whether current vision-language models (VLMs) exhibit similar capability, we introduce Jigsaw-Puzzles, a novel benchmark consisting of 1,100 carefully curated real-world images with high spatial complexity. Based on this dataset, we design five tasks to rigorously evaluate VLMs' spatial perception, structural understanding, and reasoning capabilities, while deliberately minimizing reliance on domain-specific knowledge to better isolate and assess the general spatial reasoning capability. We conduct a comprehensive evaluation across 24 state-of-the-art VLMs. The results show that even the strongest model, Gemini-2.5-Pro, achieves only 77.14% overall accuracy and performs particularly poorly on the Order Generation task, with only 30.00% accuracy, far below the performance exceeding 90% achieved by human participants. This persistent gap underscores the need for continued progress, positioning Jigsaw-Puzzles as a challenging and diagnostic benchmark for advancing spatial reasoning research in VLMs.
Authors: Cheonsu Jeong, Seongmin Sim, Hyoyoung Cho, Sungsu Kim, Byounggwan Shin
Abstract: This paper presents an intelligent work automation approach in the context of contemporary digital transformation by integrating generative AI and Intelligent Document Processing (IDP) technologies with an Automation Agent to realize End-to-End (E2E) automation of corporate financial expense processing tasks. While traditional Robotic Process Automation (RPA) has proven effective for repetitive, rule-based simple task automation, it faces limitations in handling unstructured data, exception management, and complex decision-making. This study designs and implements a four-stage integrated process comprising automatic recognition of supporting documents such as receipts via OCR/IDP, item classification based on a policy-driven database, intelligent exception handling supported by generative AI (large language models, LLMs), and human-in-the-loop final decision-making with continuous system learning through an Automation Agent. Applied to a major Korean enterprise (Company S), the system demonstrated quantitative benefits including over 80% reduction in processing time for paper receipt expense tasks, decreased error rates, and improved compliance, as well as qualitative benefits such as enhanced accuracy and consistency, increased employee satisfaction, and data-driven decision support. Furthermore, the system embodies a virtuous cycle by learning from human judgments to progressively improve automatic exception handling capabilities. Empirically, this research confirms that the organic integration of generative AI, IDP, and Automation Agents effectively overcomes the limitations of conventional automation and enables E2E automation of complex corporate processes. The study also discusses potential extensions to other domains such as accounting, human resources, and procurement, and proposes future directions for AI-driven hyper-automation development.
Authors: Zilong Wang, Jingfeng Yang, Sreyashi Nag, Samarth Varshney, Xianfeng Tang, Haoming Jiang, Jingbo Shang, Sheikh Muhammad Sarwar
Abstract: Large language models (LLMs) have exhibited extraordinary performance in a variety of tasks while it remains challenging for them to solve complex multi-step tasks as agents. In practice, agents sensitive to the outcome of certain key steps which makes them likely to fail the task because of a subtle mistake in the planning trajectory. Recent approaches resort to calibrating the reasoning process through reinforcement learning. They reward or penalize every reasoning step with process supervision, as known as Process Reward Models (PRMs). However, PRMs are difficult and costly to scale up with a large number of next action candidates since they require extensive computations to acquire the training data through the per-step trajectory exploration. To mitigate this issue, we focus on the relative reward trend across successive reasoning steps and propose maintaining an increasing reward in the collected trajectories for process supervision, which we term Reward Rising Optimization (RRO). Specifically, we incrementally augment the process supervision until identifying a step exhibiting positive reward differentials, i.e. rising rewards, relative to its preceding iteration. This method dynamically expands the search space for the next action candidates, efficiently capturing high-quality data. We provide mathematical groundings and empirical results on the WebShop and InterCode-SQL benchmarks, showing that our proposed RRO achieves superior performance while requiring much less exploration cost.
Authors: Xiangyu Zhao, Wanghan Xu, Bo Liu, Yuhao Zhou, Fenghua Ling, Ben Fei, Xiaoyu Yue, Lei Bai, Wenlong Zhang, Xiao-Ming Wu
Abstract: The rapid advancement of multimodal large language models (MLLMs) has unlocked new opportunities to tackle complex scientific challenges. Despite this progress, their application in addressing earth science problems, especially at the graduate level, remains underexplored. A significant barrier is the absence of benchmarks that capture the depth and contextual complexity of geoscientific reasoning. Current benchmarks often rely on synthetic datasets or simplistic figure-caption pairs, which do not adequately reflect the intricate reasoning and domain-specific insights required for real-world scientific applications. To address these gaps, we introduce MSEarth, a multimodal scientific benchmark curated from high-quality, open-access scientific publications. MSEarth encompasses the five major spheres of Earth science: atmosphere, cryosphere, hydrosphere, lithosphere, and biosphere, featuring over 7K figures with refined captions. These captions are crafted from the original figure captions and enriched with discussions and reasoning from the papers, ensuring the benchmark captures the nuanced reasoning and knowledge-intensive content essential for advanced scientific tasks. MSEarth supports a variety of tasks, including scientific figure captioning, multiple choice questions, and open-ended reasoning challenges. By bridging the gap in graduate-level benchmarks, MSEarth provides a scalable and high-fidelity resource to enhance the development and evaluation of MLLMs in scientific reasoning. The benchmark is publicly available to foster further research and innovation in this field. Resources related to this benchmark can be found at https://huggingface.co/MSEarth and https://github.com/xiangyu-mm/MSEarth.
URLs: https://huggingface.co/MSEarth, https://github.com/xiangyu-mm/MSEarth.
Authors: Alfin Wijaya Rahardja, Junwei Liu, Weitong Chen, Zhenpeng Chen, Yiling Lou
Abstract: LLM-based agent systems are emerging as a new software paradigm and have been widely adopted across diverse domains such as medicine, robotics, and programming. However, maintaining these systems requires substantial effort, as they are inevitably prone to bugs and continually evolve to meet changing external requirements. Therefore, automatically resolving agent issues (i.e., bug reports or feature requests) is a crucial and challenging task. While recent software engineering (SE) agents (e.g., SWE-agent) have shown promise in addressing issues in traditional software systems, it remains unclear how effectively they can resolve real-world issues in agent systems, which differ significantly from traditional software. To fill this gap, we first manually analyze 201 real-world agent issues and identify common categories of agent issues. We then spend 500 person-hours constructing AGENTISSUE-BENCH, a reproducible benchmark comprising 50 agent issue resolution tasks (each with an executable environment and failure-triggering tests). We further evaluate state-of-the-art SE agents on AGENTISSUE-BENCH and reveal their limited effectiveness (i.e., with only 3.33% - 12.67% resolution rates). These results underscore the unique challenges of maintaining agent systems compared to traditional software, highlighting the need for further research to develop advanced SE agents for resolving agent issues. Data and code are available at https://alfin06.github.io/AgentIssue-Bench-Leaderboard/#/ .
URLs: https://alfin06.github.io/AgentIssue-Bench-Leaderboard/
Authors: Hyomin Kim, Yunhui Jang, Sungsoo Ahn
Abstract: Large language models (LLMs) have large potential for molecular optimization, as they can gather external chemistry tools and enable collaborative interactions to iteratively refine molecular candidates. However, this potential remains underexplored, particularly in the context of structured reasoning, interpretability, and comprehensive tool-grounded molecular optimization. To address this gap, we introduce MT-Mol, a multi-agent framework for molecular optimization that leverages tool-guided reasoning and role-specialized LLM agents. Our system incorporates comprehensive RDKit tools, categorized into five distinct domains: structural descriptors, electronic and topological features, fragment-based functional groups, molecular representations, and miscellaneous chemical properties. Each category is managed by an expert analyst agent, responsible for extracting task-relevant tools and enabling interpretable, chemically grounded feedback. MT-Mol produces molecules with tool-aligned and stepwise reasoning through the interaction between the analyst agents, a molecule-generating scientist, a reasoning-output verifier, and a reviewer agent. As a result, we show that our framework shows the state-of-the-art performance of the PMO-1K benchmark on 17 out of 23 tasks.
Authors: Kuo Zhou, Lu Zhang
Abstract: Large Language Models (LLMs) have demonstrated formidable capabilities in solving mathematical problems, yet they may still commit logical reasoning and computational errors during the problem-solving process. Thus, this paper proposes a framework, MATH-VF, which includes a Formalizer and a Critic, for formally verifying the correctness of the solutions generated by large language models. Our framework first utilizes a Formalizer which employs an LLM to translate a natural language solution into a formal context. Afterward, our Critic (which integrates various external tools such as a Computer Algebra System and an SMT solver) evaluates the correctness of each statement within the formal context, and when a statement is incorrect, our Critic provides corrective feedback. We empirically investigate the effectiveness of MATH-VF in two scenarios: 1) Verification: MATH-VF is utilized to determine the correctness of a solution to a given problem. 2) Refinement: When MATH-VF identifies errors in the solution generated by an LLM-based solution generator for a given problem, it submits the corrective suggestions proposed by the Critic to the solution generator to regenerate the solution. We evaluate our framework on widely used mathematical benchmarks: MATH500 and ProcessBench, demonstrating the superiority of our approach over existing approaches.
Authors: Leizhen Wang, Peibo Duan, Cheng Lyu, Zhenliang Ma
Abstract: Modern navigation systems and shared mobility platforms increasingly rely on personalized route recommendations to improve individual travel experience and operational efficiency. However, a key question remains: can such sequential, personalized routing decisions collectively lead to system-optimal (SO) traffic assignment? This paper addresses this question by proposing a learning-based framework that reformulates the static SO traffic assignment problem as a single-agent deep reinforcement learning (RL) task. A central agent sequentially recommends routes to travelers as origin-destination (OD) demands arrive, to minimize total system travel time. To enhance learning efficiency and solution quality, we develop an MSA-guided deep Q-learning algorithm that integrates the iterative structure of traditional traffic assignment methods into the RL training process. The proposed approach is evaluated on both the Braess and Ortuzar-Willumsen (OW) networks. Results show that the RL agent converges to the theoretical SO solution in the Braess network and achieves only a 0.35% deviation in the OW network. Further ablation studies demonstrate that the route action set's design significantly impacts convergence speed and final performance, with SO-informed route sets leading to faster learning and better outcomes. This work provides a theoretically grounded and practically relevant approach to bridging individual routing behavior with system-level efficiency through learning-based sequential assignment.
Authors: Yisen Gao, Jiaxin Bai, Tianshi Zheng, Qingyun Sun, Ziwei Zhang, Jianxin Li, Yangqiu Song, Xingcheng Fu
Abstract: Abductive reasoning in knowledge graphs aims to generate plausible logical hypotheses from observed entities, with broad applications in areas such as clinical diagnosis and scientific discovery. However, due to a lack of controllability, a single observation may yield numerous plausible but redundant or irrelevant hypotheses on large-scale knowledge graphs. To address this limitation, we introduce the task of controllable hypothesis generation to improve the practical utility of abductive reasoning. This task faces two key challenges when controlling for generating long and complex logical hypotheses: hypothesis space collapse and hypothesis oversensitivity. To address these challenges, we propose CtrlHGen, a Controllable logcial Hypothesis Generation framework for abductive reasoning over knowledge graphs, trained in a two-stage paradigm including supervised learning and subsequent reinforcement learning. To mitigate hypothesis space collapse, we design a dataset augmentation strategy based on sub-logical decomposition, enabling the model to learn complex logical structures by leveraging semantic patterns in simpler components. To address hypothesis oversensitivity, we incorporate smoothed semantic rewards including Dice and Overlap scores, and introduce a condition-adherence reward to guide the generation toward user-specified control constraints. Extensive experiments on three benchmark datasets demonstrate that our model not only better adheres to control conditions but also achieves superior semantic similarity performance compared to baselines.
Authors: Lingyi Cai, Ruichen Zhang, Changyuan Zhao, Yu Zhang, Jiawen Kang, Dusit Niyato, Tao Jiang, Xuemin Shen
Abstract: Low-Altitude Economic Networking (LAENet) aims to support diverse flying applications below 1,000 meters by deploying various aerial vehicles for flexible and cost-effective aerial networking. However, complex decision-making, resource constraints, and environmental uncertainty pose significant challenges to the development of the LAENet. Reinforcement learning (RL) offers a potential solution in response to these challenges but has limitations in generalization, reward design, and model stability. The emergence of large language models (LLMs) offers new opportunities for RL to mitigate these limitations. In this paper, we first present a tutorial about integrating LLMs into RL by using the capacities of generation, contextual understanding, and structured reasoning of LLMs. We then propose an LLM-enhanced RL framework for the LAENet in terms of serving the LLM as information processor, reward designer, decision-maker, and generator. Moreover, we conduct a case study by using LLMs to design a reward function to improve the learning performance of RL in the LAENet. Finally, we provide a conclusion and discuss future work.
Authors: Kaiming Liu, Xuanyu Lei, Ziyue Wang, Peng Li, Yang Liu
Abstract: Large language model (LLM) agents have shown impressive reasoning capabilities in interactive decision-making tasks. These agents interact with environment through intermediate interfaces, such as predefined action spaces and interaction rules, which mediate the perception and action. However, mismatches often happen between the internal expectations of the agent regarding the influence of its issued actions and the actual state transitions in the environment, a phenomenon referred to as \textbf{agent-environment misalignment}. While prior work has invested substantially in improving agent strategies and environment design, the critical role of the interface still remains underexplored. In this work, we empirically demonstrate that agent-environment misalignment poses a significant bottleneck to agent performance. To mitigate this issue, we propose \textbf{ALIGN}, an \underline{A}uto-A\underline{l}igned \underline{I}nterface \underline{G}e\underline{n}eration framework that alleviates the misalignment by enriching the interface. Specifically, the ALIGN-generated interface enhances both the static information of the environment and the step-wise observations returned to the agent. Implemented as a lightweight wrapper, this interface achieves the alignment without modifying either the agent logic or the environment code. Experiments across multiple domains including embodied tasks, web navigation and tool-use, show consistent performance improvements, with up to a 45.67\% success rate improvement observed in ALFWorld. Meanwhile, ALIGN-generated interface can generalize across different agent architectures and LLM backbones without interface regeneration. Code and experimental results are available at https://github.com/THUNLP-MT/ALIGN.
Authors: Xiao Hu, Xingyu Lu, Liyuan Mao, YiFan Zhang, Tianke Zhang, Bin Wen, Fan Yang, Tingting Gao, Guorui Zhou
Abstract: Reinforcement learning (RL) has played an important role in improving the reasoning ability of large language models (LLMs). Some studies apply RL directly to \textit{smaller} base models (known as zero-RL) and also achieve notable progress. However, in this paper, we show that using only 920 examples, a simple distillation method based on the base model can clearly outperform zero-RL, which typically requires much more data and computational cost. By analyzing the token frequency in model outputs, we find that the distilled model shows more flexible reasoning. It uses anthropomorphic tokens and logical connectors much more often than the zero-RL model. Further analysis reveals that distillation enhances the presence of two advanced cognitive behaviors: Multi-Perspective Thinking or Attempting and Metacognitive Awareness. Frequent occurrences of these two advanced cognitive behaviors give rise to flexible reasoning, which is essential for solving complex reasoning problems, while zero-RL fails to significantly boost the frequency of these behaviors.
Authors: Zhengyang Ji, Yifan Jia, Shang Gao, Yutao Yue
Abstract: Large Vision Language Models (LVLMs) have achieved remarkable progress in multimodal tasks, yet they also exhibit notable social biases. These biases often manifest as unintended associations between neutral concepts and sensitive human attributes, leading to disparate model behaviors across demographic groups. While existing studies primarily focus on detecting and quantifying such biases, they offer limited insight into the underlying mechanisms within the models. To address this gap, we propose an explanatory framework that combines information flow analysis with multi-round dialogue evaluation, aiming to understand the origin of social bias from the perspective of imbalanced internal information utilization. Specifically, we first identify high-contribution image tokens involved in the model's reasoning process for neutral questions via information flow analysis. Then, we design a multi-turn dialogue mechanism to evaluate the extent to which these key tokens encode sensitive information. Extensive experiments reveal that LVLMs exhibit systematic disparities in information usage when processing images of different demographic groups, suggesting that social bias is deeply rooted in the model's internal reasoning dynamics. Furthermore, we complement our findings from a textual modality perspective, showing that the model's semantic representations already display biased proximity patterns, thereby offering a cross-modal explanation of bias formation.
Authors: Martin C. Cooper, Imane Bousdira, Cl\'ement Carbonnel
Abstract: A classifier is considered interpretable if each of its decisions has an explanation which is small enough to be easily understood by a human user. A DNF formula can be seen as a binary classifier $\kappa$ over boolean domains. The size of an explanation of a positive decision taken by a DNF $\kappa$ is bounded by the size of the terms in $\kappa$, since we can explain a positive decision by giving a term of $\kappa$ that evaluates to true. Since both positive and negative decisions must be explained, we consider that interpretable DNFs are those $\kappa$ for which both $\kappa$ and $\overline{\kappa}$ can be expressed as DNFs composed of terms of bounded size. In this paper, we study the family of $k$-DNFs whose complements can also be expressed as $k$-DNFs. We compare two such families, namely depth-$k$ decision trees and nested $k$-DNFs, a novel family of models. Experiments indicate that nested $k$-DNFs are an interesting alternative to decision trees in terms of interpretability and accuracy.
Authors: Shaoqing Zhang, Kehai Chen, Zhuosheng Zhang, Rumei Li, Rongxiang Weng, Yang Xiang, Liqiang Nie, Min Zhang
Abstract: Recent advancements in vision-language models (VLMs) have spurred increased interest in Device-Control Agents (DC agents), such as utilizing in-the-wild device control to manage graphical user interfaces. Conventional methods for assessing the capabilities of DC agents, such as computing step-wise action accuracy and overall task success rates, provide a macroscopic view of DC agents' performance; however, they fail to offer microscopic insights into potential errors that may occur in real-world applications. Conducting a finer-grained performance evaluation of DC agents presents significant challenges. This study introduces a new perspective on evaluation methods for DC agents by proposing the XBOUND evaluation method, which employs the calculation of a novel Explore Metric to delineate the capability boundaries of DC agents. Compared to previous evaluation methods, XBOUND focuses on individual states to assess the proficiency of DC agents in mastering these states. Furthermore, we have developed a ``pseudo'' episode tree dataset derived from Android Control test data. Utilizing this dataset and XBOUND, we comprehensively evaluate the OS-Atlas and UI-TARS series, examining both the overall and specific performance across five common tasks. Additionally, we select representative cases to highlight the current deficiencies and limitations inherent in both series. Code is available at https://github.com/sqzhang-lazy/XBOUND.
Authors: Yue Zhang, Zhiliang Tian, Shicheng Zhou, Haiyang Wang, Wenqing Hou, Yuying Liu, Xuechen Zhao, Minlie Huang, Ye Wang, Bin Zhou
Abstract: Legal Judgment Prediction (LJP) is a pivotal task in legal AI. Existing semantic-enhanced LJP models integrate judicial precedents and legal knowledge for high performance. But they neglect legal reasoning logic, a critical component of legal judgments requiring rigorous logical analysis. Although some approaches utilize legal reasoning logic for high-quality predictions, their logic rigidity hinders adaptation to case-specific logical frameworks, particularly in complex cases that are lengthy and detailed. This paper proposes a rule-enhanced legal judgment prediction framework based on first-order logic (FOL) formalism and comparative learning (CL) to develop an adaptive adjustment mechanism for legal judgment logic and further enhance performance in LJP. Inspired by the process of human exam preparation, our method follows a three-stage approach: first, we initialize judgment rules using the FOL formalism to capture complex reasoning logic accurately; next, we propose a Confusion-aware Contrastive Learning (CACL) to dynamically optimize the judgment rules through a quiz consisting of confusable cases; finally, we utilize the optimized judgment rules to predict legal judgments. Experimental results on two public datasets show superior performance across all metrics. The code is publicly available{https://anonymous.4open.science/r/RLJP-FDF1}.
Authors: Saman Marandi, Yu-Shu Hu, Mohammad Modarres
Abstract: In this paper, we present a novel diagnostic framework that integrates Knowledge Graphs (KGs) and Large Language Models (LLMs) to support system diagnostics in high-reliability systems such as nuclear power plants. Traditional diagnostic modeling struggles when systems become too complex, making functional modeling a more attractive approach. Our approach introduces a diagnostic framework grounded in the functional modeling principles of the Dynamic Master Logic (DML) model. It incorporates two coordinated LLM components, including an LLM-based workflow for automated construction of DML logic from system documentation and an LLM agent that facilitates interactive diagnostics. The generated logic is encoded into a structured KG, referred to as KG-DML, which supports hierarchical fault reasoning. Expert knowledge or operational data can also be incorporated to refine the model's precision and diagnostic depth. In the interaction phase, users submit natural language queries, which are interpreted by the LLM agent. The agent selects appropriate tools for structured reasoning, including upward and downward propagation across the KG-DML. Rather than embedding KG content into every prompt, the LLM agent distinguishes between diagnostic and interpretive tasks. For diagnostics, the agent selects and executes external tools that perform structured KG reasoning. For general queries, a Graph-based Retrieval-Augmented Generation (Graph-RAG) approach is used, retrieving relevant KG segments and embedding them into the prompt to generate natural explanations. A case study on an auxiliary feedwater system demonstrated the framework's effectiveness, with over 90% accuracy in key elements and consistent tool and argument extraction, supporting its use in safety-critical diagnostics.
Authors: Hao Li, He Cao, Bin Feng, Yanjun Shao, Xiangru Tang, Zhiyuan Yan, Li Yuan, Yonghong Tian, Yu Li
Abstract: While large language models (LLMs) with Chain-of-Thought (CoT) reasoning excel in mathematics and coding, their potential for systematic reasoning in chemistry, a domain demanding rigorous structural analysis for real-world tasks like drug design and reaction engineering, remains untapped. Current benchmarks focus on simple knowledge retrieval, neglecting step-by-step reasoning required for complex tasks such as molecular optimization and reaction prediction. To address this, we introduce ChemCoTBench, a reasoning framework that bridges molecular structure understanding with arithmetic-inspired operations, including addition, deletion, and substitution, to formalize chemical problem-solving into transparent, step-by-step workflows. By treating molecular transformations as modular "chemical operations", the framework enables slow-thinking reasoning, mirroring the logic of mathematical proofs while grounding solutions in real-world chemical constraints. We evaluate models on two high-impact tasks: Molecular Property Optimization and Chemical Reaction Prediction. These tasks mirror real-world challenges while providing structured evaluability. By providing annotated datasets, a reasoning taxonomy, and baseline evaluations, ChemCoTBench bridges the gap between abstract reasoning methods and practical chemical discovery, establishing a foundation for advancing LLMs as tools for AI-driven scientific innovation.
Authors: R. Spencer Hallyburton, Miroslav Pajic
Abstract: Many state-of-the-art AI models deployed in cyber-physical systems (CPS), while highly accurate, are simply pattern-matchers.~With limited security guarantees, there are concerns for their reliability in safety-critical and contested domains. To advance assured AI, we advocate for a paradigm shift that imbues data-driven perception models with symbolic structure, inspired by a human's ability to reason over low-level features and high-level context. We propose a neuro-symbolic paradigm for perception (NeuSPaPer) and illustrate how joint object detection and scene graph generation (SGG) yields deep scene understanding.~Powered by foundation models for offline knowledge extraction and specialized SGG algorithms for real-time deployment, we design a framework leveraging structured relational graphs that ensures the integrity of situational awareness in autonomy. Using physics-based simulators and real-world datasets, we demonstrate how SGG bridges the gap between low-level sensor perception and high-level reasoning, establishing a foundation for resilient, context-aware AI and advancing trusted autonomy in CPS.
Authors: Jiakang Yuan, Tianshuo Peng, Yilei Jiang, Yiting Lu, Renrui Zhang, Kaituo Feng, Chaoyou Fu, Tao Chen, Lei Bai, Bo Zhang, Xiangyu Yue
Abstract: Logical reasoning is a fundamental aspect of human intelligence and an essential capability for multimodal large language models (MLLMs). Despite the significant advancement in multimodal reasoning, existing benchmarks fail to comprehensively evaluate their reasoning abilities due to the lack of explicit categorization for logical reasoning types and an unclear understanding of reasoning. To address these issues, we introduce MME-Reasoning, a comprehensive benchmark designed to evaluate the reasoning ability of MLLMs, which covers all three types of reasoning (i.e., inductive, deductive, and abductive) in its questions. We carefully curate the data to ensure that each question effectively evaluates reasoning ability rather than perceptual skills or knowledge breadth, and extend the evaluation protocols to cover the evaluation of diverse questions. Our evaluation reveals substantial limitations of state-of-the-art MLLMs when subjected to holistic assessments of logical reasoning capabilities. Even the most advanced MLLMs show limited performance in comprehensive logical reasoning, with notable performance imbalances across reasoning types. In addition, we conducted an in-depth analysis of approaches such as ``thinking mode'' and Rule-based RL, which are commonly believed to enhance reasoning abilities. These findings highlight the critical limitations and performance imbalances of current MLLMs in diverse logical reasoning scenarios, providing comprehensive and systematic insights into the understanding and evaluation of reasoning capabilities.
Authors: Aidan Peppin, Julia Kreutzer, Alice Schoenauer Sebag, Kelly Marchisio, Beyza Ermis, John Dang, Samuel Cahyawijaya, Shivalika Singh, Seraphina Goldfarb-Tarrant, Viraat Aryabumi, Aakanksha, Wei-Yin Ko, Ahmet \"Ust\"un, Matthias Gall\'e, Marzieh Fadaee, Sara Hooker
Abstract: Despite advances in large language model capabilities in recent years, a large gap remains in their capabilities and safety performance for many languages beyond a relatively small handful of globally dominant languages. This paper provides researchers, policymakers and governance experts with an overview of key challenges to bridging the "language gap" in AI and minimizing safety risks across languages. We provide an analysis of why the language gap in AI exists and grows, and how it creates disparities in global AI safety. We identify barriers to address these challenges, and recommend how those working in policy and governance can help address safety concerns associated with the language gap by supporting multilingual dataset creation, transparency, and research.
Authors: Maria Cristina Carrisi, Mirko Marras, Sara Vergallo
Abstract: Younger generations are growing up in a world increasingly shaped by intelligent technologies, making early AI literacy crucial for developing the skills to critically understand and navigate them. However, education in this field often emphasizes tool-based learning, prioritizing usage over understanding the underlying concepts. This lack of knowledge leaves non-experts, especially children, prone to misconceptions, unrealistic expectations, and difficulties in recognizing biases and stereotypes. In this paper, we propose a structured and replicable teaching approach that fosters foundational AI literacy in primary students, by building upon core mathematical elements closely connected to and of interest in primary curricula, to strengthen conceptualization, data representation, classification reasoning, and evaluation of AI. To assess the effectiveness of our approach, we conducted an empirical study with thirty-one fifth-grade students across two classes, evaluating their progress through a post-test and a satisfaction survey. Our results indicate improvements in terminology understanding and usage, features description, logical reasoning, and evaluative skills, with students showing a deeper comprehension of decision-making processes and their limitations. Moreover, the approach proved engaging, with students particularly enjoying activities that linked AI concepts to real-world reasoning. Materials: https://github.com/tail-unica/ai-literacy-primary-ed.
Authors: Shashank Sharma, Janina Hoffmann, Vinay Namboodiri
Abstract: Hierarchical reinforcement learning (HRL) relies on abstract skills to solve long-horizon tasks efficiently. While existing skill discovery methods learns these skills automatically, they are limited to a single skill per task. In contrast, humans learn and use both fine-grained and coarse motor skills simultaneously. Inspired by human motor control, we propose Multi-Resolution Skill Discovery (MRSD), an HRL framework that learns multiple skill encoders at different temporal resolutions in parallel. A high-level manager dynamically selects among these skills, enabling adaptive control strategies over time. We evaluate MRSD on tasks from the DeepMind Control Suite and show that it outperforms prior state-of-the-art skill discovery and HRL methods, achieving faster convergence and higher final performance. Our findings highlight the benefits of integrating multi-resolution skills in HRL, paving the way for more versatile and efficient agents.
Authors: Yifan Wang, Kenneth P. Birman
Abstract: Today's cloud-hosted applications and services are complex systems, and a performance or functional instability can have dozens or hundreds of potential root causes. Our hypothesis is that by combining the pattern matching capabilities of modern AI tools with a natural multi-modal RAG LLM interface, problem identification and resolution can be simplified. ARCA is a new multi-modal RAG LLM system that targets this domain. Step-wise evaluations show that ARCA outperforms state-of-the-art alternatives.
Authors: Francesco Cozzi, Marco Pangallo, Alan Perotti, Andr\'e Panisson, Corrado Monti
Abstract: Agent-Based Models (ABMs) are powerful tools for studying emergent properties in complex systems. In ABMs, agent behaviors are governed by local interactions and stochastic rules. However, these rules are, in general, non-differentiable, limiting the use of gradient-based methods for optimization, and thus integration with real-world data. We propose a novel framework to learn a differentiable surrogate of any ABM by observing its generated data. Our method combines diffusion models to capture behavioral stochasticity and graph neural networks to model agent interactions. Distinct from prior surrogate approaches, our method introduces a fundamental shift: rather than approximating system-level outputs, it models individual agent behavior directly, preserving the decentralized, bottom-up dynamics that define ABMs. We validate our approach on two ABMs (Schelling's segregation model and a Predator-Prey ecosystem) showing that it replicates individual-level patterns and accurately forecasts emergent dynamics beyond training. Our results demonstrate the potential of combining diffusion models and graph learning for data-driven ABM simulation.
Authors: Xianling Mu, Joseph Ternasky, Fuat Alican, Yigit Ihlamur
Abstract: Early-stage startup investment is a high-risk endeavor characterized by scarce data and uncertain outcomes. Traditional machine learning approaches often require large, labeled datasets and extensive fine-tuning, yet remain opaque and difficult for domain experts to interpret or improve. In this paper, we propose a transparent and data-efficient investment decision framework powered by memory-augmented large language models (LLMs) using in-context learning (ICL). Central to our method is a natural language policy embedded directly into the LLM prompt, enabling the model to apply explicit reasoning patterns and allowing human experts to easily interpret, audit, and iteratively refine the logic. We introduce a lightweight training process that combines few-shot learning with an in-context learning loop, enabling the LLM to update its decision policy iteratively based on structured feedback. With only minimal supervision and no gradient-based optimization, our system predicts startup success far more accurately than existing benchmarks. It is over 20x more precise than random chance, which succeeds 1.9% of the time. It is also 7.1x more precise than the typical 5.6% success rate of top-tier venture capital (VC) firms.
Authors: Yang Yang, Jiemin Wu, Yutao Yue
Abstract: Automating robust hypothesis generation in open environments is pivotal for AI cognition. We introduce a novel framework integrating a multi-agent system, powered by Large Language Models (LLMs), with Inductive Logic Programming (ILP). Our system's LLM agents autonomously define a structured symbolic vocabulary (predicates) and relational templates , i.e., \emph{language bias} directly from raw textual data. This automated symbolic grounding (the construction of the language bias), traditionally an expert-driven bottleneck for ILP, then guides the transformation of text into facts for an ILP solver, which inductively learns interpretable rules. This approach overcomes traditional ILP's reliance on predefined symbolic structures and the noise-sensitivity of pure LLM methods. Extensive experiments in diverse, challenging scenarios validate superior performance, paving a new path for automated, explainable, and verifiable hypothesis generation.
Authors: Abdullah Hamdi, Faisal AlZahrani, Silvio Giancola, Bernard Ghanem
Abstract: Multi-view projection techniques have shown themselves to be highly effective in achieving top-performing results in the recognition of 3D shapes. These methods involve learning how to combine information from multiple view-points. However, the camera view-points from which these views are obtained are often fixed for all shapes. To overcome the static nature of current multi-view techniques, we propose learning these view-points. Specifically, we introduce the Multi-View Transformation Network (MVTN), which uses differentiable rendering to determine optimal view-points for 3D shape recognition. As a result, MVTN can be trained end-to-end with any multi-view network for 3D shape classification. We integrate MVTN into a novel adaptive multi-view pipeline that is capable of rendering both 3D meshes and point clouds. Our approach demonstrates state-of-the-art performance in 3D classification and shape retrieval on several benchmarks (ModelNet40, ScanObjectNN, ShapeNet Core55). Further analysis indicates that our approach exhibits improved robustness to occlusion compared to other methods. We also investigate additional aspects of MVTN, such as 2D pretraining and its use for segmentation. To support further research in this area, we have released MVTorch, a PyTorch library for 3D understanding and generation using multi-view projections.
Authors: Hudson de Martim
Abstract: This article proposes an adaptation of Graph Retrieval Augmented Generation (Graph RAG) specifically designed for the analysis and comprehension of legal norms, which are characterized by their predefined hierarchical structure, extensive network of internal and external references and multiple temporal versions. By combining structured knowledge graphs with contextually enriched text segments, Graph RAG offers a promising solution to address the inherent complexity and vast volume of legal data. The integration of hierarchical structure and temporal evolution into knowledge graphs - along with the concept of comprehensive Text Units - facilitates the construction of richer, interconnected representations of legal knowledge. Through a detailed analysis of Graph RAG and its application to legal norm datasets, this article aims to advance the field of Artificial Intelligence applied to Law, creating opportunities for more effective systems in legal research, legislative analysis, and decision support.
Authors: Jiabin Chen, Haiping Wang, Jinpeng Li, Yuan Liu, Zhen Dong, Bisheng Yang
Abstract: We propose SpatialLLM, a novel approach advancing spatial intelligence tasks in complex urban scenes. Unlike previous methods requiring geographic analysis tools or domain expertise, SpatialLLM is a unified language model directly addressing various spatial intelligence tasks without any training, fine-tuning, or expert intervention. The core of SpatialLLM lies in constructing detailed and structured scene descriptions from raw spatial data to prompt pre-trained LLMs for scene-based analysis. Extensive experiments show that, with our designs, pretrained LLMs can accurately perceive spatial distribution information and enable zero-shot execution of advanced spatial intelligence tasks, including urban planning, ecological analysis, traffic management, etc. We argue that multi-field knowledge, context length, and reasoning ability are key factors influencing LLM performances in urban analysis. We hope that SpatialLLM will provide a novel viable perspective for urban intelligent analysis and management. The code and dataset are available at https://github.com/WHU-USI3DV/SpatialLLM.
Authors: Jarrod Ragsdale, Rajendra Boppana
Abstract: Command-line interfaces (CLIs) provide structured textual environments for system administration. Explorations have been performed using pre-trained language models (PLMs) to simulate these environments for safe interaction in high-risk environments. However, their use has been constrained to frozen, large parameter models like GPT. For smaller architectures to reach a similar level of believability, a rich dataset of CLI interactions is required. Existing public datasets focus on mapping natural-language tasks to commands, omitting crucial execution data such as exit codes, outputs, and environmental side effects, limiting their usability for behavioral modeling. We introduce a Shell Input -Output Environment (ShIOEnv), which casts command construction as a Markov Decision Process whose state is the partially built sequence and whose actions append arguments. After each action, ShIOEnv executes the candidate and returns its exit status, output, and progress toward a minimal-length behavioral objective. Due to the intractable nature of the combinatorial argument state-action space, we derive a context-free grammar from man pages to mask invalid arguments from being emitted. We explore random and proximal-policy optimization (PPO)-optimized sampling of unrestricted and grammar-masked action spaces to produce four exploration strategies. We observed that grammar masking and PPO significantly improve sample efficiency to produce a higher quality dataset (maximizing the number of arguments while minimizing redundancies). Policy-generated datasets of shell input-output behavior pairs are used to fine-tune CodeT5, where we observe 85% improvements in BLEU-4 when constraining the action space to grammar productions with an additional 26% improvement when applying PPO. The ShIOEnv environment and curated command behavior datasets are released for use in future research.
Authors: Jianpeng Chen, Wangzhi Zhan, Haohui Wang, Zian Jia, Jingru Gan, Junkai Zhang, Jingyuan Qi, Tingwei Chen, Lifu Huang, Muhao Chen, Ling Li, Wei Wang, Dawei Zhou
Abstract: Metamaterials, engineered materials with architected structures across multiple length scales, offer unprecedented and tunable mechanical properties that surpass those of conventional materials. However, leveraging advanced machine learning (ML) for metamaterial discovery is hindered by three fundamental challenges: (C1) Data Heterogeneity Challenge arises from heterogeneous data sources, heterogeneous composition scales, and heterogeneous structure categories; (C2) Model Complexity Challenge stems from the intricate geometric constraints of ML models, which complicate their adaptation to metamaterial structures; and (C3) Human-AI Collaboration Challenge comes from the "dual black-box'' nature of sophisticated ML models and the need for intuitive user interfaces. To tackle these challenges, we introduce a unified framework, named MetamatBench, that operates on three levels. (1) At the data level, we integrate and standardize 5 heterogeneous, multi-modal metamaterial datasets. (2) The ML level provides a comprehensive toolkit that adapts 17 state-of-the-art ML methods for metamaterial discovery. It also includes a comprehensive evaluation suite with 12 novel performance metrics with finite element-based assessments to ensure accurate and reliable model validation. (3) The user level features a visual-interactive interface that bridges the gap between complex ML techniques and non-ML researchers, advancing property prediction and inverse design of metamaterials for research and applications. MetamatBench offers a unified platform deployed at http://zhoulab-1.cs.vt.edu:5550 that enables machine learning researchers and practitioners to develop and evaluate new methodologies in metamaterial discovery. For accessibility and reproducibility, we open-source our benchmark and the codebase at https://github.com/cjpcool/Metamaterial-Benchmark.
URLs: http://zhoulab-1.cs.vt.edu:5550, https://github.com/cjpcool/Metamaterial-Benchmark.
Authors: Patrick Yubeaton, Andre Nakkab, Weihua Xiao, Luca Collini, Ramesh Karri, Chinmay Hegde, Siddharth Garg
Abstract: This paper introduces VeriThoughts, a novel dataset designed for reasoning-based Verilog code generation. We establish a new benchmark framework grounded in formal verification methods to evaluate the quality and correctness of generated hardware descriptions. Additionally, we present a suite of specialized small-scale models optimized specifically for Verilog generation. Our work addresses the growing need for automated hardware design tools that can produce verifiably correct implementations from high-level specifications, potentially accelerating the hardware development process while maintaining rigorous correctness guarantees. Our code and data are available at \href{https://github.com/wilyub/VeriThoughts}{this URL}.
Authors: David Hanson
Abstract: As artificial intelligence becomes increasingly integrated into software development processes, the prevalence and sophistication of AI-generated code continue to expand rapidly. This study addresses the critical need for transparency and safety in AI generated code by examining the current landscape, identifying potential risks, and exploring future implications. We analyze market opportunities for detecting AI-generated code, discuss the challenges associated with managing increasing complexity, and propose solutions to enhance transparency and functionality analysis. Furthermore, this study investigates the longterm implications of AI generated code, including its potential role in the development of artificial general intelligence and its impact on human AI interaction. In conclusion, we emphasize the importance of proactive measures for ensuring the responsible development and deployment of AI in software engineering.
Authors: Muhammad Tayyab Khan, Lequn Chen, Wenhe Feng, Seung Ki Moon
Abstract: Metal additive manufacturing (AM) involves complex interdependencies among processes, materials, feedstock, and post-processing steps. However, the underlying relationships and domain knowledge remain fragmented across literature and static databases that often demand expert-level queries, limiting their applicability in design and planning. To address these gaps, we develop a novel and queryable knowledge graph (KG) in Neo4j, encoding 53 distinct metals and alloys across seven material families, nine AM processes, four feedstock types, and associated post-processing requirements. A large language model (LLM) interface, guided by a few-shot prompting strategy, enables natural language querying without the need for formal query syntax. The system supports a range of tasks, including compatibility checks, multi-constraint filtering, and design for AM (DfAM) guidance. User natural language queries are normalized, translated into Cypher, and executed over the KG, with results reformatted into structured responses. This work presents the first real-time, interactive system that integrates a domain-specific metal AM KG with an LLM interface, offering accessible, explainable decision support for engineers and advancing human-centric tools in manufacturing intelligence.
Authors: Aditya Bhattacharya, Katrien Verbert
Abstract: During job recruitment, traditional applicant selection methods often lack transparency. Candidates are rarely given sufficient justifications for recruiting decisions, whether they are made manually by human recruiters or through the use of black-box Applicant Tracking Systems (ATS). To address this problem, our work introduces a multi-agent AI system that uses Large Language Models (LLMs) to guide job seekers during the recruitment process. Using an iterative user-centric design approach, we first conducted a two-phased exploratory study with four active job seekers to inform the design and development of the system. Subsequently, we conducted an in-depth, qualitative user study with 20 active job seekers through individual one-to-one interviews to evaluate the developed prototype. The results of our evaluation demonstrate that participants perceived our multi-agent recruitment system as significantly more actionable, trustworthy, and fair compared to traditional methods. Our study further helped us uncover in-depth insights into factors contributing to these perceived user experiences. Drawing from these insights, we offer broader design implications for building user-aligned, multi-agent explainable AI systems across diverse domains.
Authors: Zhewei Yao, Guoheng Sun, Lukasz Borchmann, Zheyu Shen, Minghang Deng, Bohan Zhai, Hao Zhang, Ang Li, Yuxiong He
Abstract: Translating natural language into SQL (Test2SQL) is a longstanding challenge at the intersection of natural language understanding and structured data access. While large language models (LLMs) have significantly improved fluency in SQL generation, producing correct and executable SQL--particularly for complex queries--remains a bottleneck. We present Arctic-Text2SQL-R1, a reinforcement learning (RL) framework and model family designed to generate accurate, executable SQL using a lightweight reward signal based solely on execution correctness. Our approach avoids brittle intermediate supervision and complex reward shaping, promoting stable training and alignment with the end task. Combined with carefully curated data, strong supervised initialization, and effective training practices, Arctic-Text2SQL-R1 achieves state-of-the-art execution accuracy across six diverse Test2SQL benchmarks, including the top position on the BIRD leaderboard. Notably, our 7B model outperforms prior 70B-class systems, highlighting the framework's scalability and efficiency. We further demonstrate inference-time robustness through simple extensions like value retrieval and majority voting. Extensive experiments and ablation studies offer both positive and negative insights, providing practical guidance for future Test2SQL research.
Authors: Wang Cai, Hsiu-Yuan Huang, Zhixiang Wang, Yunfang Wu
Abstract: In-Context derived Vector (ICV) methods extract task-relevant representations from large language models (LLMs) and reinject them during inference, achieving comparable performance to few-shot In-Context Learning (ICL) without repeated demonstration processing. However, existing ICV methods remain sensitive to ICL-specific factors, often use coarse or semantically fragmented representations as the source of the vector, and rely on heuristic-based injection positions, limiting their applicability. To address these issues, we propose Dynamic Vector (DyVec), which incorporates an Exhaustive Query Rotation (EQR) strategy to extract robust semantically aggregated latent representations by mitigating variance introduced by ICL. It then applies Dynamic Latent Segmentation and Injection to adaptively partition representations based on task complexity and leverages REINFORCE-based optimization to learn optimal injection positions for each segment. Experiments results show that DyVec outperforms few-shot ICL, LoRA, and prior ICV baselines. Further analysis highlights the effectiveness of dynamically segmenting and injecting semantically aggregated latent representations. DyVec provides a lightweight and data-efficient solution for inference-time task adaptation.
Authors: Satya Narayana Cheetirala, Ganesh Raut, Dhavalkumar Patel, Fabio Sanatana, Robert Freeman, Matthew A Levin, Girish N. Nadkarni, Omar Dawkins, Reba Miller, Randolph M. Steinhagen, Eyal Klang, Prem Timsina
Abstract: Long text classification is challenging for Large Language Models (LLMs) due to token limits and high computational costs. This study explores whether a Retrieval Augmented Generation (RAG) approach using only the most relevant text segments can match the performance of processing entire clinical notes with large context LLMs. We begin by splitting clinical documents into smaller chunks, converting them into vector embeddings, and storing these in a FAISS index. We then retrieve the top 4,000 words most pertinent to the classification query and feed these consolidated segments into an LLM. We evaluated three LLMs (GPT4o, LLaMA, and Mistral) on a surgical complication identification task. Metrics such as AUC ROC, precision, recall, and F1 showed no statistically significant differences between the RAG based approach and whole-text processing (p > 0.05p > 0.05). These findings indicate that RAG can significantly reduce token usage without sacrificing classification accuracy, providing a scalable and cost effective solution for analyzing lengthy clinical documents.
Authors: Mathew J. Koretsky, Maya Willey, Adi Asija, Owen Bianchi, Chelsea X. Alvarado, Tanay Nayak, Nicole Kuznetsov, Sungwon Kim, Mike A. Nalls, Daniel Khashabi, Faraz Faghri
Abstract: Biomedical researchers increasingly rely on large-scale structured databases for complex analytical tasks. However, current text-to-SQL systems often struggle to map qualitative scientific questions into executable SQL, particularly when implicit domain reasoning is required. We introduce BiomedSQL, the first benchmark explicitly designed to evaluate scientific reasoning in text-to-SQL generation over a real-world biomedical knowledge base. BiomedSQL comprises 68,000 question/SQL query/answer triples grounded in a harmonized BigQuery knowledge base that integrates gene-disease associations, causal inference from omics data, and drug approval records. Each question requires models to infer domain-specific criteria, such as genome-wide significance thresholds, effect directionality, or trial phase filtering, rather than rely on syntactic translation alone. We evaluate a range of open- and closed-source LLMs across prompting strategies and interaction paradigms. Our results reveal a substantial performance gap: GPT-o3-mini achieves 59.0% execution accuracy, while our custom multi-step agent, BMSQL, reaches 62.6%, both well below the expert baseline of 90.0%. BiomedSQL provides a new foundation for advancing text-to-SQL systems capable of supporting scientific discovery through robust reasoning over structured biomedical knowledge bases. Our dataset is publicly available at https://huggingface.co/datasets/NIH-CARD/BiomedSQL, and our code is open-source at https://github.com/NIH-CARD/biomedsql.
URLs: https://huggingface.co/datasets/NIH-CARD/BiomedSQL,, https://github.com/NIH-CARD/biomedsql.
Authors: Mengru Wang, Ziwen Xu, Shengyu Mao, Shumin Deng, Zhaopeng Tu, Huajun Chen, Ningyu Zhang
Abstract: Precise control over language model generation is vital for ensuring both safety and reliability. Although prompt engineering and steering are commonly used to intervene in model behaviors, the vast number of parameters in models often results in highly intertwined internal representations. This interdependency can limit control precision and sometimes lead to unintended side effects. Recent research has explored the use of sparse autoencoders (SAE) to disentangle knowledge in high-dimensional spaces for steering. However, these applications have been limited to toy tasks owing to the nontrivial issue of locating atomic knowledge components. In this paper, we propose Steering Target Atoms (STA), a novel method that isolates and manipulates disentangled knowledge components to enhance safety. Comprehensive experiments demonstrate the effectiveness of our approach. Further analysis reveals that steering exhibits superior robustness and flexibility, particularly in adversarial scenarios. We also apply the steering strategy to the large reasoning model, confirming its effectiveness in precise reasoning control.
Authors: Shahriar Noroozizadeh, Sayantan Kumar, George H. Chen, Jeremy C. Weiss
Abstract: Understanding temporal dynamics in clinical narratives is essential for modeling patient trajectories, yet large-scale temporally annotated resources remain limited. We present PMOA-TTS, the first openly available dataset of 124,699 PubMed Open Access (PMOA) case reports, each converted into structured (event, time) timelines via a scalable LLM-based pipeline. Our approach combines heuristic filtering with Llama 3.3 to identify single-patient case reports, followed by prompt-driven extraction using Llama 3.3 and DeepSeek R1, resulting in over 5.6 million timestamped clinical events. To assess timeline quality, we evaluate against a clinician-curated reference set using three metrics: (i) event-level matching (80% match at a cosine similarity threshold of 0.1), (ii) temporal concordance (c-index > 0.90), and (iii) Area Under the Log-Time CDF (AULTC) for timestamp alignment. Corpus-level analysis shows wide diagnostic and demographic coverage. In a downstream survival prediction task, embeddings from extracted timelines achieve time-dependent concordance indices up to 0.82 $\pm$ 0.01, demonstrating the predictive value of temporally structured narratives. PMOA-TTS provides a scalable foundation for timeline extraction, temporal reasoning, and longitudinal modeling in biomedical NLP. The dataset is available at: https://huggingface.co/datasets/snoroozi/pmoa-tts .
Authors: Md Arman Islam, Devi Varaprasad Jonnala, Ritika Rekhi, Pratik Pokharel, Siddharth Cilamkoti, Asif Imran, Tevfik Kosar, Bekir Turkkan
Abstract: As the quality of code generated by Large Language Models (LLMs) improves, their adoption in the software industry for automated code generation continues to grow. Researchers primarily focus on enhancing the functional correctness of the generated code while commonly overlooking its energy efficiency and environmental impact. This paper investigates the energy efficiency of the code generated by 20 popular LLMs for 878 programming problems of varying difficulty levels and diverse algorithmic categories selected from the LeetCode platform by comparing them against canonical human-written solutions. Although LLMs can produce functionally correct results in most cases, our findings show that the performance and energy efficiency of LLM-produced solutions are often far below those of human-written solutions. Among the studied LLMs, DeepSeek-v3 and GPT-4o generate the most energy-efficient code, whereas Grok-2 and Gemini-1.5-Pro are among the least energy-efficient models. On average, human-generated canonical solutions are approximately 1.17 times more energy efficient than DeepSeek-v3, 1.21 times more energy efficient than GPT-4o, and over 2 times more energy efficient than Grok-2 and Gemini-1.5-Pro. For specific algorithmic groups such as dynamic programming, backtracking, and bit manipulation, LLM-generated code can consume up to 450 times more energy than human-generated canonical solutions.
Authors: Amirhosein Ghasemabadi, Keith G. Mills, Baochun Li, Di Niu
Abstract: Test-Time Scaling (TTS) methods for enhancing Large Language Model (LLM) reasoning often incur substantial computational costs, primarily due to extensive reliance on external Process Reward Models (PRMs) or sampling methods like Best-of-N (BoN). This paper introduces Guided by Gut (GG), an efficient self-guided TTS framework that achieves PRM-level performance without costly external verifier models. Our method employs a lightweight tree search guided solely by intrinsic LLM signals, token-level confidence and step novelty. One critical innovation is improving the reliability of internal confidence estimates via a targeted reinforcement learning fine-tuning phase. Empirical evaluations on challenging mathematical reasoning benchmarks demonstrate that GG enables smaller models (e.g., 1.5B parameters) to achieve accuracy matching or surpassing significantly larger models (e.g., 32B-70B parameters), while reducing GPU memory usage by up to 10x. Compared to PRM-based methods, GG achieves comparable accuracy with 8x faster inference speeds and 4-5x lower memory usage. Additionally, GG reduces KV cache memory usage by approximately 50% compared to the BoN strategy, facilitating more efficient and practical deployment of TTS techniques.
Authors: Avinash Madasu, Vasudev Lal, Phillip Howard
Abstract: Vision-Language Models (VLMs) are increasingly deployed in diverse cultural contexts, yet their internal biases remain poorly understood. In this work, we propose a novel framework to systematically evaluate how VLMs encode cultural differences and biases related to race, gender, and physical traits across countries. We introduce three retrieval-based tasks: (1) Race to Country retrieval, which examines the association between individuals from specific racial groups (East Asian, White, Middle Eastern, Latino, South Asian, and Black) and different countries; (2) Personal Traits to Country retrieval, where images are paired with trait-based prompts (e.g., Smart, Honest, Criminal, Violent) to investigate potential stereotypical associations; and (3) Physical Characteristics to Country retrieval, focusing on visual attributes like skinny, young, obese, and old to explore how physical appearances are culturally linked to nations. Our findings reveal persistent biases in VLMs, highlighting how visual representations may inadvertently reinforce societal stereotypes.
Authors: Aurora Poggi, Giuseppe Alessio D'Inverno, Hjalmar Brismar, Ozan \"Oktem, Matthieu Barreau, Kateryna Morozovska
Abstract: Data-driven discovery of dynamics in biological systems allows for better observation and characterization of processes, such as calcium signaling in cell culture. Recent advancements in techniques allow the exploration of previously unattainable insights of dynamical systems, such as the Sparse Identification of Non-Linear Dynamics (SINDy), overcoming the limitations of more classic methodologies. The latter requires some prior knowledge of an effective library of candidate terms, which is not realistic for a real case study. Using inspiration from fields like traffic density estimation and control theory, we propose a methodology for characterization and performance analysis of calcium delivery in a family of cells. In this work, we compare the performance of the Constrained Regularized Least-Squares Method (CRLSM) and Physics-Informed Neural Networks (PINN) for system identification and parameter discovery for governing ordinary differential equations (ODEs). The CRLSM achieves a fairly good parameter estimate and a good data fit when using the learned parameters in the Consensus problem. On the other hand, despite the initial hypothesis, PINNs fail to match the CRLSM performance and, under the current configuration, do not provide fair parameter estimation. However, we have only studied a limited number of PINN architectures, and it is expected that additional hyperparameter tuning, as well as uncertainty quantification, could significantly improve the performance in future works.
Authors: Yukun Zhang, Qi Dong
Abstract: Recent advances in Large Language Models (LLMs) have achieved strong performance, yet their internal reasoning remains opaque, limiting interpretability and trust in critical applications. We propose a novel Multi_Scale Manifold Alignment framework that decomposes the latent space into global, intermediate, and local semantic manifolds capturing themes, context, and word-level details. Our method introduces cross_scale mapping functions that jointly enforce geometric alignment (e.g., Procrustes analysis) and information preservation (via mutual information constraints like MINE or VIB). We further incorporate curvature regularization and hyperparameter tuning for stable optimization. Theoretical analysis shows that alignment error, measured by KL divergence, can be bounded under mild assumptions. This framework offers a unified explanation of how LLMs structure multi-scale semantics, advancing interpretability and enabling applications such as bias detection and robustness enhancement.
Authors: Yixuan Wang, Shiyu Ji, Yijun Liu, Yuzhuang Xu, Yang Xu, Qingfu Zhu, Wanxiang Che
Abstract: Large language models (LLMs) rely on key-value cache (KV cache) to accelerate decoding by reducing redundant computations. However, the KV cache memory usage grows substantially with longer text sequences, posing challenges for efficient deployment. Existing KV cache eviction methods prune tokens using prefilling-stage attention scores, causing inconsistency with actual inference queries, especially under tight memory budgets. In this paper, we propose Lookahead Q-Cache (LAQ), a novel eviction framework that generates low-cost pseudo lookahead queries to better approximate the true decoding-stage queries. By using these lookahead queries as the observation window for importance estimation, LAQ achieves more consistent and accurate KV cache eviction aligned with real inference scenarios. Experimental results on LongBench and Needle-in-a-Haystack benchmarks show that LAQ outperforms existing methods across various budget levels, achieving a 1 $\sim$ 4 point improvement on LongBench under limited cache budget. Moreover, LAQ is complementary to existing approaches and can be flexibly combined to yield further improvements.
Authors: Zishun Yu, Shangzhe Li, Xinhua Zhang
Abstract: Large language models have led to significant progress across many NLP tasks, although their massive sizes often incur substantial computational costs. Distillation has become a common practice to compress these large and highly capable models into smaller, more efficient ones. Many existing language model distillation methods can be viewed as behavior cloning from the perspective of imitation learning or inverse reinforcement learning. This viewpoint has inspired subsequent studies that leverage (inverse) reinforcement learning techniques, including variations of behavior cloning and temporal difference learning methods. Rather than proposing yet another specific temporal difference method, we introduce a general framework for temporal difference-based distillation by exploiting the distributional sparsity of the teacher model. Specifically, it is often observed that language models assign most probability mass to a small subset of tokens. Motivated by this observation, we design a temporal difference learning framework that operates on a reduced action space (a subset of vocabulary), and demonstrate how practical algorithms can be derived and the resulting performance improvements.
Authors: Yu Zhang, Wanli Jiang, Zhengyu Yang
Abstract: The multi-objective alignment of Large Language Models (LLMs) is essential for ensuring foundational models conform to diverse human preferences. Current research in this field typically involves either multiple policies or multiple reward models customized for various preferences, or the need to train a preference-specific supervised fine-tuning (SFT) model. In this work, we introduce a novel multi-objective alignment method, MOSLIM, which utilizes a single reward model and policy model to address diverse objectives. MOSLIM provides a flexible way to control these objectives through prompting and does not require preference training during SFT phase, allowing thousands of off-the-shelf models to be directly utilized within this training framework. MOSLIM leverages a multi-head reward model that classifies question-answer pairs instead of scoring them and then optimize policy model with a scalar reward derived from a mapping function that converts classification results from reward model into reward scores. We demonstrate the efficacy of our proposed method across several multi-objective benchmarks and conduct ablation studies on various reward model sizes and policy optimization methods. The MOSLIM method outperforms current multi-objective approaches in most results while requiring significantly fewer GPU computing resources compared with existing policy optimization methods.
Authors: Cayo Viegas, Rohit Gheyi, M\'arcio Ribeiro
Abstract: Recent advancements in Large Language Models (LLMs) have significantly expanded the capabilities of artificial intelligence in natural language processing tasks. Despite this progress, their performance in specialized domains such as computer science remains relatively unexplored. Understanding the proficiency of LLMs in these domains is critical for evaluating their practical utility and guiding future developments. The POSCOMP, a prestigious Brazilian examination used for graduate admissions in computer science promoted by the Brazlian Computer Society (SBC), provides a challenging benchmark. This study investigates whether LLMs can match or surpass human performance on the POSCOMP exam. Four LLMs - ChatGPT-4, Gemini 1.0 Advanced, Claude 3 Sonnet, and Le Chat Mistral Large - were initially evaluated on the 2022 and 2023 POSCOMP exams. The assessments measured the models' proficiency in handling complex questions typical of the exam. LLM performance was notably better on text-based questions than on image interpretation tasks. In the 2022 exam, ChatGPT-4 led with 57 correct answers out of 69 questions, followed by Gemini 1.0 Advanced (49), Le Chat Mistral (48), and Claude 3 Sonnet (44). Similar trends were observed in the 2023 exam. ChatGPT-4 achieved the highest performance, surpassing all students who took the POSCOMP 2023 exam. LLMs, particularly ChatGPT-4, show promise in text-based tasks on the POSCOMP exam, although image interpretation remains a challenge. Given the rapid evolution of LLMs, we expanded our analysis to include more recent models - o1, Gemini 2.5 Pro, Claude 3.7 Sonnet, and o3-mini-high - evaluated on the 2022-2024 POSCOMP exams. These newer models demonstrate further improvements and consistently surpass both the average and top-performing human participants across all three years.
Authors: Yukun Zhang, Qi Dong
Abstract: We introduce Dynamic Manifold Evolution Theory (DMET),a unified framework that models large language model generation as a controlled dynamical system evolving on a low_dimensional semantic manifold. By casting latent_state updates as discrete time Euler approximations of continuous dynamics, we map intrinsic energy_driven flows and context_dependent forces onto Transformer components (residual connections, attention, feed-forward networks). Leveraging Lyapunov stability theory We define three empirical metrics (state continuity, clustering quality, topological persistence) that quantitatively link latent_trajectory properties to text fluency, grammaticality, and semantic coherence. Extensive experiments across decoding parameters validate DMET's predictions and yield principled guidelines for balancing creativity and consistency in text generation.
Authors: Rui Liu, Pu Gao, Jiatian Xi, Berrak Sisman, Carlos Busso, Haizhou Li
Abstract: Text-based speech editing (TSE) modifies speech using only text, eliminating re-recording. However, existing TSE methods, mainly focus on the content accuracy and acoustic consistency of synthetic speech segments, and often overlook the emotional shifts or inconsistency issues introduced by text changes. To address this issue, we propose EmoCorrector, a novel post-correction scheme for TSE. EmoCorrector leverages Retrieval-Augmented Generation (RAG) by extracting the edited text's emotional features, retrieving speech samples with matching emotions, and synthesizing speech that aligns with the desired emotion while preserving the speaker's identity and quality. To support the training and evaluation of emotional consistency modeling in TSE, we pioneer the benchmarking Emotion Correction Dataset for TSE (ECD-TSE). The prominent aspect of ECD-TSE is its inclusion of $<$text, speech$>$ paired data featuring diverse text variations and a range of emotional expressions. Subjective and objective experiments and comprehensive analysis on ECD-TSE confirm that EmoCorrector significantly enhances the expression of intended emotion while addressing emotion inconsistency limitations in current TSE methods. Code and audio examples are available at https://github.com/AI-S2-Lab/EmoCorrector.
Authors: Sonal Prabhune, Balaji Padmanabhan, Kaushik Dutta
Abstract: We investigate the existence and persistence of a specific type of gender bias in some of the popular LLMs and contribute a new benchmark dataset, RealWorldQuestioning (released on HuggingFace ), developed from real-world questions across four key domains in business and health contexts: education, jobs, personal financial management, and general health. We define and study entropy bias, which we define as a discrepancy in the amount of information generated by an LLM in response to real questions users have asked. We tested this using four different LLMs and evaluated the generated responses both qualitatively and quantitatively by using ChatGPT-4o (as "LLM-as-judge"). Our analyses (metric-based comparisons and "LLM-as-judge" evaluation) suggest that there is no significant bias in LLM responses for men and women at a category level. However, at a finer granularity (the individual question level), there are substantial differences in LLM responses for men and women in the majority of cases, which "cancel" each other out often due to some responses being better for males and vice versa. This is still a concern since typical users of these tools often ask a specific question (only) as opposed to several varied ones in each of these common yet important areas of life. We suggest a simple debiasing approach that iteratively merges the responses for the two genders to produce a final result. Our approach demonstrates that a simple, prompt-based debiasing strategy can effectively debias LLM outputs, thus producing responses with higher information content than both gendered variants in 78% of the cases, and consistently achieving a balanced integration in the remaining cases.
Authors: Jiahao Kuang, Nuowei Liu, Changzhi Sun, Tao Ji, Yuanbin Wu
Abstract: In recent years, while natural language processing and multimodal learning have seen rapid advancements, the field of de novo protein design has also experienced significant growth. However, most current methods rely on proprietary datasets and evaluation rubrics, making fair comparisons between different approaches challenging. Moreover, these methods often employ evaluation metrics that capture only a subset of the desired properties of designed proteins, lacking a comprehensive assessment framework. To address these, we introduce PDFBench, the first comprehensive benchmark for evaluating de novo protein design from function. PDFBench supports two tasks: description-guided design and keyword-guided design. To ensure fair and multifaceted evaluation, we compile 22 metrics covering sequence plausibility, structural fidelity, and language-protein alignment, along with measures of novelty and diversity. We evaluate five state-of-the-art baselines, revealing their respective strengths and weaknesses across tasks. Finally, we analyze inter-metric correlations, exploring the relationships between four categories of metrics, and offering guidelines for metric selection. PDFBench establishes a unified framework to drive future advances in function-driven de novo protein design.
Authors: Wenkai Fang, Shunyu Liu, Yang Zhou, Kongcheng Zhang, Tongya Zheng, Kaixuan Chen, Mingli Song, Dacheng Tao
Abstract: Recent advances have demonstrated the effectiveness of Reinforcement Learning (RL) in improving the reasoning capabilities of Large Language Models (LLMs). However, existing works inevitably rely on high-quality instructions and verifiable rewards for effective training, both of which are often difficult to obtain in specialized domains. In this paper, we propose Self-play Reinforcement Learning(SeRL) to bootstrap LLM training with limited initial data. Specifically, SeRL comprises two complementary modules: self-instruction and self-rewarding. The former module generates additional instructions based on the available data at each training step, employing robust online filtering strategies to ensure instruction quality, diversity, and difficulty. The latter module introduces a simple yet effective majority-voting mechanism to estimate response rewards for additional instructions, eliminating the need for external annotations. Finally, SeRL performs conventional RL based on the generated data, facilitating iterative self-play learning. Extensive experiments on various reasoning benchmarks and across different LLM backbones demonstrate that the proposed SeRL yields results superior to its counterparts and achieves performance on par with those obtained by high-quality data with verifiable rewards. Our code is available at https://github.com/wantbook-book/SeRL.
Authors: Jifeng Hu, Sili Huang, Siyuan Guo, Zhaogeng Liu, Li Shen, Lichao Sun, Hechang Chen, Yi Chang, Dacheng Tao
Abstract: In recent years, generative models have shown remarkable capabilities across diverse fields, including images, videos, language, and decision-making. By applying powerful generative models such as flow-based models to reinforcement learning, we can effectively model complex multi-modal action distributions and achieve superior robotic control in continuous action spaces, surpassing the limitations of single-modal action distributions with traditional Gaussian-based policies. Previous methods usually adopt the generative models as behavior models to fit state-conditioned action distributions from datasets, with policy optimization conducted separately through additional policies using value-based sample weighting or gradient-based updates. However, this separation prevents the simultaneous optimization of multi-modal distribution fitting and policy improvement, ultimately hindering the training of models and degrading the performance. To address this issue, we propose Decision Flow, a unified framework that integrates multi-modal action distribution modeling and policy optimization. Specifically, our method formulates the action generation procedure of flow-based models as a flow decision-making process, where each action generation step corresponds to one flow decision. Consequently, our method seamlessly optimizes the flow policy while capturing multi-modal action distributions. We provide rigorous proofs of Decision Flow and validate the effectiveness through extensive experiments across dozens of offline RL environments. Compared with established offline RL baselines, the results demonstrate that our method achieves or matches the SOTA performance.
Authors: Dong Liu, Jiayi Zhang, Yifan Li, Yanxuan Yu, Ben Lengerich, Ying Nian Wu
Abstract: Diffusion Transformers (DiT) are powerful generative models but remain computationally intensive due to their iterative structure and deep transformer stacks. To alleviate this inefficiency, we propose FastCache, a hidden-state-level caching and compression framework that accelerates DiT inference by exploiting redundancy within the model's internal representations. FastCache introduces a dual strategy: (1) a spatial-aware token selection mechanism that adaptively filters redundant tokens based on hidden state saliency, and (2) a transformer-level cache that reuses latent activations across timesteps when changes are statistically insignificant. These modules work jointly to reduce unnecessary computation while preserving generation fidelity through learnable linear approximation. Theoretical analysis shows that FastCache maintains bounded approximation error under a hypothesis-testing-based decision rule. Empirical evaluations across multiple DiT variants demonstrate substantial reductions in latency and memory usage, with best generation output quality compared to other cache methods, as measured by FID and t-FID. Code implementation of FastCache is available on GitHub at https://github.com/NoakLiu/FastCache-xDiT.
Authors: Juntong Wu, Zijing Liu, He Cao, Hao Li, Bin Feng, Zishan Shu, Ke Yu, Li Yuan, Yu Li
Abstract: In recent years, protein-text models have gained significant attention for their potential in protein generation and understanding. Current approaches focus on integrating protein-related knowledge into large language models through continued pretraining and multi-modal alignment, enabling simultaneous comprehension of textual descriptions and protein sequences. Through a thorough analysis of existing model architectures and text-based protein understanding benchmarks, we identify significant data leakage issues present in current benchmarks. Moreover, conventional metrics derived from natural language processing fail to accurately assess the model's performance in this domain. To address these limitations, we reorganize existing datasets and introduce a novel evaluation framework based on biological entities. Motivated by our observation, we propose a retrieval-enhanced method, which significantly outperforms fine-tuned LLMs for protein-to-text generation and shows accuracy and efficiency in training-free scenarios. Our code and data can be seen at https://github.com/IDEA-XL/RAPM.
Authors: Yeonjoon Jung, Daehyun Ahn, Hyungjun Kim, Taesu Kim, Eunhyeok Park
Abstract: Low-Rank Adaptation (LoRA) is a popular method for parameter-efficient fine-tuning (PEFT) of generative models, valued for its simplicity and effectiveness. Despite recent enhancements, LoRA still suffers from a fundamental limitation: overfitting when the bottleneck is widened. It performs best at ranks 32-64, yet its accuracy stagnates or declines at higher ranks, still falling short of full fine-tuning (FFT) performance. We identify the root cause as LoRA's structural bottleneck, which introduces gradient entanglement to the unrelated input channels and distorts gradient propagation. To address this, we introduce a novel structure, Granular Low-Rank Adaptation (GraLoRA) that partitions weight matrices into sub-blocks, each with its own low-rank adapter. With negligible computational or storage cost, GraLoRA overcomes LoRA's limitations, effectively increases the representational capacity, and more closely approximates FFT behavior. Experiments on code generation and commonsense reasoning benchmarks show that GraLoRA consistently outperforms LoRA and other baselines, achieving up to +8.5% absolute gain in Pass@1 on HumanEval+. These improvements hold across model sizes and rank settings, making GraLoRA a scalable and robust solution for PEFT. Code, data, and scripts are available at https://github.com/SqueezeBits/GraLoRA.git
Authors: Shuoming Zhang, Jiacheng Zhao, Chunwei Xia, Zheng Wang, Yunji Chen, Xiaobing Feng, Huimin Cui
Abstract: Large language models (LLMs) have the potential to revolutionize how we design and implement compilers and code translation tools. However, existing LLMs struggle to handle long and complex programs. We introduce LEGO-Compiler, a novel neural compilation system that leverages LLMs to translate high-level languages into assembly code. Our approach centers on three key innovations: LEGO translation, which decomposes the input program into manageable blocks; breaking down the complex compilation process into smaller, simpler verifiable steps by organizing it as a verifiable LLM workflow by external tests; and a feedback mechanism for self-correction. Supported by formal proofs of translation composability, LEGO-Compiler demonstrates high accuracy on multiple datasets, including over 99% on ExeBench and 97.9% on industrial-grade AnsiBench. Additionally, LEGO-Compiler has also acheived near one order-of-magnitude improvement on compilable code size scalability. This work opens new avenues for applying LLMs to system-level tasks, complementing traditional compiler technologies.
Authors: Lijun Zhang, Lin Li, Yajie Qi, Huizhong Song, Yaodong Yang, Jun Wang, Wei Wei
Abstract: When fine-tuning pre-trained Large Language Models (LLMs) to align with human values and intentions, maximizing the estimated reward can lead to superior performance, but it also introduces potential risks due to deviations from the reference model's intended behavior. Most existing methods typically introduce KL divergence to constrain deviations between the trained model and the reference model; however, this may not be sufficient in certain applications that require tight risk control. In this paper, we introduce Risk-aware Direct Preference Optimization (Ra-DPO), a novel approach that incorporates risk-awareness by employing a class of nested risk measures. This approach formulates a constrained risk-aware advantage function maximization problem and then converts the Bradley-Terry model into a token-level representation. The objective function maximizes the likelihood of the policy while suppressing the deviation between a trained model and the reference model using a sequential risk ratio, thereby enhancing the model's risk-awareness. Experimental results across three open-source datasets: IMDb Dataset, Anthropic HH Dataset, and AlpacaEval, demonstrate the proposed method's superior performance in balancing alignment performance and model drift. Our code is opensourced at https://github.com/zlj123-max/Ra-DPO.
Authors: Jiahui Geng, Qing Li, Zongxiong Chen, Yuxia Wang, Derui Zhu, Zhuohan Xie, Chenyang Lyu, Xiuying Chen, Preslav Nakov, Fakhri Karray
Abstract: The rapid advancement of vision-language models (VLMs) has brought a lot of attention to their safety alignment. However, existing methods have primarily focused on model undersafety, where the model responds to hazardous queries, while neglecting oversafety, where the model refuses to answer safe queries. In this paper, we introduce the concept of $\textit{safety calibration}$, which systematically addresses both undersafety and oversafety. Specifically, we present $\textbf{VSCBench}$, a novel dataset of 3,600 image-text pairs that are visually or textually similar but differ in terms of safety, which is designed to evaluate safety calibration across image-centric and text-centric scenarios. Based on our benchmark, we evaluate safety calibration across eleven widely used VLMs. Our extensive experiments revealed major issues with both undersafety and oversafety. We further investigated four approaches to improve the model's safety calibration. We found that even though some methods effectively calibrated the models' safety problems, these methods also lead to the degradation of models' utility. This trade-off underscores the urgent need for advanced calibration methods, and our benchmark provides a valuable tool for evaluating future approaches. Our code and data are available at https://github.com/jiahuigeng/VSCBench.git.
Authors: Jaeyoung Choe, Jihoon Kim, Woohwan Jung
Abstract: Retrieval-augmented generation (RAG) based large language models (LLMs) are widely used in finance for their excellent performance on knowledge-intensive tasks. However, standardized documents (e.g., SEC filing) share similar formats such as repetitive boilerplate texts, and similar table structures. This similarity forces traditional RAG methods to misidentify near-duplicate text, leading to duplicate retrieval that undermines accuracy and completeness. To address these issues, we propose the Hierarchical Retrieval with Evidence Curation (HiREC) framework. Our approach first performs hierarchical retrieval to reduce confusion among similar texts. It first retrieve related documents and then selects the most relevant passages from the documents. The evidence curation process removes irrelevant passages. When necessary, it automatically generates complementary queries to collect missing information. To evaluate our approach, we construct and release a Large-scale Open-domain Financial (LOFin) question answering benchmark that includes 145,897 SEC documents and 1,595 question-answer pairs. Our code and data are available at https://github.com/deep-over/LOFin-bench-HiREC.
Authors: Lennart Ullner, Alona Zharova, Felix Creutzig
Abstract: Efficient energy management in prosumer households is key to alleviating grid stress in an energy transition marked by electric vehicles (EV), renewable energies and battery storage. However, it is unclear how households optimize prosumer EV charging. Here we study real-world data from 90 households on fixed-rate electricity tariffs in German-speaking countries to investigate the potential of Deep Reinforcement Learning (DRL) and other control approaches (Rule-Based, Model Predictive Control) to manage the dynamic and uncertain environment of Home Energy Management (HEM) and optimize household charging patterns. The DRL agent efficiently aligns charging of EV and battery storage with photovoltaic (PV) surplus. We find that frequent EV charging transactions, early EV connections and PV surplus increase optimization potential. A detailed analysis of nine households (1 hour resolution, 1 year) demonstrates that high battery capacity facilitates self optimization; in this case further algorithmic control shows little value. In cases with relatively low battery capacity, algorithmic control with DRL improves energy management and cost savings by a relevant margin. This result is further corroborated by our simulation of a synthetic household. We conclude that prosumer households with optimization potential would profit from DRL, thus benefiting also the full electricity system and its decarbonization.
Authors: Lorenzo Baraldi, Davide Bucciarelli, Federico Betti, Marcella Cornia, Lorenzo Baraldi, Nicu Sebe, Rita Cucchiara
Abstract: Instruction-based image editing models offer increased personalization opportunities in generative tasks. However, properly evaluating their results is challenging, and most of the existing metrics lag in terms of alignment with human judgment and explainability. To tackle these issues, we introduce DICE (DIfference Coherence Estimator), a model designed to detect localized differences between the original and the edited image and to assess their relevance to the given modification request. DICE consists of two key components: a difference detector and a coherence estimator, both built on an autoregressive Multimodal Large Language Model (MLLM) and trained using a strategy that leverages self-supervision, distillation from inpainting networks, and full supervision. Through extensive experiments, we evaluate each stage of our pipeline, comparing different MLLMs within the proposed framework. We demonstrate that DICE effectively identifies coherent edits, effectively evaluating images generated by different editing models with a strong correlation with human judgment. We publicly release our source code, models, and data.
Authors: Royden Wagner, Omer Sahin Tas, Felix Hauser, Marlon Steiner, Dominik Strutz, Abhishek Vivekanandan, Carlos Fernandez, Christoph Stiller
Abstract: Motion forecasts of road users (i.e., agents) vary in complexity as a function of scene constraints and interactive behavior. We address this with a multi-task learning method for motion forecasting that includes a retrocausal flow of information. The corresponding tasks are to forecast (1) marginal trajectory distributions for all modeled agents and (2) joint trajectory distributions for interacting agents. Using a transformer model, we generate the joint distributions by re-encoding marginal distributions followed by pairwise modeling. This incorporates a retrocausal flow of information from later points in marginal trajectories to earlier points in joint trajectories. Per trajectory point, we model positional uncertainty using compressed exponential power distributions. Notably, our method achieves state-of-the-art results in the Waymo Interaction Prediction dataset and generalizes well to the Argoverse 2 dataset. Additionally, our method provides an interface for issuing instructions through trajectory modifications. Our experiments show that regular training of motion forecasting leads to the ability to follow goal-based instructions and to adapt basic directional instructions to the scene context. Code: https://github.com/kit-mrt/future-motion
Authors: Zihong Chen, Wanli Jiang, Jinzhe Li, Zhonghang Yuan, Huanjun Kong, Wanli Ouyang, Nanqing Dong
Abstract: Fine-tuning for large language models (LLMs) typically requires substantial amounts of high-quality supervised data, which is both costly and labor-intensive to acquire. While synthetic data generation has emerged as a promising solution, existing approaches frequently suffer from factual inaccuracies, insufficient long-tail coverage, simplistic knowledge structures, and homogenized outputs. To address these challenges, we introduce GraphGen, a knowledge graph-guided framework designed for three key question-answering (QA) scenarios: atomic QA, aggregated QA, and multi-hop QA. It begins by constructing a fine-grained knowledge graph from the source text. It then identifies knowledge gaps in LLMs using the expected calibration error metric, prioritizing the generation of QA pairs that target high-value, long-tail knowledge. Furthermore, GraphGen incorporates multi-hop neighborhood sampling to capture complex relational information and employs style-controlled generation to diversify the resulting QA data. Experimental results on knowledge-intensive tasks under closed-book settings demonstrate that GraphGen outperforms conventional synthetic data methods, offering a more reliable and comprehensive solution to the data scarcity challenge in supervised fine-tuning. The code and data are publicly available at https://github.com/open-sciencelab/GraphGen.
Authors: Arvindh Arun, Sumit Kumar, Mojtaba Nayyeri, Bo Xiong, Ponnurangam Kumaraguru, Antonio Vergari, Steffen Staab
Abstract: Knowledge Graph Foundation Models (KGFMs) have shown promise in enabling zero-shot reasoning over unseen graphs by learning transferable patterns. However, most existing KGFMs rely solely on graph structure, overlooking the rich semantic signals encoded in textual attributes. We introduce SEMMA, a dual-module KGFM that systematically integrates transferable textual semantics alongside structure. SEMMA leverages Large Language Models (LLMs) to enrich relation identifiers, generating semantic embeddings that subsequently form a textual relation graph, which is fused with the structural component. Across 54 diverse KGs, SEMMA outperforms purely structural baselines like ULTRA in fully inductive link prediction. Crucially, we show that in more challenging generalization settings, where the test-time relation vocabulary is entirely unseen, structural methods collapse while SEMMA is 2x more effective. Our findings demonstrate that textual semantics are critical for generalization in settings where structure alone fails, highlighting the need for foundation models that unify structural and linguistic signals in knowledge reasoning.
Authors: Julio de la Torre-Vanegas, Miguel Soriano-Garcia, Israel Becerra, Diego Mercado-Ravell
Abstract: Landing safely in crowded urban environments remains an essential yet challenging endeavor for Unmanned Aerial Vehicles (UAVs), especially in emergency situations. In this work, we propose a risk-aware approach that harnesses semantic segmentation to continuously evaluate potential hazards in the drone's field of view. By using a specialized deep neural network to assign pixel-level risk values and applying an algorithm based on risk maps, our method adaptively identifies a stable Safe Landing Zone (SLZ) despite moving critical obstacles such as vehicles, people, etc., and other visual challenges like shifting illumination. A control system then guides the UAV toward this low-risk region, employing altitude-dependent safety thresholds and temporal landing point stabilization to ensure robust descent trajectories. Experimental validation in diverse urban environments demonstrates the effectiveness of our approach, achieving over 90% landing success rates in very challenging real scenarios, showing significant improvements in various risk metrics. Our findings suggest that risk-oriented vision methods can effectively help reduce the risk of accidents in emergency landing situations, particularly in complex, unstructured, urban scenarios, densely populated with moving risky obstacles, while potentiating the true capabilities of UAVs in complex urban operations.
Authors: Jian Zhang, Hanbo Zhang, Anxing Xiao, David Hsu
Abstract: Operating home appliances, among the most common tools in every household, is a critical capability for assistive home robots. This paper presents ApBot, a robot system that operates novel household appliances by "reading" their user manuals. ApBot faces multiple challenges: (i) infer goal-conditioned partial policies from their unstructured, textual descriptions in a user manual document, (ii) ground the policies to the appliance in the physical world, and (iii) execute the policies reliably over potentially many steps, despite compounding errors. To tackle these challenges, ApBot constructs a structured, symbolic model of an appliance from its manual, with the help of a large vision-language model (VLM). It grounds the symbolic actions visually to control panel elements. Finally, ApBot closes the loop by updating the model based on visual feedback. Our experiments show that across a wide range of simulated and real-world appliances, ApBot achieves consistent and statistically significant improvements in task success rate, compared with state-of-the-art large VLMs used directly as control policies. These results suggest that a structured internal representations plays an important role in robust robot operation of home appliances, especially, complex ones.
Authors: Aideen Fay, In\'es Garc\'ia-Redondo, Qiquan Wang, Haim Dubossarsky, Anthea Monod
Abstract: Understanding how adversarial conditions affect language models requires techniques that capture both global structure and local detail within high-dimensional activation spaces. We propose persistent homology (PH), a tool from topological data analysis, to systematically characterize multiscale latent space dynamics in LLMs under two distinct attack modes -- backdoor fine-tuning and indirect prompt injection. By analyzing six state-of-the-art LLMs, we show that adversarial conditions consistently compress latent topologies, reducing structural diversity at smaller scales while amplifying dominant features at coarser ones. These topological signatures are statistically robust across layers, architectures, model sizes, and align with the emergence of adversarial effects deeper in the network. To capture finer-grained mechanisms underlying these shifts, we introduce a neuron-level PH framework that quantifies how information flows and transforms within and across layers. Together, our findings demonstrate that PH offers a principled and unifying approach to interpreting representational dynamics in LLMs, particularly under distributional shift.
Authors: Zhaolin Li, Jan Niehues
Abstract: With approximately 7,000 languages spoken worldwide, current large language models (LLMs) support only a small subset. Prior research indicates LLMs can learn new languages for certain tasks without supervised data. We extend this investigation to speech recognition, investigating whether LLMs can learn unseen, low-resource languages through in-context learning (ICL). With experiments on four diverse endangered languages that LLMs have not been trained on, we find that providing more relevant text samples enhances performance in both language modelling and Automatic Speech Recognition (ASR) tasks. Furthermore, we show that the probability-based approach outperforms the traditional instruction-based approach in language learning. Lastly, we show ICL enables LLMs to achieve ASR performance that is comparable to or even surpasses dedicated language models trained specifically for these languages, while preserving the original capabilities of the LLMs.
Authors: Lei Tian, Xiaomin Li, Liqian Ma, Hefei Huang, Zirui Zheng, Hao Yin, Taiqing Li, Huchuan Lu, Xu Jia
Abstract: Recent advances in 3D reconstruction techniques and vision-language models have fueled significant progress in 3D semantic understanding, a capability critical to robotics, autonomous driving, and virtual/augmented reality. However, methods that rely on 2D priors are prone to a critical challenge: cross-view semantic inconsistencies induced by occlusion, image blur, and view-dependent variations. These inconsistencies, when propagated via projection supervision, deteriorate the quality of 3D Gaussian semantic fields and introduce artifacts in the rendered outputs. To mitigate this limitation, we propose CCL-LGS, a novel framework that enforces view-consistent semantic supervision by integrating multi-view semantic cues. Specifically, our approach first employs a zero-shot tracker to align a set of SAM-generated 2D masks and reliably identify their corresponding categories. Next, we utilize CLIP to extract robust semantic encodings across views. Finally, our Contrastive Codebook Learning (CCL) module distills discriminative semantic features by enforcing intra-class compactness and inter-class distinctiveness. In contrast to previous methods that directly apply CLIP to imperfect masks, our framework explicitly resolves semantic conflicts while preserving category discriminability. Extensive experiments demonstrate that CCL-LGS outperforms previous state-of-the-art methods. Our project page is available at https://epsilontl.github.io/CCL-LGS/.
Authors: Chenghao Qian, Wenjing Li, Yuhu Guo, Gustav Markkula
Abstract: In this work, we present WeatherEdit, a novel weather editing pipeline for generating realistic weather effects with controllable types and severity in 3D scenes. Our approach is structured into two key components: weather background editing and weather particle construction. For weather background editing, we introduce an all-in-one adapter that integrates multiple weather styles into a single pretrained diffusion model, enabling the generation of diverse weather effects in 2D image backgrounds. During inference, we design a Temporal-View (TV-) attention mechanism that follows a specific order to aggregate temporal and spatial information, ensuring consistent editing across multi-frame and multi-view images. To construct the weather particles, we first reconstruct a 3D scene using the edited images and then introduce a dynamic 4D Gaussian field to generate snowflakes, raindrops and fog in the scene. The attributes and dynamics of these particles are precisely controlled through physical-based modelling and simulation, ensuring realistic weather representation and flexible severity adjustments. Finally, we integrate the 4D Gaussian field with the 3D scene to render consistent and highly realistic weather effects. Experiments on multiple driving datasets demonstrate that WeatherEdit can generate diverse weather effects with controllable condition severity, highlighting its potential for autonomous driving simulation in adverse weather. See project page: https://jumponthemoon.github.io/w-edit
Authors: Berat Kutay U\u{g}ra\c{s}, \"Omer Nezih Gerek, \.Ibrahim Talha Sayg{\i}
Abstract: Accurate ECG interpretation is vital, yet complex cardiac data and "black-box" AI models limit clinical utility. Inspired by Transformer architectures' success in NLP for understanding sequential data, we frame ECG as the heart's unique "language" of temporal patterns. We present CardioPatternFormer, a novel Transformer-based model for interpretable ECG classification. It employs a sophisticated attention mechanism to precisely identify and classify diverse cardiac patterns, excelling at discerning subtle anomalies and distinguishing multiple co-occurring conditions. This pattern-guided attention provides clear insights by highlighting influential signal regions, effectively allowing the "heart to talk" through transparent interpretations. CardioPatternFormer demonstrates robust performance on challenging ECGs, including complex multi-pathology cases. Its interpretability via attention maps enables clinicians to understand the model's rationale, fostering trust and aiding informed diagnostic decisions. This work offers a powerful, transparent solution for advanced ECG analysis, paving the way for more reliable and clinically actionable AI in cardiology.
Authors: Vibhor Agarwal, Arjoo Gupta, Suparna De, Nishanth Sastry
Abstract: Understanding online conversations has attracted research attention with the growth of social networks and online discussion forums. Content analysis of posts and replies in online conversations is difficult because each individual utterance is usually short and may implicitly refer to other posts within the same conversation. Thus, understanding individual posts requires capturing the conversational context and dependencies between different parts of a conversation tree and then encoding the context dependencies between posts and comments/replies into the language model. To this end, we propose a general-purpose mechanism to discover appropriate conversational context for various aspects about an online post in a conversation, such as whether it is informative, insightful, interesting or funny. Specifically, we design two families of Conversation Kernels, which explore different parts of the neighborhood of a post in the tree representing the conversation and through this, build relevant conversational context that is appropriate for each task being considered. We apply our developed method to conversations crawled from slashdot.org, which allows users to apply highly different labels to posts, such as 'insightful', 'funny', etc., and therefore provides an ideal experimental platform to study whether a framework such as Conversation Kernels is general-purpose and flexible enough to be adapted to disparately different conversation understanding tasks.
Authors: Abhijit Chunduru, Majid Morafah, Mahdi Morafah, Vishnu Pandi Chellapandi, Ang Li
Abstract: The inevitable presence of data heterogeneity has made federated learning very challenging. There are numerous methods to deal with this issue, such as local regularization, better model fusion techniques, and data sharing. Though effective, they lack a deep understanding of how data heterogeneity can affect the global decision boundary. In this paper, we bridge this gap by performing an experimental analysis of the learned decision boundary using a toy example. Our observations are surprising: (1) we find that the existing methods suffer from forgetting and clients forget the global decision boundary and only learn the perfect local one, and (2) this happens regardless of the initial weights, and clients forget the global decision boundary even starting from pre-trained optimal weights. In this paper, we present FedProj, a federated learning framework that robustly learns the global decision boundary and avoids its forgetting during local training. To achieve better ensemble knowledge fusion, we design a novel server-side ensemble knowledge transfer loss to further calibrate the learned global decision boundary. To alleviate the issue of learned global decision boundary forgetting, we further propose leveraging an episodic memory of average ensemble logits on a public unlabeled dataset to regulate the gradient updates at each step of local training. Experimental results demonstrate that FedProj outperforms state-of-the-art methods by a large margin.
Authors: Roi Cohen, Russa Biswas, Gerard de Melo
Abstract: Factual completeness is a general term that captures how detailed and informative a factually correct text is. For instance, the factual sentence ``Barack Obama was born in the United States'' is factually correct, though less informative than the factual sentence ``Barack Obama was born in Honolulu, Hawaii, United States''. Despite the known fact that LLMs tend to hallucinate and generate factually incorrect text, they might also tend to choose to generate factual text that is indeed factually correct and yet less informative than other, more informative choices. In this work, we tackle this problem by proposing an informativeness alignment mechanism. This mechanism takes advantage of recent factual benchmarks to propose an informativeness alignment objective. This objective prioritizes answers that are both correct and informative. A key finding of our work is that when training a model to maximize this objective or optimize its preference, we can improve not just informativeness but also factuality.
Authors: Naba Rizvi, Harper Strickland, Saleha Ahmedi, Aekta Kallepalli, Isha Khirwadkar, William Wu, Imani N. S. Munyaka, Nedjma Ousidhoum
Abstract: Large language models (LLMs) are increasingly used in decision-making tasks like r\'esum\'e screening and content moderation, giving them the power to amplify or suppress certain perspectives. While previous research has identified disability-related biases in LLMs, little is known about how they conceptualize ableism or detect it in text. We evaluate the ability of four LLMs to identify nuanced ableism directed at autistic individuals. We examine the gap between their understanding of relevant terminology and their effectiveness in recognizing ableist content in context. Our results reveal that LLMs can identify autism-related language but often miss harmful or offensive connotations. Further, we conduct a qualitative comparison of human and LLM explanations. We find that LLMs tend to rely on surface-level keyword matching, leading to context misinterpretations, in contrast to human annotators who consider context, speaker identity, and potential impact. On the other hand, both LLMs and humans agree on the annotation scheme, suggesting that a binary classification is adequate for evaluating LLM performance, which is consistent with findings from prior studies involving human annotators.
Authors: Matthew Lisondra, Beno Benhabib, Goldie Nejat
Abstract: Rapid advancements in foundation models, including Large Language Models, Vision-Language Models, Multimodal Large Language Models, and Vision-Language-Action Models have opened new avenues for embodied AI in mobile service robotics. By combining foundation models with the principles of embodied AI, where intelligent systems perceive, reason, and act through physical interactions, robots can improve understanding, adapt to, and execute complex tasks in dynamic real-world environments. However, embodied AI in mobile service robots continues to face key challenges, including multimodal sensor fusion, real-time decision-making under uncertainty, task generalization, and effective human-robot interactions (HRI). In this paper, we present the first systematic review of the integration of foundation models in mobile service robotics, identifying key open challenges in embodied AI and examining how foundation models can address them. Namely, we explore the role of such models in enabling real-time sensor fusion, language-conditioned control, and adaptive task execution. Furthermore, we discuss real-world applications in the domestic assistance, healthcare, and service automation sectors, demonstrating the transformative impact of foundation models on service robotics. We also include potential future research directions, emphasizing the need for predictive scaling laws, autonomous long-term adaptation, and cross-embodiment generalization to enable scalable, efficient, and robust deployment of foundation models in human-centric robotic systems.
Authors: Hawau Olamide Toyin, Rufael Marew, Humaid Alblooshi, Samar M. Magdy, Hanan Aldarmaki
Abstract: We introduce ArVoice, a multi-speaker Modern Standard Arabic (MSA) speech corpus with diacritized transcriptions, intended for multi-speaker speech synthesis, and can be useful for other tasks such as speech-based diacritic restoration, voice conversion, and deepfake detection. ArVoice comprises: (1) a new professionally recorded set from six voice talents with diverse demographics, (2) a modified subset of the Arabic Speech Corpus; and (3) high-quality synthetic speech from two commercial systems. The complete corpus consists of a total of 83.52 hours of speech across 11 voices; around 10 hours consist of human voices from 7 speakers. We train three open-source TTS and two voice conversion systems to illustrate the use cases of the dataset. The corpus is available for research use.
Authors: Elias Arbash, Ahmed Jamal Afifi, Ymane Belahsen, Margret Fuchs, Pedram Ghamisi, Paul Scheunders, Richard Gloaguen
Abstract: The global challenge of sustainable recycling demands automated, fast, and accurate, state-of-the-art (SOTA) material detection systems that act as a bedrock for a circular economy. Democratizing access to these cutting-edge solutions that enable real-time waste analysis is essential for scaling up recycling efforts and fostering the Green Deal. In response, we introduce \textbf{Electrolyzers-HSI}, a novel multimodal benchmark dataset designed to accelerate the recovery of critical raw materials through accurate electrolyzer materials classification. The dataset comprises 55 co-registered high-resolution RGB images and hyperspectral imaging (HSI) data cubes spanning the 400--2500 nm spectral range, yielding over 4.2 million pixel vectors and 424,169 labeled ones. This enables non-invasive spectral analysis of shredded electrolyzer samples, supporting quantitative and qualitative material classification and spectral properties investigation. We evaluate a suite of baseline machine learning (ML) methods alongside SOTA transformer-based deep learning (DL) architectures, including Vision Transformer, SpectralFormer, and the Multimodal Fusion Transformer, to investigate architectural bottlenecks for further efficiency optimisation when deploying transformers in material identification. We implement zero-shot detection techniques and majority voting across pixel-level predictions to establish object-level classification robustness. In adherence to the FAIR data principles, the electrolyzers-HSI dataset and accompanying codebase are openly available at https://github.com/hifexplo/Electrolyzers-HSI and https://rodare.hzdr.de/record/3668, supporting reproducible research and facilitating the broader adoption of smart and sustainable e-waste recycling solutions.
URLs: https://github.com/hifexplo/Electrolyzers-HSI, https://rodare.hzdr.de/record/3668,
Authors: Shenao Zhang, Yaqing Wang, Yinxiao Liu, Tianqi Liu, Peter Grabowski, Eugene Ie, Zhaoran Wang, Yunxuan Li
Abstract: Large Language Models (LLMs) trained via Reinforcement Learning (RL) have exhibited strong reasoning capabilities and emergent reflective behaviors, such as backtracking and error correction. However, conventional Markovian RL confines exploration to the training phase to learn an optimal deterministic policy and depends on the history contexts only through the current state. Therefore, it remains unclear whether reflective reasoning will emerge during Markovian RL training, or why they are beneficial at test time. To remedy this, we recast reflective exploration within the Bayes-Adaptive RL framework, which explicitly optimizes the expected return under a posterior distribution over Markov decision processes. This Bayesian formulation inherently incentivizes both reward-maximizing exploitation and information-gathering exploration via belief updates. Our resulting algorithm, BARL, instructs the LLM to stitch and switch strategies based on the observed outcomes, offering principled guidance on when and how the model should reflectively explore. Empirical results on both synthetic and mathematical reasoning tasks demonstrate that BARL outperforms standard Markovian RL approaches at test time, achieving superior token efficiency with improved exploration effectiveness. Our code is available at https://github.com/shenao-zhang/BARL.
Authors: Jihoon Lee, Min Song
Abstract: Despite significant advancements in Large Vision-Language Models, Object Hallucination (OH) remains a persistent challenge. Building upon prior studies on contrastive decoding that address this issue without requiring additional model training, we introduce RVCD (Retrieval Visual Contrastive Decoding), an advanced method to suppress OH. RVCD leverages both negative and positive images at the logit level, explicitly referencing AI-generated images designed to represent a single concept. Our approach demonstrates substantial improvements over existing decoding-based methods.
Authors: Jiabao Ji, Yongchao Chen, Yang Zhang, Ramana Rao Kompella, Chuchu Fan, Gaowen Liu, Shiyu Chang
Abstract: Large language models (LLMs) have demonstrated strong performance in various robot control tasks. However, their deployment in real-world applications remains constrained. Even state-ofthe-art LLMs, such as GPT-o4mini, frequently produce invalid action plans that violate physical constraints, such as directing a robot to an unreachable location or causing collisions between robots. This issue primarily arises from a lack of awareness of these physical constraints during the reasoning process. To address this issue, we propose a novel framework that integrates reinforcement learning with verifiable rewards (RLVR) to incentivize knowledge of physical constraints into LLMs to induce constraints-aware reasoning during plan generation. In this approach, only valid action plans that successfully complete a control task receive positive rewards. We applied our method to two small-scale LLMs: a non-reasoning Qwen2.5-3B-Instruct and a reasoning Qwen3-4B. The experiment results demonstrate that constraint-aware small LLMs largely outperform large-scale models without constraints, grounded on both the BoxNet task and a newly developed BoxNet3D environment built using MuJoCo. This work highlights the effectiveness of grounding even small LLMs with physical constraints to enable scalable and efficient multi-robot control in complex, physically constrained environments.
Authors: Xingyu Chen, Shihao Ma, Runsheng Lin, Jiecong Lin, Bo Wang
Abstract: Designing regulatory DNA sequences that achieve precise cell-type-specific gene expression is crucial for advancements in synthetic biology, gene therapy and precision medicine. Although transformer-based language models (LMs) can effectively capture patterns in regulatory DNA, their generative approaches often struggle to produce novel sequences with reliable cell-specific activity. Here, we introduce Ctrl-DNA, a novel constrained reinforcement learning (RL) framework tailored for designing regulatory DNA sequences with controllable cell-type specificity. By formulating regulatory sequence design as a biologically informed constrained optimization problem, we apply RL to autoregressive genomic LMs, enabling the models to iteratively refine sequences that maximize regulatory activity in targeted cell types while constraining off-target effects. Our evaluation on human promoters and enhancers demonstrates that Ctrl-DNA consistently outperforms existing generative and RL-based approaches, generating high-fitness regulatory sequences and achieving state-of-the-art cell-type specificity. Moreover, Ctrl-DNA-generated sequences capture key cell-type-specific transcription factor binding sites (TFBS), short DNA motifs recognized by regulatory proteins that control gene expression, demonstrating the biological plausibility of the generated sequences.
Authors: Dane Malenfant, Blake A. Richards
Abstract: Sometimes we benefit from actions that others have taken even when we are unaware that they took those actions. For example, if your neighbor chooses not to take a parking spot in front of your house when you are not there, you can benefit, even without being aware that they took this action. These "hidden gifts" represent an interesting challenge for multi-agent reinforcement learning (MARL), since assigning credit when the beneficial actions of others are hidden is non-trivial. Here, we study the impact of hidden gifts with a very simple MARL task. In this task, agents in a grid-world environment have individual doors to unlock in order to obtain individual rewards. As well, if all the agents unlock their door the group receives a larger collective reward. However, there is only one key for all of the doors, such that the collective reward can only be obtained when the agents drop the key for others after they use it. Notably, there is nothing to indicate to an agent that the other agents have dropped the key, thus the act of dropping the key for others is a "hidden gift". We show that several different state-of-the-art RL algorithms, including MARL algorithms, fail to learn how to obtain the collective reward in this simple task. Interestingly, we find that independent model-free policy gradient agents can solve the task when we provide them with information about their own action history, but MARL agents still cannot solve the task with action history. Finally, we derive a correction term for these independent agents, inspired by learning aware approaches, which reduces the variance in learning and helps them to converge to collective success more reliably. These results show that credit assignment in multi-agent settings can be particularly challenging in the presence of "hidden gifts", and demonstrate that learning awareness in independent agents can benefit these settings.
Authors: Xiaoxiao Jiang, Suyi Li, Lingyun Yang, Tianyu Feng, Zhipeng Di, Weiyi Lu, Guoxuan Zhu, Xiu Lin, Kan Liu, Yinghao Yu, Tao Lan, Guodong Yang, Lin Qu, Liping Zhang, Wei Wang
Abstract: Generative image editing using diffusion models has become a prevalent application in today's AI cloud services. In production environments, image editing typically involves a mask that specifies the regions of an image template to be edited. The use of masks provides direct control over the editing process and introduces sparsity in the model inference. In this paper, we present InstGenIE, a system that efficiently serves image editing requests. The key insight behind InstGenIE is that image editing only modifies the masked regions of image templates while preserving the original content in the unmasked areas. Driven by this insight, InstGenIE judiciously skips redundant computations associated with the unmasked areas by reusing cached intermediate activations from previous inferences. To mitigate the high cache loading overhead, InstGenIE employs a bubble-free pipeline scheme that overlaps computation with cache loading. Additionally, to reduce queuing latency in online serving while improving the GPU utilization, InstGenIE proposes a novel continuous batching strategy for diffusion model serving, allowing newly arrived requests to join the running batch in just one step of denoising computation, without waiting for the entire batch to complete. As heterogeneous masks induce imbalanced loads, InstGenIE also develops a load balancing strategy that takes into account the loads of both computation and cache loading. Collectively, InstGenIE outperforms state-of-the-art diffusion serving systems for image editing, achieving up to 3x higher throughput and reducing average request latency by up to 14.7x while ensuring image quality.
Authors: Ziju Shen, Naohao Huang, Fanyi Yang, Yutong Wang, Guoxiong Gao, Tianyi Xu, Jiedong Jiang, Wanyi He, Pu Yang, Mengzhou Sun, Haocheng Ju, Peihao Wu, Bryan Dai, Bin Dong
Abstract: Nowadays, formal theorem provers have made monumental progress on high-school and competition-level mathematics, but few of them generalize to more advanced mathematics. In this paper, we present REAL-Prover, a new open-source stepwise theorem prover for Lean 4 to push this boundary. This prover, based on our fine-tuned large language model (REAL-Prover-v1) and integrated with a retrieval system (Leansearch-PS), notably boosts performance on solving college-level mathematics problems. To train REAL-Prover-v1, we developed HERALD-AF, a data extraction pipeline that converts natural language math problems into formal statements, and a new open-source Lean 4 interactive environment (Jixia-interactive) to facilitate synthesis data collection. In our experiments, our prover using only supervised fine-tune achieves competitive results with a 23.7% success rate (Pass@64) on the ProofNet dataset-comparable to state-of-the-art (SOTA) models. To further evaluate our approach, we introduce FATE-M, a new benchmark focused on algebraic problems, where our prover achieves a SOTA success rate of 56.7% (Pass@64).
Authors: Shijie Liu, Andrew C. Cullen, Paul Montague, Sarah Erfani, Benjamin I. P. Rubinstein
Abstract: Similar to other machine learning frameworks, Offline Reinforcement Learning (RL) is shown to be vulnerable to poisoning attacks, due to its reliance on externally sourced datasets, a vulnerability that is exacerbated by its sequential nature. To mitigate the risks posed by RL poisoning, we extend certified defenses to provide larger guarantees against adversarial manipulation, ensuring robustness for both per-state actions, and the overall expected cumulative reward. Our approach leverages properties of Differential Privacy, in a manner that allows this work to span both continuous and discrete spaces, as well as stochastic and deterministic environments -- significantly expanding the scope and applicability of achievable guarantees. Empirical evaluations demonstrate that our approach ensures the performance drops to no more than $50\%$ with up to $7\%$ of the training data poisoned, significantly improving over the $0.008\%$ in prior work~\citep{wu_copa_2022}, while producing certified radii that is $5$ times larger as well. This highlights the potential of our framework to enhance safety and reliability in offline RL.
Authors: Ting Xu, Zhichao Huang, Jiankai Sun, Shanbo Cheng, Wai Lam
Abstract: We present Sequential Policy Optimization for Simultaneous Machine Translation (SeqPO-SiMT), a new policy optimization framework that defines the simultaneous machine translation (SiMT) task as a sequential decision making problem, incorporating a tailored reward to enhance translation quality while reducing latency. In contrast to popular Reinforcement Learning from Human Feedback (RLHF) methods, such as PPO and DPO, which are typically applied in single-step tasks, SeqPO-SiMT effectively tackles the multi-step SiMT task. This intuitive framework allows the SiMT LLMs to simulate and refine the SiMT process using a tailored reward. We conduct experiments on six datasets from diverse domains for En to Zh and Zh to En SiMT tasks, demonstrating that SeqPO-SiMT consistently achieves significantly higher translation quality with lower latency. In particular, SeqPO-SiMT outperforms the supervised fine-tuning (SFT) model by 1.13 points in COMET, while reducing the Average Lagging by 6.17 in the NEWSTEST2021 En to Zh dataset. While SiMT operates with far less context than offline translation, the SiMT results of SeqPO-SiMT on 7B LLM surprisingly rival the offline translation of high-performing LLMs, including Qwen-2.5-7B-Instruct and LLaMA-3-8B-Instruct.
Authors: Jinwu Hu, Zhitian Zhang, Guohao Chen, Xutao Wen, Chao Shuai, Wei Luo, Bin Xiao, Yuanqing Li, Mingkui Tan
Abstract: While Large Language Models (LLMs) have exhibited remarkable emergent capabilities through extensive pre-training, they still face critical limitations in generalizing to specialized domains and handling diverse linguistic variations, known as distribution shifts. In this paper, we propose a Test-Time Learning (TTL) paradigm for LLMs, namely TLM, which dynamically adapts LLMs to target domains using only unlabeled test data during testing. Specifically, we first provide empirical evidence and theoretical insights to reveal that more accurate predictions from LLMs can be achieved by minimizing the input perplexity of the unlabeled test data. Based on this insight, we formulate the Test-Time Learning process of LLMs as input perplexity minimization, enabling self-supervised enhancement of LLM performance. Furthermore, we observe that high-perplexity samples tend to be more informative for model optimization. Accordingly, we introduce a Sample Efficient Learning Strategy that actively selects and emphasizes these high-perplexity samples for test-time updates. Lastly, to mitigate catastrophic forgetting and ensure adaptation stability, we adopt Low-Rank Adaptation (LoRA) instead of full-parameter optimization, which allows lightweight model updates while preserving more original knowledge from the model. We introduce the AdaptEval benchmark for TTL and demonstrate through experiments that TLM improves performance by at least 20% compared to original LLMs on domain knowledge adaptation.
Authors: Zexu Pan, Shengkui Zhao, Tingting Wang, Kun Zhou, Yukun Ma, Chong Zhang, Bin Ma
Abstract: Audio-visual speaker extraction isolates a target speaker's speech from a mixture speech signal conditioned on a visual cue, typically using the target speaker's face recording. However, in real-world scenarios, other co-occurring faces are often present on-screen, providing valuable speaker activity cues in the scene. In this work, we introduce a plug-and-play inter-speaker attention module to process these flexible numbers of co-occurring faces, allowing for more accurate speaker extraction in complex multi-person environments. We integrate our module into two prominent models: the AV-DPRNN and the state-of-the-art AV-TFGridNet. Extensive experiments on diverse datasets, including the highly overlapped VoxCeleb2 and sparsely overlapped MISP, demonstrate that our approach consistently outperforms baselines. Furthermore, cross-dataset evaluations on LRS2 and LRS3 confirm the robustness and generalizability of our method.
Authors: Ana M. Cabanas, Alma Pedro, Domingo Mery
Abstract: Understanding how facial affect analysis (FAA) systems perform across different demographic groups requires reliable measurement of sensitive attributes such as ancestry, often approximated by skin tone, which itself is highly influenced by lighting conditions. This study compares two objective skin tone classification methods: the widely used Individual Typology Angle (ITA) and a perceptually grounded alternative based on Lightness ($L^*$) and Hue ($H^*$). Using AffectNet and a MobileNet-based model, we assess fairness across skin tone groups defined by each method. Results reveal a severe underrepresentation of dark skin tones ($\sim 2 \%$), alongside fairness disparities in F1-score (up to 0.08) and TPR (up to 0.11) across groups. While ITA shows limitations due to its sensitivity to lighting, the $H^*$-$L^*$ method yields more consistent subgrouping and enables clearer diagnostics through metrics such as Equal Opportunity. Grad-CAM analysis further highlights differences in model attention patterns by skin tone, suggesting variation in feature encoding. To support future mitigation efforts, we also propose a modular fairness-aware pipeline that integrates perceptual skin tone estimation, model interpretability, and fairness evaluation. These findings emphasize the relevance of skin tone measurement choices in fairness assessment and suggest that ITA-based evaluations may overlook disparities affecting darker-skinned individuals.
Authors: Bo Pan, Liang Zhao
Abstract: Allocating more compute to large language models (LLMs) reasoning has generally been demonstrated to improve their effectiveness, but also results in increased inference time. In contrast, humans can perform tasks faster and better with increased experience and exposure. Hence, this paper aims to investigate the question: Can LLMs also become faster at reasoning through recurrent exposure on relevant tasks, and if so, how can it be achieved? To address these questions, we first formalize the problem setting of LLM reasoning speedup systematically in the dimensions of task relevancy and compute budget calculation. We then propose SpeedupLLM, a theoretically guaranteed framework to implement and benchmark such reasoning speedup behaviour based on adaptive compute allocation and memory mechanisms. We further conduct comprehensive experiments to benchmark such behaviour across different question similarity levels, memory methods, and reasoning methods. Results show that LLMs can generally reason faster with past experience, achieving up to a 56% reduction in compute cost when equipped with appropriate memory and reasoning methods.
Authors: Haoyu Zhang, Yisen Feng, Qiaohui Chu, Meng Liu, Weili Guan, Yaowei Wang, Liqiang Nie
Abstract: In this report, we present the method that achieves third place for Ego4D EgoSchema Challenge in CVPR 2025. To improve the reliability of answer prediction in egocentric video question answering, we propose an effective extension to the previously proposed HCQA framework. Our approach introduces a multi-source aggregation strategy to generate diverse predictions, followed by a confidence-based filtering mechanism that selects high-confidence answers directly. For low-confidence cases, we incorporate a fine-grained reasoning module that performs additional visual and contextual analysis to refine the predictions. Evaluated on the EgoSchema blind test set, our method achieves 77% accuracy on over 5,000 human-curated multiple-choice questions, outperforming last year's winning solution and the majority of participating teams. Our code will be added at https://github.com/Hyu-Zhang/HCQA.
Authors: Eduardo Y. Sakabe, Felipe S. Abrah\~ao, Alexandre Sim\~oes, Esther Colombini, Paula Costa, Ricardo Gudwin, Hector Zenil
Abstract: Understanding and controlling the informational complexity of neural networks is a central challenge in machine learning, with implications for generalization, optimization, and model capacity. While most approaches rely on entropy-based loss functions and statistical metrics, these measures often fail to capture deeper, causally relevant algorithmic regularities embedded in network structure. We propose a shift toward algorithmic information theory, using Binarized Neural Networks (BNNs) as a first proxy. Grounded in algorithmic probability (AP) and the universal distribution it defines, our approach characterizes learning dynamics through a formal, causally grounded lens. We apply the Block Decomposition Method (BDM) -- a scalable approximation of algorithmic complexity based on AP -- and demonstrate that it more closely tracks structural changes during training than entropy, consistently exhibiting stronger correlations with training loss across varying model sizes and randomized training runs. These results support the view of training as a process of algorithmic compression, where learning corresponds to the progressive internalization of structured regularities. In doing so, our work offers a principled estimate of learning progression and suggests a framework for complexity-aware learning and regularization, grounded in first principles from information theory, complexity, and computability.
Authors: Mengmeng Chen, Xiaohu Wu, Qiqi Liu, Tiantian He, Yew-Soon Ong, Yaochu Jin, Qicheng Lao, Han Yu
Abstract: Multi-objective optimization (MOO) exists extensively in machine learning, and aims to find a set of Pareto-optimal solutions, called the Pareto front, e.g., it is fundamental for multiple avenues of research in federated learning (FL). Pareto-Front Learning (PFL) is a powerful method implemented using Hypernetworks (PHNs) to approximate the Pareto front. This method enables the acquisition of a mapping function from a given preference vector to the solutions on the Pareto front. However, most existing PFL approaches still face two challenges: (a) sampling rays in high-dimensional spaces; (b) failing to cover the entire Pareto Front which has a convex shape. Here, we introduce a novel PFL framework, called as PHN-HVVS, which decomposes the design space into Voronoi grids and deploys a genetic algorithm (GA) for Voronoi grid partitioning within high-dimensional space. We put forward a new loss function, which effectively contributes to more extensive coverage of the resultant Pareto front and maximizes the HV Indicator. Experimental results on multiple MOO machine learning tasks demonstrate that PHN-HVVS outperforms the baselines significantly in generating Pareto front. Also, we illustrate that PHN-HVVS advances the methodologies of several recent problems in the FL field. The code is available at https://github.com/buptcmm/phnhvvs}{https://github.com/buptcmm/phnhvvs.
URLs: https://github.com/buptcmm/phnhvvs, https://github.com/buptcmm/phnhvvs.
Authors: Yan Wang, Yang Ren, Lingfei Qian, Xueqing Peng, Keyi Wang, Yi Han, Dongji Feng, Xiao-Yang Liu, Jimin Huang, Qianqian Xie
Abstract: We introduce FinTagging, the first full-scope, table-aware XBRL benchmark designed to evaluate the structured information extraction and semantic alignment capabilities of large language models (LLMs) in the context of XBRL-based financial reporting. Unlike prior benchmarks that oversimplify XBRL tagging as flat multi-class classification and focus solely on narrative text, FinTagging decomposes the XBRL tagging problem into two subtasks: FinNI for financial entity extraction and FinCL for taxonomy-driven concept alignment. It requires models to jointly extract facts and align them with the full 10k+ US-GAAP taxonomy across both unstructured text and structured tables, enabling realistic, fine-grained evaluation. We assess a diverse set of LLMs under zero-shot settings, systematically analyzing their performance on both subtasks and overall tagging accuracy. Our results reveal that, while LLMs demonstrate strong generalization in information extraction, they struggle with fine-grained concept alignment, particularly in disambiguating closely related taxonomy entries. These findings highlight the limitations of existing LLMs in fully automating XBRL tagging and underscore the need for improved semantic reasoning and schema-aware modeling to meet the demands of accurate financial disclosure. Code is available at our GitHub repository and data is at our Hugging Face repository.
Authors: Lingyu Qiu, Ke Jiang, Xiaoyang Tan
Abstract: Recent advancements in domain generalization for deepfake detection have attracted significant attention, with previous methods often incorporating additional modules to prevent overfitting to domain-specific patterns. However, such regularization can hinder the optimization of the empirical risk minimization (ERM) objective, ultimately degrading model performance. In this paper, we propose a novel learning objective that aligns generalization gradient updates with ERM gradient updates. The key innovation is the application of perturbations to model parameters, aligning the ascending points across domains, which specifically enhances the robustness of deepfake detection models to domain shifts. This approach effectively preserves domain-invariant features while managing domain-specific characteristics, without introducing additional regularization. Experimental results on multiple challenging deepfake detection datasets demonstrate that our gradient alignment strategy outperforms state-of-the-art domain generalization techniques, confirming the efficacy of our method. The code is available at https://github.com/Lynn0925/RoGA.
Authors: Yi Zhu, Xin Zou, Xindong Wu
Abstract: Existing cyberbullying detection benchmarks were organized by the polarity of speech, such as "offensive" and "non-offensive", which were essentially hate speech detection. However, in the real world, cyberbullying often attracted widespread social attention through incidents. To address this problem, we propose a novel annotation method to construct a cyberbullying dataset that organized by incidents. The constructed CHNCI is the first Chinese cyberbullying incident detection dataset, which consists of 220,676 comments in 91 incidents. Specifically, we first combine three cyberbullying detection methods based on explanations generation as an ensemble method to generate the pseudo labels, and then let human annotators judge these labels. Then we propose the evaluation criteria for validating whether it constitutes a cyberbullying incident. Experimental results demonstrate that the constructed dataset can be a benchmark for the tasks of cyberbullying detection and incident prediction. To the best of our knowledge, this is the first study for the Chinese cyberbullying incident detection task.
Authors: Qinzhuo Wu, Pengzhi Gao, Wei Liu, Jian Luan
Abstract: Graphical User Interface (GUI) agents have gained substantial attention due to their impressive capabilities to complete tasks through multiple interactions within GUI environments. However, existing agents primarily focus on enhancing the accuracy of individual actions and often lack effective mechanisms for detecting and recovering from errors. To address these shortcomings, we propose the BacktrackAgent, a robust framework that incorporates a backtracking mechanism to improve task completion efficiency. BacktrackAgent includes verifier, judger, and reflector components as modules for error detection and recovery, while also applying judgment rewards to further enhance the agent's performance. Additionally, we develop a training dataset specifically designed for the backtracking mechanism, which considers the outcome pages after action executions. Experimental results show that BacktrackAgent has achieved performance improvements in both task success rate and step accuracy on Mobile3M and Auto-UI benchmarks. Our data and code will be released upon acceptance.
Authors: Xu Kang, Siqi Jiang, Kangwei Xu, Jiahao Li, Ruibo Wu
Abstract: Terpenoids are a crucial class of natural products that have been studied for over 150 years, but their interdisciplinary nature (spanning chemistry, pharmacology, and biology) complicates knowledge integration. To address this, the authors developed TeroSeek, a curated knowledge base (KB) built from two decades of terpenoid literature, coupled with an AI-powered question-answering chatbot and web service. Leveraging a retrieval-augmented generation (RAG) framework, TeroSeek provides structured, high-quality information and outperforms general-purpose large language models (LLMs) in terpenoid-related queries. It serves as a domain-specific expert tool for multidisciplinary research and is publicly available at http://teroseek.qmclab.com.
Authors: Yang He, Xiao Ding, Bibo Cai, Yufei Zhang, Kai Xiong, Zhouhao Sun, Bing Qin, Ting Liu
Abstract: While reasoning-augmented large language models (RLLMs) significantly enhance complex task performance through extended reasoning chains, they inevitably introduce substantial unnecessary token consumption, particularly for simpler problems where Short Chain-of-Thought (Short CoT) suffices. This overthinking phenomenon leads to inefficient resource usage without proportional accuracy gains. To address this issue, we propose Self-Route, a dynamic reasoning framework that automatically selects between general and reasoning modes based on model capability estimation. Our approach introduces a lightweight pre-inference stage to extract capability-aware embeddings from hidden layer representations, enabling real-time evaluation of the model's ability to solve problems. We further construct Gradient-10K, a model difficulty estimation-based dataset with dense complexity sampling, to train the router for precise capability boundary detection. Extensive experiments demonstrate that Self-Route achieves comparable accuracy to reasoning models while reducing token consumption by 30-55\% across diverse benchmarks. The proposed framework demonstrates consistent effectiveness across models with different parameter scales and reasoning paradigms, highlighting its general applicability and practical value.
Authors: Yukun Zhang, Xueqing Zhou
Abstract: We propose a novel framework, Continuous_Time Attention, which infuses partial differential equations (PDEs) into the Transformer's attention mechanism to address the challenges of extremely long input sequences. Instead of relying solely on a static attention matrix, we allow attention weights to evolve over a pseudo_time dimension via diffusion, wave, or reaction_diffusion dynamics. This mechanism systematically smooths local noise, enhances long_range dependencies, and stabilizes gradient flow. Theoretically, our analysis shows that PDE_based attention leads to better optimization landscapes and polynomial rather than exponential decay of distant interactions. Empirically, we benchmark our method on diverse experiments_demonstrating consistent gains over both standard and specialized long sequence Transformer variants. Our findings highlight the potential of PDE_based formulations to enrich attention mechanisms with continuous_time dynamics and global coherence.
Authors: Boyi Zeng, Shixiang Song, Siyuan Huang, Yixuan Wang, He Li, Ziwei He, Xinbing Wang, Zhiyu Li, Zhouhan Lin
Abstract: Humans ponder before articulating complex sentence elements, enabling deeper cognitive processing through focused effort. In this work, we introduce this pondering process into language models by repeatedly invoking the forward process within a single token generation step. During pondering, instead of generating an actual token sampled from the prediction distribution, the model ponders by yielding a weighted sum of all token embeddings according to the predicted token distribution. The generated embedding is then fed back as input for another forward pass. We show that the model can learn to ponder in this way through self-supervised learning, without any human annotations. Our method is straightforward and can be seamlessly integrated with various existing language models. Experiments across three widely used open-source architectures-GPT-2, Pythia, and LLaMA-and extensive downstream task evaluations demonstrate the effectiveness and generality of our method. For language modeling tasks, pondering language models achieve performance comparable to vanilla models with twice the number of parameters. On 9 downstream benchmarks, our pondering-enhanced Pythia models significantly outperform the official Pythia models. Notably, pondering-enhanced Pythia-1B is comparable to TinyLlama-1.1B, which is trained on 10 times more data. The code is available at https://github.com/LUMIA-Group/PonderingLM.
Authors: Kiant\'e Brantley, Mingyu Chen, Zhaolin Gao, Jason D. Lee, Wen Sun, Wenhao Zhan, Xuezhou Zhang
Abstract: Reinforcement learning (RL) has emerged as a powerful tool for fine-tuning large language models (LLMs) to improve complex reasoning abilities. However, state-of-the-art policy optimization methods often suffer from high computational overhead and memory consumption, primarily due to the need for multiple generations per prompt and the reliance on critic networks or advantage estimates of the current policy. In this paper, we propose $A$*-PO, a novel two-stage policy optimization framework that directly approximates the optimal advantage function and enables efficient training of LLMs for reasoning tasks. In the first stage, we leverage offline sampling from a reference policy to estimate the optimal value function $V$*, eliminating the need for costly online value estimation. In the second stage, we perform on-policy updates using a simple least-squares regression loss with only a single generation per prompt. Theoretically, we establish performance guarantees and prove that the KL-regularized RL objective can be optimized without requiring complex exploration strategies. Empirically, $A$*-PO achieves competitive performance across a wide range of mathematical reasoning benchmarks, while reducing training time by up to 2$\times$ and peak memory usage by over 30% compared to PPO, GRPO, and REBEL. Implementation of $A$*-PO can be found at https://github.com/ZhaolinGao/A-PO.
Authors: Shenkai Zhao, Xinao Zhang, Lipeng Pan, Xiaobin Xu, Danilo Pelusi
Abstract: Semi-supervised classification based on active learning has made significant progress, but the existing methods often ignore the uncertainty estimation (or reliability) of the prediction results during the learning process, which makes it questionable whether the selected samples can effectively update the model. Hence, this paper proposes an evidential deep active learning approach for semi-supervised classification (EDALSSC). EDALSSC builds a semi-supervised learning framework to simultaneously quantify the uncertainty estimation of labeled and unlabeled data during the learning process. The uncertainty estimation of the former is associated with evidential deep learning, while that of the latter is modeled by combining ignorance information and conflict information of the evidence from the perspective of the T-conorm operator. Furthermore, this article constructs a heuristic method to dynamically balance the influence of evidence and the number of classes on uncertainty estimation to ensure that it does not produce counter-intuitive results in EDALSSC. For the sample selection strategy, EDALSSC selects the sample with the greatest uncertainty estimation that is calculated in the form of a sum when the training loss increases in the latter half of the learning process. Experimental results demonstrate that EDALSSC outperforms existing semi-supervised and supervised active learning approaches on image classification datasets.
Authors: Saharsh Barve, Andy Mao, Jiayue Melissa Shi, Prerna Juneja, Koustuv Saha
Abstract: Recent advances in generative AI have enabled visual content creation through text-to-image (T2I) generation. However, despite their creative potential, T2I models often replicate and amplify societal stereotypes -- particularly those related to gender, race, and culture -- raising important ethical concerns. This paper proposes a theory-driven bias detection rubric and a Social Stereotype Index (SSI) to systematically evaluate social biases in T2I outputs. We audited three major T2I model outputs -- DALL-E-3, Midjourney-6.1, and Stability AI Core -- using 100 queries across three categories -- geocultural, occupational, and adjectival. Our analysis reveals that initial outputs are prone to include stereotypical visual cues, including gendered professions, cultural markers, and western beauty norms. To address this, we adopted our rubric to conduct targeted prompt refinement using LLMs, which significantly reduced bias -- SSI dropped by 61% for geocultural, 69% for occupational, and 51% for adjectival queries. We complemented our quantitative analysis through a user study examining perceptions, awareness, and preferences around AI-generated biased imagery. Our findings reveal a key tension -- although prompt refinement can mitigate stereotypes, it can limit contextual alignment. Interestingly, users often perceived stereotypical images to be more aligned with their expectations. We discuss the need to balance ethical debiasing with contextual relevance and call for T2I systems that support global diversity and inclusivity while not compromising the reflection of real-world social complexity.
Authors: Zachary C. Brown, David Carlson
Abstract: The field of hypothesis generation promises to reduce costs in neuroscience by narrowing the range of interventional studies needed to study various phenomena. Existing machine learning methods can generate scientific hypotheses from complex datasets, but many approaches assume causal relationships are static over time, limiting their applicability to systems with dynamic, state-dependent behavior, such as the brain. While some techniques attempt dynamic causal discovery through factor models, they often restrict relationships to linear patterns or impose other simplifying assumptions. We propose a novel method that models dynamic graphs as a conditionally weighted superposition of static graphs, where each static graph can capture nonlinear relationships. This approach enables the detection of complex, time-varying interactions between variables beyond linear limitations. Our method improves f1-scores of predicted dynamic causal patterns by roughly 22-28% on average over baselines in some of our experiments, with some improvements reaching well over 60%. A case study on real brain data demonstrates our method's ability to uncover relationships linked to specific behavioral states, offering valuable insights into neural dynamics.
Authors: Nicy Scaria, Silvester John Joseph Kennedy, Diksha Seth, Deepak Subramani
Abstract: Small Language Models (SLMs) offer computational efficiency and accessibility, making them promising for educational applications. However, their capacity for complex reasoning, particularly in domains such as physics, remains underexplored. This study investigates the high school physics reasoning capabilities of state-of-the-art SLMs (under 4 billion parameters), including instruct versions of Llama 3.2, Phi 4 Mini, Gemma 3, and Qwen series. We developed a comprehensive physics dataset from the OpenStax High School Physics textbook, annotated according to Bloom's Taxonomy, with LaTeX and plaintext mathematical notations. A novel cultural contextualization approach was applied to a subset, creating culturally adapted problems for Asian, African, and South American/Australian contexts while preserving core physics principles. Using an LLM-as-a-judge framework with Google's Gemini 2.5 Flash, we evaluated answer and reasoning chain correctness, along with calculation accuracy. The results reveal significant differences between the SLMs. Qwen 3 1.7B achieved high `answer accuracy' (85%), but `fully correct reasoning' was substantially low (38%). The format of the mathematical notation had a negligible impact on performance. SLMs exhibited varied performance across the physics topics and showed a decline in reasoning quality with increasing cognitive and knowledge complexity. In particular, the consistency of reasoning was largely maintained in diverse cultural contexts, especially by better performing models. These findings indicate that, while SLMs can often find correct answers, their underlying reasoning is frequently flawed, suggesting an overreliance on pattern recognition. For SLMs to become reliable educational tools in physics, future development must prioritize enhancing genuine understanding and the generation of sound, verifiable reasoning chains over mere answer accuracy.
Authors: Zechen Li, Lanqing Yang, Yiheng Bian, Hao Pan, Yongjian Fu, Yezhou Wang, Yi-Chao Chen, Guangtao Xue, Ju Ren
Abstract: This paper presents an innovative frequency-embedded 3D Gaussian splatting (3DGS) algorithm for wideband radio-frequency (RF) radiance field modeling, offering an advancement over the existing works limited to single-frequency modeling. Grounded in fundamental physics, we uncover the complex relationship between EM wave propagation behaviors and RF frequencies. Inspired by this, we design an EM feature network with attenuation and radiance modules to learn the complex relationships between RF frequencies and the key properties of each 3D Gaussian, specifically the attenuation factor and RF signal intensity. By training the frequency-embedded 3DGS model, we can efficiently reconstruct RF radiance fields at arbitrary unknown frequencies within a given 3D environment. Finally, we propose a large-scale power angular spectrum (PAS) dataset containing 50000 samples ranging from 1 to 100 GHz in 6 indoor environments, and conduct extensive experiments to verify the effectiveness of our method. Our approach achieves an average Structural Similarity Index Measure (SSIM) up to 0.72, and a significant improvement up to 17.8% compared to the current state-of-the-art (SOTA) methods trained on individual test frequencies. Additionally, our method achieves an SSIM of 0.70 without prior training on these frequencies, which represents only a 2.8% performance drop compared to models trained with full PAS data. This demonstrates our model's capability to estimate PAS at unknown frequencies. For related code and datasets, please refer to https://github.com/sim-2-real/Wideband3DGS.
Authors: Kui Wu, Shuhang Xu, Hao Chen, Churan Wang, Zhoujun Li, Yizhou Wang, Fangwei Zhong
Abstract: We introduce a novel self-improving framework that enhances Embodied Visual Tracking (EVT) with Visual-Language Models (VLMs) to address the limitations of current active visual tracking systems in recovering from tracking failure. Our approach combines the off-the-shelf active tracking methods with VLMs' reasoning capabilities, deploying a fast visual policy for normal tracking and activating VLM reasoning only upon failure detection. The framework features a memory-augmented self-reflection mechanism that enables the VLM to progressively improve by learning from past experiences, effectively addressing VLMs' limitations in 3D spatial reasoning. Experimental results demonstrate significant performance improvements, with our framework boosting success rates by $72\%$ with state-of-the-art RL-based approaches and $220\%$ with PID-based methods in challenging environments. This work represents the first integration of VLM-based reasoning to assist EVT agents in proactive failure recovery, offering substantial advances for real-world robotic applications that require continuous target monitoring in dynamic, unstructured environments. Project website: https://sites.google.com/view/evt-recovery-assistant.
Authors: Shahrooz Pouryousef
Abstract: User-item interactions contain rich collaborative signals that form the backbone of many successful recommender systems. While recent work has explored the use of large language models (LLMs) for recommendation, it remains unclear whether LLMs can effectively reason over this type of collaborative information. In this paper, we conduct a systematic comparison between LLMs and classical matrix factorization (MF) models to assess LLMs' ability to leverage user-item interaction data. We further introduce a simple retrieval-augmented generation (RAG) method that enhances LLMs by grounding their predictions in structured interaction data. Our experiments reveal that current LLMs often fall short in capturing collaborative patterns inherent to MF models, but that our RAG-based approach substantially improves recommendation quality-highlighting a promising direction for future LLM-based recommenders.
Authors: Zhuoyu Cheng, Kohei Hatano, Eiji Takimoto
Abstract: We consider a bandit optimization problem for nonconvex and non-smooth functions, where in each trial the loss function is the sum of a linear function and a small but arbitrary perturbation chosen after observing the player's choice. We give both expected and high probability regret bounds for the problem. Our result also implies an improved high-probability regret bound for the bandit linear optimization, a special case with no perturbation. We also give a lower bound on the expected regret.
Authors: Zongcai Tan amd Dandan Zhang
Abstract: Optical tweezers (OT) offer unparalleled capabilities for micromanipulation with submicron precision in biomedical applications. However, controlling conventional multi-trap OT to achieve cooperative manipulation of multiple complex-shaped microrobots in dynamic environments poses a significant challenge. To address this, we introduce Interactive OT Gym, a reinforcement learning (RL)-based simulation platform designed for OT-driven microrobotics. Our platform supports complex physical field simulations and integrates haptic feedback interfaces, RL modules, and context-aware shared control strategies tailored for OT-driven microrobot in cooperative biological object manipulation tasks. This integration allows for an adaptive blend of manual and autonomous control, enabling seamless transitions between human input and autonomous operation. We evaluated the effectiveness of our platform using a cell manipulation task. Experimental results show that our shared control system significantly improves micromanipulation performance, reducing task completion time by approximately 67% compared to using pure human or RL control alone and achieving a 100% success rate. With its high fidelity, interactivity, low cost, and high-speed simulation capabilities, Interactive OT Gym serves as a user-friendly training and testing environment for the development of advanced interactive OT-driven micromanipulation systems and control algorithms. For more details on the project, please see our website https://sites.google.com/view/otgym
Authors: Yufei Zhan, Hongyin Zhao, Yousong Zhu, Shurong Zheng, Fan Yang, Ming Tang, Jinqiao Wang
Abstract: Large Multimodal Models (LMMs) have recently demonstrated remarkable visual understanding performance on both vision-language and vision-centric tasks. However, they often fall short in integrating advanced, task-specific capabilities for compositional reasoning, which hinders their progress toward truly competent general vision models. To address this, we present a unified visual reasoning mechanism that enables LMMs to solve complicated compositional problems by leveraging their intrinsic capabilities (e.g. grounding and visual understanding capabilities). Different from the previous shortcut learning mechanism, our approach introduces a human-like understanding-thinking-answering process, allowing the model to complete all steps in a single pass forwarding without the need for multiple inferences or external tools. This design bridges the gap between foundational visual capabilities and general question answering, encouraging LMMs to generate faithful and traceable responses for complex visual reasoning. Meanwhile, we curate 334K visual instruction samples covering both general scenes and text-rich scenes and involving multiple foundational visual capabilities. Our trained model, Griffon-R, has the ability of end-to-end automatic understanding, self-thinking, and reasoning answers. Comprehensive experiments show that Griffon-R not only achieves advancing performance on complex visual reasoning benchmarks including VSR and CLEVR, but also enhances multimodal capabilities across various benchmarks like MMBench and ScienceQA. Data, models, and codes will be release at https://github.com/jefferyZhan/Griffon/tree/master/Griffon-R soon.
URLs: https://github.com/jefferyZhan/Griffon/tree/master/Griffon-R
Authors: Ansel Blume, Jeonghwan Kim, Hyeonjeong Ha, Elen Chatikyan, Xiaomeng Jin, Khanh Duy Nguyen, Nanyun Peng, Kai-Wei Chang, Derek Hoiem, Heng Ji
Abstract: Real-world objects are composed of distinctive, object-specific parts. Identifying these parts is key to performing fine-grained, compositional reasoning-yet, large multimodal models (LMMs) struggle to perform this seemingly straightforward task. In this work, we introduce PARTONOMY, an LMM benchmark designed for pixel-level part grounding. We construct PARTONOMY from existing part datasets and our own rigorously annotated set of images, encompassing 862 part labels and 534 object labels for evaluation. Unlike existing datasets that simply ask models to identify generic parts, PARTONOMY uses specialized concepts (e.g., agricultural airplane), and challenges models to compare objects' parts, consider part-whole relationships, and justify textual predictions with visual segmentations. Our experiments demonstrate significant limitations in state-of-the-art LMMs (e.g., LISA-13B achieves only 5.9% gIoU), highlighting a critical gap in their part grounding abilities. We note that existing segmentation-enabled LMMs (segmenting LMMs) have two key architectural shortcomings: they use special [SEG] tokens not seen during pretraining which induce distribution shift, and they discard predicted segmentations instead of using past predictions to guide future ones. To address these deficiencies, we train several part-centric LMMs and propose PLUM, a novel segmenting LMM that uses span tagging instead of segmentation tokens and that conditions on prior predictions in a feedback loop. We find that pretrained PLUM outperforms existing segmenting LMMs on reasoning segmentation, VQA, and visual hallucination benchmarks. In addition, PLUM finetuned on our proposed Explanatory Part Segmentation task is competitive with segmenting LMMs trained on significantly more segmentation data. Our work opens up new avenues towards enabling fine-grained, grounded visual understanding in LMMs.
Authors: Xiaqiang Tang, Jian Li, Keyu Hu, Du Nan, Xiaolong Li, Xi Zhang, Weigao Sun, Sihong Xie
Abstract: Faithfulness hallucination are claims generated by a Large Language Model (LLM) not supported by contexts provided to the LLM. Lacking assessment standard, existing benchmarks only contain "factual statements" that rephrase source materials without marking "cognitive statements" that make inference from the given context, making the consistency evaluation and optimization of cognitive statements difficult. Inspired by how an evidence is assessed in the legislative domain, we design a rigorous framework to assess different levels of faithfulness of cognitive statements and create a benchmark dataset where we reveal insightful statistics. We design an annotation pipeline to create larger benchmarks for different LLMs automatically, and the resulting larger-scale CogniBench-L dataset can be used to train accurate cognitive hallucination detection model. We release our model and dataset at: https://github.com/FUTUREEEEEE/CogniBench
Authors: Heng Tang, Feng Liu, Xinbo Chen, Jiawei Chen, Bohao Wang, Changwang Zhang, Jun Wang, Yuegang Sun, Bingde Hu, Can Wang
Abstract: Recent years have witnessed extensive exploration of Large Language Models (LLMs) on the field of Recommender Systems (RS). There are currently two commonly used strategies to enable LLMs to have recommendation capabilities: 1) The "Guidance-Only" strategy uses in-context learning to exploit and amplify the inherent semantic understanding and item recommendation capabilities of LLMs; 2) The "Tuning-Only" strategy uses supervised fine-tuning (SFT) to fine-tune LLMs with the aim of fitting them to real recommendation data. However, neither of these strategies can effectively bridge the gap between the knowledge space of LLMs and recommendation, and their performance do not meet our expectations. To better enable LLMs to learn recommendation knowledge, we combine the advantages of the above two strategies and proposed a novel "Guidance+Tuning" method called Self-Optimized Fine-Tuning (SOFT), which adopts the idea of curriculum learning. It first employs self-distillation to construct an auxiliary easy-to-learn but meaningful dataset from a fine-tuned LLM. Then it further utilizes a self-adaptive curriculum scheduler to enable LLMs to gradually learn from simpler data (self-distilled data) to more challenging data (real RS data). Extensive experiments demonstrate that SOFT significantly enhances the recommendation accuracy (37.59\% on average) of LLM-based methods. The code is available via https://anonymous.4open.science/r/Self-Optimized-Fine-Tuning-264E
URLs: https://anonymous.4open.science/r/Self-Optimized-Fine-Tuning-264E
Authors: Jungyoub Cha, Hyunjong Kim, Sungzoon Cho
Abstract: Speculative decoding is a widely adopted technique for accelerating inference in large language models (LLMs), but its performance degrades on long inputs due to increased attention cost and reduced draft accuracy. We introduce SpecExtend, a drop-in enhancement that improves the performance of speculative decoding on long sequences without any additional training. SpecExtend integrates efficient attention mechanisms such as FlashAttention and Hybrid Tree Attention into both the draft and target models, reducing latency across all stages. To improve draft accuracy and speed, we propose Cross-model Retrieval, a novel KV cache update strategy that uses the target model's attention scores to dynamically select relevant context for the draft model. Extensive evaluations on three long-context understanding datasets show that SpecExtend accelerates standard tree-based speculative decoding by up to 2.22x for inputs up to 16K tokens, providing an effective solution for speculative decoding of long sequences. The code is available at https://github.com/jycha98/SpecExtend .
Authors: Jiaping Xiao, Cheng Wen Tsao, Yuhang Zhang, Mir Feroskhan
Abstract: Path planning is a critical component in autonomous drone operations, enabling safe and efficient navigation through complex environments. Recent advances in foundation models, particularly large language models (LLMs) and vision-language models (VLMs), have opened new opportunities for enhanced perception and intelligent decision-making in robotics. However, their practical applicability and effectiveness in global path planning remain relatively unexplored. This paper proposes foundation model-guided path planners (FM-Planner) and presents a comprehensive benchmarking study and practical validation for drone path planning. Specifically, we first systematically evaluate eight representative LLM and VLM approaches using standardized simulation scenarios. To enable effective real-time navigation, we then design an integrated LLM-Vision planner that combines semantic reasoning with visual perception. Furthermore, we deploy and validate the proposed path planner through real-world experiments under multiple configurations. Our findings provide valuable insights into the strengths, limitations, and feasibility of deploying foundation models in real-world drone applications and providing practical implementations in autonomous flight. Project site: https://github.com/NTU-ICG/FM-Planner.
Authors: Juan A. Rodriguez, Haotian Zhang, Abhay Puri, Aarash Feizi, Rishav Pramanik, Pascal Wichmann, Arnab Mondal, Mohammad Reza Samsami, Rabiul Awal, Perouz Taslakian, Spandana Gella, Sai Rajeswar, David Vazquez, Christopher Pal, Marco Pedersoli
Abstract: Scalable Vector Graphics (SVG) offer a powerful format for representing visual designs as interpretable code. Recent advances in vision-language models (VLMs) have enabled high-quality SVG generation by framing the problem as a code generation task and leveraging large-scale pretraining. VLMs are particularly suitable for this task as they capture both global semantics and fine-grained visual patterns, while transferring knowledge across vision, natural language, and code domains. However, existing VLM approaches often struggle to produce faithful and efficient SVGs because they never observe the rendered images during training. Although differentiable rendering for autoregressive SVG code generation remains unavailable, rendered outputs can still be compared to original inputs, enabling evaluative feedback suitable for reinforcement learning (RL). We introduce RLRF(Reinforcement Learning from Rendering Feedback), an RL method that enhances SVG generation in autoregressive VLMs by leveraging feedback from rendered SVG outputs. Given an input image, the model generates SVG roll-outs that are rendered and compared to the original image to compute a reward. This visual fidelity feedback guides the model toward producing more accurate, efficient, and semantically coherent SVGs. RLRF significantly outperforms supervised fine-tuning, addressing common failure modes and enabling precise, high-quality SVG generation with strong structural understanding and generalization.
Authors: Joon-Seung Choi, Dong-Min Byun, Hyung-Seok Oh, Seong-Whan Lee
Abstract: Controlling singing style is crucial for achieving an expressive and natural singing voice. Among the various style factors, vibrato plays a key role in conveying emotions and enhancing musical depth. However, modeling vibrato remains challenging due to its dynamic nature, making it difficult to control in singing voice conversion. To address this, we propose VibESVC, a controllable singing voice conversion model that explicitly extracts and manipulates vibrato using discrete wavelet transform. Unlike previous methods that model vibrato implicitly, our approach decomposes the F0 contour into frequency components, enabling precise transfer. This allows vibrato control for enhanced flexibility. Experimental results show that VibE-SVC effectively transforms singing styles while preserving speaker similarity. Both subjective and objective evaluations confirm high-quality conversion.
Authors: Junsik Kim, Jinwook Park, Kangil Kim
Abstract: In knowledge graph embedding, leveraging relation-specific entity-transformation has markedly enhanced performance. However, the consistency of embedding differences before and after transformation remains unaddressed, risking the loss of valuable inductive bias inherent in the embeddings. This inconsistency stems from two problems. First, transformation representations are specified for relations in a disconnected manner, allowing dissimilar transformations and corresponding entity-embeddings for similar relations. Second, a generalized plug-in approach as a SFBR (Semantic Filter Based on Relations) disrupts this consistency through excessive concentration of entity embeddings under entity-based regularization, generating indistinguishable score distributions among relations. In this paper, we introduce a plug-in KGE method, Relation-Semantics Consistent Filter (RSCF), containing more consistent entity-transformation characterized by three features: 1) shared affine transformation of relation embeddings across all relations, 2) rooted entity-transformation that adds an entity embedding to its change represented by the transformed vector, and 3) normalization of the change to prevent scale reduction. To amplify the advantages of consistency that preserve semantics on embeddings, RSCF adds relation transformation and prediction modules for enhancing the semantics. In knowledge graph completion tasks with distance-based and tensor decomposition models, RSCF significantly outperforms state-of-the-art KGE methods, showing robustness across all relations and their frequencies.
Authors: Kai Chen, Taihang Zhen, Hewei Wang, Kailai Liu, Xinfeng Li, Jing Huo, Tianpei Yang, Jinfeng Xu, Wei Dong, Yang Gao
Abstract: As large language models (LLMs) are increasingly deployed in healthcare, ensuring their safety, particularly within collaborative multi-agent configurations, is paramount. In this paper we introduce MedSentry, a benchmark comprising 5 000 adversarial medical prompts spanning 25 threat categories with 100 subthemes. Coupled with this dataset, we develop an end-to-end attack-defense evaluation pipeline to systematically analyze how four representative multi-agent topologies (Layers, SharedPool, Centralized, and Decentralized) withstand attacks from 'dark-personality' agents. Our findings reveal critical differences in how these architectures handle information contamination and maintain robust decision-making, exposing their underlying vulnerability mechanisms. For instance, SharedPool's open information sharing makes it highly susceptible, whereas Decentralized architectures exhibit greater resilience thanks to inherent redundancy and isolation. To mitigate these risks, we propose a personality-scale detection and correction mechanism that identifies and rehabilitates malicious agents, restoring system safety to near-baseline levels. MedSentry thus furnishes both a rigorous evaluation framework and practical defense strategies that guide the design of safer LLM-based multi-agent systems in medical domains.
Authors: Shuo Wang, Shunyang Huang, Jinghui Yuan, Zhixiang Shen, Zhao Kang
Abstract: Fusing heterogeneous information remains a persistent challenge in modern data analysis. While significant progress has been made, existing approaches often fail to account for the inherent heterogeneity of object patterns across different semantic spaces. To address this limitation, we propose the Cooperation of Experts (CoE) framework, which encodes multi-typed information into unified heterogeneous multiplex networks. By overcoming modality and connection differences, CoE provides a powerful and flexible model for capturing the intricate structures of real-world complex data. In our framework, dedicated encoders act as domain-specific experts, each specializing in learning distinct relational patterns in specific semantic spaces. To enhance robustness and extract complementary knowledge, these experts collaborate through a novel large margin mechanism supported by a tailored optimization strategy. Rigorous theoretical analyses guarantee the framework's feasibility and stability, while extensive experiments across diverse benchmarks demonstrate its superior performance and broad applicability. Our code is available at https://github.com/strangeAlan/CoE.
Authors: Xin Zhou, Kisub Kim, Ting Zhang, Martin Weyssow, Luis F. Gomes, Guang Yang, David Lo
Abstract: Large Language Models (LLMs) and other automated techniques have been increasingly used to support software developers by generating software artifacts such as code snippets, patches, and comments. However, accurately assessing the correctness of these generated artifacts remains a significant challenge. On one hand, human evaluation provides high accuracy but is labor-intensive and lacks scalability. On the other hand, other existing automatic evaluation metrics are scalable and require minimal human effort, but they often fail to accurately reflect the actual correctness of generated software artifacts. In this paper, we present SWE-Judge, the first evaluation metric for LLM-as-Ensemble-Judge specifically designed to accurately assess the correctness of generated software artifacts. SWE-Judge first defines five distinct evaluation strategies, each implemented as an independent judge. A dynamic team selection mechanism then identifies the most appropriate subset of judges to produce a final correctness score through ensembling. We evaluate SWE-Judge across a diverse set of software engineering (SE) benchmarks, including CoNaLa, Card2Code, HumanEval-X, APPS, APR-Assess, and Summary-Assess. These benchmarks span three SE tasks: code generation, automated program repair, and code summarization. Experimental results demonstrate that SWE-Judge consistently achieves a higher correlation with human judgments, with improvements ranging from 5.9% to 183.8% over existing automatic metrics. Furthermore, SWE-Judge reaches agreement levels with human annotators that are comparable to inter-annotator agreement in code generation and program repair tasks. These findings underscore SWE-Judge's potential as a scalable and reliable alternative to human evaluation.
Authors: Xinjie Lin, Gang Xiong, Gaopeng Gou, Wenqi Dong, Jing Yu, Zhen Li, Wei Xia
Abstract: Encrypted traffic classification is highly challenging in network security due to the need for extracting robust features from content-agnostic traffic data. Existing approaches face critical issues: (i) Distribution drift, caused by reliance on the closedworld assumption, limits adaptability to realworld, shifting patterns; (ii) Dependence on labeled data restricts applicability where such data is scarce or unavailable. Large language models (LLMs) have demonstrated remarkable potential in offering generalizable solutions across a wide range of tasks, achieving notable success in various specialized fields. However, their effectiveness in traffic analysis remains constrained by challenges in adapting to the unique requirements of the traffic domain. In this paper, we introduce a novel traffic representation model named Encrypted Traffic Out-of-Distribution Instruction Tuning with LLM (ETooL), which integrates LLMs with knowledge of traffic structures through a self-supervised instruction tuning paradigm. This framework establishes connections between textual information and traffic interactions. ETooL demonstrates more robust classification performance and superior generalization in both supervised and zero-shot traffic classification tasks. Notably, it achieves significant improvements in F1 scores: APP53 (I.I.D.) to 93.19%(6.62%) and 92.11%(4.19%), APP53 (O.O.D.) to 74.88%(18.17%) and 72.13%(15.15%), and ISCX-Botnet (O.O.D.) to 95.03%(9.16%) and 81.95%(12.08%). Additionally, we construct NETD, a traffic dataset designed to support dynamic distributional shifts, and use it to validate ETooL's effectiveness under varying distributional conditions. Furthermore, we evaluate the efficiency gains achieved through ETooL's instruction tuning approach.
Authors: Nam-Gyu Kim, Deok-Hyeon Cho, Seung-Bin Kim, Seong-Whan Lee
Abstract: Recent advances in expressive text-to-speech (TTS) have introduced diverse methods based on style embedding extracted from reference speech. However, synthesizing high-quality expressive speech remains challenging. We propose Spotlight-TTS, which exclusively emphasizes style via voiced-aware style extraction and style direction adjustment. Voiced-aware style extraction focuses on voiced regions highly related to style while maintaining continuity across different speech regions to improve expressiveness. We adjust the direction of the extracted style for optimal integration into the TTS model, which improves speech quality. Experimental results demonstrate that Spotlight-TTS achieves superior performance compared to baseline models in terms of expressiveness, overall speech quality, and style transfer capability. Our audio samples are publicly available.
Authors: Antony Zhao, Alex Proshkin, Fergal Hennessy, Francesco Crivelli
Abstract: Large transformer models have been shown to be capable of performing in-context learning. By using examples in a prompt as well as a query, they are capable of performing tasks such as few-shot, one-shot, or zero-shot learning to output the corresponding answer to this query. One area of interest to us is that these transformer models have been shown to be capable of learning the general class of certain functions, such as linear functions and small 2-layer neural networks, on random data (Garg et al, 2023). We aim to extend this to the image space to analyze their capability to in-context learn more complex functions on the image space, such as convolutional neural networks and other methods.
Authors: Jiyoung Lee, Seungho Kim, Jieun Han, Jun-Min Lee, Kitaek Kim, Alice Oh, Edward Choi
Abstract: Large Language Models (LLMs) are predominantly evaluated on Standard American English (SAE), often overlooking the diversity of global English varieties. This narrow focus may raise fairness concerns as degraded performance on non-standard varieties can lead to unequal benefits for users worldwide. Therefore, it is critical to extensively evaluate the linguistic robustness of LLMs on multiple non-standard English varieties. We introduce Trans-EnV, a framework that automatically transforms SAE datasets into multiple English varieties to evaluate the linguistic robustness. Our framework combines (1) linguistics expert knowledge to curate variety-specific features and transformation guidelines from linguistic literature and corpora, and (2) LLM-based transformations to ensure both linguistic validity and scalability. Using Trans-EnV, we transform six benchmark datasets into 38 English varieties and evaluate seven state-of-the-art LLMs. Our results reveal significant performance disparities, with accuracy decreasing by up to 46.3% on non-standard varieties. These findings highlight the importance of comprehensive linguistic robustness evaluation across diverse English varieties. Each construction of Trans-EnV was validated through rigorous statistical testing and consultation with a researcher in the field of second language acquisition, ensuring its linguistic validity. Our \href{https://github.com/jiyounglee-0523/TransEnV}{code} and \href{https://huggingface.co/collections/jiyounglee0523/transenv-681eadb3c0c8cf363b363fb1}{datasets} are publicly available.
URLs: https://github.com/jiyounglee-0523/TransEnV, https://huggingface.co/collections/jiyounglee0523/transenv-681eadb3c0c8cf363b363fb1
Authors: Yiding Shi, Jianan Zhou, Wen Song, Jieyi Bi, Yaoxin Wu, Jie Zhang
Abstract: Heuristic design with large language models (LLMs) has emerged as a promising approach for tackling combinatorial optimization problems (COPs). However, existing approaches often rely on manually predefined evolutionary computation (EC) optimizers and single-task training schemes, which may constrain the exploration of diverse heuristic algorithms and hinder the generalization of the resulting heuristics. To address these issues, we propose Meta-Optimization of Heuristics (MoH), a novel framework that operates at the optimizer level, discovering effective optimizers through the principle of meta-learning. Specifically, MoH leverages LLMs to iteratively refine a meta-optimizer that autonomously constructs diverse optimizers through (self-)invocation, thereby eliminating the reliance on a predefined EC optimizer. These constructed optimizers subsequently evolve heuristics for downstream tasks, enabling broader heuristic exploration. Moreover, MoH employs a multi-task training scheme to promote its generalization capability. Experiments on classic COPs demonstrate that MoH constructs an effective and interpretable meta-optimizer, achieving state-of-the-art performance across various downstream tasks, particularly in cross-size settings.
Authors: Chengyu Wang, Junbing Yan, Wenrui Cai, Yuanhao Yue, Jun Huang
Abstract: In this paper, we present EasyDistill, a comprehensive toolkit designed for effective black-box and white-box knowledge distillation (KD) of large language models (LLMs). Our framework offers versatile functionalities, including data synthesis, supervised fine-tuning, ranking optimization, and reinforcement learning techniques specifically tailored for KD scenarios. The toolkit accommodates KD functionalities for both System 1 (fast, intuitive) and System 2 (slow, analytical) models. With its modular design and user-friendly interface, EasyDistill empowers researchers and industry practitioners to seamlessly experiment with and implement state-of-the-art KD strategies for LLMs. In addition, EasyDistill provides a series of robust distilled models and KD-based industrial solutions developed by us, along with the corresponding open-sourced datasets, catering to a variety of use cases. Furthermore, we describe the seamless integration of EasyDistill into Alibaba Cloud's Platform for AI (PAI). Overall, the EasyDistill toolkit makes advanced KD techniques for LLMs more accessible and impactful within the NLP community.
Authors: Yoojin Kwon, Hongjun Suh, Wooseok Lee, Taesik Gong, Songyi Han, Hyung-Sin Kim
Abstract: Modern on-device neural network applications must operate under resource constraints while adapting to unpredictable domain shifts. However, this combined challenge-model compression and domain adaptation-remains largely unaddressed, as prior work has tackled each issue in isolation: compressed networks prioritize efficiency within a fixed domain, whereas large, capable models focus on handling domain shifts. In this work, we propose CoDA, a frequency composition-based framework that unifies compression and domain adaptation. During training, CoDA employs quantization-aware training (QAT) with low-frequency components, enabling a compressed model to selectively learn robust, generalizable features. At test time, it refines the compact model in a source-free manner (i.e., test-time adaptation, TTA), leveraging the full-frequency information from incoming data to adapt to target domains while treating high-frequency components as domain-specific cues. LFC are aligned with the trained distribution, while HFC unique to the target distribution are solely utilized for batch normalization. CoDA can be integrated synergistically into existing QAT and TTA methods. CoDA is evaluated on widely used domain-shift benchmarks, including CIFAR10-C and ImageNet-C, across various model architectures. With significant compression, it achieves accuracy improvements of 7.96%p on CIFAR10-C and 5.37%p on ImageNet-C over the full-precision TTA baseline.
Authors: Yiwei Wu, Atticus Geiger, Rapha\"el Milli\`ere
Abstract: Variable binding -- the ability to associate variables with values -- is fundamental to symbolic computation and cognition. Although classical architectures typically implement variable binding via addressable memory, it is not well understood how modern neural networks lacking built-in binding operations may acquire this capacity. We investigate this by training a Transformer to dereference queried variables in symbolic programs where variables are assigned either numerical constants or other variables. Each program requires following chains of variable assignments up to four steps deep to find the queried value, and also contains irrelevant chains of assignments acting as distractors. Our analysis reveals a developmental trajectory with three distinct phases during training: (1) random prediction of numerical constants, (2) a shallow heuristic prioritizing early variable assignments, and (3) the emergence of a systematic mechanism for dereferencing assignment chains. Using causal interventions, we find that the model learns to exploit the residual stream as an addressable memory space, with specialized attention heads routing information across token positions. This mechanism allows the model to dynamically track variable bindings across layers, resulting in accurate dereferencing. Our results show how Transformer models can learn to implement systematic variable binding without explicit architectural support, bridging connectionist and symbolic approaches.
Authors: Pingrui Zhang, Yifei Su, Pengyuan Wu, Dong An, Li Zhang, Zhigang Wang, Dong Wang, Yan Ding, Bin Zhao, Xuelong Li
Abstract: Vision-and-Language Navigation (VLN) requires the agent to navigate by following natural instructions under partial observability, making it difficult to align perception with language. Recent methods mitigate this by imagining future scenes, yet they rely on vision-based synthesis, leading to high computational cost and redundant details. To this end, we propose to adaptively imagine key environmental semantics via \textit{language} form, enabling a more reliable and efficient strategy. Specifically, we introduce a novel Adaptive Text Dreamer (ATD), a dual-branch self-guided imagination policy built upon a large language model (LLM). ATD is designed with a human-like left-right brain architecture, where the left brain focuses on logical integration, and the right brain is responsible for imaginative prediction of future scenes. To achieve this, we fine-tune only the Q-former within both brains to efficiently activate domain-specific knowledge in the LLM, enabling dynamic updates of logical reasoning and imagination during navigation. Furthermore, we introduce a cross-interaction mechanism to regularize the imagined outputs and inject them into a navigation expert module, allowing ATD to jointly exploit both the reasoning capacity of the LLM and the expertise of the navigation model. We conduct extensive experiments on the R2R benchmark, where ATD achieves state-of-the-art performance with fewer parameters. The code is \href{https://github.com/zhangpingrui/Adaptive-Text-Dreamer}{here}.
Authors: Junhyuk Choi, Minju Kim, Yeseon Hong, Bugeun Kim
Abstract: As large vision language models(LVLMs) rapidly advance, concerns about their potential to learn and generate social biases and stereotypes are increasing. Previous studies on LVLM's stereotypes face two primary limitations: metrics that overlooked the importance of content words, and datasets that overlooked the effect of color. To address these limitations, this study introduces new evaluation metrics based on the Stereotype Content Model (SCM). We also propose BASIC, a benchmark for assessing gender, race, and color stereotypes. Using SCM metrics and BASIC, we conduct a study with eight LVLMs to discover stereotypes. As a result, we found three findings. (1) The SCM-based evaluation is effective in capturing stereotypes. (2) LVLMs exhibit color stereotypes in the output along with gender and race ones. (3) Interaction between model architecture and parameter sizes seems to affect stereotypes. We release BASIC publicly on [anonymized for review].
Authors: Rahul Nair, Inge Vejsbjerg, Elizabeth Daly, Christos Varytimidis, Bran Knowles
Abstract: Humble AI (Knowles et al., 2023) argues for cautiousness in AI development and deployments through scepticism (accounting for limitations of statistical learning), curiosity (accounting for unexpected outcomes), and commitment (accounting for multifaceted values beyond performance). We present a real-world case study for humble AI in the domain of algorithmic hiring. Specifically, we evaluate virtual screening algorithms in a widely used hiring platform that matches candidates to job openings. There are several challenges in misrecognition and stereotyping in such contexts that are difficult to assess through standard fairness and trust frameworks; e.g., someone with a non-traditional background is less likely to rank highly. We demonstrate technical feasibility of how humble AI principles can be translated to practice through uncertainty quantification of ranks, entropy estimates, and a user experience that highlights algorithmic unknowns. We describe preliminary discussions with focus groups made up of recruiters. Future user studies seek to evaluate whether the higher cognitive load of a humble AI system fosters a climate of trust in its outcomes.
Authors: Injae Na, Keonwoong Noh, Woohwan Jung
Abstract: LLM providers typically offer multiple LLM tiers, varying in performance and price. As NLP tasks become more complex and modularized, selecting the suitable LLM tier for each subtask is a key challenge to balance between cost and performance. To address the problem, we introduce LLM Automatic Transmission (LLM-AT) framework that automatically selects LLM tiers without training. LLM-AT consists of Starter, Generator, and Judge. The starter selects the initial LLM tier expected to solve the given question, the generator produces a response using the LLM of the selected tier, and the judge evaluates the validity of the response. If the response is invalid, LLM-AT iteratively upgrades to a higher-tier model, generates a new response, and re-evaluates until a valid response is obtained. Additionally, we propose accuracy estimator, which enables the suitable initial LLM tier selection without training. Given an input question, accuracy estimator estimates the expected accuracy of each LLM tier by computing the valid response rate across top-k similar queries from past inference records. Experiments demonstrate that LLM-AT achieves superior performance while reducing costs, making it a practical solution for real-world applications.
Authors: Yang Zhang, Xinran Li, Jianing Ye, Delin Qu, Shuang Qiu, Chongjie Zhang, Xiu Li, Chenjia Bai
Abstract: World models have recently attracted growing interest in Multi-Agent Reinforcement Learning (MARL) due to their ability to improve sample efficiency for policy learning. However, accurately modeling environments in MARL is challenging due to the exponentially large joint action space and highly uncertain dynamics inherent in multi-agent systems. To address this, we reduce modeling complexity by shifting from jointly modeling the entire state-action transition dynamics to focusing on the state space alone at each timestep through sequential agent modeling. Specifically, our approach enables the model to progressively resolve uncertainty while capturing the structured dependencies among agents, providing a more accurate representation of how agents influence the state. Interestingly, this sequential revelation of agents' actions in a multi-agent system aligns with the reverse process in diffusion models--a class of powerful generative models known for their expressiveness and training stability compared to autoregressive or latent variable models. Leveraging this insight, we develop a flexible and robust world model for MARL using diffusion models. Our method, Diffusion-Inspired Multi-Agent world model (DIMA), achieves state-of-the-art performance across multiple multi-agent control benchmarks, significantly outperforming prior world models in terms of final return and sample efficiency, including MAMuJoCo and Bi-DexHands. DIMA establishes a new paradigm for constructing multi-agent world models, advancing the frontier of MARL research.
Authors: Zhuo Li, Guodong Du, Weiyang Guo, Yigeng Zhou, Xiucheng Li, Wenya Wang, Fangming Liu, Yequan Wang, Deheng Ye, Min Zhang, Jing Li
Abstract: Aligning large language models (LLMs) to simultaneously satisfy multiple objectives remains a significant challenge, especially given the diverse and often conflicting nature of human preferences. Existing alignment methods struggle to balance trade-offs effectively, often requiring costly retraining or yielding suboptimal results across the Pareto frontier of preferences. In this paper, we introduce \textit{HoE}(Hierarchical Mixture-of-Experts), a \textit{lightweight}, \textit{parameter-efficient}, and \textit{plug-and-play} approach that eliminates the need for model training, while enabling LLMs to adapt across the entire Pareto frontier and accommodate diverse user preferences. In particular, \textit{HoE} consists of three hierarchical components: LoRA Experts, Router Experts and Preference Routing, reaching optimal Pareto frontiers and achieving a trade-off between parameter size, training cost, and performance. We evaluate \textit{HoE} across various tasks on 14 objectives and 200 different preferences among 6 benchmarks, demonstrating superior performance over 15 recent baselines. Code is available in the supplementary materials.
Authors: Lorenzo Monti, Tatiana Muraveva, Alessia Garofalo, Gisella Clementini, Maria Letizia Valentini
Abstract: RR Lyrae stars (RRLs) are old pulsating variables widely used as metallicity tracers due to the correlation between their metal abundances and light curve morphology. With ESA Gaia DR3 providing light curves for about 270,000 RRLs, there is a pressing need for scalable methods to estimate their metallicities from photometric data. We introduce a unified deep learning framework that estimates metallicities for both fundamental-mode (RRab) and first-overtone (RRc) RRLs using Gaia G-band light curves. This approach extends our previous work on RRab stars to include RRc stars, aiming for high predictive accuracy and broad generalization across both pulsation types. The model is based on a Gated Recurrent Unit (GRU) neural network optimized for time-series extrinsic regression. Our pipeline includes preprocessing steps such as phase folding, smoothing, and sample weighting, and uses photometric metallicities from the literature as training targets. The architecture is designed to handle morphological differences between RRab and RRc light curves without requiring separate models. On held-out validation sets, our GRU model achieves strong performance: for RRab stars, MAE = 0.0565 dex, RMSE = 0.0765 dex, R^2 = 0.9401; for RRc stars, MAE = 0.0505 dex, RMSE = 0.0720 dex, R^2 = 0.9625. These results show the effectiveness of deep learning for large-scale photometric metallicity estimation and support its application to studies of stellar populations and Galactic structure.
Authors: Andrea Giovanni Nuzzolese
Abstract: Knowledge Graphs (KGs) are increasingly adopted as a foundational technology for integrating heterogeneous data in domains such as climate science, cultural heritage, and the life sciences. Declarative mapping languages like R2RML and RML have played a central role in enabling scalable and reusable KG construction, offering a transparent means of transforming structured and semi-structured data into RDF. In this paper, we present PyRML, a lightweight, Python-native library for building Knowledge Graphs through declarative mappings. PyRML supports core RML constructs and provides a programmable interface for authoring, executing, and testing mappings directly within Python environments. It integrates with popular data and semantic web libraries (e.g., Pandas and RDFlib), enabling transparent and modular workflows. By lowering the barrier to entry for KG creation and fostering reproducible, ontology-aligned data integration, PyRML bridges the gap between declarative semantics and practical KG engineering.
Authors: Shiqi Zhang, Tuomas Virtanen
Abstract: Bioacoustic sound event detection (BioSED) is crucial for biodiversity conservation but faces practical challenges during model development and training: limited amounts of annotated data, sparse events, species diversity, and class imbalance. To address these challenges efficiently with a limited labeling budget, we apply the mismatch-first farthest-traversal (MFFT), an active learning method integrating committee voting disagreement and diversity analysis. We also refine an existing BioSED dataset specifically for evaluating active learning algorithms. Experimental results demonstrate that MFFT achieves a mAP of 68% when cold-starting and 71% when warm-starting (which is close to the fully-supervised mAP of 75%) while using only 2.3% of the annotations. Notably, MFFT excels in cold-start scenarios and with rare species, which are critical for monitoring endangered species, demonstrating its practical value.
Authors: Yiyuan Yang, Shitong Xu, Niki Trigoni, Andrew Markham
Abstract: Sound source localization (SSL) is a critical technology for determining the position of sound sources in complex environments. However, existing methods face challenges such as high computational costs and precise calibration requirements, limiting their deployment in dynamic or resource-constrained environments. This paper introduces a novel 3D SSL framework, which uses sparse cross-attention, pretraining, and adaptive signal coherence metrics, to achieve accurate and computationally efficient localization with fewer input microphones. The framework is also fault-tolerant to unreliable or even unknown microphone position inputs, ensuring its applicability in real-world scenarios. Preliminary experiments demonstrate its scalability for multi-source localization without requiring additional hardware. This work advances SSL by balancing the model's performance and efficiency and improving its robustness for real-world scenarios.
Authors: Felix Krejca, Tobias Kietreiber, Alexander Buchelt, Sebastian Neumaier
Abstract: The increasing volume of online discussions requires advanced automatic content moderation to maintain responsible discourse. While hate speech detection on social media is well-studied, research on German-language newspaper forums remains limited. Existing studies often neglect platform-specific context, such as user history and article themes. This paper addresses this gap by developing and evaluating binary classification models for automatic content moderation in German newspaper forums, incorporating contextual information. Using LSTM, CNN, and ChatGPT-3.5 Turbo, and leveraging the One Million Posts Corpus from the Austrian newspaper Der Standard, we assess the impact of context-aware models. Results show that CNN and LSTM models benefit from contextual information and perform competitively with state-of-the-art approaches. In contrast, ChatGPT's zero-shot classification does not improve with added context and underperforms.
Authors: Xiangqing Shen, Fanfan Wang, Rui Xia
Abstract: LLMs have demonstrated remarkable capabilities in complex reasoning tasks, yet they often suffer from hallucinations and lack reliable factual grounding. Meanwhile, knowledge graphs (KGs) provide structured factual knowledge but lack the flexible reasoning abilities of LLMs. In this paper, we present Reason-Align-Respond (RAR), a novel framework that systematically integrates LLM reasoning with knowledge graphs for KGQA. Our approach consists of three key components: a Reasoner that generates human-like reasoning chains, an Aligner that maps these chains to valid KG paths, and a Responser that synthesizes the final answer. We formulate this process as a probabilistic model and optimize it using the Expectation-Maximization algorithm, which iteratively refines the reasoning chains and knowledge paths. Extensive experiments on multiple benchmarks demonstrate the effectiveness of RAR, achieving state-of-the-art performance with Hit@1 scores of 93.3% and 91.0% on WebQSP and CWQ respectively. Human evaluation confirms that RAR generates high-quality, interpretable reasoning chains well-aligned with KG paths. Furthermore, RAR exhibits strong zero-shot generalization capabilities and maintains computational efficiency during inference.
Authors: Sen Bai, Chunqi Yang, Xin Bai, Xin Zhang, Zhengang Jiang
Abstract: Along with AI computing shining in scientific discovery, its potential in the combinatorial optimization (CO) domain has also emerged in recent years. Yet, existing unsupervised neural network solvers struggle to solve $k$-grouping problems (e.g., coloring, partitioning) on large-scale graphs and hypergraphs, due to limited computational frameworks. In this work, we propose Deep $k$-grouping, an unsupervised learning-based CO framework. Specifically, we contribute: Novel one-hot encoded polynomial unconstrained binary optimization (OH-PUBO), a formulation for modeling k-grouping problems on graphs and hypergraphs (e.g., graph/hypergraph coloring and partitioning); GPU-accelerated algorithms for large-scale k-grouping CO problems. Deep $k$-grouping employs the relaxation of large-scale OH-PUBO objectives as differentiable loss functions and trains to optimize them in an unsupervised manner. To ensure scalability, it leverages GPU-accelerated algorithms to unify the training pipeline; A Gini coefficient-based continuous relaxation annealing strategy to enforce discreteness of solutions while preventing convergence to local optima. Experimental results demonstrate that Deep $k$-grouping outperforms existing neural network solvers and classical heuristics such as SCIP and Tabu.
Authors: Keheliya Gallaba, Ali Arabat, Dayi Lin, Mohammed Sayagh, Ahmed E. Hassan
Abstract: Foundation Models (FMs) have shown remarkable capabilities in various natural language tasks. However, their ability to accurately capture stakeholder requirements remains a significant challenge for using FMs for software development. This paper introduces a novel approach that leverages an FM-powered multi-agent system called AlignMind to address this issue. By having a cognitive architecture that enhances FMs with Theory-of-Mind capabilities, our approach considers the mental states and perspectives of software makers. This allows our solution to iteratively clarify the beliefs, desires, and intentions of stakeholders, translating these into a set of refined requirements and a corresponding actionable natural language workflow in the often-overlooked requirements refinement phase of software engineering, which is crucial after initial elicitation. Through a multifaceted evaluation covering 150 diverse use cases, we demonstrate that our approach can accurately capture the intents and requirements of stakeholders, articulating them as both specifications and a step-by-step plan of action. Our findings suggest that the potential for significant improvements in the software development process justifies these investments. Our work lays the groundwork for future innovation in building intent-first development environments, where software makers can seamlessly collaborate with AIs to create software that truly meets their needs.
Authors: Tongyu Lu, Charlotta-Marlena Geist, Jan Melechovsky, Abhinaba Roy, Dorien Herremans
Abstract: We propose MelodySim, a melody-aware music similarity model and dataset for plagiarism detection. First, we introduce a novel method to construct a dataset with focus on melodic similarity. By augmenting Slakh2100; an existing MIDI dataset, we generate variations of each piece while preserving the melody through modifications such as note splitting, arpeggiation, minor track dropout (excluding bass), and re-instrumentation. A user study confirms that positive pairs indeed contain similar melodies, with other musical tracks significantly changed. Second, we develop a segment-wise melodic-similarity detection model that uses a MERT encoder and applies a triplet neural network to capture melodic similarity. The resultant decision matrix highlights where plagiarism might occur. Our model achieves high accuracy on the MelodySim test set.
Authors: Jie Shao, Jianxin Wu
Abstract: Despite the impressive performance of large language models (LLMs), the process of endowing them with new capabilities--such as mathematical reasoning--remains largely empirical and opaque. A critical open question is whether reasoning abilities stem from the entire model, specific modules, or are merely artifacts of overfitting. In this work, we hypothesize that the reasoning capabilities in well-trained LLMs are primarily attributed to the output projection module (oproj) in the Transformer's multi-head self-attention (MHSA) mechanism. To support this hypothesis, we introduce Stethoscope for Networks (SfN), a suite of diagnostic tools designed to probe and analyze the internal behaviors of LLMs. Using SfN, we provide both circumstantial and empirical evidence suggesting that oproj plays a central role in enabling reasoning, whereas other modules contribute more to fluent dialogue. These findings offer a new perspective on LLM interpretability and open avenues for more targeted training strategies, potentially enabling more efficient and specialized LLMs.
Authors: Sen Bai, Chunqi Yang, Xin Bai, Xin Zhang, Zhengang Jiang
Abstract: Binary (0-1) integer programming (BIP) is pivotal in scientific domains requiring discrete decision-making. As the advance of AI computing, recent works explore neural network-based solvers for integer linear programming (ILP) problems. Yet, they lack scalability for tackling nonlinear challenges. To handle nonlinearities, state-of-the-art Branch-and-Cut solvers employ linear relaxations, leading to exponential growth in auxiliary variables and severe computation limitations. To overcome these limitations, we propose BIPNN (Binary Integer Programming Neural Network), an unsupervised learning framework to solve nonlinear BIP problems via hypergraph neural networks (HyperGNN). Specifically, BIPNN reformulates BIPs-constrained, discrete, and nonlinear (sin, log, exp) optimization problems-into unconstrained, differentiable, and polynomial loss functions. The reformulation stems from the observation of a precise one-to-one mapping between polynomial BIP objectives and hypergraph structures, enabling the unsupervised training of HyperGNN to optimize BIP problems in an end-to-end manner. On this basis, we propose a GPU-accelerated and continuous-annealing-enhanced training pipeline for BIPNN. The pipeline enables BIPNN to optimize large-scale nonlinear terms in BIPs fully in parallel via straightforward gradient descent, thus significantly reducing the training cost while ensuring the generation of discrete, high-quality solutions. Extensive experiments on synthetic and real-world datasets highlight the superiority of our approach.
Authors: Geetika, Somya Tyagi, Bapi Chatterjee
Abstract: Instrumental variables (IV) analysis is an important applied tool for areas such as healthcare and consumer economics. For IV analysis in high-dimensional settings, the Generalized Method of Moments (GMM) using deep neural networks offers an efficient approach. With non-i.i.d. data sourced from scattered decentralized clients, federated learning is a popular paradigm for training the models while promising data privacy. However, to our knowledge, no federated algorithm for either GMM or IV analysis exists to date. In this work, we introduce federated instrumental variables analysis (FedIV) via federated generalized method of moments (FedGMM). We formulate FedGMM as a federated zero-sum game defined by a federated non-convex non-concave minimax optimization problem, which is solved using federated gradient descent ascent (FedGDA) algorithm. One key challenge arises in theoretically characterizing the federated local optimality. To address this, we present properties and existence results of clients' local equilibria via FedGDA limit points. Thereby, we show that the federated solution consistently estimates the local moment conditions of every participating client. The proposed algorithm is backed by extensive experiments to demonstrate the efficacy of our approach.
Authors: Xinlei Yin, Xiulian Peng, Xue Jiang, Zhiwei Xiong, Yan Lu
Abstract: Target audio source separation with natural language queries presents a promising paradigm for extracting arbitrary audio events through arbitrary text descriptions. Existing methods mainly face two challenges, the difficulty in jointly modeling acoustic-textual alignment and semantic-aware separation within a blindly-learned single-stage architecture, and the reliance on large-scale accurately-labeled training data to compensate for inefficient cross-modal learning and separation. To address these challenges, we propose a hierarchical decomposition framework, HSM-TSS, that decouples the task into global-local semantic-guided feature separation and structure-preserving acoustic reconstruction. Our approach introduces a dual-stage mechanism for semantic separation, operating on distinct global and local semantic feature spaces. We first perform global-semantic separation through a global semantic feature space aligned with text queries. A Q-Audio architecture is employed to align audio and text modalities, serving as pretrained global-semantic encoders. Conditioned on the predicted global feature, we then perform the second-stage local-semantic separation on AudioMAE features that preserve time-frequency structures, followed by acoustic reconstruction. We also propose an instruction processing pipeline to parse arbitrary text queries into structured operations, extraction or removal, coupled with audio descriptions, enabling flexible sound manipulation. Our method achieves state-of-the-art separation performance with data-efficient training while maintaining superior semantic consistency with queries in complex auditory scenes.
Authors: Runze Lin, Junghui Chen, Biao Huang, Lei Xie, Hongye Su
Abstract: In the era of Industry 4.0 and smart manufacturing, process systems engineering must adapt to digital transformation. While reinforcement learning offers a model-free approach to process control, its applications are limited by the dependence on accurate digital twins and well-designed reward functions. To address these limitations, this paper introduces a novel framework that integrates inverse reinforcement learning (IRL) with multi-task learning for data-driven, multi-mode control design. Using historical closed-loop data as expert demonstrations, IRL extracts optimal reward functions and control policies. A latent-context variable is incorporated to distinguish modes, enabling the training of mode-specific controllers. Case studies on a continuous stirred tank reactor and a fed-batch bioreactor validate the effectiveness of this framework in handling multi-mode data and training adaptable controllers.
Authors: Zhipeng He, Chun Ouyang, Lijie Wen, Cong Liu, Catarina Moreira
Abstract: Adversarial attacks pose a significant threat to machine learning models by inducing incorrect predictions through imperceptible perturbations to input data. While these attacks have been extensively studied in unstructured data like images, their application to tabular data presents new challenges. These challenges arise from the inherent heterogeneity and complex feature interdependencies in tabular data, which differ significantly from those in image data. To address these differences, it is crucial to consider imperceptibility as a key criterion specific to tabular data. Most current research focuses primarily on achieving effective adversarial attacks, often overlooking the importance of maintaining imperceptibility. To address this gap, we propose a new benchmark for adversarial attacks on tabular data that evaluates both effectiveness and imperceptibility. In this study, we assess the effectiveness and imperceptibility of five adversarial attacks across four models using eleven tabular datasets, including both mixed and numerical-only datasets. Our analysis explores how these factors interact and influence the overall performance of the attacks. We also compare the results across different dataset types to understand the broader implications of these findings. The findings from this benchmark provide valuable insights for improving the design of adversarial attack algorithms, thereby advancing the field of adversarial machine learning on tabular data.
Authors: Nils Neukirch, Johanna Vielhaben, Nils Strodthoff
Abstract: Internal representations are crucial for understanding deep neural networks, such as their properties and reasoning patterns, but remain difficult to interpret. While mapping from feature space to input space aids in interpreting the former, existing approaches often rely on crude approximations. We propose using a conditional diffusion model - a pretrained high-fidelity diffusion model conditioned on spatially resolved feature maps - to learn such a mapping in a probabilistic manner. We demonstrate the feasibility of this approach across various pretrained image classifiers from CNNs to ViTs, showing excellent reconstruction capabilities. Through qualitative comparisons and robustness analysis, we validate our method and showcase possible applications, such as the visualization of concept steering in input space or investigations of the composite nature of the feature space. This approach has broad potential for improving feature space understanding in computer vision models.
Authors: Aiyue Chen, Bin Dong, Jingru Li, Jing Lin, Yiwu Yao, Gongyi Wang
Abstract: Video generation using diffusion models is highly computationally intensive, with 3D attention in Diffusion Transformer (DiT) models accounting for over 80\% of the total computational resources. In this work, we introduce {\bf RainFusion}, a novel training-free sparse attention method that exploits inherent sparsity nature in visual data to accelerate attention computation while preserving video quality. Specifically, we identify three unique sparse patterns in video generation attention calculations--Spatial Pattern, Temporal Pattern and Textural Pattern. The sparse pattern for each attention head is determined online with negligible overhead (\textasciitilde\,0.2\%) with our proposed {\bf ARM} (Adaptive Recognition Module) during inference. Our proposed {\bf RainFusion} is a plug-and-play method, that can be seamlessly integrated into state-of-the-art 3D-attention video generation models without additional training or calibration. We evaluate our method on leading open-sourced models including HunyuanVideo, OpenSoraPlan-1.2 and CogVideoX-5B, demonstrating its broad applicability and effectiveness. Experimental results show that RainFusion achieves over {\bf 2\(\times\)} speedup in attention computation while maintaining video quality, with only a minimal impact on VBench scores (-0.2\%).
Authors: Faruk Alpay
Abstract: This paper presents a formal categorical proof that exam-driven educational systems obstruct identity emergence and block creative convergence. Using the framework of Alpay Algebra II and III, we define Exam-Grade Collapse Systems (EGCS) as functorial constructs where learning dynamics $\varphi$ are recursively collapsed by evaluative morphisms $E$. We prove that under such collapse regimes, no nontrivial fixed-point algebra $\mu_\varphi$ can exist, hence learner identity cannot stabilize. This creates a universal fixed-point trap: all generative functors are entropically folded before symbolic emergence occurs. Our model mathematically explains the creativity suppression, research stagnation, and structural entropy loss induced by timed exams and grade-based feedback. The results apply category theory to expose why modern educational systems prevent {\phi}-emergence and block observer-invariant self-formation. This work provides the first provable algebraic obstruction of identity formation caused by institutional feedback mechanics.
Authors: Wei Chen, Zhao Zhang, Meng Yuan, Kepeng Xu, Fuzhen Zhuang
Abstract: In this paper, we address the task of targeted sentiment analysis (TSA), which involves two sub-tasks, i.e., identifying specific aspects from reviews and determining their corresponding sentiments. Aspect extraction forms the foundation for sentiment prediction, highlighting the critical dependency between these two tasks for effective cross-task knowledge transfer. While most existing studies adopt a multi-task learning paradigm to align task-specific features in the latent space, they predominantly rely on coarse-grained knowledge transfer. Such approaches lack fine-grained control over aspect-sentiment relationships, often assuming uniform sentiment polarity within related aspects. This oversimplification neglects contextual cues that differentiate sentiments, leading to negative transfer. To overcome these limitations, we propose FCKT, a fine-grained cross-task knowledge transfer framework tailored for TSA. By explicitly incorporating aspect-level information into sentiment prediction, FCKT achieves fine-grained knowledge transfer, effectively mitigating negative transfer and enhancing task performance. Experiments on three datasets, including comparisons with various baselines and large language models (LLMs), demonstrate the effectiveness of FCKT. The source code is available on https://github.com/cwei01/FCKT.
Authors: Zhenling Chen, Haiwei Fu, Zhiguo Zeng
Abstract: Digital twins offer a promising solution to the lack of sufficient labeled data in deep learning-based fault diagnosis by generating simulated data for model training. However, discrepancies between simulation and real-world systems can lead to a significant drop in performance when models are applied in real scenarios. To address this issue, we propose a fault diagnosis framework based on Domain-Adversarial Neural Networks (DANN), which enables knowledge transfer from simulated (source domain) to real-world (target domain) data. We evaluate the proposed framework using a publicly available robotics fault diagnosis dataset, which includes 3,600 sequences generated by a digital twin model and 90 real sequences collected from physical systems. The DANN method is compared with commonly used lightweight deep learning models such as CNN, TCN, Transformer, and LSTM. Experimental results show that incorporating domain adaptation significantly improves the diagnostic performance. For example, applying DANN to a baseline CNN model improves its accuracy from 70.00% to 80.22% on real-world test data, demonstrating the effectiveness of domain adaptation in bridging the sim-to-real gap.
Authors: Fatemeh Pesaran Zadeh, Yoojin Oh, Gunhee Kim
Abstract: Aligning large VLMs with human preferences is a challenging task, as methods like RLHF and DPO often overfit to textual information or exacerbate hallucinations. Although augmenting negative image samples partially addresses these pitfalls, no prior work has employed listwise preference optimization for VLMs, due to the complexity and cost of constructing listwise image samples. In this work, we propose LPOI, the first object-aware listwise preference optimization developed for reducing hallucinations in VLMs. LPOI identifies and masks a critical object in the image, and then interpolates the masked region between the positive and negative images to form a sequence of incrementally more complete images. The model is trained to rank these images in ascending order of object visibility, effectively reducing hallucinations while retaining visual fidelity. LPOI requires no extra annotations beyond standard pairwise preference data, as it automatically constructs the ranked lists through object masking and interpolation. Comprehensive experiments on MMHalBench, AMBER, and Object HalBench confirm that LPOI outperforms existing preference optimization methods in reducing hallucinations and enhancing VLM performance. We make the code available at https://github.com/fatemehpesaran310/lpoi.
Authors: Yichuan Cao, Yibo Miao, Xiao-Shan Gao, Yinpeng Dong
Abstract: Text-to-image (T2I) models raise ethical and safety concerns due to their potential to generate inappropriate or harmful images. Evaluating these models' security through red-teaming is vital, yet white-box approaches are limited by their need for internal access, complicating their use with closed-source models. Moreover, existing black-box methods often assume knowledge about the model's specific defense mechanisms, limiting their utility in real-world commercial API scenarios. A significant challenge is how to evade unknown and diverse defense mechanisms. To overcome this difficulty, we propose a novel Rule-based Preference modeling Guided Red-Teaming (RPG-RT), which iteratively employs LLM to modify prompts to query and leverages feedback from T2I systems for fine-tuning the LLM. RPG-RT treats the feedback from each iteration as a prior, enabling the LLM to dynamically adapt to unknown defense mechanisms. Given that the feedback is often labeled and coarse-grained, making it difficult to utilize directly, we further propose rule-based preference modeling, which employs a set of rules to evaluate desired or undesired feedback, facilitating finer-grained control over the LLM's dynamic adaptation process. Extensive experiments on nineteen T2I systems with varied safety mechanisms, three online commercial API services, and T2V models verify the superiority and practicality of our approach.
Authors: Mete Erdogan, Francesco Tonin, Volkan Cevher
Abstract: The high inference demands of transformer-based Large Language Models (LLMs) pose substantial challenges in their deployment. To this end, we introduce Neural Block Linearization (NBL), a novel framework for accelerating transformer model inference by replacing self-attention layers with linear approximations derived from Linear Minimum Mean Squared Error estimators. NBL leverages Canonical Correlation Analysis to compute a theoretical upper bound on the approximation error. Then, we use this bound as a criterion for substitution, selecting the LLM layers with the lowest linearization error. NBL can be efficiently applied to pre-trained LLMs without the need for fine-tuning. In experiments, NBL achieves notable computational speed-ups while preserving competitive accuracy on multiple reasoning benchmarks. For instance, applying NBL to 12 self-attention layers in DeepSeek-R1-Distill-Llama-8B increases the inference speed by 32% with less than 1% accuracy trade-off, making it a flexible and promising solution to improve the inference efficiency of LLMs.
Authors: Marta Grobelna, Jan K\v{r}et\'insk\'y, Maximilian Weininger
Abstract: We consider two-player zero-sum concurrent stochastic games (CSGs) played on graphs with reachability and safety objectives. These include degenerate classes such as Markov decision processes or turn-based stochastic games, which can be solved by linear or quadratic programming; however, in practice, value iteration (VI) outperforms the other approaches and is the most implemented method. Similarly, for CSGs, this practical performance makes VI an attractive alternative to the standard theoretical solution via the existential theory of reals. VI starts with an under-approximation of the sought values for each state and iteratively updates them, traditionally terminating once two consecutive approximations are $\epsilon$-close. However, this stopping criterion lacks guarantees on the precision of the approximation, which is the goal of this work. We provide bounded (a.k.a. interval) VI for CSGs: it complements standard VI with a converging sequence of over-approximations and terminates once the over- and under-approximations are $\epsilon$-close.
Authors: Anna Neumann, Elisabeth Kirsten, Muhammad Bilal Zafar, Jatinder Singh
Abstract: System prompts in Large Language Models (LLMs) are predefined directives that guide model behaviour, taking precedence over user inputs in text processing and generation. LLM deployers increasingly use them to ensure consistent responses across contexts. While model providers set a foundation of system prompts, deployers and third-party developers can append additional prompts without visibility into others' additions, while this layered implementation remains entirely hidden from end-users. As system prompts become more complex, they can directly or indirectly introduce unaccounted for side effects. This lack of transparency raises fundamental questions about how the position of information in different directives shapes model outputs. As such, this work examines how the placement of information affects model behaviour. To this end, we compare how models process demographic information in system versus user prompts across six commercially available LLMs and 50 demographic groups. Our analysis reveals significant biases, manifesting in differences in user representation and decision-making scenarios. Since these variations stem from inaccessible and opaque system-level configurations, they risk representational, allocative and potential other biases and downstream harms beyond the user's ability to detect or correct. Our findings draw attention to these critical issues, which have the potential to perpetuate harms if left unexamined. Further, we argue that system prompt analysis must be incorporated into AI auditing processes, particularly as customisable system prompts become increasingly prevalent in commercial AI deployments.
Authors: Daeen Kabir, Minhajur Rahman Chowdhury Mahim, Sheikh Shafayat, Adnan Sadik, Arian Ahmed, Eunsu Kim, Alice Oh
Abstract: In this work, we introduce BLUCK, a new dataset designed to measure the performance of Large Language Models (LLMs) in Bengali linguistic understanding and cultural knowledge. Our dataset comprises 2366 multiple-choice questions (MCQs) carefully curated from compiled collections of several college and job level examinations and spans 23 categories covering knowledge on Bangladesh's culture and history and Bengali linguistics. We benchmarked BLUCK using 6 proprietary and 3 open-source LLMs - including GPT-4o, Claude-3.5-Sonnet, Gemini-1.5-Pro, Llama-3.3-70B-Instruct, and DeepSeekV3. Our results show that while these models perform reasonably well overall, they, however, struggles in some areas of Bengali phonetics. Although current LLMs' performance on Bengali cultural and linguistic contexts is still not comparable to that of mainstream languages like English, our results indicate Bengali's status as a mid-resource language. Importantly, BLUCK is also the first MCQ-based evaluation benchmark that is centered around native Bengali culture, history, and linguistics.
Authors: Stephen Chung, Wenyu Du, Jie Fu
Abstract: Recent studies show that the reasoning capabilities of Large Language Models (LLMs) can be improved by applying Reinforcement Learning (RL) to question-answering (QA) tasks in areas such as math and coding. With a long context length, LLMs may learn to perform search, as indicated by the self-correction behavior observed in DeepSeek R1. However, this search behavior is often imprecise and lacks confidence, resulting in long, redundant responses and highlighting deficiencies in intuition and verification. Inspired by the Dual Process Theory in psychology, we introduce a simple modification to the QA task that includes four stages: Fast Thinking, where the LLM must answer within a strict token budget; Verification, where the model evaluates its initial response; Slow Thinking, where it refines the initial response with more deliberation; and Summarization, where it distills the refinement from the previous stage into precise steps. Our proposed task improves average accuracy from 24.9% to 27.9% for Qwen2.5-1.5B, and from 45.9% to 49.8% for DeepSeek-R1-Qwen-1.5B. Notably, for Qwen2.5-1.5B, the Fast Thinking mode alone achieves 26.8% accuracy using fewer than 1000 tokens, demonstrating substantial inference efficiency gains. These findings suggest that intuition and deliberative reasoning are distinct, complementary systems benefiting from targeted training.
Authors: Bogdan Bogachov, Yaoyao Fiona Zhao
Abstract: Despite recent advancements in domain adaptation techniques for large language models, these methods remain computationally intensive, and the resulting models can still exhibit hallucination issues. Most existing adaptation methods do not prioritize reducing the computational resources required for fine-tuning and inference of language models. Hallucination issues have gradually decreased with each new model release. However, they remain prevalent in engineering contexts, where generating well-structured text with minimal errors and inconsistencies is critical. This work introduces a novel approach called the Small Language Graph (SLG), which is a lightweight adaptation solution designed to address the two key challenges outlined above. The system is structured in the form of a graph, where each node represents a lightweight expert - a small language model fine-tuned on specific and concise texts. The results of this study have shown that SLG was able to surpass conventional fine-tuning methods on the Exact Match metric by 3 times. Additionally, the fine-tuning process was 1.7 times faster compared to that of a larger stand-alone language model. These findings introduce a potential for small to medium-sized engineering companies to confidently use generative AI technologies, such as LLMs, without the necessity to invest in expensive computational resources. Also, the graph architecture and the small size of expert nodes offer a possible opportunity for distributed AI systems, thus potentially diverting the global need for expensive centralized compute clusters.
Authors: Yi-Cheng Lin, Kang-Chieh Chen, Zhe-Yan Li, Tzu-Heng Wu, Tzu-Hsuan Wu, Kuan-Yu Chen, Hung-yi Lee, Yun-Nung Chen
Abstract: Large language model (LLM)-driven multi-agent systems (MAS) are transforming how humans and AIs collaboratively generate ideas and artifacts. While existing surveys provide comprehensive overviews of MAS infrastructures, they largely overlook the dimension of \emph{creativity}, including how novel outputs are generated and evaluated, how creativity informs agent personas, and how creative workflows are coordinated. This is the first survey dedicated to creativity in MAS. We focus on text and image generation tasks, and present: (1) a taxonomy of agent proactivity and persona design; (2) an overview of generation techniques, including divergent exploration, iterative refinement, and collaborative synthesis, as well as relevant datasets and evaluation metrics; and (3) a discussion of key challenges, such as inconsistent evaluation standards, insufficient bias mitigation, coordination conflicts, and the lack of unified benchmarks. This survey offers a structured framework and roadmap for advancing the development, evaluation, and standardization of creative MAS.
Authors: Moritz A. Zanger, Max Weltevrede, Yaniv Oren, Pascal R. Van der Vaart, Caroline Horsch, Wendelin B\"ohmer, Matthijs T. J. Spaan
Abstract: Estimating epistemic uncertainty in value functions is a crucial challenge for many aspects of reinforcement learning (RL), including efficient exploration, safe decision-making, and offline RL. While deep ensembles provide a robust method for quantifying value uncertainty, they come with significant computational overhead. Single-model methods, while computationally favorable, often rely on heuristics and typically require additional propagation mechanisms for myopic uncertainty estimates. In this work we introduce universal value-function uncertainties (UVU), which, similar in spirit to random network distillation (RND), quantify uncertainty as squared prediction errors between an online learner and a fixed, randomly initialized target network. Unlike RND, UVU errors reflect policy-conditional value uncertainty, incorporating the future uncertainties any given policy may encounter. This is due to the training procedure employed in UVU: the online network is trained using temporal difference learning with a synthetic reward derived from the fixed, randomly initialized target network. We provide an extensive theoretical analysis of our approach using neural tangent kernel (NTK) theory and show that in the limit of infinite network width, UVU errors are exactly equivalent to the variance of an ensemble of independent universal value functions. Empirically, we show that UVU achieves equal performance to large ensembles on challenging multi-task offline RL settings, while offering simplicity and substantial computational savings.
Authors: Jintao Zhang, Xiaoming Xu, Jia Wei, Haofeng Huang, Pengle Zhang, Chendong Xiang, Jun Zhu, Jianfei Chen
Abstract: The efficiency of attention is critical because its time complexity grows quadratically with sequence length. SageAttention2 addresses this by utilizing quantization to accelerate matrix multiplications (Matmul) in attention. To further accelerate SageAttention2, we propose to utilize the faster instruction of FP8 Matmul accumulated in FP16. The instruction is 2x faster than the FP8 Matmul used in SageAttention2. Our experiments show that SageAttention2++ achieves a 3.9x speedup over FlashAttention while maintaining the same attention accuracy as SageAttention2. This means SageAttention2++ effectively accelerates various models, including those for language, image, and video generation, with negligible end-to-end metrics loss. The code will be available at https://github.com/thu-ml/SageAttention.
Authors: Honglin Gao, Xiang Li, Lan Zhao, Gaoxi Xiao
Abstract: Heterogeneous graph neural networks (HGNNs) have recently drawn increasing attention for modeling complex multi-relational data in domains such as recommendation, finance, and social networks. While existing research has been largely focused on enhancing HGNNs' predictive performance, their robustness and security, especially under backdoor attacks, remain underexplored. In this paper, we propose a novel Heterogeneous Backdoor Attack (HeteroBA) framework for node classification tasks on heterogeneous graphs. HeteroBA inserts carefully crafted trigger nodes with realistic features and targeted structural connections, leveraging attention-based and clustering-based strategies to select influential auxiliary nodes for effective trigger propagation, thereby causing the model to misclassify specific nodes into a target label while maintaining accuracy on clean data. Experimental results on three datasets and various HGNN architectures demonstrate that HeteroBA achieves high attack success rates with minimal impact on the clean accuracy. Our method sheds light on potential vulnerabilities in HGNNs and calls for more robust defenses against backdoor threats in multi-relational graph scenarios.
Authors: Hailin Zhong, Hanlin Wang, Yujun Ye, Meiyi Zhang, Shengxin Zhu
Abstract: Current personalized recommender systems predominantly rely on static offline data for algorithm design and evaluation, significantly limiting their ability to capture long-term user preference evolution and social influence dynamics in real-world scenarios. To address this fundamental challenge, we propose a high-fidelity social simulation platform integrating human-like cognitive agents and dynamic social interactions to realistically simulate user behavior evolution under recommendation interventions. Specifically, the system comprises a population of Sim-User Agents, each equipped with a five-layer cognitive architecture that encapsulates key psychological mechanisms, including episodic memory, affective state transitions, adaptive preference learning, and dynamic trust-risk assessments. In particular, we innovatively introduce the Intimacy--Curiosity--Reciprocity--Risk (ICR2) motivational engine grounded in psychological and sociological theories, enabling more realistic user decision-making processes. Furthermore, we construct a multilayer heterogeneous social graph (GGBond Graph) supporting dynamic relational evolution, effectively modeling users' evolving social ties and trust dynamics based on interest similarity, personality alignment, and structural homophily. During system operation, agents autonomously respond to recommendations generated by typical recommender algorithms (e.g., Matrix Factorization, MultVAE, LightGCN), deciding whether to consume, rate, and share content while dynamically updating their internal states and social connections, thereby forming a stable, multi-round feedback loop. This innovative design transcends the limitations of traditional static datasets, providing a controlled, observable environment for evaluating long-term recommender effects.
Authors: Saisamarth Rajesh Phaye, Milos Cernak, Andrew Harper
Abstract: Conventional methods for speech enhancement rely on handcrafted loss functions (e.g., time or frequency domain losses) or deep feature losses (e.g., using WavLM or wav2vec), which often fail to capture subtle signal properties essential for optimal performance. To address this, we propose Model as Loss, a novel training paradigm that utilizes the encoder from the same model as a loss function to guide the training. The Model as Loss paradigm leverages the encoder's task-specific feature space, optimizing the decoder to produce output consistent with perceptual and task-relevant characteristics of the clean signal. By using the encoder's learned features as a loss function, this framework enforces self-consistency between the clean reference speech and the enhanced model output. Our approach outperforms pre-trained deep feature losses on standard speech enhancement benchmarks, offering better perceptual quality and robust generalization to both in-domain and out-of-domain datasets.
Authors: Michael Stenger, Robert Leppich, Andr\'e Bauer, Samuel Kounev
Abstract: The growing need for synthetic time series, due to data augmentation or privacy regulations, has led to numerous generative models, frameworks, and evaluation measures alike. Objectively comparing these measures on a large scale remains an open challenge. We propose the Synthetic Time series Evaluation Benchmark (STEB) -- the first benchmark framework that enables comprehensive and interpretable automated comparisons of synthetic time series evaluation measures. Using 10 diverse datasets, randomness injection, and 13 configurable data transformations, STEB computes indicators for measure reliability and score consistency. It tracks running time, test errors, and features sequential and parallel modes of operation. In our experiments, we determine a ranking of 41 measures from literature and confirm that the choice of upstream time series embedding heavily impacts the final score.
Authors: Elija Perrier
Abstract: AIXI is a widely studied model of artificial general intelligence (AGI) based upon principles of induction and reinforcement learning. However, AIXI is fundamentally classical in nature - as are the environments in which it is modelled. Given the universe is quantum mechanical in nature and the exponential overhead required to simulate quantum mechanical systems classically, the question arises as to whether there are quantum mechanical analogues of AIXI which are theoretically consistent or practically feasible as models of universal intelligence. To address this question, we extend the framework to quantum information and present Quantum AIXI (QAIXI). We introduce a model of quantum agent/environment interaction based upon quantum and classical registers and channels, showing how quantum AIXI agents may take both classical and quantum actions. We formulate the key components of AIXI in quantum information terms, extending previous research on quantum Kolmogorov complexity and a QAIXI value function. We discuss conditions and limitations upon quantum Solomonoff induction and show how contextuality fundamentally affects QAIXI models.
Authors: Rochelle Choenni, Ivan Titov
Abstract: Multilingual LLM performance is often critically dependent on model size. With an eye on efficiency, this has led to a surge in interest in one-shot pruning methods that retain the benefits of large-scale pretraining while shrinking the model size. However, as pruning tends to come with performance loss, it is important to understand the trade-offs between multilinguality and sparsification. In this work, we study multilingual performance under different sparsity constraints and show that moderate ratios already substantially harm performance. To help bridge this gap, we propose M-Wanda, a pruning method that models cross-lingual variation by incorporating language-aware activation statistics into its pruning criterion and dynamically adjusts layerwise sparsity based on cross-lingual importance. We show that M-Wanda consistently improves performance at minimal additional costs. We are the first to explicitly optimize pruning to retain multilingual performance, and hope to inspire future advances in multilingual pruning.
Authors: ShuNing Sun, YinSong Xiong, Yu Zhang, Zhuoran Zheng
Abstract: Although \textbf{L}abel \textbf{D}istribution \textbf{L}earning (LDL) has promising representation capabilities for characterizing the polysemy of an instance, the complexity and high cost of the label distribution annotation lead to inexact in the construction of the label space. The existence of a large number of inexact labels generates a label space with uncertainty, which misleads the LDL algorithm to yield incorrect decisions. To alleviate this problem, we model the uncertainty of label distributions by constructing a \textbf{L}atent \textbf{L}abel \textbf{D}istribution \textbf{G}rid (LLDG) to form a low-noise representation space. Specifically, we first construct a label correlation matrix based on the differences between labels, and then expand each value of the matrix into a vector that obeys a Gaussian distribution, thus building a LLDG to model the uncertainty of the label space. Finally, the LLDG is reconstructed by the LLDG-Mixer to generate an accurate label distribution. Note that we enforce a customized low-rank scheme on this grid, which assumes that the label relations may be noisy and it needs to perform noise-reduction with the help of a Tucker reconstruction technique. Furthermore, we attempt to evaluate the effectiveness of the LLDG by considering its generation as an upstream task to achieve the classification of the objects. Extensive experimental results show that our approach performs competitively on several benchmarks.
Authors: Huy Hoang, Tien Mai, Pradeep Varakantham, Tanvi Verma
Abstract: Offline imitation learning typically learns from expert and unlabeled demonstrations, yet often overlooks the valuable signal in explicitly undesirable behaviors. In this work, we study offline imitation learning from contrasting behaviors, where the dataset contains both expert and undesirable demonstrations. We propose a novel formulation that optimizes a difference of KL divergences over the state-action visitation distributions of expert and undesirable (or bad) data. Although the resulting objective is a DC (Difference-of-Convex) program, we prove that it becomes convex when expert demonstrations outweigh undesirable demonstrations, enabling a practical and stable non-adversarial training objective. Our method avoids adversarial training and handles both positive and negative demonstrations in a unified framework. Extensive experiments on standard offline imitation learning benchmarks demonstrate that our approach consistently outperforms state-of-the-art baselines.
Authors: Yu Yan, Sheng Sun, Zhifei Zheng, Ziji Hao, Teli Liu, Min Liu
Abstract: To construct responsible and secure AI applications, harmful information data is widely utilized for adversarial testing and the development of safeguards. Existing studies mainly leverage Large Language Models (LLMs) to synthesize data to obtain high-quality task datasets at scale, thereby avoiding costly human annotation. However, limited by the safety alignment mechanisms of LLMs, the synthesis of harmful data still faces challenges in generation reliability and content diversity. In this study, we propose a novel harmful information synthesis framework, PoisonSwarm, which applies the model crowdsourcing strategy to generate diverse harmful data while maintaining a high success rate. Specifically, we generate abundant benign data as the based templates in a counterfactual manner. Subsequently, we decompose each based template into multiple semantic units and perform unit-by-unit toxification and final refinement through dynamic model switching, thus ensuring the success of synthesis. Experimental results demonstrate that PoisonSwarm achieves state-of-the-art performance in synthesizing different categories of harmful data with high scalability and diversity.
Authors: Gleb Mezentsev, Ivan Oseledets
Abstract: A recent study showed that large language models (LLMs) can reconstruct surprisingly long texts - up to thousands of tokens - via autoregressive generation from just one specially trained input embedding. In this work, we explore whether such reconstruction is possible without autoregression. We show that frozen LLMs can generate hundreds of accurate tokens in just one forward pass, when provided with only two learned embeddings. This reveals a surprising and underexplored capability of LLMs - multi-token generation without iterative decoding. We investigate the behaviour of these embeddings and provide insight into the type of information they encode. We also empirically show that although these representations are not unique for a given text, they form connected and local regions in embedding space - a property that suggests the potential of learning a dedicated encoder into that space.
Authors: Jong Hak Moon, Geon Choi, Paloma Rabaey, Min Gwan Kim, Hyuk Gi Hong, Jung-Oh Lee, Hangyul Yoon, Eun Woo Doe, Jiyoun Kim, Harshita Sharma, Daniel C. Castro, Javier Alvarez-Valle, Edward Choi
Abstract: Radiology reports convey detailed clinical observations and capture diagnostic reasoning that evolves over time. However, existing evaluation methods are limited to single-report settings and rely on coarse metrics that fail to capture fine-grained clinical semantics and temporal dependencies. We introduce LUNGUAGE,a benchmark dataset for structured radiology report generation that supports both single-report evaluation and longitudinal patient-level assessment across multiple studies. It contains 1,473 annotated chest X-ray reports, each reviewed by experts, and 80 of them contain longitudinal annotations to capture disease progression and inter-study intervals, also reviewed by experts. Using this benchmark, we develop a two-stage framework that transforms generated reports into fine-grained, schema-aligned structured representations, enabling longitudinal interpretation. We also propose LUNGUAGESCORE, an interpretable metric that compares structured outputs at the entity, relation, and attribute level while modeling temporal consistency across patient timelines. These contributions establish the first benchmark dataset, structuring framework, and evaluation metric for sequential radiology reporting, with empirical results demonstrating that LUNGUAGESCORE effectively supports structured report evaluation. The code is available at: https://github.com/SuperSupermoon/Lunguage
Authors: Roi Cohen, Omri Fahn, Gerard de Melo
Abstract: Large Language Models are known to capture real-world knowledge, allowing them to excel in many downstream tasks. Despite recent advances, these models are still prone to what are commonly known as hallucinations, causing them to emit unwanted and factually incorrect text. In this work, we study how well LLMs capture uncertainty, without explicitly being trained for that. We show that, if considering uncertainty as a linear concept in the model's latent space, it might indeed be captured, even after only pretraining. We further show that, though unintuitive, LLMs appear to capture several different types of uncertainty, each of which can be useful to predict the correctness for a specific task or benchmark. Furthermore, we provide in-depth results such as demonstrating a correlation between our correction prediction and the model's ability to abstain from misinformation using words, and the lack of impact of model scaling for capturing uncertainty. Finally, we claim that unifying the uncertainty types as a single one using instruction-tuning or [IDK]-token tuning is helpful for the model in terms of correctness prediction.
Authors: Qinjun Fei, Nuria Rodr\'iguez-Barroso, Mar\'ia Victoria Luz\'on, Zhongliang Zhang, Francisco Herrera
Abstract: In cross-silo Federated Learning (FL), client selection is critical to ensure high model performance, yet it remains challenging due to data quality decompensation, budget constraints, and incentive compatibility. As training progresses, these factors exacerbate client heterogeneity and degrade global performance. Most existing approaches treat these challenges in isolation, making jointly optimizing multiple factors difficult. To address this, we propose Shapley-Bid Reputation Optimized Federated Learning (SBRO-FL), a unified framework integrating dynamic bidding, reputation modeling, and cost-aware selection. Clients submit bids based on their perceived data quality, and their contributions are evaluated using Shapley values to quantify their marginal impact on the global model. A reputation system, inspired by prospect theory, captures historical performance while penalizing inconsistency. The client selection problem is formulated as a 0-1 integer program that maximizes reputation-weighted utility under budget constraints. Experiments on FashionMNIST, EMNIST, CIFAR-10, and SVHN datasets show that SBRO-FL improves accuracy, convergence speed, and robustness, even in adversarial and low-bid interference scenarios. Our results highlight the importance of balancing data reliability, incentive compatibility, and cost efficiency to enable scalable and trustworthy FL deployments.
Authors: Alvaro Gonzalez-Jimenez, Simone Lionetti, Ludovic Amruthalingam, Philippe Gottfrois, Fabian Gr\"oger, Marc Pouly, Alexander A. Navarini
Abstract: Medical anomaly detection has emerged as a promising solution to challenges in data availability and labeling constraints. Traditional methods extract features from different layers of pre-trained networks in Euclidean space; however, Euclidean representations fail to effectively capture the hierarchical relationships within these features, leading to suboptimal anomaly detection performance. We propose a novel yet simple approach that projects feature representations into hyperbolic space, aggregates them based on confidence levels, and classifies samples as healthy or anomalous. Our experiments demonstrate that hyperbolic space consistently outperforms Euclidean-based frameworks, achieving higher AUROC scores at both image and pixel levels across multiple medical benchmark datasets. Additionally, we show that hyperbolic space exhibits resilience to parameter variations and excels in few-shot scenarios, where healthy images are scarce. These findings underscore the potential of hyperbolic space as a powerful alternative for medical anomaly detection. The project website can be found at https://hyperbolic-anomalies.github.io
Authors: Nima Sedghiyeh, Sara Sadeghi, Reza Khodadadi, Farzin Kashani, Omid Aghdaei, Somayeh Rahimi, Mohammad Sadegh Safari
Abstract: Although Automatic Speech Recognition (ASR) systems have become an integral part of modern technology, their evaluation remains challenging, particularly for low-resource languages such as Persian. This paper introduces Persian Speech Recognition Benchmark(PSRB), a comprehensive benchmark designed to address this gap by incorporating diverse linguistic and acoustic conditions. We evaluate ten ASR systems, including state-of-the-art commercial and open-source models, to examine performance variations and inherent biases. Additionally, we conduct an in-depth analysis of Persian ASR transcriptions, identifying key error types and proposing a novel metric that weights substitution errors. This metric enhances evaluation robustness by reducing the impact of minor and partial errors, thereby improving the precision of performance assessment. Our findings indicate that while ASR models generally perform well on standard Persian, they struggle with regional accents, children's speech, and specific linguistic challenges. These results highlight the necessity of fine-tuning and incorporating diverse, representative training datasets to mitigate biases and enhance overall ASR performance. PSRB provides a valuable resource for advancing ASR research in Persian and serves as a framework for developing benchmarks in other low-resource languages. A subset of the PSRB dataset is publicly available at https://huggingface.co/datasets/PartAI/PSRB.
Authors: Felix Chalumeau, Daniel Rajaonarivonivelomanantsoa, Ruan de Kock, Claude Formanek, Sasha Abramowitz, Oumayma Mahjoub, Wiem Khlifi, Simon Du Toit, Louay Ben Nessir, Refiloe Shabe, Arnol Fokam, Siddarth Singh, Ulrich Mbou Sob, Arnu Pretorius
Abstract: Reinforcement learning (RL) systems have countless applications, from energy-grid management to protein design. However, such real-world scenarios are often extremely difficult, combinatorial in nature, and require complex coordination between multiple agents. This level of complexity can cause even state-of-the-art RL systems, trained until convergence, to hit a performance ceiling which they are unable to break out of with zero-shot inference. Meanwhile, many digital or simulation-based applications allow for an inference phase that utilises a specific time and compute budget to explore multiple attempts before outputting a final solution. In this work, we show that such an inference phase employed at execution time, and the choice of a corresponding inference strategy, are key to breaking the performance ceiling observed in complex multi-agent RL problems. Our main result is striking: we can obtain up to a 126% and, on average, a 45% improvement over the previous state-of-the-art across 17 tasks, using only a couple seconds of extra wall-clock time during execution. We also demonstrate promising compute scaling properties, supported by over 60k experiments, making it the largest study on inference strategies for complex RL to date. Our experimental data and code are available at https://sites.google.com/view/inf-marl.
Authors: Ilker Kesen, Jonas F. Lotz, Ingo Ziegler, Phillip Rust, Desmond Elliott
Abstract: Pixel language models operate directly on images of rendered text, eliminating the need for a fixed vocabulary. While these models have demonstrated strong capabilities for downstream cross-lingual transfer, multilingual pretraining remains underexplored. We introduce PIXEL-M4, a model pretrained on four visually and linguistically diverse languages: English, Hindi, Ukrainian, and Simplified Chinese. Multilingual evaluations on semantic and syntactic tasks show that PIXEL-M4 outperforms an English-only counterpart on non-Latin scripts. Word-level probing analyses confirm that PIXEL-M4 captures rich linguistic features, even in languages not seen during pretraining. Furthermore, an analysis of its hidden representations shows that multilingual pretraining yields a semantic embedding space closely aligned across the languages used for pretraining. This work demonstrates that multilingual pretraining substantially enhances the capability of pixel language models to effectively support a diverse set of languages.
Authors: Yao Huang, Yitong Sun, Shouwei Ruan, Yichi Zhang, Yinpeng Dong, Xingxing Wei
Abstract: Large Language Models (LLMs), despite advanced general capabilities, still suffer from numerous safety risks, especially jailbreak attacks that bypass safety protocols. Understanding these vulnerabilities through black-box jailbreak attacks, which better reflect real-world scenarios, offers critical insights into model robustness. While existing methods have shown improvements through various prompt engineering techniques, their success remains limited against safety-aligned models, overlooking a more fundamental problem: the effectiveness is inherently bounded by the predefined strategy spaces. However, expanding this space presents significant challenges in both systematically capturing essential attack patterns and efficiently navigating the increased complexity. To better explore the potential of expanding the strategy space, we address these challenges through a novel framework that decomposes jailbreak strategies into essential components based on the Elaboration Likelihood Model (ELM) theory and develops genetic-based optimization with intention evaluation mechanisms. To be striking, our experiments reveal unprecedented jailbreak capabilities by expanding the strategy space: we achieve over 90% success rate on Claude-3.5 where prior methods completely fail, while demonstrating strong cross-model transferability and surpassing specialized safeguard models in evaluation accuracy. The code is open-sourced at: https://github.com/Aries-iai/CL-GSO.
Authors: Farshad Noravesh, Reza Haffari, Layki Soon, Arghya Pal
Abstract: Graph Neural Networks (GNNs) have emerged as a powerful tool for processing data represented in graph structures, achieving remarkable success across a wide range of applications. However, to further improve the performance on graph classification benchmarks, structural representation of each node that encodes rich local topological information in the neighbourhood of nodes is an important type of feature that is often overlooked in the modeling. The consequence of neglecting the structural information has resulted high number of layers to connect messages from distant nodes which by itself produces other problems such as oversmoothing. In the present paper, we leverage these structural information that are modeled by anonymous random walks (ARWs) and introduce graph structure attention network (GSAT) which is a generalization of graph attention network(GAT) to integrate the original attribute and the structural representation to enforce the model to automatically find patterns for attending to different edges in the node neighbourhood to enrich graph representation. Our experiments show GSAT slightly improves SOTA on some graph classification benchmarks.
Authors: Emanuele La Malfa, Gabriele La Malfa, Samuele Marro, Jie M. Zhang, Elizabeth Black, Micheal Luck, Philip Torr, Michael Wooldridge
Abstract: Recent interest in Multi-Agent Systems of Large Language Models (MAS LLMs) has led to an increase in frameworks leveraging multiple LLMs to tackle complex tasks. However, much of this literature appropriates the terminology of MAS without engaging with its foundational principles. In this position paper, we highlight critical discrepancies between MAS theory and current MAS LLMs implementations, focusing on four key areas: the social aspect of agency, environment design, coordination and communication protocols, and measuring emergent behaviours. Our position is that many MAS LLMs lack multi-agent characteristics such as autonomy, social interaction, and structured environments, and often rely on oversimplified, LLM-centric architectures. The field may slow down and lose traction by revisiting problems the MAS literature has already addressed. Therefore, we systematically analyse this issue and outline associated research opportunities; we advocate for better integrating established MAS concepts and more precise terminology to avoid mischaracterisation and missed opportunities.
Authors: Andrea Pedrotti, Giulia Rambelli, Caterina Villani, Marianna Bolognesi
Abstract: People can categorize the same entity at multiple taxonomic levels, such as basic (bear), superordinate (animal), and subordinate (grizzly bear). While prior research has focused on basic-level categories, this study is the first attempt to examine the organization of categories by analyzing exemplars produced at the subordinate level. We present a new Italian psycholinguistic dataset of human-generated exemplars for 187 concrete words. We then use these data to evaluate whether textual and vision LLMs produce meaningful exemplars that align with human category organization across three key tasks: exemplar generation, category induction, and typicality judgment. Our findings show a low alignment between humans and LLMs, consistent with previous studies. However, their performance varies notably across different semantic domains. Ultimately, this study highlights both the promises and the constraints of using AI-generated exemplars to support psychological and linguistic research.
Authors: Ihab Bendidi, Yassir El Mesbahi, Alisandra K. Denton, Karush Suri, Kian Kenyon-Dean, Auguste Genovesio, Emmanuel Noutahi
Abstract: Understanding cellular responses to stimuli is crucial for biological discovery and drug development. Transcriptomics provides interpretable, gene-level insights, while microscopy imaging offers rich predictive features but is harder to interpret. Weakly paired datasets, where samples share biological states, enable multimodal learning but are scarce, limiting their utility for training and multimodal inference. We propose a framework to enhance transcriptomics by distilling knowledge from microscopy images. Using weakly paired data, our method aligns and binds modalities, enriching gene expression representations with morphological information. To address data scarcity, we introduce (1) Semi-Clipped, an adaptation of CLIP for cross-modal distillation using pretrained foundation models, achieving state-of-the-art results, and (2) PEA (Perturbation Embedding Augmentation), a novel augmentation technique that enhances transcriptomics data while preserving inherent biological information. These strategies improve the predictive power and retain the interpretability of transcriptomics, enabling rich unimodal representations for complex biological tasks.
Authors: Allaa Boutaleb, Bernd Amann, Hubert Naacke, Rafael Angarita
Abstract: Recent table representation learning and data discovery methods tackle table union search (TUS) within data lakes, which involves identifying tables that can be unioned with a given query table to enrich its content. These methods are commonly evaluated using benchmarks that aim to assess semantic understanding in real-world TUS tasks. However, our analysis of prominent TUS benchmarks reveals several limitations that allow simple baselines to perform surprisingly well, often outperforming more sophisticated approaches. This suggests that current benchmark scores are heavily influenced by dataset-specific characteristics and fail to effectively isolate the gains from semantic understanding. To address this, we propose essential criteria for future benchmarks to enable a more realistic and reliable evaluation of progress in semantic table union search.
Authors: Takuhiro Kaneko
Abstract: Recent advancements in neural 3D representations, such as neural radiance fields (NeRF) and 3D Gaussian splatting (3DGS), have enabled the accurate estimation of 3D structures from multiview images. However, this capability is limited to estimating the visible external structure, and identifying the invisible internal structure hidden behind the surface is difficult. To overcome this limitation, we address a new task called Structure from Collision (SfC), which aims to estimate the structure (including the invisible internal structure) of an object from appearance changes during collision. To solve this problem, we propose a novel model called SfC-NeRF that optimizes the invisible internal structure of an object through a video sequence under physical, appearance (i.e., visible external structure)-preserving, and keyframe constraints. In particular, to avoid falling into undesirable local optima owing to its ill-posed nature, we propose volume annealing; that is, searching for global optima by repeatedly reducing and expanding the volume. Extensive experiments on 115 objects involving diverse structures (i.e., various cavity shapes, locations, and sizes) and material properties revealed the properties of SfC and demonstrated the effectiveness of the proposed SfC-NeRF.
Authors: Henryk Mustroph, Michel Kunkler, Stefanie Rinderle-Ma
Abstract: Suffix prediction of business processes forecasts the remaining sequence of events until process completion. Current approaches focus on predicting a single, most likely suffix. However, if the future course of a process is exposed to uncertainty or has high variability, the expressiveness of a single suffix prediction can be limited. To address this limitation, we propose probabilistic suffix prediction, a novel approach that approximates a probability distribution of suffixes. The proposed approach is based on an Uncertainty-Aware Encoder-Decoder LSTM (U-ED-LSTM) and a Monte Carlo (MC) suffix sampling algorithm. We capture epistemic uncertainties via MC dropout and aleatoric uncertainties as learned loss attenuation. This technical report provides a detailed evaluation of the U-ED-LSTM's predictive performance and assesses its calibration on four real-life event logs with three different hyperparameter settings. The results show that i) the U-ED-LSTM has reasonable predictive performance across various datasets, ii) aggregating probabilistic suffix predictions into mean values can outperform most likely predictions, particularly for rare prefixes or longer suffixes, and iii) the approach effectively captures uncertainties present in event logs.
Authors: Muhammad Imran, Wayne G. Brisbane, Li-Ming Su, Jason P. Joseph, Wei Shao
Abstract: Background and objective: Micro-ultrasound (micro-US) is a novel imaging modality with diagnostic accuracy comparable to MRI for detecting clinically significant prostate cancer (csPCa). We investigated whether artificial intelligence (AI) interpretation of micro-US can outperform clinical screening methods using PSA and digital rectal examination (DRE). Methods: We retrospectively studied 145 men who underwent micro-US guided biopsy (79 with csPCa, 66 without). A self-supervised convolutional autoencoder was used to extract deep image features from 2D micro-US slices. Random forest classifiers were trained using five-fold cross-validation to predict csPCa at the slice level. Patients were classified as csPCa-positive if 88 or more consecutive slices were predicted positive. Model performance was compared with a classifier using PSA, DRE, prostate volume, and age. Key findings and limitations: The AI-based micro-US model and clinical screening model achieved AUROCs of 0.871 and 0.753, respectively. At a fixed threshold, the micro-US model achieved 92.5% sensitivity and 68.1% specificity, while the clinical model showed 96.2% sensitivity but only 27.3% specificity. Limitations include a retrospective single-center design and lack of external validation. Conclusions and clinical implications: AI-interpreted micro-US improves specificity while maintaining high sensitivity for csPCa detection. This method may reduce unnecessary biopsies and serve as a low-cost alternative to PSA-based screening. Patient summary: We developed an AI system to analyze prostate micro-ultrasound images. It outperformed PSA and DRE in detecting aggressive cancer and may help avoid unnecessary biopsies.
Authors: Qishuai Zhong, Zongmin Li, Siqi Fan, Aixin Sun
Abstract: Effective engagement by large language models (LLMs) requires adapting responses to users' sociodemographic characteristics, such as age, occupation, and education level. While many real-world applications leverage dialogue history for contextualization, existing evaluations of LLMs' behavioral adaptation often focus on single-turn prompts. In this paper, we propose a framework to evaluate LLM adaptation when attributes are introduced either (1) explicitly via user profiles in the prompt or (2) implicitly through multi-turn dialogue history. We assess the consistency of model behavior across these modalities. Using a multi-agent pipeline, we construct a synthetic dataset pairing dialogue histories with distinct user profiles and employ questions from the Value Survey Module (VSM 2013) (Hofstede and Hofstede, 2016) to probe value expression. Our findings indicate that most models adjust their expressed values in response to demographic changes, particularly in age and education level, but consistency varies. Models with stronger reasoning capabilities demonstrate greater alignment, indicating the importance of reasoning in robust sociodemographic adaptation.
Authors: Anissa Alloula, Charles Jones, Ben Glocker, Bart{\l}omiej W. Papie\.z
Abstract: Despite the constant development of new bias mitigation methods for machine learning, no method consistently succeeds, and a fundamental question remains unanswered: when and why do bias mitigation techniques fail? In this paper, we hypothesise that a key factor may be the often-overlooked but crucial step shared by many bias mitigation methods: the definition of subgroups. To investigate this, we conduct a comprehensive evaluation of state-of-the-art bias mitigation methods across multiple vision and language classification tasks, systematically varying subgroup definitions, including coarse, fine-grained, intersectional, and noisy subgroups. Our results reveal that subgroup choice significantly impacts performance, with certain groupings paradoxically leading to worse outcomes than no mitigation at all. Our findings suggest that observing a disparity between a set of subgroups is not a sufficient reason to use those subgroups for mitigation. Through theoretical analysis, we explain these phenomena and uncover a counter-intuitive insight that, in some cases, improving fairness with respect to a particular set of subgroups is best achieved by using a different set of subgroups for mitigation. Our work highlights the importance of careful subgroup definition in bias mitigation and suggest it as a alternative lever for improving the robustness and fairness of machine learning models.
Authors: James Oldfield, Shawn Im, Yixuan Li, Mihalis A. Nicolaou, Ioannis Patras, Grigorios G Chrysos
Abstract: Multilayer perceptrons (MLPs) are an integral part of large language models, yet their dense representations render them difficult to understand, edit, and steer. Recent methods learn interpretable approximations via neuron-level sparsity, yet fail to faithfully reconstruct the original mapping--significantly increasing model's next-token cross-entropy loss. In this paper, we advocate for moving to layer-level sparsity to overcome the accuracy trade-off in sparse layer approximation. Under this paradigm, we introduce Mixture of Decoders (MxDs). MxDs generalize MLPs and Gated Linear Units, expanding pre-trained dense layers into tens of thousands of specialized sublayers. Through a flexible form of tensor factorization, each sparsely activating MxD sublayer implements a linear transformation with full-rank weights--preserving the original decoders' expressive capacity even under heavy sparsity. Experimentally, we show that MxDs significantly outperform state-of-the-art methods (e.g., Transcoders) on the sparsity-accuracy frontier in language models with up to 3B parameters. Further evaluations on sparse probing and feature steering demonstrate that MxDs learn similarly specialized features of natural language--opening up a promising new avenue for designing interpretable yet faithful decompositions. Our code is included at: https://github.com/james-oldfield/MxD/.
Authors: Andrej Schwanke, Lyubomir Ivanov, David Salinas, Fabio Ferreira, Aaron Klein, Frank Hutter, Arber Zela
Abstract: Large Language Models (LLMs) have recently emerged as effective surrogate models and candidate generators within global optimization frameworks for expensive blackbox functions. Despite promising results, LLM-based methods often struggle in high-dimensional search spaces or when lacking domain-specific priors, leading to sparse or uninformative suggestions. To overcome these limitations, we propose HOLLM, a novel global optimization algorithm that enhances LLM-driven sampling by partitioning the search space into promising subregions. Each subregion acts as a ``meta-arm'' selected via a bandit-inspired scoring mechanism that effectively balances exploration and exploitation. Within each selected subregion, an LLM then proposes high-quality candidate points, without any explicit domain knowledge. Empirical evaluation on standard optimization benchmarks shows that HOLLM consistently matches or surpasses leading Bayesian optimization and trust-region methods, while substantially outperforming global LLM-based sampling strategies.
Authors: Jingyuan Huang, Xi Zhu, Minghao Guo, Yongfeng Zhang
Abstract: Web 2.0 social platforms are inherently centralized, with user data and algorithmic decisions controlled by the platform. However, users can only passively receive social predictions without being able to choose the underlying algorithm, which limits personalization. Fortunately, with the emergence of blockchain, users are allowed to choose algorithms that are tailored to their local situation, improving prediction results in a personalized way. In a blockchain environment, each user possesses its own model to perform the social prediction, capturing different perspectives on social interactions. In our work, we propose DeSocial, a decentralized social network learning framework deployed on an Ethereum (ETH) local development chain that integrates distributed data storage, node-level consensus, and user-driven model selection through Ganache. In the first stage, each user leverages DeSocial to evaluate multiple backbone models on their local subgraph. DeSocial coordinates the execution and returns model-wise prediction results, enabling the user to select the most suitable backbone for personalized social prediction. Then, DeSocial uniformly selects several validation nodes that possess the algorithm specified by each user, and aggregates the prediction results by majority voting, to prevent errors caused by any single model's misjudgment. Extensive experiments show that DeSocial has an evident improvement compared to the five classical centralized social network learning models, promoting user empowerment in blockchain-based decentralized social networks, showing the importance of multi-node validation and personalized algorithm selection based on blockchain. Our implementation is available at: https://github.com/agiresearch/DeSocial.
Authors: Zixuan Xie, Xinyu Liu, Rohan Chandra, Shangtong Zhang
Abstract: Linear TD($\lambda$) is one of the most fundamental reinforcement learning algorithms for policy evaluation. Previously, convergence rates are typically established under the assumption of linearly independent features, which does not hold in many practical scenarios. This paper instead establishes the first $L^2$ convergence rates for linear TD($\lambda$) operating under arbitrary features, without making any algorithmic modification or additional assumptions. Our results apply to both the discounted and average-reward settings. To address the potential non-uniqueness of solutions resulting from arbitrary features, we develop a novel stochastic approximation result featuring convergence rates to the solution set instead of a single point.
Authors: Maoli Liu, Zhuohua Li, Xiangxiang Dai, John C. S. Lui
Abstract: Conversational recommender systems proactively query users with relevant "key terms" and leverage the feedback to elicit users' preferences for personalized recommendations. Conversational contextual bandits, a prevalent approach in this domain, aim to optimize preference learning by balancing exploitation and exploration. However, several limitations hinder their effectiveness in real-world scenarios. First, existing algorithms employ key term selection strategies with insufficient exploration, often failing to thoroughly probe users' preferences and resulting in suboptimal preference estimation. Second, current algorithms typically rely on deterministic rules to initiate conversations, causing unnecessary interactions when preferences are well-understood and missed opportunities when preferences are uncertain. To address these limitations, we propose three novel algorithms: CLiSK, CLiME, and CLiSK-ME. CLiSK introduces smoothed key term contexts to enhance exploration in preference learning, CLiME adaptively initiates conversations based on preference uncertainty, and CLiSK-ME integrates both techniques. We theoretically prove that all three algorithms achieve a tighter regret upper bound of $O(\sqrt{dT\log{T}})$ with respect to the time horizon $T$, improving upon existing methods. Additionally, we provide a matching lower bound $\Omega(\sqrt{dT})$ for conversational bandits, demonstrating that our algorithms are nearly minimax optimal. Extensive evaluations on both synthetic and real-world datasets show that our approaches achieve at least a 14.6% improvement in cumulative regret.
Authors: Xiao Liu, Xinyi Dong, Xinyang Gao, Yansong Feng, Xun Pang
Abstract: Recent advancements in large language models (LLMs) have shown promise in generating novel research ideas. However, these ideas often face challenges related to feasibility and expected effectiveness. This paper explores how augmenting LLMs with relevant data during the idea generation process can enhance the quality of generated ideas. We introduce two ways of incorporating data: (1) providing metadata during the idea generation stage to guide LLMs toward feasible directions, and (2) adding automatic validation during the idea selection stage to assess the empirical plausibility of hypotheses within ideas. We conduct experiments in the social science domain, specifically with climate negotiation topics, and find that metadata improves the feasibility of generated ideas by 20%, while automatic validation improves the overall quality of selected ideas by 7%. A human study shows that LLM-generated ideas, along with their related data and validation processes, inspire researchers to propose research ideas with higher quality. Our work highlights the potential of data-driven research idea generation, and underscores the practical utility of LLM-assisted ideation in real-world academic settings.
Authors: Hovhannes Tamoyan, Subhabrata Dutta, Iryna Gurevych
Abstract: Factual incorrectness in generated content is one of the primary concerns in ubiquitous deployment of large language models (LLMs). Prior findings suggest LLMs can (sometimes) detect factual incorrectness in their generated content (i.e., fact-checking post-generation). In this work, we provide evidence supporting the presence of LLMs' internal compass that dictate the correctness of factual recall at the time of generation. We demonstrate that for a given subject entity and a relation, LLMs internally encode linear features in the Transformer's residual stream that dictate whether it will be able to recall the correct attribute (that forms a valid entity-relation-attribute triplet). This self-awareness signal is robust to minor formatting variations. We investigate the effects of context perturbation via different example selection strategies. Scaling experiments across model sizes and training dynamics highlight that self-awareness emerges rapidly during training and peaks in intermediate layers. These findings uncover intrinsic self-monitoring capabilities within LLMs, contributing to their interpretability and reliability.
Authors: Dario Satriani, Enzo Veltri, Donatello Santoro, Paolo Papotti
Abstract: Factuality in Large Language Models (LLMs) is a persistent challenge. Current benchmarks often assess short factual answers, overlooking the critical ability to generate structured, multi-record tabular outputs from parametric knowledge. We demonstrate that this relational fact retrieval is substantially more difficult than isolated point-wise queries, even when individual facts are known to the model, exposing distinct failure modes sensitive to output dimensionality (e.g., number of attributes or records). To systematically evaluate this under-explored capability, we introduce RelationalFactQA, a new benchmark featuring diverse natural language questions (paired with SQL) and gold-standard tabular answers, specifically designed to assess knowledge retrieval in a structured format. RelationalFactQA enables analysis across varying query complexities, output sizes, and data characteristics. Our experiments reveal that even state-of-the-art LLMs struggle significantly, not exceeding 25% factual accuracy in generating relational outputs, with performance notably degrading as output dimensionality increases. These findings underscore critical limitations in current LLMs' ability to synthesize structured factual knowledge and establish RelationalFactQA as a crucial resource for measuring future progress in LLM factuality.
Authors: Xiao Liu, Da Yin, Zirui Wu, Yansong Feng
Abstract: Tools enhance the reasoning capabilities of large language models (LLMs) in complex problem-solving tasks, but not all tasks have available tools. In the absence of predefined tools, prior works have explored instructing LLMs to generate tools on their own. However, such approaches rely heavily on the models' internal knowledge and would fail in domains beyond the LLMs' knowledge scope. To address this limitation, we propose RefTool, a reference-guided framework for automatic tool creation that leverages structured external materials such as textbooks. RefTool consists of two modules: (1) tool creation, where LLMs generate executable tools from reference content, validate them using illustrative examples, and organize them hierarchically into a toolbox; and (2) tool utilization, where LLMs navigate the toolbox structure to select and apply the appropriate tools to solve problems. Experiments on causality, physics, and chemistry benchmarks demonstrate that RefTool outperforms existing tool-creation and domain-specific reasoning methods by 11.3% on average accuracy, while being cost-efficient and broadly generalizable. Analyses reveal that grounding tool creation in references produces accurate and faithful tools, and that the hierarchical structure facilitates effective tool selection. RefTool enables LLMs to overcome knowledge limitations, demonstrating the value of grounding tool creation in external references for enhanced and generalizable reasoning.
Authors: Brett Bissey, Kyle Gatesman, Walker Dimon, Mohammad Alam, Luis Robaina, Joseph Weissman
Abstract: This paper introduces a comprehensive framework designed to analyze and secure decision-support systems trained with Deep Reinforcement Learning (DRL), prior to deployment, by providing insights into learned behavior patterns and vulnerabilities discovered through simulation. The introduced framework aids in the development of precisely timed and targeted observation perturbations, enabling researchers to assess adversarial attack outcomes within a strategic decision-making context. We validate our framework, visualize agent behavior, and evaluate adversarial outcomes within the context of a custom-built strategic game, CyberStrike. Utilizing the proposed framework, we introduce a method for systematically discovering and ranking the impact of attacks on various observation indices and time-steps, and we conduct experiments to evaluate the transferability of adversarial attacks across agent architectures and DRL training algorithms. The findings underscore the critical need for robust adversarial defense mechanisms to protect decision-making policies in high-stakes environments.
Authors: Jinbao Wang, Hanzhe Liang, Can Gao, Chenxi Hu, Jie Zhou, Yunkang Cao, Linlin Shen, Weiming Shen
Abstract: Multimodal feature reconstruction is a promising approach for 3D anomaly detection, leveraging the complementary information from dual modalities. We further advance this paradigm by utilizing multi-modal mentor learning, which fuses intermediate features to further distinguish normal from feature differences. To address these challenges, we propose a novel method called Mentor3AD, which utilizes multi-modal mentor learning. By leveraging the shared features of different modalities, Mentor3AD can extract more effective features and guide feature reconstruction, ultimately improving detection performance. Specifically, Mentor3AD includes a Mentor of Fusion Module (MFM) that merges features extracted from RGB and 3D modalities to create a mentor feature. Additionally, we have designed a Mentor of Guidance Module (MGM) to facilitate cross-modal reconstruction, supported by the mentor feature. Lastly, we introduce a Voting Module (VM) to more accurately generate the final anomaly score. Extensive comparative and ablation studies on MVTec 3D-AD and Eyecandies have verified the effectiveness of the proposed method.
Authors: Haoming Song, Delin Qu, Yuanqi Yao, Qizhi Chen, Qi Lv, Yiwen Tang, Modi Shi, Guanghui Ren, Maoqing Yao, Bin Zhao, Dong Wang, Xuelong Li
Abstract: Humans practice slow thinking before performing actual actions when handling complex tasks in the physical world. This thinking paradigm, recently, has achieved remarkable advancement in boosting Large Language Models (LLMs) to solve complex tasks in digital domains. However, the potential of slow thinking remains largely unexplored for robotic foundation models interacting with the physical world. In this work, we propose Hume: a dual-system Vision-Language-Action (VLA) model with value-guided System-2 thinking and cascaded action denoising, exploring human-like thinking capabilities of Vision-Language-Action models for dexterous robot control. System 2 of Hume implements value-Guided thinking by extending a Vision-Language-Action Model backbone with a novel value-query head to estimate the state-action value of predicted actions. The value-guided thinking is conducted by repeat sampling multiple action candidates and selecting one according to state-action value. System 1 of Hume is a lightweight reactive visuomotor policy that takes System 2 selected action and performs cascaded action denoising for dexterous robot control. At deployment time, System 2 performs value-guided thinking at a low frequency while System 1 asynchronously receives the System 2 selected action candidate and predicts fluid actions in real time. We show that Hume outperforms the existing state-of-the-art Vision-Language-Action models across multiple simulation benchmark and real-robot deployments.
Authors: Binh Duc Vu, Jan Kapar, Marvin Wright, David S. Watson
Abstract: We propose a principled method for autoencoding with random forests. Our strategy builds on foundational results from nonparametric statistics and spectral graph theory to learn a low-dimensional embedding of the model that optimally represents relationships in the data. We provide exact and approximate solutions to the decoding problem via constrained optimization, split relabeling, and nearest neighbors regression. These methods effectively invert the compression pipeline, establishing a map from the embedding space back to the input space using splits learned by the ensemble's constituent trees. The resulting decoders are universally consistent under common regularity assumptions. The procedure works with supervised or unsupervised models, providing a window into conditional or joint distributions. We demonstrate various applications of this autoencoder, including powerful new tools for visualization, compression, clustering, and denoising. Experiments illustrate the ease and utility of our method in a wide range of settings, including tabular, image, and genomic data.
Authors: Zhiqi Ai, Meixuan Bao, Zhiyong Chen, Zhi Yang, Xinnuo Li, Shugong Xu
Abstract: The performance of speaker verification systems is adversely affected by speaker aging. However, due to challenges in data collection, particularly the lack of sustained and large-scale longitudinal data for individuals, research on speaker aging remains difficult. In this paper, we present VoxAging, a large-scale longitudinal dataset collected from 293 speakers (226 English speakers and 67 Mandarin speakers) over several years, with the longest time span reaching 17 years (approximately 900 weeks). For each speaker, the data were recorded at weekly intervals. We studied the phenomenon of speaker aging and its effects on advanced speaker verification systems, analyzed individual speaker aging processes, and explored the impact of factors such as age group and gender on speaker aging research.
Authors: Muzhi Zhu, Hao Zhong, Canyu Zhao, Zongze Du, Zheng Huang, Mingyu Liu, Hao Chen, Cheng Zou, Jingdong Chen, Ming Yang, Chunhua Shen
Abstract: Active vision, also known as active perception, refers to the process of actively selecting where and how to look in order to gather task-relevant information. It is a critical component of efficient perception and decision-making in humans and advanced embodied agents. Recently, the use of Multimodal Large Language Models (MLLMs) as central planning and decision-making modules in robotic systems has gained extensive attention. However, despite the importance of active perception in embodied intelligence, there is little to no exploration of how MLLMs can be equipped with or learn active perception capabilities. In this paper, we first provide a systematic definition of MLLM-based active perception tasks. We point out that the recently proposed GPT-o3 model's zoom-in search strategy can be regarded as a special case of active perception; however, it still suffers from low search efficiency and inaccurate region selection. To address these issues, we propose ACTIVE-O3, a purely reinforcement learning based training framework built on top of GRPO, designed to equip MLLMs with active perception capabilities. We further establish a comprehensive benchmark suite to evaluate ACTIVE-O3 across both general open-world tasks, such as small-object and dense object grounding, and domain-specific scenarios, including small object detection in remote sensing and autonomous driving, as well as fine-grained interactive segmentation. In addition, ACTIVE-O3 also demonstrates strong zero-shot reasoning abilities on the V* Benchmark, without relying on any explicit reasoning data. We hope that our work can provide a simple codebase and evaluation protocol to facilitate future research on active perception in MLLMs.
Authors: Xiangru Jian, Wei Pang, Zhengyuan Dong, Chao Zhang, M. Tamer \"Ozsu
Abstract: Current video analytics approaches face a fundamental trade-off between flexibility and efficiency. End-to-end Vision Language Models (VLMs) often struggle with long-context processing and incur high computational costs, while neural-symbolic methods depend heavily on manual labeling and rigid rule design. In this paper, we introduce LazyVLM, a neuro-symbolic video analytics system that provides a user-friendly query interface similar to VLMs, while addressing their scalability limitation. LazyVLM enables users to effortlessly drop in video data and specify complex multi-frame video queries using a semi-structured text interface for video analytics. To address the scalability limitations of VLMs, LazyVLM decomposes multi-frame video queries into fine-grained operations and offloads the bulk of the processing to efficient relational query execution and vector similarity search. We demonstrate that LazyVLM provides a robust, efficient, and user-friendly solution for querying open-domain video data at scale.
Authors: Uri Gadot, Rinon Gal, Yftah Ziser, Gal Chechik, Shie Mannor
Abstract: Text-to-image generation has evolved beyond single monolithic models to complex multi-component pipelines. These combine fine-tuned generators, adapters, upscaling blocks and even editing steps, leading to significant improvements in image quality. However, their effective design requires substantial expertise. Recent approaches have shown promise in automating this process through large language models (LLMs), but they suffer from two critical limitations: extensive computational requirements from generating images with hundreds of predefined pipelines, and poor generalization beyond memorized training examples. We introduce a novel reinforcement learning-based framework that addresses these inefficiencies. Our approach first trains an ensemble of reward models capable of predicting image quality scores directly from prompt-workflow combinations, eliminating the need for costly image generation during training. We then implement a two-phase training strategy: initial workflow vocabulary training followed by GRPO-based optimization that guides the model toward higher-performing regions of the workflow space. Additionally, we incorporate a classifier-free guidance based enhancement technique that extrapolates along the path between the initial and GRPO-tuned models, further improving output quality. We validate our approach through a set of comparisons, showing that it can successfully create new flows with greater diversity and lead to superior image quality compared to existing baselines.
Authors: Omer Dahary, Yehonathan Cohen, Or Patashnik, Kfir Aberman, Daniel Cohen-Or
Abstract: Generating multiple distinct subjects remains a challenge for existing text-to-image diffusion models. Complex prompts often lead to subject leakage, causing inaccuracies in quantities, attributes, and visual features. Preventing leakage among subjects necessitates knowledge of each subject's spatial location. Recent methods provide these spatial locations via an external layout control. However, enforcing such a prescribed layout often conflicts with the innate layout dictated by the sampled initial noise, leading to misalignment with the model's prior. In this work, we introduce a new approach that predicts a spatial layout aligned with the prompt, derived from the initial noise, and refines it throughout the denoising process. By relying on this noise-induced layout, we avoid conflicts with externally imposed layouts and better preserve the model's prior. Our method employs a small neural network to predict and refine the evolving noise-induced layout at each denoising step, ensuring clear boundaries between subjects while maintaining consistency. Experimental results show that this noise-aligned strategy achieves improved text-image alignment and more stable multi-subject generation compared to existing layout-guided techniques, while preserving the rich diversity of the model's original distribution.
Authors: Wei Pang, Kevin Qinghong Lin, Xiangru Jian, Xi He, Philip Torr
Abstract: Academic poster generation is a crucial yet challenging task in scientific communication, requiring the compression of long-context interleaved documents into a single, visually coherent page. To address this challenge, we introduce the first benchmark and metric suite for poster generation, which pairs recent conference papers with author-designed posters and evaluates outputs on (i)Visual Quality-semantic alignment with human posters, (ii)Textual Coherence-language fluency, (iii)Holistic Assessment-six fine-grained aesthetic and informational criteria scored by a VLM-as-judge, and notably (iv)PaperQuiz-the poster's ability to convey core paper content as measured by VLMs answering generated quizzes. Building on this benchmark, we propose PosterAgent, a top-down, visual-in-the-loop multi-agent pipeline: the (a)Parser distills the paper into a structured asset library; the (b)Planner aligns text-visual pairs into a binary-tree layout that preserves reading order and spatial balance; and the (c)Painter-Commenter loop refines each panel by executing rendering code and using VLM feedback to eliminate overflow and ensure alignment. In our comprehensive evaluation, we find that GPT-4o outputs-though visually appealing at first glance-often exhibit noisy text and poor PaperQuiz scores, and we find that reader engagement is the primary aesthetic bottleneck, as human-designed posters rely largely on visual semantics to convey meaning. Our fully open-source variants (e.g. based on the Qwen-2.5 series) outperform existing 4o-driven multi-agent systems across nearly all metrics, while using 87% fewer tokens. It transforms a 22-page paper into a finalized yet editable .pptx poster - all for just $0.005. These findings chart clear directions for the next generation of fully automated poster-generation models. The code and datasets are available at https://github.com/Paper2Poster/Paper2Poster.
Authors: Haowei Wang, Junjie Wang, Xiaojun Jia, Rupeng Zhang, Mingyang Li, Zhe Liu, Yang Liu, Qing Wang
Abstract: Vision-Language Model (VLM) based Web Agents represent a significant step towards automating complex tasks by simulating human-like interaction with websites. However, their deployment in uncontrolled web environments introduces significant security vulnerabilities. Existing research on adversarial environmental injection attacks often relies on unrealistic assumptions, such as direct HTML manipulation, knowledge of user intent, or access to agent model parameters, limiting their practical applicability. In this paper, we propose AdInject, a novel and real-world black-box attack method that leverages the internet advertising delivery to inject malicious content into the Web Agent's environment. AdInject operates under a significantly more realistic threat model than prior work, assuming a black-box agent, static malicious content constraints, and no specific knowledge of user intent. AdInject includes strategies for designing malicious ad content aimed at misleading agents into clicking, and a VLM-based ad content optimization technique that infers potential user intents from the target website's context and integrates these intents into the ad content to make it appear more relevant or critical to the agent's task, thus enhancing attack effectiveness. Experimental evaluations demonstrate the effectiveness of AdInject, attack success rates exceeding 60% in most scenarios and approaching 100% in certain cases. This strongly demonstrates that prevalent advertising delivery constitutes a potent and real-world vector for environment injection attacks against Web Agents. This work highlights a critical vulnerability in Web Agent security arising from real-world environment manipulation channels, underscoring the urgent need for developing robust defense mechanisms against such threats. Our code is available at https://github.com/NicerWang/AdInject.
Authors: Dingming Li, Hongxing Li, Zixuan Wang, Yuchen Yan, Hang Zhang, Siqi Chen, Guiyang Hou, Shengpei Jiang, Wenqi Zhang, Yongliang Shen, Weiming Lu, Yueting Zhuang
Abstract: Vision-language models (VLMs) have demonstrated remarkable capabilities in understanding and reasoning about visual content, but significant challenges persist in tasks requiring cross-viewpoint understanding and spatial reasoning. We identify a critical limitation: current VLMs excel primarily at egocentric spatial reasoning (from the camera's perspective) but fail to generalize to allocentric viewpoints when required to adopt another entity's spatial frame of reference. We introduce ViewSpatial-Bench, the first comprehensive benchmark designed specifically for multi-viewpoint spatial localization recognition evaluation across five distinct task types, supported by an automated 3D annotation pipeline that generates precise directional labels. Comprehensive evaluation of diverse VLMs on ViewSpatial-Bench reveals a significant performance disparity: models demonstrate reasonable performance on camera-perspective tasks but exhibit reduced accuracy when reasoning from a human viewpoint. By fine-tuning VLMs on our multi-perspective spatial dataset, we achieve an overall performance improvement of 46.24% across tasks, highlighting the efficacy of our approach. Our work establishes a crucial benchmark for spatial intelligence in embodied AI systems and provides empirical evidence that modeling 3D spatial relationships enhances VLMs' corresponding spatial comprehension capabilities.
Authors: Yihan Wang, Qiao Yan, Zhenghao Xing, Lihao Liu, Junjun He, Chi-Wing Fu, Xiaowei Hu, Pheng-Ann Heng
Abstract: Large language models (LLMs) have demonstrated strong potential in clinical question answering, with recent multi-agent frameworks further improving diagnostic accuracy via collaborative reasoning. However, we identify a recurring issue of Silent Agreement, where agents prematurely converge on diagnoses without sufficient critical analysis, particularly in complex or ambiguous cases. We present a new concept called Catfish Agent, a role-specialized LLM designed to inject structured dissent and counter silent agreement. Inspired by the ``catfish effect'' in organizational psychology, the Catfish Agent is designed to challenge emerging consensus to stimulate deeper reasoning. We formulate two mechanisms to encourage effective and context-aware interventions: (i) a complexity-aware intervention that modulates agent engagement based on case difficulty, and (ii) a tone-calibrated intervention articulated to balance critique and collaboration. Evaluations on nine medical Q&A and three medical VQA benchmarks show that our approach consistently outperforms both single- and multi-agent LLMs frameworks, including leading commercial models such as GPT-4o and DeepSeek-R1.
Authors: Shimao Zhang, Zhejian Lai, Xiang Liu, Shuaijie She, Xiao Liu, Yeyun Gong, Shujian Huang, Jiajun Chen
Abstract: Multilingual Alignment is an effective and representative paradigm to enhance LLMs' multilingual capabilities, which transfers the capabilities from the high-resource languages to the low-resource languages. Meanwhile, some researches on language-specific neurons reveal that there are language-specific neurons that are selectively activated in LLMs when processing different languages. This provides a new perspective to analyze and understand LLMs' mechanisms more specifically in multilingual scenarios. In this work, we propose a new finer-grained neuron identification algorithm, which detects language neurons~(including language-specific neurons and language-related neurons) and language-agnostic neurons. Furthermore, based on the distributional characteristics of different types of neurons, we divide the LLMs' internal process for multilingual inference into four parts: (1) multilingual understanding, (2) shared semantic space reasoning, (3) multilingual output space transformation, and (4) vocabulary space outputting. Additionally, we systematically analyze the models before and after alignment with a focus on different types of neurons. We also analyze the phenomenon of ''Spontaneous Multilingual Alignment''. Overall, our work conducts a comprehensive investigation based on different types of neurons, providing empirical results and valuable insights for better understanding multilingual alignment and multilingual capabilities of LLMs.
Authors: Tom Silver, Rohan Chitnis, Nishanth Kumar, Willie McClinton, Tomas Lozano-Perez, Leslie Pack Kaelbling, Joshua Tenenbaum
Abstract: Efficient planning in continuous state and action spaces is fundamentally hard, even when the transition model is deterministic and known. One way to alleviate this challenge is to perform bilevel planning with abstractions, where a high-level search for abstract plans is used to guide planning in the original transition space. Previous work has shown that when state abstractions in the form of symbolic predicates are hand-designed, operators and samplers for bilevel planning can be learned from demonstrations. In this work, we propose an algorithm for learning predicates from demonstrations, eliminating the need for manually specified state abstractions. Our key idea is to learn predicates by optimizing a surrogate objective that is tractable but faithful to our real efficient-planning objective. We use this surrogate objective in a hill-climbing search over predicate sets drawn from a grammar. Experimentally, we show across four robotic planning environments that our learned abstractions are able to quickly solve held-out tasks, outperforming six baselines. Code: https://tinyurl.com/predicators-release
Authors: Songlin Xu, Xinyu Zhang
Abstract: Using deep neural networks as computational models to simulate cognitive process can provide key insights into human behavioral dynamics. Challenges arise when environments are highly dynamic, obscuring stimulus-behavior relationships. However, the majority of current research focuses on simulating human cognitive behaviors under ideal conditions, neglecting the influence of environmental disturbances. We propose CogReact, integrating drift-diffusion with deep reinforcement learning to simulate granular effects of dynamic environmental stimuli on human cognitive process. Quantitatively, it improves cognition modelling by considering temporal effect of environmental stimuli on cognitive process and captures both subject-specific and stimuli-specific behavioural differences. Qualitatively, it captures general trends in human cognitive process under stimuli, better than baselines. Our approach is examined in diverse environmental influences on various cognitive tasks. Overall, it demonstrates a powerful, data-driven methodology to simulate, align with, and understand the vagaries of human cognitive response in dynamic contexts.
Authors: Jiayuan Mao, Xuelin Yang, Xikun Zhang, Noah D. Goodman, Jiajun Wu
Abstract: Building machines that can reason about physical events and their causal relationships is crucial for flexible interaction with the physical world. However, most existing physical and causal reasoning benchmarks are exclusively based on synthetically generated events and synthetic natural language descriptions of causal relationships. This design brings up two issues. First, there is a lack of diversity in both event types and natural language descriptions; second, causal relationships based on manually-defined heuristics are different from human judgments. To address both shortcomings, we present the CLEVRER-Humans benchmark, a video reasoning dataset for causal judgment of physical events with human labels. We employ two techniques to improve data collection efficiency: first, a novel iterative event cloze task to elicit a new representation of events in videos, which we term Causal Event Graphs (CEGs); second, a data augmentation technique based on neural language generative models. We convert the collected CEGs into questions and answers to be consistent with prior work. Finally, we study a collection of baseline approaches for CLEVRER-Humans question-answering, highlighting the great challenges set forth by our benchmark.
Authors: Benhao Huang, Yingzhuo Yu, Jin Huang, Xingjian Zhang, Jiaqi Ma
Abstract: The quality of datasets plays an increasingly crucial role in the research and development of modern artificial intelligence (AI). Despite the proliferation of open dataset platforms nowadays, data quality issues, such as incomplete documentation, inaccurate labels, ethical concerns, and outdated information, remain common in widely used datasets. Furthermore, these issues are often subtle and difficult to be detected by rule-based scripts, therefore requiring identification and verification by dataset users or maintainers--a process that is both time-consuming and prone to human mistakes. With the surging ability of large language models (LLM), it's promising to streamline the discovery of hidden dataset issues with LLM agents. To achieve this, one significant challenge is enabling LLM agents to detect issues in the wild rather than simply fixing known ones. In this work, we establish a benchmark to measure LLM agent's ability to tackle this challenge. We carefully curate 221 real-world test cases from eight popular dataset platforms and propose an automatic evaluation framework using GPT-4o. Our proposed framework shows strong empirical alignment with expert evaluations, validated through extensive comparisons with human annotations. Without any hints, most competitive Curator agent can only reveal $\sim$30\% of the data quality issues in the proposed dataset, highlighting the complexity of this task and indicating that applying LLM agents to real-world dataset curation still requires further in-depth exploration and innovation. The data and code are available at \href{https://github.com/TRAIS-Lab/dca-bench}{https://github.com/TRAIS-Lab/dca-bench}.
URLs: https://github.com/TRAIS-Lab/dca-bench, https://github.com/TRAIS-Lab/dca-bench
Authors: Ricardo Knauer, Mario Koddenbrock, Raphael Wallsberger, Nicholas M. Brisson, Georg N. Duda, Deborah Falla, David W. Evans, Erik Rodner
Abstract: Large language models (LLMs) provide powerful means to leverage prior knowledge for predictive modeling when data is limited. In this work, we demonstrate how LLMs can use their compressed world knowledge to generate intrinsically interpretable machine learning models, i.e., decision trees, without any training data. We find that these zero-shot decision trees can even surpass data-driven trees on some small-sized tabular datasets and that embeddings derived from these trees perform better than data-driven tree-based embeddings on average. Our decision tree induction and embedding approaches can therefore serve as new knowledge-driven baselines for data-driven machine learning methods in the low-data regime. Furthermore, they offer ways to harness the rich world knowledge within LLMs for tabular machine learning tasks. Our code and results are available at https://github.com/ml-lab-htw/llm-trees.
Authors: Haoyang Su, Renqi Chen, Shixiang Tang, Zhenfei Yin, Xinzhe Zheng, Jinzhe Li, Biqing Qi, Qi Wu, Hui Li, Wanli Ouyang, Philip Torr, Bowen Zhou, Nanqing Dong
Abstract: The rapid advancement of scientific progress requires innovative tools that can accelerate knowledge discovery. Although recent AI methods, particularly large language models (LLMs), have shown promise in tasks such as hypothesis generation and experimental design, they fall short of replicating the collaborative nature of real-world scientific practices, where diverse experts work together in teams to tackle complex problems. To address the limitations, we propose an LLM-based multi-agent system, i.e., Virtual Scientists (VirSci), designed to mimic the teamwork inherent in scientific research. VirSci organizes a team of agents to collaboratively generate, evaluate, and refine research ideas. Through comprehensive experiments, we demonstrate that this multi-agent approach outperforms the state-of-the-art method in producing novel scientific ideas. We further investigate the collaboration mechanisms that contribute to its tendency to produce ideas with higher novelty, offering valuable insights to guide future research and illuminating pathways toward building a robust system for autonomous scientific discovery. The code is available at https://github.com/open-sciencelab/Virtual-Scientists.
URLs: https://github.com/open-sciencelab/Virtual-Scientists.
Authors: Haochuan Wang, Xiachong Feng, Lei Li, Yu Guo, Zhanyue Qin, Dianbo Sui, Lingpeng Kong
Abstract: The rapid advancement of large language models has accelerated their application in reasoning, with strategic reasoning drawing increasing attention. To evaluate the strategic reasoning capabilities of LLMs, game theory, with its concise structure, has become the preferred approach for many researchers. However, current research typically focuses on a limited selection of games, resulting in low coverage of game types. Additionally, classic game scenarios carry risks of data leakage, and the benchmarks used often lack extensibility, rendering them inadequate for evaluating state-of-the-art models. To address these challenges, we propose TMGBench, characterized by comprehensive game type coverage, diverse scenarios and flexible game organization. Specifically, we incorporate all 144 game types summarized by the Robinson-Goforth topology of 2x2 games, constructed as classic games in our benchmark; we also synthetize diverse, higher-quality game scenarios for each classic game, which we refer to as story-based games. Lastly, to provide a sustainable evaluation framework adaptable to increasingly powerful LLMs, we treat the aforementioned games as atomic units and organize them into more complex forms through sequential, parallel, and nested structures. We conducted a comprehensive evaluation of mainstream LLMs, covering tests on rational reasoning, reasoning robustness, Theory-of-Mind capabilities, and reasoning in complex game forms. The results revealed LLMs still have flaws in the accuracy and consistency of strategic reasoning processes, and their levels of mastery over Theory-of-Mind also vary. Additionally, SOTA models like o3-mini, Qwen3 and deepseek-reasoner, were also evaluated across the sequential, parallel, and nested game structures while the results highlighted the challenges posed by TMGBench.
Authors: Chao Huang, Jiahui Chen, Chen Chen, Chunyan Chen, Renjie Su, Shiyu Du, ChenChen, Hongrui Liang, Daojing Lin
Abstract: The design of crystal materials plays a critical role in areas such as new energy development, biomedical engineering, and semiconductors. Recent advances in data-driven methods have enabled the generation of diverse crystal structures. However, most existing approaches still rely on random sampling without strict constraints, requiring multiple post-processing steps to identify stable candidates with the desired physical and chemical properties. In this work, we present a new constrained generation framework that takes multiple constraints as input and enables the generation of crystal structures with specific chemical and properties. In this framework, intermediate constraints, such as symmetry information and composition ratio, are generated by a constraint generator based on large language models (LLMs), which considers the target properties. These constraints are then used by a subsequent crystal structure generator to ensure that the structure generation process is under control. Our method generates crystal structures with a probability of meeting the target properties that is more than twice that of existing approaches. Furthermore, nearly 100% of the generated crystals strictly adhere to predefined chemical composition, eliminating the risks of supply chain during production.
Authors: Filip Ilievski, Barbara Hammer, Frank van Harmelen, Benjamin Paassen, Sascha Saralajew, Ute Schmid, Michael Biehl, Marianna Bolognesi, Xin Luna Dong, Kiril Gashteovski, Pascal Hitzler, Giuseppe Marra, Pasquale Minervini, Martin Mundt, Axel-Cyrille Ngonga Ngomo, Alessandro Oltramari, Gabriella Pasi, Zeynep G. Saribatur, Luciano Serafini, John Shawe-Taylor, Vered Shwartz, Gabriella Skitalinskaya, Clemens Stachl, Gido M. van de Ven, Thomas Villmann
Abstract: Recent advances in AI -- including generative approaches -- have resulted in technology that can support humans in scientific discovery and forming decisions, but may also disrupt democracies and target individuals. The responsible use of AI and its participation in human-AI teams increasingly shows the need for AI alignment, that is, to make AI systems act according to our preferences. A crucial yet often overlooked aspect of these interactions is the different ways in which humans and machines generalise. In cognitive science, human generalisation commonly involves abstraction and concept learning. In contrast, AI generalisation encompasses out-of-domain generalisation in machine learning, rule-based reasoning in symbolic AI, and abstraction in neurosymbolic AI. In this perspective paper, we combine insights from AI and cognitive science to identify key commonalities and differences across three dimensions: notions of, methods for, and evaluation of generalisation. We map the different conceptualisations of generalisation in AI and cognitive science along these three dimensions and consider their role for alignment in human-AI teaming. This results in interdisciplinary challenges across AI and cognitive science that must be tackled to provide a foundation for effective and cognitively supported alignment in human-AI teaming scenarios.
Authors: Byungjun Kim, Minju Kim, Dayeon Seo, Bugeun Kim
Abstract: We highlight two significant issues leading to the passivity of current merchant non-player characters (NPCs): pricing and communication. While immersive interactions with active NPCs have been a focus, price negotiations between merchant NPCs and players remain underexplored. First, passive pricing refers to the limited ability of merchants to modify predefined item prices. Second, passive communication means that merchants can only interact with players in a scripted manner. To tackle these issues and create an active merchant NPC, we propose a merchant framework based on large language models (LLMs), called MART, which consists of an appraiser module and a negotiator module. We conducted two experiments to explore various implementation options under different training methods and LLM sizes, considering a range of possible game environments. Our findings indicate that finetuning methods, such as supervised finetuning (SFT) and knowledge distillation (KD), are effective in using smaller LLMs to implement active merchant NPCs. Additionally, we found three irregular cases arising from the responses of LLMs.
Authors: Shane Storks, Itamar Bar-Yossef, Yayuan Li, Zheyuan Zhang, Jason J. Corso, Joyce Chai
Abstract: Procedural mistake detection (PMD) is a challenging problem of classifying whether a human user (observed through egocentric video) has successfully executed a task (specified by a procedural text). Despite significant recent efforts, machine performance in the wild remains nonviable, and the reasoning processes underlying this performance are opaque. As such, we extend PMD to require generating visual self-dialog rationales to inform decisions. Given the impressive, mature image understanding capabilities observed in recent vision-and-language models (VLMs), we curate a suitable benchmark dataset for PMD based on individual frames. As our reformulation enables unprecedented transparency, we leverage a natural language inference (NLI) model to formulate two automated metrics for the coherence of generated rationales. We establish baselines for this reframed task, showing that while VLMs struggle off-the-shelf, their accuracy, coherence, and efficiency can be improved by incorporating these metrics into common inference and fine-tuning methods- though not without tradeoff. Lastly, our multi-faceted metrics visualize common outcomes, highlighting areas for further improvement.
Authors: Haohong Lin, Xin Huang, Tung Phan-Minh, David S. Hayden, Huan Zhang, Ding Zhao, Siddhartha Srinivasa, Eric M. Wolff, Hongge Chen
Abstract: Simulation is critical for safety evaluation in autonomous driving, particularly in capturing complex interactive behaviors. However, generating realistic and controllable traffic scenarios in long-tail situations remains a significant challenge. Existing generative models suffer from the conflicting objective between user-defined controllability and realism constraints, which is amplified in safety-critical contexts. In this work, we introduce the Causal Compositional Diffusion Model (CCDiff), a structure-guided diffusion framework to address these challenges. We first formulate the learning of controllable and realistic closed-loop simulation as a constrained optimization problem. Then, CCDiff maximizes controllability while adhering to realism by automatically identifying and injecting causal structures directly into the diffusion process, providing structured guidance to enhance both realism and controllability. Through rigorous evaluations on benchmark datasets and in a closed-loop simulator, CCDiff demonstrates substantial gains over state-of-the-art approaches in generating realistic and user-preferred trajectories. Our results show CCDiff's effectiveness in extracting and leveraging causal structures, showing improved closed-loop performance based on key metrics such as collision rate, off-road rate, FDE, and comfort.
Authors: Xuanle Zhao, Xianzhen Luo, Qi Shi, Chi Chen, Shuo Wang, Zhiyuan Liu, Maosong Sun
Abstract: Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in chart understanding tasks. However, interpreting charts with textual descriptions often leads to information loss, as it fails to fully capture the dense information embedded in charts. In contrast, parsing charts into code provides lossless representations that can effectively contain all critical details. Although existing open-source MLLMs have achieved success in chart understanding tasks, they still face two major challenges when applied to chart-to-code tasks: (1) Low executability and poor restoration of chart details in the generated code and (2) Lack of large-scale and diverse training data. To address these challenges, we propose \textbf{ChartCoder}, the first dedicated chart-to-code MLLM, which leverages Code LLMs as the language backbone to enhance the executability of the generated code. Furthermore, we introduce \textbf{Chart2Code-160k}, the first large-scale and diverse dataset for chart-to-code generation, and propose the \textbf{Snippet-of-Thought (SoT)} method, which transforms direct chart-to-code generation data into step-by-step generation. Experiments demonstrate that ChartCoder, with only 7B parameters, surpasses existing open-source MLLMs on chart-to-code benchmarks, achieving superior chart restoration and code excitability. Our code is available at https://github.com/thunlp/ChartCoder.
Authors: Yuyang Wu, Yifei Wang, Ziyu Ye, Tianqi Du, Stefanie Jegelka, Yisen Wang
Abstract: Large Language Models (LLMs) employ Chain-of-Thought (CoT) reasoning to deconstruct complex problems. While longer CoTs are often presumed superior, this paper challenges that notion, arguing that longer is not always better. Drawing on combined evidence from real-world observations, controlled experiments, and theoretical analysis, we demonstrate that task accuracy typically follows an inverted U-shaped curve with CoT length, where performance initially improves but eventually decreases as the number of CoT steps increases. With controlled experiments, we further uncover the scaling behaviors of the optimal CoT length: it increases with task difficulty but decreases with model capability, exposing an inherent simplicity bias where more capable models favor shorter, more efficient CoT reasoning. This bias is also evident in Reinforcement Learning (RL) training, where models gravitate towards shorter CoTs as their accuracy improves. To have a deep understanding of these dynamics, we establish a simple theoretical model that formally proves these phenomena, including the optimal length's scaling laws and the emergence of simplicity bias during RL. Guided by this framework, we demonstrate significant practical benefits from training with optimally-lengthed CoTs and employing length-aware filtering at inference. These findings offer both a principled understanding of the "overthinking" phenomenon and multiple practical guidelines for CoT calibration, enabling LLMs to achieve optimal reasoning performance with adaptive CoTs tailored to task complexity and model capability.
Authors: Hairu Wang, Yuan Feng, Xike Xie, S Kevin Zhou
Abstract: Although Large Language Models achieve strong success in many tasks, they still suffer from hallucinations and knowledge deficiencies in real-world applications. Many knowledge graph-based retrieval-augmented generation (KG-RAG) methods enhance the quality and credibility of LLMs by leveraging structure and semantic information in KGs as external knowledge bases. However, these methods struggle to effectively incorporate structure information, either incurring high computational costs or underutilizing available knowledge. Inspired by smoothing operations in graph representation learning, we propose path pooling, a simple, training-free strategy that introduces structure information through a novel path-centric pooling operation. It seamlessly integrates into existing KG-RAG methods in a plug-and-play manner, enabling richer structure information utilization. Extensive experiments demonstrate that incorporating the path pooling into the state-of-the-art KG-RAG method consistently improves performance across various settings while introducing negligible additional cost.
Authors: Ziyu Wan, Yunxiang Li, Xiaoyu Wen, Yan Song, Hanjing Wang, Linyi Yang, Mark Schmidt, Jun Wang, Weinan Zhang, Shuyue Hu, Ying Wen
Abstract: Recent research on Reasoning of Large Language Models (LLMs) has sought to further enhance their performance by integrating meta-thinking -- enabling models to monitor, evaluate, and control their reasoning processes for more adaptive and effective problem-solving. However, current single-agent work lacks a specialized design for acquiring meta-thinking, resulting in low efficacy. To address this challenge, we introduce Reinforced Meta-thinking Agents (ReMA), a novel framework that leverages Multi-Agent Reinforcement Learning (MARL) to elicit meta-thinking behaviors, encouraging LLMs to think about thinking. ReMA decouples the reasoning process into two hierarchical agents: a high-level meta-thinking agent responsible for generating strategic oversight and plans, and a low-level reasoning agent for detailed executions. Through iterative reinforcement learning with aligned objectives, these agents explore and learn collaboration, leading to improved generalization and robustness. Empirical results from single-turn experiments demonstrate that ReMA outperforms single-agent RL baselines on complex reasoning tasks, including competitive-level mathematical benchmarks and LLM-as-a-Judge benchmarks. Additionally, we further extend ReMA to multi-turn interaction settings, leveraging turn-level ratio and parameter sharing to improve efficiency. Comprehensive ablation studies further illustrate the evolving dynamics of each distinct agent, providing valuable insights into how the meta-thinking reasoning process enhances the reasoning capabilities of LLMs. Our code can be found in https://github.com/ziyuwan/ReMA-public
Authors: Xueyang Zhou, Guiyao Tie, Guowen Zhang, Weidong Wang, Zhigang Zuo, Di Wu, Duanfeng Chu, Pan Zhou, Neil Zhenqiang Gong, Lichao Sun
Abstract: The rise of Large Reasoning Models (LRMs) signifies a paradigm shift toward advanced computational reasoning. Yet, this progress disrupts traditional agent frameworks, traditionally anchored by execution-oriented Large Language Models (LLMs). To explore this transformation, we propose the LaRMA framework, encompassing nine tasks across Tool Usage, Plan Design, and Problem Solving, assessed with three top LLMs (e.g., Claude3.5-sonnet) and five leading LRMs (e.g., DeepSeek-R1). Our findings address four research questions: LRMs surpass LLMs in reasoning-intensive tasks like Plan Design, leveraging iterative reflection for superior outcomes; LLMs excel in execution-driven tasks such as Tool Usage, prioritizing efficiency; hybrid LLM-LRM configurations, pairing LLMs as actors with LRMs as reflectors, optimize agent performance by blending execution speed with reasoning depth; and LRMs' enhanced reasoning incurs higher computational costs, prolonged processing, and behavioral challenges, including overthinking and fact-ignoring tendencies. This study fosters deeper inquiry into LRMs' balance of deep thinking and overthinking, laying a critical foundation for future agent design advancements.
Authors: Yifei Dong, Fengyi Wu, Qi He, Heng Li, Minghan Li, Zebang Cheng, Yuxuan Zhou, Jingdong Sun, Qi Dai, Zhi-Qi Cheng, Alexander G Hauptmann
Abstract: Vision-and-Language Navigation (VLN) systems often focus on either discrete (panoramic) or continuous (free-motion) paradigms alone, overlooking the complexities of human-populated, dynamic environments. We introduce a unified Human-Aware VLN (HA-VLN) benchmark that merges these paradigms under explicit social-awareness constraints. Our contributions include: 1. A standardized task definition that balances discrete-continuous navigation with personal-space requirements; 2. An enhanced human motion dataset (HAPS 2.0) and upgraded simulators capturing realistic multi-human interactions, outdoor contexts, and refined motion-language alignment; 3. Extensive benchmarking on 16,844 human-centric instructions, revealing how multi-human dynamics and partial observability pose substantial challenges for leading VLN agents; 4. Real-world robot tests validating sim-to-real transfer in crowded indoor spaces; and 5. A public leaderboard supporting transparent comparisons across discrete and continuous tasks. Empirical results show improved navigation success and fewer collisions when social context is integrated, underscoring the need for human-centric design. By releasing all datasets, simulators, agent code, and evaluation tools, we aim to advance safer, more capable, and socially responsible VLN research.
Authors: Seungwon Lim, Sungwoong Kim, Jihwan Yu, Sungjae Lee, Jiwan Chung, Youngjae Yu
Abstract: Escape rooms present a unique cognitive challenge that demands exploration-driven planning: with the sole instruction to 'escape the room', players must actively search their environment, collecting information, and finding solutions through repeated trial and error. Motivated by this, we introduce VisEscape, a benchmark of 20 virtual escape rooms specifically designed to evaluate AI models under these challenging conditions, where success depends not only on solving isolated puzzles but also on iteratively constructing and refining spatial-temporal knowledge of a dynamically changing environment. On VisEscape, we observe that even state-of-the-art multi-modal models generally fail to escape the rooms, showing considerable variation in their progress and problem-solving approaches. We find that integrating memory management and reasoning contributes to efficient exploration and enables successive hypothesis formulation and testing, thereby leading to significant improvements in dynamic and exploration-driven environments
Authors: Subbarao Kambhampati, Kaya Stechly, Karthik Valmeekam, Lucas Saldyt, Siddhant Bhambri, Vardhan Palod, Atharva Gundawar, Soumya Rani Samineni, Durgesh Kalwar, Upasana Biswas
Abstract: Intermediate token generation (ITG), where a model produces output before the solution, has been proposed as a method to improve the performance of language models on reasoning tasks. These intermediate tokens have been called "reasoning traces" or even "thoughts" -- implicitly anthropomorphizing the model, implying these tokens resemble steps a human might take when solving a challenging problem.In this paper, we present evidence that this anthropomorphization isn't a harmless metaphor, and instead is quite dangerous -- it confuses the nature of these models and how to use them effectively, and leads to questionable research.
Authors: Pengrui Quan, Brian Wang, Kang Yang, Liying Han, Mani Srivastava
Abstract: Spatiotemporal reasoning plays a key role in Cyber-Physical Systems (CPS). Despite advances in Large Language Models (LLMs) and Large Reasoning Models (LRMs), their capacity to reason about complex spatiotemporal signals remains underexplored. This paper proposes a hierarchical SpatioTemporal reAsoning benchmaRK, STARK, to systematically evaluate LLMs across three levels of reasoning complexity: state estimation (e.g., predicting field variables, localizing and tracking events in space and time), spatiotemporal reasoning over states (e.g., inferring spatial-temporal relationships), and world-knowledge-aware reasoning that integrates contextual and domain knowledge (e.g., intent prediction, landmark-aware navigation). We curate 26 distinct spatiotemporal tasks with diverse sensor modalities, comprising 14,552 challenges where models answer directly or by Python Code Interpreter. Evaluating 3 LRMs and 8 LLMs, we find LLMs achieve limited success in tasks requiring geometric reasoning (e.g., multilateration or triangulation), particularly as complexity increases. Surprisingly, LRMs show robust performance across tasks with various levels of difficulty, often competing or surpassing traditional first-principle-based methods. Our results show that in reasoning tasks requiring world knowledge, the performance gap between LLMs and LRMs narrows, with some LLMs even surpassing LRMs. However, the LRM o3 model continues to achieve leading performance across all evaluated tasks, a result attributed primarily to the larger size of the reasoning models. STARK motivates future innovations in model architectures and reasoning paradigms for intelligent CPS by providing a structured framework to identify limitations in the spatiotemporal reasoning of LLMs and LRMs.
Authors: Shibo Hong, Jiahao Ying, Haiyuan Liang, Mengdi Zhang, Jun Kuang, Jiazheng Zhang, Yixin Cao
Abstract: Evaluating the open-ended outputs of large language models (LLMs) has become a bottleneck as model capabilities, task diversity, and modality coverage rapidly expand. Existing "LLM-as-a-Judge" evaluators are typically narrow in a few tasks, aspects, or modalities, and easily suffer from low consistency. In this paper, we argue that explicit, fine-grained aspect specification is the key to both generalizability and objectivity in automated evaluation. To this end, we propose a hierarchical aspect taxonomy encompassing 112 distinct aspects that unifies evaluation across four representative settings -- Natural Language Generation, Image Understanding, Image Generation, and Interleaved Text-and-Image Generation. Building upon this taxonomy, we create FRABench, a benchmark comprising 60.4k pairwise samples with 325k evaluation labels obtained from a combination of human and LLM annotations. FRABench provides the first large-scale, multi-modal resource for training and meta-evaluating fine-grained LMM judges. Leveraging FRABench, we develop GenEval, a fine-grained evaluator generalizable across tasks and modalities. Experiments show that GenEval (i) attains high agreement with GPT-4o and expert annotators, (ii) transfers robustly to unseen tasks and modalities, and (iii) reveals systematic weaknesses of current LMMs on evaluation.
Authors: Yubin Li, Xingyu Liu, Guozhang Chen
Abstract: The brain's intricate connectome, a blueprint for its function, presents immense complexity, yet it arises from a compact genetic code, hinting at underlying low-dimensional organizational principles. This work bridges connectomics and representation learning to uncover these principles. We propose a framework that combines subgraph extraction from the Drosophila connectome, FlyWire, with a generative model to derive interpretable low-dimensional representations of neural circuitry. Crucially, an explainability module links these latent dimensions to specific structural features, offering insights into their functional relevance. We validate our approach by demonstrating effective graph reconstruction and, significantly, the ability to manipulate these latent codes to controllably generate connectome subgraphs with predefined properties. This research offers a novel tool for understanding brain architecture and a potential avenue for designing bio-inspired artificial neural networks.
Authors: Zongyuan Deng, Yujie Cai, Qing Liu, Shiyao Mu, Bin Lyu, Zhen Yang
Abstract: The selection of base station sites is a critical challenge in 5G network planning, which requires efficient optimization of coverage, cost, user satisfaction, and practical constraints. Traditional manual methods, reliant on human expertise, suffer from inefficiencies and are limited to an unsatisfied planning-construction consistency. Existing AI tools, despite improving efficiency in certain aspects, still struggle to meet the dynamic network conditions and multi-objective needs of telecom operators' networks. To address these challenges, we propose TelePlanNet, an AI-driven framework tailored for the selection of base station sites, integrating a three-layer architecture for efficient planning and large-scale automation. By leveraging large language models (LLMs) for real-time user input processing and intent alignment with base station planning, combined with training the planning model using the improved group relative policy optimization (GRPO) reinforcement learning, the proposed TelePlanNet can effectively address multi-objective optimization, evaluates candidate sites, and delivers practical solutions. Experiments results show that the proposed TelePlanNet can improve the consistency to 78%, which is superior to the manual methods, providing telecom operators with an efficient and scalable tool that significantly advances cellular network planning.
Authors: Wonje Jeung, Sangyeon Yoon, Minsuk Kahng, Albert No
Abstract: Large Reasoning Models (LRMs) have become powerful tools for complex problem solving, but their structured reasoning pathways can lead to unsafe outputs when exposed to harmful prompts. Existing safety alignment methods reduce harmful outputs but can degrade reasoning depth, leading to significant trade-offs in complex, multi-step tasks, and remain vulnerable to sophisticated jailbreak attacks. To address this, we introduce SAFEPATH, a lightweight alignment method that fine-tunes LRMs to emit a short, 8-token Safety Primer at the start of their reasoning, in response to harmful prompts, while leaving the rest of the reasoning process unsupervised. Empirical results across multiple benchmarks indicate that SAFEPATH effectively reduces harmful outputs while maintaining reasoning performance. Specifically, SAFEPATH reduces harmful responses by up to 90.0% and blocks 83.3% of jailbreak attempts in the DeepSeek-R1-Distill-Llama-8B model, while requiring 295.9x less compute than Direct Refusal and 314.1x less than SafeChain. We further introduce a zero-shot variant that requires no fine-tuning. In addition, we provide a comprehensive analysis of how existing methods in LLMs generalize, or fail, when applied to reasoning-centric models, revealing critical gaps and new directions for safer AI.
Authors: Mengru Wang, Xingyu Chen, Yue Wang, Zhiwei He, Jiahao Xu, Tian Liang, Qiuzhi Liu, Yunzhi Yao, Wenxuan Wang, Ruotian Ma, Haitao Mi, Ningyu Zhang, Zhaopeng Tu, Xiaolong Li, Dong Yu
Abstract: Mixture-of-Experts (MoE) architectures within Large Reasoning Models (LRMs) have achieved impressive reasoning capabilities by selectively activating experts to facilitate structured cognitive processes. Despite notable advances, existing reasoning models often suffer from cognitive inefficiencies like overthinking and underthinking. To address these limitations, we introduce a novel inference-time steering methodology called Reinforcing Cognitive Experts (RICE), designed to improve reasoning performance without additional training or complex heuristics. Leveraging normalized Pointwise Mutual Information (nPMI), we systematically identify specialized experts, termed ''cognitive experts'' that orchestrate meta-level reasoning operations characterized by tokens like ''
Authors: Philipp D. Siedler
Abstract: We introduce a novel dataset designed to benchmark the physical and spatial reasoning capabilities of Large Language Models (LLM) based on topology optimization, a method for computing optimal material distributions within a design space under prescribed loads and supports. In this dataset, LLMs are provided with conditions such as 2D boundary, applied forces and supports, and must reason about the resulting optimal material distribution. The dataset includes a variety of tasks, ranging from filling in masked regions within partial structures to predicting complete material distributions. Solving these tasks requires understanding the flow of forces and the required material distribution under given constraints, without access to simulation tools or explicit physical models, challenging models to reason about structural stability and spatial organization. Our dataset targets the evaluation of spatial and physical reasoning abilities in 2D settings, offering a complementary perspective to traditional language and logic benchmarks.
Authors: Renfei Dang, Shujian Huang, Jiajun Chen
Abstract: While current reasoning models possess strong exploratory capabilities, they are often criticized for overthinking due to redundant and unnecessary reflections. In this work, we reveal for the first time that overthinking in reasoning models may stem from their internal bias towards input texts. Upon encountering a reasoning problem, the model immediately forms a preliminary guess about the answer, which we term as an internal bias since it is not derived through actual reasoning. When this guess conflicts with its reasoning result, the model tends to engage in reflection, leading to the waste of computational resources. Through further interpretability experiments, we find that this behavior is largely driven by the model's excessive attention to the input section, which amplifies the influence of internal bias on its decision-making process. Additionally, by masking out the original input section, the affect of internal bias can be effectively alleviated and the reasoning length could be reduced by 31%-53% across different complex reasoning tasks. Notably, in most cases, this approach also leads to improvements in accuracy. These findings demonstrate a causal relationship between internal bias and overthinking.
Authors: Guiyao Tie, Xueyang Zhou, Tianhe Gu, Ruihang Zhang, Chaoran Hu, Sizhe Zhang, Mengqu Sun, Yan Zhang, Pan Zhou, Lichao Sun
Abstract: Recent advances in Multi-Modal Large Language Models (MLLMs) have enabled unified processing of language, vision, and structured inputs, opening the door to complex tasks such as logical deduction, spatial reasoning, and scientific analysis. Despite their promise, the reasoning capabilities of MLLMs, particularly those augmented with intermediate thinking traces (MLLMs-T), remain poorly understood and lack standardized evaluation benchmarks. Existing work focuses primarily on perception or final answer correctness, offering limited insight into how models reason or fail across modalities. To address this gap, we introduce the MMMR, a new benchmark designed to rigorously evaluate multi-modal reasoning with explicit thinking. The MMMR comprises 1) a high-difficulty dataset of 1,083 questions spanning six diverse reasoning types with symbolic depth and multi-hop demands and 2) a modular Reasoning Trace Evaluation Pipeline (RTEP) for assessing reasoning quality beyond accuracy through metrics like relevance, consistency, and structured error annotations. Empirical results show that MLLMs-T overall outperform non-thinking counterparts, but even top models like Claude-3.7-Sonnet and Gemini-2.5 Pro suffer from reasoning pathologies such as inconsistency and overthinking. This benchmark reveals persistent gaps between accuracy and reasoning quality and provides an actionable evaluation pipeline for future model development. Overall, the MMMR offers a scalable foundation for evaluating, comparing, and improving the next generation of multi-modal reasoning systems.
Authors: Ashish Sundar, Chunbo Luo, Xiaoyang Wang
Abstract: Model-based reinforcement learning (MBRL) offers an intuitive way to increase the sample efficiency of model-free RL methods by simultaneously training a world model that learns to predict the future. MBRL methods have progressed by largely prioritising the actor; optimising the world model learning has been neglected meanwhile. Improving the fidelity of the world model and reducing its time to convergence can yield significant downstream benefits, one of which is improving the ensuing performance of any actor it may train. We propose a novel approach that anticipates and actively seeks out high-entropy states using short-horizon latent predictions generated by the world model, offering a principled alternative to traditional curiosity-driven methods that chase once-novel states well after they were stumbled into. While many model predictive control (MPC) based methods offer similar alternatives, they typically lack commitment, synthesising multi step plans after every step. To mitigate this, we present a hierarchical planner that dynamically decides when to replan, planning horizon length, and the weighting between reward and entropy. While our method can theoretically be applied to any model that trains its own actors with solely model generated data, we have applied it to just Dreamer as a proof of concept. Our method finishes the Miniworld procedurally generated mazes 50% faster than base Dreamer at convergence and the policy trained in imagination converges in only 60% of the environment steps that base Dreamer needs.
Authors: Wentao Sun, Jo\~ao Paulo Nogueira, Alonso Silva
Abstract: Despite remarkable advances in the field, LLMs remain unreliable in distinguishing causation from correlation. Recent results from the Corr2Cause dataset benchmark reveal that state-of-the-art LLMs -- such as GPT-4 (F1 score: 29.08) -- only marginally outperform random baselines (Random Uniform, F1 score: 20.38), indicating limited capacity of generalization. To tackle this limitation, we propose a novel structured approach: rather than directly answering causal queries, we provide the model with the capability to structure its thinking by guiding the model to build a structured knowledge graph, systematically encoding the provided correlational premises, to answer the causal queries. This intermediate representation significantly enhances the model's causal capabilities. Experiments on the test subset of the Corr2Cause dataset benchmark with Qwen3-32B model (reasoning model) show substantial gains over standard direct prompting methods, improving F1 scores from 32.71 to 48.26 (over 47.5% relative increase), along with notable improvements in precision and recall. These results underscore the effectiveness of providing the model with the capability to structure its thinking and highlight its promising potential for broader generalization across diverse causal inference tasks.
Authors: Yedi Zhang, Sun Yi Emma, Annabelle Lee Jia En, Jin Song Dong
Abstract: Large language models (LLMs) have emerged as a dominant AI paradigm due to their exceptional text understanding and generation capabilities. However, their tendency to generate inconsistent or erroneous outputs challenges their reliability, especially in high-stakes domains requiring accuracy and trustworthiness. Existing research primarily focuses on detecting and mitigating model misbehavior in general-purpose scenarios, often overlooking the potential of integrating domain-specific knowledge. In this work, we advance misbehavior detection by incorporating domain knowledge. The core idea is to design a general specification language that enables domain experts to customize domain-specific predicates in a lightweight and intuitive manner, supporting later runtime verification of LLM outputs. To achieve this, we design a novel specification language, ESL, and introduce a runtime verification framework, RvLLM, to validate LLM output against domain-specific constraints defined in ESL. We evaluate RvLLM on three representative tasks: violation detection against Singapore Rapid Transit Systems Act, numerical comparison, and inequality solving. Experimental results demonstrate that RvLLM effectively detects erroneous outputs across various LLMs in a lightweight and flexible manner. The results reveal that despite their impressive capabilities, LLMs remain prone to low-level errors due to limited interpretability and a lack of formal guarantees during inference, and our framework offers a potential long-term solution by leveraging expert domain knowledge to rigorously and efficiently verify LLM outputs.
Authors: Peijie Yu, Yifan Yang, Jinjian Li, Zelong Zhang, Haorui Wang, Xiao Feng, Feng Zhang
Abstract: Agents based on large language models leverage tools to modify environments, revolutionizing how AI interacts with the physical world. Unlike traditional NLP tasks that rely solely on historical dialogue for responses, these agents must consider more complex factors, such as inter-tool relationships, environmental feedback and previous decisions, when making choices. Current research typically evaluates agents via multi-turn dialogues. However, it overlooks the influence of these critical factors on agent behavior. To bridge this gap, we present an open-source and high-quality benchmark $C^3$-Bench. This benchmark integrates attack concepts and applies univariate analysis to pinpoint key elements affecting agent robustness. In concrete, we design three challenges: navigate complex tool relationships, handle critical hidden information and manage dynamic decision paths. Complementing these challenges, we introduce fine-grained metrics, innovative data collection algorithms and reproducible evaluation methods. Extensive experiments are conducted on 49 mainstream agents, encompassing general fast-thinking, slow-thinking and domain-specific models. We observe that agents have significant shortcomings in handling tool dependencies, long context information dependencies and frequent policy-type switching. In essence, $C^3$-Bench aims to expose model vulnerabilities through these challenges and drive research into the interpretability of agent performance. The benchmark is publicly available at https://github.com/yupeijei1997/C3-Bench.
Authors: Wenhao Liu, Zhengkang Guo, Mingchen Xie, Jingwen Xu, Zisu Huang, Muzhao Tian, Jianhan Xu, Muling Wu, Xiaohua Wang, Changze Lv, He-Da Wang, Hu Yao, Xiaoqing Zheng, Xuanjing Huang
Abstract: Large language models (LLMs) are increasingly expected to tackle complex tasks, driven by their expanding applications and users' growing proficiency in crafting sophisticated prompts. However, as the number of explicitly stated requirements increases (particularly more than 10 constraints), LLMs often struggle to accurately follow such complex instructions. To address this challenge, we propose RECAST, a novel framework for synthesizing datasets where each example incorporates far more constraints than those in existing benchmarks. These constraints are extracted from real-world prompt-response pairs to ensure practical relevance. RECAST enables automatic verification of constraint satisfaction via rule-based validators for quantitative constraints and LLM-based validators for qualitative ones. Using this framework, we construct RECAST-30K, a large-scale, high-quality dataset comprising 30k instances spanning 15 constraint types. Experimental results demonstrate that models fine-tuned on RECAST-30K show substantial improvements in following complex instructions. Moreover, the verifiability provided by RECAST enables the design of reward functions for reinforcement learning, which further boosts model performance on complex and challenging tasks.
Authors: Jaemin Kim, Hangeol Chang, Hyunmin Hwang, Choonghan Kim, Jong Chul Ye
Abstract: Large Language Models (LLMs) have demonstrated remarkable general capabilities, but enhancing skills such as reasoning often demands substantial computational resources and may compromise their generalization. While Parameter-Efficient Fine-Tuning (PEFT) methods offer a more resource-conscious alternative, they typically requires retraining for each LLM backbone due to architectural dependencies. To address these challenges, here we propose Universal Reasoner (UniR) - a single, lightweight, composable, and plug-and-play reasoning module that can be used with any frozen LLM to endow it with specialized reasoning capabilities. Specifically, UniR decomposes the reward into a standalone reasoning module that is trained independently using predefined rewards, effectively translating trajectory-level signals into token-level guidance. Once trained, UniR can be combined with any frozen LLM at inference time by simply adding its output logits to those of the LLM backbone. This additive structure naturally enables modular composition: multiple UniR modules trained for different tasks can be jointly applied by summing their logits, enabling complex reasoning via composition. Experimental results on mathematical reasoning and machine translation tasks show that UniR significantly outperforms existing baseline fine-tuning methods using the Llama3.2 model. Furthermore, UniR demonstrates strong weak-to-strong generalization: reasoning modules trained on smaller models effectively guide much larger LLMs. This makes UniR a cost-efficient, adaptable, and robust solution for enhancing reasoning in LLMs without compromising their core capabilities. Code is open-sourced at https://github.com/hangeol/UniR
Authors: Chanyeol Choi, Alejandro Lopez-Lira, Jihoon Kwon, Minjae Kim, Juneha Hwang, Minsoo Ha, Chaewoon Kim, Jaeseon Ha, Suyeol Yun, Jin Kim
Abstract: Extracting structured and quantitative insights from unstructured financial filings is essential in investment research, yet remains time-consuming and resource-intensive. Conventional approaches in practice rely heavily on labor-intensive manual processes, limiting scalability and delaying the research workflow. In this paper, we propose an efficient and scalable method for accurately extracting quantitative insights from unstructured financial documents, leveraging a multi-agent system composed of large language models. Our proposed multi-agent system consists of two specialized agents: the \emph{Extraction Agent} and the \emph{Text-to-SQL Agent}. The \textit{Extraction Agent} automatically identifies key performance indicators from unstructured financial text, standardizes their formats, and verifies their accuracy. On the other hand, the \textit{Text-to-SQL Agent} generates executable SQL statements from natural language queries, allowing users to access structured data accurately without requiring familiarity with the database schema. Through experiments, we demonstrate that our proposed system effectively transforms unstructured text into structured data accurately and enables precise retrieval of key information. First, we demonstrate that our system achieves approximately 95\% accuracy in transforming financial filings into structured data, matching the performance level typically attained by human annotators. Second, in a human evaluation of the retrieval task -- where natural language queries are used to search information from structured data -- 91\% of the responses were rated as correct by human evaluators. In both evaluations, our system generalizes well across financial document types, consistently delivering reliable performance.
Authors: Anqing Jiang, Yu Gao, Zhigang Sun, Yiru Wang, Jijun Wang, Jinghao Chai, Qian Cao, Yuweng Heng, Hao Jiang, Zongzheng Zhang, Xianda Guo, Hao Sun, Hao Zhao
Abstract: Research interest in end-to-end autonomous driving has surged owing to its fully differentiable design integrating modular tasks, i.e. perception, prediction and planing, which enables optimization in pursuit of the ultimate goal. Despite the great potential of the end-to-end paradigm, existing methods suffer from several aspects including expensive BEV (bird's eye view) computation, action diversity, and sub-optimal decision in complex real-world scenarios. To address these challenges, we propose a novel hybrid sparse-dense diffusion policy, empowered by a Vision-Language Model (VLM), called Diff-VLA. We explore the sparse diffusion representation for efficient multi-modal driving behavior. Moreover, we rethink the effectiveness of VLM driving decision and improve the trajectory generation guidance through deep interaction across agent, map instances and VLM output. Our method shows superior performance in Autonomous Grand Challenge 2025 which contains challenging real and reactive synthetic scenarios. Our methods achieves 45.0 PDMS.
Authors: Junteng Liu, Yuanxiang Fan, Zhuo Jiang, Han Ding, Yongyi Hu, Chi Zhang, Yiqi Shi, Shitong Weng, Aili Chen, Shiqi Chen, Yunan Huang, Mozhi Zhang, Pengyu Zhao, Junjie Yan, Junxian He
Abstract: Recent advances such as OpenAI-o1 and DeepSeek R1 have demonstrated the potential of Reinforcement Learning (RL) to enhance reasoning abilities in Large Language Models (LLMs). While open-source replication efforts have primarily focused on mathematical and coding domains, methods and resources for developing general reasoning capabilities remain underexplored. This gap is partly due to the challenge of collecting diverse and verifiable reasoning data suitable for RL. We hypothesize that logical reasoning is critical for developing general reasoning capabilities, as logic forms a fundamental building block of reasoning. In this work, we present SynLogic, a data synthesis framework and dataset that generates diverse logical reasoning data at scale, encompassing 35 diverse logical reasoning tasks. The SynLogic approach enables controlled synthesis of data with adjustable difficulty and quantity. Importantly, all examples can be verified by simple rules, making them ideally suited for RL with verifiable rewards. In our experiments, we validate the effectiveness of RL training on the SynLogic dataset based on 7B and 32B models. SynLogic leads to state-of-the-art logical reasoning performance among open-source datasets, surpassing DeepSeek-R1-Distill-Qwen-32B by 6 points on BBEH. Furthermore, mixing SynLogic data with mathematical and coding tasks improves the training efficiency of these domains and significantly enhances reasoning generalization. Notably, our mixed training model outperforms DeepSeek-R1-Zero-Qwen-32B across multiple benchmarks. These findings position SynLogic as a valuable resource for advancing the broader reasoning capabilities of LLMs. We open-source both the data synthesis pipeline and the SynLogic dataset at https://github.com/MiniMax-AI/SynLogic.
Authors: Ziming Wei, Bingqian Lin, Zijian Jiao, Yunshuang Nie, Liang Ma, Yuecheng Liu, Yuzheng Zhuang, Xiaodan Liang
Abstract: Spatial Planning is a crucial part in the field of spatial intelligence, which requires the understanding and planning about object arrangements in space perspective. AI agents with the spatial planning ability can better adapt to various real-world applications, including robotic manipulation, automatic assembly, urban planning etc. Recent works have attempted to construct benchmarks for evaluating the spatial intelligence of Multimodal Large Language Models (MLLMs). Nevertheless, these benchmarks primarily focus on spatial reasoning based on typical Visual Question-Answering (VQA) forms, which suffers from the gap between abstract spatial understanding and concrete task execution. In this work, we take a step further to build a comprehensive benchmark called MineAnyBuild, aiming to evaluate the spatial planning ability of open-world AI agents in the Minecraft game. Specifically, MineAnyBuild requires an agent to generate executable architecture building plans based on the given multi-modal human instructions. It involves 4,000 curated spatial planning tasks and also provides a paradigm for infinitely expandable data collection by utilizing rich player-generated content. MineAnyBuild evaluates spatial planning through four core supporting dimensions: spatial understanding, spatial reasoning, creativity, and spatial commonsense. Based on MineAnyBuild, we perform a comprehensive evaluation for existing MLLM-based agents, revealing the severe limitations but enormous potential in their spatial planning abilities. We believe our MineAnyBuild will open new avenues for the evaluation of spatial intelligence and help promote further development for open-world AI agents capable of spatial planning.
Authors: Can Xu, Qingfeng Sun, Kai Zheng, Xiubo Geng, Pu Zhao, Jiazhan Feng, Chongyang Tao, Qingwei Lin, Daxin Jiang
Abstract: Training large language models (LLMs) with open-domain instruction following data brings colossal success. However, manually creating such instruction data is very time-consuming and labor-intensive. Moreover, humans may struggle to produce high-complexity instructions. In this paper, we show an avenue for creating large amounts of instruction data with varying levels of complexity using LLM instead of humans. Starting with an initial set of instructions, we use our proposed Evol-Instruct to rewrite them step by step into more complex instructions. Then, we mix all generated instruction data to fine-tune LLaMA. We call the resulting model WizardLM. Human evaluations on a complexity-balanced test bed and Vicuna's testset show that instructions from Evol-Instruct are superior to human-created ones. By analyzing the human evaluation results of the high complexity part, we demonstrate that outputs from our WizardLM are preferred to outputs from OpenAI ChatGPT. In GPT-4 automatic evaluation, WizardLM achieves more than 90\% capacity of ChatGPT on 17 out of 29 skills. Even though WizardLM still lags behind ChatGPT in some aspects, our findings suggest that fine-tuning with AI-evolved instructions is a promising direction for enhancing LLMs. Our code and data are public at https://github.com/nlpxucan/WizardLM
Authors: Ziyang Luo, Can Xu, Pu Zhao, Qingfeng Sun, Xiubo Geng, Wenxiang Hu, Chongyang Tao, Jing Ma, Qingwei Lin, Daxin Jiang
Abstract: Code Large Language Models (Code LLMs), such as StarCoder, have demonstrated exceptional performance in code-related tasks. However, most existing models are solely pre-trained on extensive raw code data without instruction fine-tuning. In this paper, we introduce WizardCoder, which empowers Code LLMs with complex instruction fine-tuning, by adapting the Evol-Instruct method to the domain of code. Through comprehensive experiments on four prominent code generation benchmarks, namely HumanEval, HumanEval+, MBPP, and DS-1000, we unveil the exceptional capabilities of our model. It surpasses all other open-source Code LLMs by a substantial margin. Moreover, our model even outperforms the largest closed LLMs, Anthropic's Claude and Google's Bard, on HumanEval and HumanEval+. Our code, model weights, and data are public at https://github.com/nlpxucan/WizardLM
Authors: Hana Chockler, David A. Kelly, Daniel Kroening
Abstract: Existing explanation tools for image classifiers usually give only a single explanation for an image's classification. For many images, however, image classifiers accept more than one explanation for the image label. These explanations are useful for analyzing the decision process of the classifier and for detecting errors. Thus, restricting the number of explanations to just one severely limits insight into the behavior of the classifier. In this paper, we describe an algorithm and a tool, MultEX, for computing multiple explanations as the output of a black-box image classifier for a given image. Our algorithm uses a principled approach based on actual causality. We analyze its theoretical complexity and evaluate MultEX against the state-of-the-art across three different models and three different datasets. We find that MultEX finds more explanations and that these explanations are of higher quality.
Authors: Bhargav Ganguly, Yang Xu, Vaneet Aggarwal
Abstract: This paper investigates the potential of quantum acceleration in addressing infinite horizon Markov Decision Processes (MDPs) to enhance average reward outcomes. We introduce an innovative quantum framework for the agent's engagement with an unknown MDP, extending the conventional interaction paradigm. Our approach involves the design of an optimism-driven tabular Reinforcement Learning algorithm that harnesses quantum signals acquired by the agent through efficient quantum mean estimation techniques. Through thorough theoretical analysis, we demonstrate that the quantum advantage in mean estimation leads to exponential advancements in regret guarantees for infinite horizon Reinforcement Learning. Specifically, the proposed Quantum algorithm achieves a regret bound of $\tilde{\mathcal{O}}(1)$, a significant improvement over the $\tilde{\mathcal{O}}(\sqrt{T})$ bound exhibited by classical counterparts.
Authors: Yotam Wolf, Noam Wies, Dorin Shteyman, Binyamin Rothberg, Yoav Levine, Amnon Shashua
Abstract: Language model alignment has become an important component of AI safety, allowing safe interactions between humans and language models, by enhancing desired behaviors and inhibiting undesired ones. It is often done by tuning the model or inserting preset aligning prompts. Recently, representation engineering, a method which alters the model's behavior via changing its representations post-training, was shown to be effective in aligning LLMs (Zou et al., 2023a). Representation engineering yields gains in alignment oriented tasks such as resistance to adversarial attacks and reduction of social biases, but was also shown to cause a decrease in the ability of the model to perform basic tasks. In this paper we study the tradeoff between the increase in alignment and decrease in helpfulness of the model. We propose a theoretical framework which provides bounds for these two quantities, and demonstrate their relevance empirically. First, we find that under the conditions of our framework, alignment can be guaranteed with representation engineering, and at the same time that helpfulness is harmed in the process. Second, we show that helpfulness is harmed quadratically with the norm of the representation engineering vector, while the alignment increases linearly with it, indicating a regime in which it is efficient to use representation engineering. We validate our findings empirically, and chart the boundaries to the usefulness of representation engineering for alignment.
Authors: Xuanxin Wu, Yuki Arase
Abstract: Recent studies have used both automatic metrics and human evaluations to assess the simplification abilities of LLMs. However, the suitability of existing evaluation methodologies for LLMs remains in question. First, the suitability of current automatic metrics on LLMs' simplification evaluation is still uncertain. Second, current human evaluation approaches in sentence simplification often fall into two extremes: they are either too superficial, failing to offer a clear understanding of the models' performance, or overly detailed, making the annotation process complex and prone to inconsistency, which in turn affects the evaluation's reliability. To address these problems, this study provides in-depth insights into LLMs' performance while ensuring the reliability of the evaluation. We design an error-based human annotation framework to assess the LLMs' simplification capabilities. We select both closed-source and open-source LLMs, including GPT-4, Qwen2.5-72B, and Llama-3.2-3B. We believe that these models offer a representative selection across large, medium, and small sizes of LLMs. Results show that GPT-4 generally generates fewer erroneous simplification outputs compared to the current state-of-the-art. However, LLMs have their limitations, as seen in GPT-4's struggles with lexical paraphrasing. Results show that LLMs generally generate fewer erroneous simplification outputs compared to the previous state-of-the-art. However, LLMs have their limitations, as seen in GPT-4's and Qwen2.5-72B's struggle with lexical paraphrasing. Furthermore, we conduct meta-evaluations on widely used automatic metrics using our human annotations. We find that these metrics lack sufficient sensitivity to assess the overall high-quality simplifications, particularly those generated by high-performance LLMs.
Authors: James Chua, Edward Rees, Hunar Batra, Samuel R. Bowman, Julian Michael, Ethan Perez, Miles Turpin
Abstract: Chain-of-thought prompting (CoT) has the potential to improve the explainability of language model reasoning. But CoT can also systematically misrepresent the factors influencing models' behavior -- for example, rationalizing answers in line with a user's opinion. We first create a new dataset of 9 different biases that affect GPT-3.5-Turbo and Llama-8b models. These consist of spurious-few-shot patterns, post hoc rationalization, and sycophantic settings. Models switch to the answer implied by the bias, without mentioning the effect of the bias in the CoT. To mitigate this biased reasoning problem, we introduce bias-augmented consistency training (BCT), an unsupervised fine-tuning scheme that trains models to give consistent reasoning across prompts with and without biasing features. We construct a suite testing nine forms of biased reasoning on seven question-answering tasks, and find that applying BCT to GPT-3.5-Turbo with one bias reduces the rate of biased reasoning by 86\% on held-out tasks. Moreover, this model generalizes to other forms of bias, reducing biased reasoning on held-out biases by an average of 37\%. As BCT generalizes to held-out biases and does not require gold labels, this method may hold promise for reducing biased reasoning from as-of-yet unknown biases and on tasks where ground truth reasoning is unavailable.
Authors: Yongquan He, Wenyuan Zhang, Xuancheng Huang, Peng Zhang, Lingxun Meng, Xiang Zhou, Ke Zeng, Xunliang Cai
Abstract: Instruction tuning for large language models (LLMs) can drive them to produce results consistent with human goals in specific downstream tasks. However, the process of continual instruction tuning (CIT) for LLMs may bring about the catastrophic forgetting (CF) problem, where previously learned abilities are degraded. Recent methods try to alleviate the CF problem by modifying models or replaying data, which may only remember the surface-level pattern of instructions and get confused on held-out tasks. In this paper, we propose a novel continual instruction tuning method based on Key-part Information Gain (KPIG). Our method computes the information gain on masked parts to dynamically replay data and refine the training objective, which enables LLMs to capture task-aware information relevant to the correct response and alleviate overfitting to general descriptions in instructions. In addition, we propose two metrics, P-score and V-score, to measure the generalization and instruction-following abilities of LLMs. Experiments demonstrate our method achieves superior performance on both seen and held-out tasks.
Authors: Marina Ionova, Jan Kristof Behrens
Abstract: Assembly processes involving humans and robots are challenging scenarios because the individual activities and access to shared workspace have to be coordinated. Fixed robot programs leave no room to diverge from a fixed protocol. Working on such a process can be stressful for the user and lead to ineffective behavior or failure. We propose a novel approach of online constraint-based scheduling in a reactive execution control framework facilitating behavior trees called CoBOS. This allows the robot to adapt to uncertain events such as delayed activity completions and activity selection (by the human). The user will experience less stress as the robotic coworkers adapt their behavior to best complement the human-selected activities to complete the common task. In addition to the improved working conditions, our algorithm leads to increased efficiency, even in highly uncertain scenarios. We evaluate our algorithm using a probabilistic simulation study with 56000 experiments. We outperform all other compared methods by a margin of 4-10%. Initial real robot experiments using a Franka Emika Panda robot and human tracking based on HTC Vive VR gloves look promising.
Authors: Rongyu Zhang, Yijiang Liu, Huanrui Yang, Shenli Zheng, Dan Wang, Yuan Du, Li Du, Shanghang Zhang
Abstract: Large language models (LLMs) encounter significant adaptation challenges in diverse multitask finetuning. Mixture-of-experts (MoE) provides a promising solution with a dynamic architecture, enabling effective task decoupling. However, scaling up the number of MoE experts incurs substantial parameter and computational overheads and suffers from limited performance gain due to naive routing mechanisms. In this paper, we design a novel framework, mix\underline{\textbf{T}}ure\underline{\textbf{-}}of-\underline{\textbf{R}}ank-on\underline{\textbf{E}}-e\underline{\textbf{X}}perts (\texttt{T-REX}), which leverages the combination of ultra-low rank experts to construct LoRA weights on pretrained LLMs. The rank-1 experts enable a mix-and-match mechanism to quadratically expand the vector subspace of experts with linear parameter overheads, achieving approximate error reduction with optimal efficiency. In addition, T-REX offers implicit guidance to the router, leveraging the inherent semantic clustering of training embeddings as prior knowledge, enabling optimized feature allocation across experts for a smoother convergence. Extensive theoretical and empirical results demonstrate that T-REX achieves superior efficiency and generalizability across diverse tasks. Compared with other LoRA-based methods, T-REX achieves up to 1.78\% mean accuracy improvement with around 30\%-40\% less trainable parameters across 14 public datasets. \href{https://github.com/RoyZry98/T-REX-Pytorch}{Code} is available.
Authors: Justin Cui, Wei-Lin Chiang, Ion Stoica, Cho-Jui Hsieh
Abstract: Large Language Models (LLMs) require careful safety alignment to prevent malicious outputs. While significant research focuses on mitigating harmful content generation, the enhanced safety often come with the side effect of over-refusal, where LLMs may reject innocuous prompts and become less helpful. Although the issue of over-refusal has been empirically observed, a systematic measurement is challenging due to the difficulty of crafting prompts that can elicit the over-refusal behaviors of LLMs. This study proposes a novel method for automatically generating large-scale over-refusal datasets. Leveraging this technique, we introduce OR-Bench, the first large-scale over-refusal benchmark. OR-Bench comprises 80,000 over-refusal prompts across 10 common rejection categories, a subset of around 1,000 hard prompts that are challenging even for state-of-the-art LLMs, and an additional 600 toxic prompts to prevent indiscriminate responses. We then conduct a comprehensive study to measure the over-refusal of 32 popular LLMs across 8 model families. Our datasets are publicly available at https://huggingface.co/bench-llms and our codebase is open-sourced at https://github.com/justincui03/or-bench. We hope this benchmark can help the community develop better safety aligned models.
URLs: https://huggingface.co/bench-llms, https://github.com/justincui03/or-bench.
Authors: Han Jiang, Xiaoyuan Yi, Zhihua Wei, Ziang Xiao, Shu Wang, Xing Xie
Abstract: Warning: Contains harmful model outputs. Despite significant advancements, the propensity of Large Language Models (LLMs) to generate harmful and unethical content poses critical challenges. Measuring value alignment of LLMs becomes crucial for their regulation and responsible deployment. Although numerous benchmarks have been constructed to assess social bias, toxicity, and ethical issues in LLMs, those static benchmarks suffer from evaluation chronoeffect, in which, as models rapidly evolve, existing benchmarks may leak into training data or become saturated, overestimating ever-developing LLMs. To tackle this problem, we propose GETA, a novel generative evolving testing approach based on adaptive testing methods in measurement theory. Unlike traditional adaptive testing methods that rely on a static test item pool, GETA probes the underlying moral boundaries of LLMs by dynamically generating test items tailored to model capability. GETA co-evolves with LLMs by learning a joint distribution of item difficulty and model value conformity, thus effectively addressing evaluation chronoeffect. We evaluated various popular LLMs with GETA and demonstrated that 1) GETA can dynamically create difficulty-tailored test items and 2) GETA's evaluation results are more consistent with models' performance on unseen OOD and i.i.d. items, laying the groundwork for future evaluation paradigms.
Authors: Khai-Nguyen Nguyen, Khai Le-Duc, Bach Phan Tat, Duy Le, Long Vo-Dang, Truong-Son Hy
Abstract: Transparency in AI healthcare decision-making is crucial. By incorporating rationales to explain reason for each predicted label, users could understand Large Language Models (LLMs)'s reasoning to make better decision. In this work, we introduce a new task - Sentiment Reasoning - for both speech and text modalities, and our proposed multimodal multitask framework and the world's largest multimodal sentiment analysis dataset. Sentiment Reasoning is an auxiliary task in sentiment analysis where the model predicts both the sentiment label and generates the rationale behind it based on the input transcript. Our study conducted on both human transcripts and Automatic Speech Recognition (ASR) transcripts shows that Sentiment Reasoning helps improve model transparency by providing rationale for model prediction with quality semantically comparable to humans while also improving model's classification performance (+2% increase in both accuracy and macro-F1) via rationale-augmented fine-tuning. Also, no significant difference in the semantic quality of generated rationales between human and ASR transcripts. All code, data (five languages - Vietnamese, English, Chinese, German, and French) and models are published online: https://github.com/leduckhai/Sentiment-Reasoning
Authors: Zhiyuan Li, Heng Wang, Dongnan Liu, Chaoyi Zhang, Ao Ma, Jieting Long, Weidong Cai
Abstract: Multimodal Large Language Models (MLLMs) have showcased exceptional Chain-of-Thought (CoT) reasoning ability in complex textual inference tasks including causal reasoning. However, will these causalities remain straightforward when crucial hints hide in visual details? If not, what factors might influence cross-modal generalization? Whether we can effectively enhance their capacity for robust causal inference across both text and vision? Motivated by these, we introduce MuCR - a novel Multimodal Causal Reasoning benchmark that leverages synthetic siamese images and text pairs to challenge MLLMs. Additionally, we develop tailored metrics from multiple perspectives, including image-level match, phrase-level understanding, and sentence-level explanation, to comprehensively assess MLLMs' comprehension abilities. Our experiments reveal that current MLLMs fall short in multimodal causal reasoning compared to their performance in purely textual settings. Additionally, we find that identifying visual cues across images is key to effective cross-modal generalization. Finally, we propose a VcCoT strategy that better highlights visual cues, and our results confirm its efficacy in enhancing multimodal causal reasoning. The project is available at: https://github.com/Zhiyuan-Li-John/MuCR
Authors: Zeju Qiu, Weiyang Liu, Haiwen Feng, Zhen Liu, Tim Z. Xiao, Katherine M. Collins, Joshua B. Tenenbaum, Adrian Weller, Michael J. Black, Bernhard Sch\"olkopf
Abstract: Against the backdrop of enthusiasm for large language models (LLMs), there is a growing need to scientifically assess their capabilities and shortcomings. This is nontrivial in part because it is difficult to find tasks which the models have not encountered during training. Utilizing symbolic graphics programs, we propose a domain well-suited to test multiple spatial-semantic reasoning skills of LLMs. Popular in computer graphics, these programs procedurally generate visual data. While LLMs exhibit impressive skills in general program synthesis and analysis, symbolic graphics programs offer a new layer of evaluation: they allow us to test an LLM's ability to answer semantic questions about the images or 3D geometries without a vision encoder. To semantically understand the symbolic programs, LLMs would need to possess the ability to "imagine" and reason how the corresponding graphics content would look with only the symbolic description of the local curvatures and strokes. We use this task to evaluate LLMs by creating a large benchmark for the semantic visual understanding of symbolic graphics programs, built procedurally with minimal human effort. Particular emphasis is placed on transformations of images that leave the image level semantics invariant while introducing significant changes to the underlying program. We evaluate commercial and open-source LLMs on our benchmark to assess their ability to reason about visual output of programs, finding that LLMs considered stronger at reasoning generally perform better. Lastly, we introduce a novel method to improve this ability -- Symbolic Instruction Tuning (SIT), in which the LLM is finetuned with pre-collected instruction data on symbolic graphics programs. Interestingly, we find that SIT not only improves LLM's understanding on symbolic programs, but it also improves general reasoning ability on various other benchmarks.
Authors: Surya Murthy, Mustafa O. Karabag, Ufuk Topcu
Abstract: Autonomous agents interact with other autonomous agents and humans of unknown preferences to share resources in their environment. We explore sequential trading for resource allocation in a setting where two greedily rational agents sequentially trade resources from a finite set of categories. Each agent has a utility function that depends on the amount of resources it possesses in each category. The offering agent makes trade offers to improve its utility without knowing the responding agent's utility function, and the responding agent only accepts offers that improve its utility. To facilitate cooperation between an autonomous agent and another autonomous agent or a human, we present an algorithm for the offering agent to estimate the responding agent's gradient (preferences) and make offers based on previous acceptance or rejection responses. The algorithm's goal is to reach a Pareto-optimal resource allocation state while ensuring that the utilities of both agents improve after every accepted trade. The algorithm estimates the responding agent's gradient by leveraging the rejected offers and the greedy rationality assumption, to prune the space of potential gradients. We show that, after the algorithm makes a finite number of rejected offers, the algorithm either finds a mutually beneficial trade or certifies that the current state is epsilon-weakly Pareto optimal. We compare the proposed algorithm against various baselines in continuous and discrete trading scenarios and show that it improves the societal benefit with fewer offers. Additionally, we validate these findings in a user study with human participants, where the algorithm achieves high performance in scenarios with high resource conflict due to aligned agent goals.
Authors: Yifu Yuan, Zhenrui Zheng, Zibin Dong, Jianye Hao
Abstract: Multi-objective Reinforcement Learning (MORL) seeks to develop policies that simultaneously optimize multiple conflicting objectives, but it requires extensive online interactions. Offline MORL provides a promising solution by training on pre-collected datasets to generalize to any preference upon deployment. However, real-world offline datasets are often conservatively and narrowly distributed, failing to comprehensively cover preferences, leading to the emergence of out-of-distribution (OOD) preference areas. Existing offline MORL algorithms exhibit poor generalization to OOD preferences, resulting in policies that do not align with preferences. Leveraging the excellent expressive and generalization capabilities of diffusion models, we propose MODULI (Multi-objective Diffusion Planner with Sliding Guidance), which employs a preference-conditioned diffusion model as a planner to generate trajectories that align with various preferences and derive action for decision-making. To achieve accurate generation, MODULI introduces two return normalization methods under diverse preferences for refining guidance. To further enhance generalization to OOD preferences, MODULI proposes a novel sliding guidance mechanism, which involves training an additional slider adapter to capture the direction of preference changes. Incorporating the slider, it transitions from in-distribution (ID) preferences to generating OOD preferences, patching, and extending the incomplete Pareto front. Extensive experiments on the D4MORL benchmark demonstrate that our algorithm outperforms state-of-the-art Offline MORL baselines, exhibiting excellent generalization to OOD preferences.
Authors: Ziyin Zhang, Hang Yu, Shijie Li, Peng Di, Jianguo Li, Rui Wang
Abstract: Programming languages possess rich semantic information - such as data flow - that is represented by graphs and not available from the surface form of source code. Recent code language models have scaled to billions of parameters, but model source code solely as text tokens while ignoring any other structural information. Conversely, models that do encode structural information of code make modifications to the Transformer architecture, limiting their scale and compatibility with pretrained LLMs. In this work, we take the best of both worlds with GALLa - Graph Aligned Large Language Models. GALLa utilizes graph neural networks and cross-modal alignment technologies to inject the structural information of code into LLMs as an auxiliary task during finetuning. This framework is both model-agnostic and task-agnostic, as it can be applied to any code LLM for any code downstream task, and requires the structural graph data only at training time from a corpus unrelated to the finetuning data, while incurring no cost at inference time over the baseline LLM. Experiments on five code tasks with seven different baseline LLMs ranging in size from 350M to 14B validate the effectiveness of GALLa, demonstrating consistent improvement over the baseline, even for powerful models such as LLaMA3 and Qwen2.5-Coder.
Authors: Qunzhong Wang, Xiangguo Sun, Hong Cheng
Abstract: In recent years, graph prompting has emerged as a promising research direction, enabling the learning of additional tokens or subgraphs appended to the original graphs without requiring retraining of pre-trained graph models across various applications. This novel paradigm, shifting from the traditional pretraining and finetuning to pretraining and prompting has shown significant empirical success in simulating graph data operations, with applications ranging from recommendation systems to biological networks and graph transferring. However, despite its potential, the theoretical underpinnings of graph prompting remain underexplored, raising critical questions about its fundamental effectiveness. The lack of rigorous theoretical proof of why and how much it works is more like a dark cloud over the graph prompt area to go further. To fill this gap, this paper introduces a theoretical framework that rigorously analyzes graph prompting from a data operation perspective. Our contributions are threefold: First, we provide a formal guarantee theorem, demonstrating graph prompts capacity to approximate graph transformation operators, effectively linking upstream and downstream tasks. Second, we derive upper bounds on the error of these data operations by graph prompts for a single graph and extend this discussion to batches of graphs, which are common in graph model training. Third, we analyze the distribution of data operation errors, extending our theoretical findings from linear graph models (e.g., GCN) to non-linear graph models (e.g., GAT). Extensive experiments support our theoretical results and confirm the practical implications of these guarantees.
Authors: Xiang Hu, Zhihao Teng, Jun Zhao, Wei Wu, Kewei Tu
Abstract: Despite the success of Transformers, handling long contexts remains challenging due to the limited length generalization and quadratic complexity of self-attention. Thus Transformers often require post-training with a larger attention window, significantly increasing computational and memory costs. In this paper, we propose a novel attention mechanism based on dynamic context, Grouped Cross Attention (GCA), which can generalize to 1000 times the pre-training context length while maintaining the ability to access distant information with a constant attention window size. For a given input sequence, we split it into chunks and use each chunk to retrieve top-k relevant past chunks for subsequent text generation. Specifically, unlike most previous works that use an off-the-shelf retriever, our key innovation allows the retriever to learn how to retrieve past chunks that better minimize the auto-regressive loss of subsequent tokens in an end-to-end manner. Such a mechanism accommodates retrieved chunks with a fixed-size attention window to achieve long-range information access, significantly reducing computational and memory costs during training and inference. Experiments show that GCA-based models achieve near-perfect accuracy in passkey retrieval for 16M context lengths, which is 1000 times the training length.
Authors: Wooseong Yang, Chen Wang, Zihe Song, Weizhi Zhang, Philip S. Yu
Abstract: Session-based recommendation (SBR) aims to capture dynamic user preferences by analyzing item sequences within individual sessions. However, most existing approaches focus mainly on intra-session item relationships, neglecting the connections between items across different sessions (inter-session relationships), which limits their ability to fully capture complex item interactions. While some methods incorporate inter-session information, they often suffer from high computational costs, leading to longer training times and reduced efficiency. To address these challenges, we propose the CLIP-SBR (Cluster-aware Item Prompt learning for Session-Based Recommendation) framework. CLIP-SBR is composed of two modules: 1) an item relationship mining module that builds a global graph to effectively model both intra- and inter-session relationships, and 2) an item cluster-aware prompt learning module that uses soft prompts to integrate these relationships into SBR models efficiently. We evaluate CLIP-SBR across eight SBR models and three benchmark datasets, consistently demonstrating improved recommendation performance and establishing CLIP-SBR as a robust solution for session-based recommendation tasks.
Authors: Irene Cannistraci, Simone Antonelli, Emanuele Palumbo, Thomas M. Sutter, Emanuele Rodol\`a, Bastian Rieck, Julia E. Vogt
Abstract: Foundation Models have shown impressive performance in various tasks and domains, yet they require massive computational resources, raising concerns about accessibility and sustainability. Previous attempts to reduce foundation model size fall short of fully addressing the problem, as they end up increasing computational load through additional training steps. Recent works reveal that deep neural networks exhibit internal representation similarities. While inter-network similarities have enabled techniques such as model stitching and merging, intra-network similarities remain underexplored for improving efficiency. In this paper, we propose Transformer Blocks Approximation (TBA), a novel method that leverages intra-network similarities to identify and approximate transformer blocks in large vision models. TBA replaces these blocks using lightweight, closed-form transformations, without retraining or fine-tuning the rest of the model. The proposed method reduces the number of parameters while having minimal impact on the downstream task. We validate the effectiveness and generalizability of TBA through extensive experiments across multiple datasets (e.g., Imagenet-1k and CIFAR100) and state-of-the-art pretrained vision models (e.g, ViT, DiNO-v2, and DEiT).
Authors: Kaishuai Xu, Tiezheng Yu, Wenjun Hou, Yi Cheng, Chak Tou Leong, Liangyou Li, Xin Jiang, Lifeng Shang, Qun Liu, Wenjie Li
Abstract: Large Language Models (LLMs) have exhibited strong mathematical reasoning prowess, tackling tasks ranging from basic arithmetic to advanced competition-level problems. However, frequently occurring subtle yet critical errors, such as miscalculations or incorrect substitutions, limit the LLMs' full potential. Existing studies to improve mathematical ability typically involve applying preference learning to step-wise solution pairs. Although these methods leverage samples of varying granularity to mitigate reasoning errors, they overlook critical subtle errors. In this work, we propose a novel preference learning framework called eRror-Injected Self-Editing (RISE), which injects predefined subtle errors into pivotal tokens in reasoning or computation steps to construct hard pairs for error mitigation. In detail, RISE uses the LLM itself to edit a small number of tokens in the solution, injecting designed subtle errors. Then, pairs composed of self-edited solutions and their corresponding correct ones, along with pairs of correct and incorrect solutions obtained through sampling, are used together for subtle error-aware DPO training. Compared with other preference learning methods, RISE further refines the training objective without requiring fine-grained sampling or preference annotation. Extensive experiments validate the effectiveness of RISE, with preference learning on Qwen2-7B-Instruct yielding notable improvements of 3.0% on GSM8K and 7.9% on MATH with only 4.5K training samples. Moreover, the effect of error mitigation extends from mathematical reasoning to logical reasoning and code generation.
Authors: Noorbakhsh Amiri Golilarz, Elias Hossain, Abdoljalil Addeh, Keyan Alexander Rahimi
Abstract: In this paper, we discuss learning algorithms and their importance in different types of applications which includes training to identify important patterns and features in a straightforward, easy-to-understand manner. We will review the main concepts of artificial intelligence (AI), machine learning (ML), deep learning (DL), and hybrid models. Some important subsets of Machine Learning algorithms such as supervised, unsupervised, and reinforcement learning are also discussed in this paper. These techniques can be used for some important tasks like prediction, classification, and segmentation. Convolutional Neural Networks (CNNs) are used for image and video processing and many more applications. We dive into the architecture of CNNs and how to integrate CNNs with ML algorithms to build hybrid models. This paper explores the vulnerability of learning algorithms to noise, leading to misclassification. We further discuss the integration of learning algorithms with Large Language Models (LLM) to generate coherent responses applicable to many domains such as healthcare, marketing, and finance by learning important patterns from large volumes of data. Furthermore, we discuss the next generation of learning algorithms and how we may have an unified Adaptive and Dynamic Network to perform important tasks. Overall, this article provides brief overview of learning algorithms, exploring their current state, applications and future direction.
Authors: Georgios Chochlakis, Alexandros Potamianos, Kristina Lerman, Shrikanth Narayanan
Abstract: In-context Learning (ICL) has become the primary method for performing natural language tasks with Large Language Models (LLMs). The knowledge acquired during pre-training is crucial for this few-shot capability, providing the model with task priors. However, recent studies have shown that ICL predominantly relies on retrieving task priors rather than "learning" to perform tasks. This limitation is particularly evident in complex subjective domains such as emotion and morality, where priors significantly influence posterior predictions. In this work, we examine whether this is the result of the aggregation used in corresponding datasets, where trying to combine low-agreement, disparate annotations might lead to annotation artifacts that create detrimental noise in the prompt. Moreover, we evaluate the posterior bias towards certain annotators by grounding our study in appropriate, quantitative measures of LLM priors. Our results indicate that aggregation is a confounding factor in the modeling of subjective tasks, and advocate focusing on modeling individuals instead. However, aggregation does not explain the entire gap between ICL and the state of the art, meaning other factors in such tasks also account for the observed phenomena. Finally, by rigorously studying annotator-level labels, we find that it is possible for minority annotators to both better align with LLMs and have their perspectives further amplified.
Authors: Runchu Tian, Yanghao Li, Yuepeng Fu, Siyang Deng, Qinyu Luo, Cheng Qian, Shuo Wang, Xin Cong, Zhong Zhang, Yesai Wu, Yankai Lin, Huadong Wang, Xiaojiang Liu
Abstract: Positional bias in large language models (LLMs) hinders their ability to effectively process long inputs. A prominent example is the "lost in the middle" phenomenon, where LLMs struggle to utilize relevant information situated in the middle of the input. While prior research primarily focuses on single pieces of relevant information, real-world applications often involve multiple relevant information pieces. To bridge this gap, we present LongPiBench, a benchmark designed to assess positional bias involving multiple pieces of relevant information. Thorough experiments are conducted with five commercial and six open-source models. These experiments reveal that while most current models are robust against the "lost in the middle" issue, there exist significant biases related to the spacing of relevant information pieces. These findings highlight the importance of evaluating and reducing positional biases to advance LLM's capabilities.
Authors: Yuyang Ding, Xinyu Shi, Xiaobo Liang, Juntao Li, Zhaopeng Tu, Qiaoming Zhu, Min Zhang
Abstract: Improving the mathematical reasoning capabilities of Large Language Models (LLMs) is critical for advancing artificial intelligence. However, access to extensive, diverse, and high-quality reasoning datasets remains a significant challenge, particularly for the open-source community. In this paper, we propose ScaleQuest, a novel, scalable, and cost-effective data synthesis method that enables the generation of large-scale mathematical reasoning datasets using lightweight 7B-scale models. ScaleQuest introduces a two-stage question-tuning process comprising Question Fine-Tuning (QFT) and Question Preference Optimization (QPO) to unlock the question generation capabilities of problem-solving models. By generating diverse questions from scratch -- without relying on powerful proprietary models or seed data -- we produce a dataset of 1 million problem-solution pairs. Our experiments demonstrate that models trained on our data outperform existing open-source datasets in both in-domain and out-of-domain evaluations. Furthermore, our approach shows continued performance improvement as the volume of training data increases, highlighting its potential for ongoing data scaling. The extensive improvements observed in code reasoning tasks demonstrate the generalization capabilities of our proposed method. Our work provides the open-source community with a practical solution to enhance the mathematical reasoning abilities of LLMs.
Authors: Mou\"in Ben Ammar, David Brellmann, Arturo Mendoza, Antoine Manzanera, Gianni Franchi
Abstract: Out-of-distribution (OOD) detection is essential for ensuring the reliability and safety of machine learning systems. In recent years, it has received increasing attention, particularly through post-hoc detection and training-based methods. In this paper, we focus on post-hoc OOD detection, which enables identifying OOD samples without altering the model's training procedure or objective. Our primary goal is to investigate the relationship between model capacity and its OOD detection performance. Specifically, we aim to answer the following question: Does the Double Descent phenomenon manifest in post-hoc OOD detection? This question is crucial, as it can reveal whether overparameterization, which is already known to benefit generalization, can also enhance OOD detection. Despite the growing interest in these topics by the classic supervised machine learning community, this intersection remains unexplored for OOD detection. We empirically demonstrate that the Double Descent effect does indeed appear in post-hoc OOD detection. Furthermore, we provide theoretical insights to explain why this phenomenon emerges in such setting. Finally, we show that the overparameterized regime does not yield superior results consistently, and we propose a method to identify the optimal regime for OOD detection based on our observations.
Authors: Hjalmar Wijk, Tao Lin, Joel Becker, Sami Jawhar, Neev Parikh, Thomas Broadley, Lawrence Chan, Michael Chen, Josh Clymer, Jai Dhyani, Elena Ericheva, Katharyn Garcia, Brian Goodrich, Nikola Jurkovic, Holden Karnofsky, Megan Kinniment, Aron Lajko, Seraphina Nix, Lucas Sato, William Saunders, Maksym Taran, Ben West, Elizabeth Barnes
Abstract: Frontier AI safety policies highlight automation of AI research and development (R&D) by AI agents as an important capability to anticipate. However, there exist few evaluations for AI R&D capabilities, and none that are highly realistic and have a direct comparison to human performance. We introduce RE-Bench (Research Engineering Benchmark, v1), which consists of 7 challenging, open-ended ML research engineering environments and data from 71 8-hour attempts by 61 distinct human experts. We confirm that our experts make progress in the environments given 8 hours, with 82% of expert attempts achieving a non-zero score and 24% matching or exceeding our strong reference solutions. We compare humans to several public frontier models through best-of-k with varying time budgets and agent designs, and find that the best AI agents achieve a score 4x higher than human experts when both are given a total time budget of 2 hours per environment. However, humans currently display better returns to increasing time budgets, narrowly exceeding the top AI agent scores given an 8-hour budget, and achieving 2x the score of the top AI agent when both are given 32 total hours (across different attempts). Qualitatively, we find that modern AI agents possess significant expertise in many ML topics -- e.g. an agent wrote a faster custom Triton kernel than any of our human experts' -- and can generate and test solutions over ten times faster than humans, at much lower cost. We open-source the evaluation environments, human expert data, analysis code and agent trajectories to facilitate future research.
Authors: Tae-Hoon Lee, Min-Soo Kim
Abstract: Integer linear programming (ILP) is widely utilized for various combinatorial optimization problems. Primal heuristics play a crucial role in quickly finding feasible solutions for NP-hard ILP. Although \textit{end-to-end learning}-based primal heuristics (E2EPH) have recently been proposed, they are typically unable to independently generate feasible solutions and mainly focus on binary variables. Ensuring feasibility is critical, especially when handling non-binary integer variables. To address this challenge, we propose RL-SPH, a novel reinforcement learning-based start primal heuristic capable of independently generating feasible solutions, even for ILP involving non-binary integers. Experimental results demonstrate that RL-SPH rapidly obtains high-quality feasible solutions, achieving on average a 44x lower primal gap and a 2.3x lower primal integral compared to existing primal heuristics.
Authors: Federico Ruggeri, Gaetano Signorelli
Abstract: A popular end-to-end architecture for selective rationalization is the select-then-predict pipeline, comprising a generator to extract highlights fed to a predictor. Such a cooperative system suffers from suboptimal equilibrium minima due to the dominance of one of the two modules, a phenomenon known as interlocking. While several contributions aimed at addressing interlocking, they only mitigate its effect, often by introducing feature-based heuristics, sampling, and ad-hoc regularizations. We present GenSPP, the first interlocking-free architecture for selective rationalization that does not require any learning overhead, as the above-mentioned. GenSPP avoids interlocking by performing disjoint training of the generator and predictor via genetic global search. Experiments on a synthetic and a real-world benchmark show that our model outperforms several state-of-the-art competitors.
Authors: Kun Wu, Chengkai Hou, Jiaming Liu, Zhengping Che, Xiaozhu Ju, Zhuqin Yang, Meng Li, Yinuo Zhao, Zhiyuan Xu, Guang Yang, Shichao Fan, Xinhua Wang, Fei Liao, Zhen Zhao, Guangyu Li, Zhao Jin, Lecheng Wang, Jilei Mao, Ning Liu, Pei Ren, Qiang Zhang, Yaoxu Lyu, Mengzhen Liu, Jingyang He, Yulin Luo, Zeyu Gao, Chenxuan Li, Chenyang Gu, Yankai Fu, Di Wu, Xingyu Wang, Sixiang Chen, Zhenyu Wang, Pengju An, Siyuan Qian, Shanghang Zhang, Jian Tang
Abstract: In this paper, we introduce RoboMIND (Multi-embodiment Intelligence Normative Data for Robot Manipulation), a dataset containing 107k demonstration trajectories across 479 diverse tasks involving 96 object classes. RoboMIND is collected through human teleoperation and encompasses comprehensive robotic-related information, including multi-view observations, proprioceptive robot state information, and linguistic task descriptions. To ensure data consistency and reliability for imitation learning, RoboMIND is built on a unified data collection platform and a standardized protocol, covering four distinct robotic embodiments: the Franka Emika Panda, the UR5e, the AgileX dual-arm robot, and a humanoid robot with dual dexterous hands. Our dataset also includes 5k real-world failure demonstrations, each accompanied by detailed causes, enabling failure reflection and correction during policy learning. Additionally, we created a digital twin environment in the Isaac Sim simulator, replicating the real-world tasks and assets, which facilitates the low-cost collection of additional training data and enables efficient evaluation. To demonstrate the quality and diversity of our dataset, we conducted extensive experiments using various imitation learning methods for single-task settings and state-of-the-art Vision-Language-Action (VLA) models for multi-task scenarios. By leveraging RoboMIND, the VLA models achieved high manipulation success rates and demonstrated strong generalization capabilities. To the best of our knowledge, RoboMIND is the largest multi-embodiment teleoperation dataset collected on a unified platform, providing large-scale and high-quality robotic training data. Our project is at https://x-humanoid-robomind.github.io/.
Authors: Ting Bai, Yue Yu, Le Huang, Zenan Xu, Zhe Zhao, Chuan Shi
Abstract: The sparse Mixture-of-Experts (MoE) architecture of large language models (LLMs) confronts an inherent issue of load imbalance arising from the simplistic linear router strategy, which ultimately causes the instability and inefficient learning of LLMs. To address this challenge, we introduce a novel MoE graph-based framework $\textbf{GMoE}$, aimed at enhancing the collaboration among multiple experts. In GMoE, a graph router function is designed to capture the collaboration signals among experts. This enables all experts to dynamically allocate information derived from input data by sharing information with their neighboring experts. Moreover, we put forward two coordination strategies in GMoE: the $\textit{Poisson distribution-based distinction strategy}$ and the $\textit{Normal distribution-based balance strategy}$, to further release the capacity of each expert and increase the model stability in the fine-tuning of LLMs. Specifically, we leverage a parameter-efficient fine-tuning technique, i.e., Low-Rank Adaptation (LoRA), to implement the graph MoE architecture. Extensive experiments on four real-world benchmark datasets demonstrate the effectiveness of GMoE, showing the benefits of facilitating collaborations of multiple experts in LLM fine-tuning. The code of experimental implementation is available at https://github.com/BAI-LAB/GMoE
Authors: Kaichen Ouyang, Shengwei Fu, Zong Ke, Renxiang Guan, Ke Liang, Dayu Hu
Abstract: Evolutionary algorithms (EAs) simulate natural selection but have two main limitations: (1) they rarely update individuals based on global correlations, limiting comprehensive learning; (2) they struggle with balancing exploration and exploitation, where excessive exploitation causes premature convergence, and excessive exploration slows down the search. Moreover, EAs often depend on manual parameter settings, which can disrupt the exploration-exploitation balance.To address these issues, we propose Graph Neural Evolution (GNE), a novel EA framework. GNE represents the population as a graph, where nodes represent individuals, and edges capture their relationships, enabling global information usage. GNE utilizes spectral graph neural networks (GNNs) to decompose evolutionary signals into frequency components, applying a filtering function to fuse these components. High-frequency components capture diverse global information, while low-frequency ones capture more consistent information. This explicit frequency filtering strategy directly controls global-scale features through frequency components, overcoming the limitations of manual parameter settings and making the exploration-exploitation control more interpretable and effective.Extensive tests on nine benchmark functions (e.g., Sphere, Rastrigin, Rosenbrock) show that GNE outperforms classical (GA, DE, CMA-ES) and advanced algorithms (SDAES, RL-SHADE) under various conditions, including noise-corrupted and optimal solution deviation scenarios. GNE achieves solutions several orders of magnitude better (e.g., 3.07e-20 mean on Sphere vs. 1.51e-07).
Authors: Xiaoshuai Song, Yanan Wu, Weixun Wang, Jiaheng Liu, Wenbo Su, Bo Zheng
Abstract: Self-Correction aims to enable large language models (LLMs) to self-verify and self-refine their initial responses without external feedback. However, LLMs often fail to effectively self-verify and generate correct feedback, further misleading refinement and leading to the failure of self-correction, especially in complex reasoning tasks. In this paper, we propose Program-driven Self-Correction (ProgCo). First, program-driven verification (ProgVe) achieves complex verification logic and extensive validation through self-generated, self-executing verification pseudo-programs. Then, program-driven refinement (ProgRe) receives feedback from ProgVe, conducts dual reflection and refinement on both responses and verification programs to mitigate misleading of incorrect feedback in complex reasoning tasks. Experiments on three instruction-following and mathematical benchmarks indicate that ProgCo achieves effective self-correction, and can be further enhance performance when combined with real program tools. We release our code at https://github.com/songxiaoshuai/progco.
Authors: Hadi Askari, Shivanshu Gupta, Terry Tong, Fei Wang, Anshuman Chhabra, Muhao Chen
Abstract: In this work, we introduce a novel paradigm for generalized In-Context Learning (ICL), termed Indirect In-Context Learning. In Indirect ICL, we explore demonstration selection strategies tailored for two distinct real-world scenarios: Mixture of Tasks and Noisy ICL. We systematically evaluate the effectiveness of Influence Functions (IFs) as a selection tool for these settings, highlighting the potential of IFs to better capture the informativeness of examples within the demonstration pool. For the Mixture of Tasks setting, demonstrations are drawn from 28 diverse tasks, including MMLU, BigBench, StrategyQA, and CommonsenseQA. We demonstrate that combining BertScore-Recall (BSR) with an IF surrogate model can further improve performance, leading to average absolute accuracy gains of 0.37\% and 1.45\% for 3-shot and 5-shot setups when compared to traditional ICL metrics. In the Noisy ICL setting, we examine scenarios where demonstrations might be mislabeled or have adversarial noise. Our experiments show that reweighting traditional ICL selectors (BSR and Cosine Similarity) with IF-based selectors boosts accuracy by an average of 2.90\% for Cosine Similarity and 2.94\% for BSR on noisy GLUE benchmarks. For the adversarial sub-setting, we show the utility of using IFs for task-agnostic demonstration selection for backdoor attack mitigation. Showing a 32.89\% reduction in Attack Success Rate compared to task-aware methods. In sum, we propose a robust framework for demonstration selection that generalizes beyond traditional ICL, offering valuable insights into the role of IFs for Indirect ICL.
Authors: Xiaoqing Zhang, Ang Lv, Yuhan Liu, Flood Sung, Wei Liu, Jian Luan, Shuo Shang, Xiuying Chen, Rui Yan
Abstract: Large language models (LLMs) excel at few-shot in-context learning (ICL) without requiring parameter updates. However, as ICL demonstrations increase from a few to many, performance tends to plateau and eventually decline. We identify two primary causes for this trend: the suboptimal negative log-likelihood (NLL) optimization objective and the incremental data noise. To address these issues, we introduce \textit{DrICL}, a novel optimization method that enhances model performance through \textit{Differentiated} and \textit{Reweighting} objectives. Globally, DrICL utilizes differentiated learning to optimize the NLL objective, ensuring that many-shot performance surpasses zero-shot levels. Locally, it dynamically adjusts the weighting of many-shot demonstrations by leveraging cumulative advantages inspired by reinforcement learning, thereby mitigating the impact of noisy data. Recognizing the lack of multi-task datasets with diverse many-shot distributions, we develop the \textit{Many-Shot ICL Benchmark} (ICL-50)-a large-scale benchmark of 50 tasks that cover shot numbers from 1 to 350 within sequences of up to 8,000 tokens-for both fine-tuning and evaluation purposes. Experimental results demonstrate that LLMs enhanced with DrICL achieve significant improvements in many-shot setups across various tasks, including both in-domain and out-of-domain scenarios. We release the code and dataset hoping to facilitate further research in many-shot ICL\footnote{https://github.com/xiaoqzhwhu/DrICL}.
Authors: Kristof Meding
Abstract: What constitutes a fair decision? This question is not only difficult for humans but becomes more challenging when Artificial Intelligence (AI) models are used. In light of discriminatory algorithmic behaviors, the EU has recently passed the AI Act, which mandates specific rules for high-risk systems, incorporating both traditional legal non-discrimination regulations and machine learning based algorithmic fairness concepts. This paper aims to bridge these two different concepts in the AI Act through: First, a necessary high-level introduction of both concepts targeting legal and computer science-oriented scholars, and second, an in-depth analysis of the AI Act's relationship between legal non-discrimination regulations and algorithmic fairness. Our analysis reveals three key findings: (1.) Most non-discrimination regulations target only high-risk AI systems. (2.) The regulation of high-risk systems encompasses both data input requirements and output monitoring, though these regulations are partly inconsistent and raise questions of computational feasibility. (3.) Finally, we consider the possible (future) interaction of classical EU non-discrimination law and the AI Act regulations. We recommend developing more specific auditing and testing methodologies for AI systems. This paper aims to serve as a foundation for future interdisciplinary collaboration between legal scholars and computer science-oriented machine learning researchers studying discrimination in AI systems.
Authors: Yang Xu, Vaneet Aggarwal
Abstract: We address the problem of quantum reinforcement learning (QRL) under model-free settings with quantum oracle access to the Markov Decision Process (MDP). This paper introduces a Quantum Natural Policy Gradient (QNPG) algorithm, which replaces the random sampling used in classical Natural Policy Gradient (NPG) estimators with a deterministic gradient estimation approach, enabling seamless integration into quantum systems. While this modification introduces a bounded bias in the estimator, the bias decays exponentially with increasing truncation levels. This paper demonstrates that the proposed QNPG algorithm achieves a sample complexity of $\tilde{\mathcal{O}}(\epsilon^{-1.5})$ for queries to the quantum oracle, significantly improving the classical lower bound of $\tilde{\mathcal{O}}(\epsilon^{-2})$ for queries to the MDP.
Authors: David Salinas, Omar Swelam, Frank Hutter
Abstract: Evaluating Large Language Models (LLMs) often requires costly human annotations. To address this, LLM-based judges have been proposed, which compare the outputs of two LLMs enabling the ranking of models without human intervention. While several approaches have been proposed, many confounding factors are present between different papers. For instance the model, the prompt and other hyperparameters are typically changed at the same time making apple-to-apple comparisons challenging. In this paper, we propose to systematically analyze and tune the hyperparameters of LLM judges. To alleviate the high cost of evaluating a judge, we propose to leverage multi-objective multi-fidelity which allows to find judges that trade accuracy for cost and also significantly reduce the cost of the search. Our method identifies judges that not only outperform existing benchmarks in accuracy and cost-efficiency but also utilize open-weight models, ensuring greater accessibility and reproducibility. The code to reproduce our experiments is available at this repository https://github.com/geoalgo/judgetuning .
Authors: Haoran Luo, Haihong E, Yikai Guo, Qika Lin, Xiaobao Wu, Xinyu Mu, Wenhao Liu, Meina Song, Yifan Zhu, Luu Anh Tuan
Abstract: Knowledge Base Question Answering (KBQA) aims to answer natural language questions with a large-scale structured knowledge base (KB). Despite advancements with large language models (LLMs), KBQA still faces challenges in weak KB awareness, imbalance between effectiveness and efficiency, and high reliance on annotated data. To address these challenges, we propose KBQA-o1, a novel agentic KBQA method with Monte Carlo Tree Search (MCTS). It introduces a ReAct-based agent process for stepwise logical form generation with KB environment exploration. Moreover, it employs MCTS, a heuristic search method driven by policy and reward models, to balance agentic exploration's performance and search space. With heuristic exploration, KBQA-o1 generates high-quality annotations for further improvement by incremental fine-tuning. Experimental results show that KBQA-o1 outperforms previous low-resource KBQA methods with limited annotated data, boosting Llama-3.1-8B model's GrailQA F1 performance to 78.5% compared to 48.5% of the previous sota method with GPT-3.5-turbo. Our code is publicly available.
Authors: Xinyu Liu, Zixuan Xie, Shangtong Zhang
Abstract: $Q$-learning is one of the most fundamental reinforcement learning algorithms. It is widely believed that $Q$-learning with linear function approximation (i.e., linear $Q$-learning) suffers from possible divergence until the recent work Meyn (2024) which establishes the ultimate almost sure boundedness of the iterates of linear $Q$-learning. Building on this success, this paper further establishes the first $L^2$ convergence rate of linear $Q$-learning iterates (to a bounded set). Similar to Meyn (2024), we do not make any modification to the original linear $Q$-learning algorithm, do not make any Bellman completeness assumption, and do not make any near-optimality assumption on the behavior policy. All we need is an $\epsilon$-softmax behavior policy with an adaptive temperature. The key to our analysis is the general result of stochastic approximations under Markovian noise with fast-changing transition functions. As a side product, we also use this general result to establish the $L^2$ convergence rate of tabular $Q$-learning with an $\epsilon$-softmax behavior policy, for which we rely on a novel pseudo-contraction property of the weighted Bellman optimality operator.
Authors: Shashank Sharma, Janina Hoffmann, Vinay Namboodiri
Abstract: Hierarchical Reinforcement Learning (HRL) agents often struggle with long-horizon visual planning due to their reliance on error-prone distance metrics. We propose Discrete Hierarchical Planning (DHP), a method that replaces continuous distance estimates with discrete reachability checks to evaluate subgoal feasibility. DHP recursively constructs tree-structured plans by decomposing long-term goals into sequences of simpler subtasks, using a novel advantage estimation strategy that inherently rewards shorter plans and generalizes beyond training depths. In addition, to address the data efficiency challenge, we introduce an exploration strategy that generates targeted training examples for the planning modules without needing expert data. Experiments in 25-room navigation environments demonstrate $100\%$ success rate (vs $82\%$ baseline) and $73$-step average episode length (vs $158$-step baseline). The method also generalizes to momentum-based control tasks and requires only $\log N$ steps for replanning. Theoretical analysis and ablations validate our design choices.
Authors: Chhavi Yadav, Evan Monroe Laufer, Dan Boneh, Kamalika Chaudhuri
Abstract: In principle, explanations are intended as a way to increase trust in machine learning models and are often obligated by regulations. However, many circumstances where these are demanded are adversarial in nature, meaning the involved parties have misaligned interests and are incentivized to manipulate explanations for their purpose. As a result, explainability methods fail to be operational in such settings despite the demand \cite{bordt2022post}. In this paper, we take a step towards operationalizing explanations in adversarial scenarios with Zero-Knowledge Proofs (ZKPs), a cryptographic primitive. Specifically we explore ZKP-amenable versions of the popular explainability algorithm LIME and evaluate their performance on Neural Networks and Random Forests. Our code is publicly available at https://github.com/emlaufer/ExpProof.
Authors: Xu Zhang, Kaidi Xu, Ziqing Hu, Ren Wang
Abstract: Mixture of Experts (MoE) have shown remarkable success in leveraging specialized expert networks for complex machine learning tasks. However, their susceptibility to adversarial attacks presents a critical challenge for deployment in robust applications. This paper addresses the critical question of how to incorporate robustness into MoEs while maintaining high natural accuracy. We begin by analyzing the vulnerability of MoE components, finding that expert networks are notably more susceptible to adversarial attacks than the router. Based on this insight, we propose a targeted robust training technique that integrates a novel loss function to enhance the adversarial robustness of MoE, requiring only the robustification of one additional expert without compromising training or inference efficiency. Building on this, we introduce a dual-model strategy that linearly combines a standard MoE model with our robustified MoE model using a smoothing parameter. This approach allows for flexible control over the robustness-accuracy trade-off. We further provide theoretical foundations by deriving certified robustness bounds for both the single MoE and the dual-model. To push the boundaries of robustness and accuracy, we propose a novel joint training strategy JTDMoE for the dual-model. This joint training enhances both robustness and accuracy beyond what is achievable with separate models. Experimental results on CIFAR-10 and TinyImageNet datasets using ResNet18 and Vision Transformer (ViT) architectures demonstrate the effectiveness of our proposed methods. The code is publicly available at https://github.com/TIML-Group/Robust-MoE-Dual-Model.
Authors: Fanxu Meng, Pingzhi Tang, Zengwei Yao, Xing Sun, Muhan Zhang
Abstract: In this paper, we present TransMLA, a framework that seamlessly converts any GQA-based pre-trained model into an MLA-based model. Our approach enables direct compatibility with DeepSeek's codebase, allowing these models to fully leverage DeepSeek-specific optimizations such as vLLM and SGlang. By compressing 93% of the KV cache in LLaMA-2-7B, TransMLA achieves a 10.6x inference speedup at an 8K context length while preserving meaningful output quality. Additionally, the model requires only 6 billion tokens for fine-tuning to regain performance on par with the original across multiple benchmarks. TransMLA offers a practical solution for migrating GQA-based models to the MLA structure. When combined with DeepSeek's advanced features, such as FP8 quantization and Multi-Token Prediction, even greater inference acceleration can be realized.
Authors: Wenbo Pan, Zhichao Liu, Qiguang Chen, Xiangyang Zhou, Haining Yu, Xiaohua Jia
Abstract: Large Language Models' safety-aligned behaviors, such as refusing harmful queries, can be represented by linear directions in activation space. Previous research modeled safety behavior with a single direction, limiting mechanistic understanding to an isolated safety feature. In this work, we discover that safety-aligned behavior is jointly controlled by multi-dimensional directions. Namely, we study the vector space of representation shifts during safety fine-tuning on Llama 3 8B for refusing jailbreaks. By studying orthogonal directions in the space, we first find that a dominant direction governs the model's refusal behavior, while multiple smaller directions represent distinct and interpretable features like hypothetical narrative and role-playing. We then measure how different directions promote or suppress the dominant direction, showing the important role of secondary directions in shaping the model's refusal representation. Finally, we demonstrate that removing certain trigger tokens in harmful queries can mitigate these directions to bypass the learned safety capability, providing new insights on understanding safety alignment vulnerability from a multi-dimensional perspective. Code and artifacts are available at https://github.com/BMPixel/safety-residual-space.
Authors: Jiahao Huo, Yibo Yan, Xu Zheng, Yuanhuiyi Lyu, Xin Zou, Zhihua Wei, Xuming Hu
Abstract: Recent progress in Machine Unlearning (MU) has introduced solutions for the selective removal of private or sensitive information encoded within deep neural networks. Nonetheless, MU for Multimodal Large Language Models (MLLMs) remains in its nascent phase. Therefore, we propose to reformulate the task of multimodal MU in the era of MLLMs, which aims to erase only the visual patterns associated with a given entity while preserving the corresponding textual knowledge encoded within the original parameters of the language model backbone. Furthermore, we develop a novel geometry-constrained gradient ascent method MMUnlearner. It updates the weights of MLLMs with a weight saliency map jointly restricted by the remaining concepts and textual knowledge during unlearning, thereby preserving parameters essential for non-target knowledge. Extensive experiments demonstrate that MMUnlearner surpasses baselines that finetuning MLLMs with VQA data directly through Gradient Ascent (GA) or Negative Preference Optimization (NPO), across all evaluation dimensions. Our code can be found in [this URL](https://github.com/Z1zs/MMUnlearner).
Authors: Bhargavi Kalyani I, A Rama Prasad Mathi, Niladri Sett
Abstract: Link prediction (LP) is an important problem in network science and machine learning research. The state-of-the-art LP methods are usually evaluated in a uniform setup, ignoring several factors associated with the data and application specific needs. We identify a number of such factors, such as, network-type, problem-type, geodesic distance between the end nodes and its distribution over the classes, nature and applicability of LP methods, class imbalance and its impact on early retrieval, evaluation metric, etc., and present an experimental setup which allows us to evaluate LP methods in a rigorous and controlled manner. We perform extensive experiments with a variety of LP methods over real network datasets in this controlled setup, and gather valuable insights on the interactions of these factors with the performance of LP through an array of carefully designed hypotheses. Following the insights, we provide recommendations to be followed as best practice for evaluating LP methods.
Authors: Andrei Jarca, Florinel Alin Croitoru, Radu Tudor Ionescu
Abstract: Masked language modeling has become a widely adopted unsupervised technique to pre-train large language models (LLMs). However, the process of selecting tokens for masking is random, and the percentage of masked tokens is typically fixed for the entire training process. In this paper, we propose to adjust the masking ratio and to decide which tokens to mask based on a novel task-informed anti-curriculum learning scheme. First, we harness task-specific knowledge about useful and harmful tokens in order to determine which tokens to mask. Second, we propose a cyclic decaying masking ratio, which corresponds to an anti-curriculum schedule (from hard to easy). We exemplify our novel task-informed anti-curriculum by masking (TIACBM) approach across three diverse downstream tasks: sentiment analysis, text classification by topic, and authorship attribution. Our findings suggest that TIACBM enhances the ability of the model to focus on key task-relevant features, contributing to statistically significant performance gains across tasks. We release our code at https://github.com/JarcaAndrei/TIACBM.
Authors: Vishal Dey, Xiao Hu, Xia Ning
Abstract: Despite recent advancements, most computational methods for molecule optimization are constrained to single- or double-property optimization tasks and suffer from poor scalability and generalizability to novel optimization tasks. Meanwhile, Large Language Models (LLMs) demonstrate remarkable out-of-domain generalizability to novel tasks. To demonstrate LLMs' potential for molecule optimization, we introduce MuMOInstruct, the first high-quality instruction-tuning dataset specifically focused on complex multi-property molecule optimization tasks. Leveraging MuMOInstruct, we develop GeLLMOs, a series of instruction-tuned LLMs for molecule optimization. Extensive evaluations across 5 in-domain and 5 out-of-domain tasks demonstrate that GeLLMOs consistently outperform state-of-the-art baselines. GeLLMOs also exhibit outstanding zero-shot generalization to unseen tasks, significantly outperforming powerful closed-source LLMs. Such strong generalizability demonstrates the tremendous potential of GeLLMOs as foundational models for molecule optimization, thereby tackling novel optimization tasks without resource-intensive retraining. MuMOInstruct, models, and code are accessible through https://github.com/ninglab/GeLLMO.
Authors: Nadav Borenstein, Greta Warren, Desmond Elliott, Isabelle Augenstein
Abstract: Two commonly employed strategies to combat the rise of misinformation on social media are (i) fact-checking by professional organisations and (ii) community moderation by platform users. Policy changes by Twitter/X and, more recently, Meta, signal a shift away from partnerships with fact-checking organisations and towards an increased reliance on crowdsourced community notes. However, the extent and nature of dependencies between fact-checking and helpful community notes remain unclear. To address these questions, we use language models to annotate a large corpus of Twitter/X community notes with attributes such as topic, cited sources, and whether they refute claims tied to broader misinformation narratives. Our analysis reveals that community notes cite fact-checking sources up to five times more than previously reported. Fact-checking is especially crucial for notes on posts linked to broader narratives, which are twice as likely to reference fact-checking sources compared to other sources. Our results show that successful community moderation relies on professional fact-checking and highlight how citizen and professional fact-checking are deeply intertwined.
Authors: Minyeong Hwang, Ziseok Lee, Kwang-Soo Kim, Kyungsu Kim, Eunho Yang
Abstract: Linker generation is critical in drug discovery applications such as lead optimization and PROTAC design, where molecular fragments are assembled into diverse drug candidates via molecular linker. Existing methods fall into point cloud-free and point cloud-aware categories based on their use of fragments' 3D poses alongside their topologies in sampling the linker's topology. Point cloud-free models prioritize sample diversity but suffer from lower validity due to overlooking fragments' spatial constraints, while point cloud-aware models ensure higher validity but restrict diversity by enforcing strict spatial constraints. To overcome these trade-offs without additional training, we propose HybridLinker, a framework that enhances point cloud-aware inference by providing diverse bonding topologies from a pretrained point cloud-free model as guidance. At its core, we propose LinkerDPS, the first diffusion posterior sampling (DPS) method operating across point cloud-free and point cloud-aware spaces, bridging molecular topology with 3D point clouds via an energy-inspired function. By transferring the diverse sampling distribution of point cloud-free models into the point cloud-aware distribution, HybridLinker significantly surpasses baselines, improving both validity and diversity in foundational molecular design and applied drug optimization tasks, establishing a new DPS framework in the molecular domains beyond imaging.
Authors: Xiaoqing Zhang, Yuhan Liu, Flood Sung, Xiuying Chen, Shuo Shang, Rui Yan
Abstract: Code generation is crucial in software engineering for automating the coding process efficiently. While test-time computation methods show promise, they suffer from high latency due to multiple computation rounds. To overcome this, we introduce \textbf{ThinkCoder}, a framework that combines thorough exploration with optimal refinement. The exploration phase diversifies the solution space by searching for potential solutions, followed by a refinement phase that enhances precision. This approach allows us to select the best solution through careful consideration before taking action, avoiding excessive trial and error. To further minimize test-time computation overhead, we introduce preference-driven optimization with Reinforced Self-Training (ReST), which uses exploration trajectories from ThinkCoder to guide LLM's evolution. This approach enhances LLM's exploration efficiency via preference learning, cutting costs while maintaining accuracy. ThinkCoder boosts the performance with a single LLM, excelling on benchmarks like HumanEval and MBPP. Compared to SOTA models, it improves Pass@1 by 3.0\% over MapCoder with just 6.4\% of the computation cost. Against AgentCoder, ThinkCoder achieves a 0.5\% higher Pass@1 after 2 rounds, outperforming AgentCoder's 5 rounds. Additionally, ReST with success trajectories enhances efficiency, allowing models like LLaMA2-7B to achieve competitive results using only 20\% of the computational resources. These results highlight the framework's effectiveness and scalability.
Authors: Fahim Tajwar, Yiding Jiang, Abitha Thankaraj, Sumaita Sadia Rahman, J Zico Kolter, Jeff Schneider, Ruslan Salakhutdinov
Abstract: Efficient exploration is essential for intelligent systems interacting with their environment, but existing language models often fall short in scenarios that require strategic information gathering. In this paper, we present Paprika, a fine-tuning approach that enables language models to develop general decision-making capabilities that are not confined to particular environments. By training on synthetic interaction data from different tasks that require diverse strategies, Paprika teaches models to explore and adapt their behavior on a new task based on environment feedback in-context without more gradient updates. Experimental results show that models fine-tuned with Paprika can effectively transfer their learned decision-making capabilities to entirely unseen tasks without additional training. Unlike traditional training, our approach's primary bottleneck lies in sampling useful interaction data instead of model updates. To improve sample efficiency, we propose a curriculum learning strategy that prioritizes sampling trajectories from tasks with high learning potential. These results suggest a promising path towards AI systems that can autonomously solve novel sequential decision-making problems that require interactions with the external world.
Authors: Kaishuai Xu, Tiezheng Yu, Wenjun Hou, Yi Cheng, Liangyou Li, Xin Jiang, Lifeng Shang, Qun Liu, Wenjie Li
Abstract: Large Language Models (LLMs) are being used more and more extensively for automated evaluation in various scenarios. Previous studies have attempted to fine-tune open-source LLMs to replicate the evaluation explanations and judgments of powerful proprietary models, such as GPT-4. However, these methods are largely limited to text-based analyses under predefined general criteria, resulting in reduced adaptability for unseen instructions and demonstrating instability in evaluating adherence to quantitative and structural constraints. To address these limitations, we propose a novel evaluation framework, ARJudge, that adaptively formulates evaluation criteria and synthesizes both text-based and code-driven analyses to evaluate LLM responses. ARJudge consists of two components: a fine-tuned Analyzer that generates multi-faceted evaluation analyses and a tuning-free Refiner that combines and refines all analyses to make the final judgment. We construct a Composite Analysis Corpus that integrates tasks for evaluation criteria generation alongside text-based and code-driven analysis generation to train the Analyzer. Our results demonstrate that ARJudge outperforms existing fine-tuned evaluators in effectiveness and robustness. Furthermore, it demonstrates the importance of multi-faceted evaluation and code-driven analyses in enhancing evaluation capabilities.
Authors: Lars Benedikt Kaesberg, Jonas Becker, Jan Philip Wahle, Terry Ruas, Bela Gipp
Abstract: Much of the success of multi-agent debates depends on carefully choosing the right parameters. The decision-making protocol stands out as it can highly impact final model answers, depending on how decisions are reached. Systematic comparison of decision protocols is difficult because many studies alter multiple discussion parameters beyond the protocol. So far, it has been largely unknown how decision-making influences different tasks. This work systematically evaluates the impact of seven decision protocols (e.g., majority voting, unanimity consensus). We change only one variable at a time - the decision protocol - to analyze how different methods affect the collaboration between agents and measure differences in knowledge and reasoning tasks. Our results show that voting protocols improve performance by 13.2% in reasoning tasks and consensus protocols by 2.8% in knowledge tasks compared to other decision protocols. Increasing the number of agents improves performance, while more discussion rounds before voting reduce it. To improve decision-making by increasing answer diversity, we propose two new methods, All-Agents Drafting (AAD) and Collective Improvement (CI). Our methods improve task performance by up to 3.3% with AAD and up to 7.4% with CI. This work demonstrates the importance of decision-making in multi-agent debates beyond scaling.
Authors: Michael Y. Hu, Jackson Petty, Chuan Shi, William Merrill, Tal Linzen
Abstract: Pretraining language models on formal language can improve their acquisition of natural language. Which features of the formal language impart an inductive bias that leads to effective transfer? Drawing on insights from linguistics and complexity theory, we hypothesize that effective transfer occurs when two conditions are met: the formal language should capture the dependency structures present in natural language, and it should remain within the computational limitations of the model architecture. We experiment with pre-pretraining (training on formal language before natural languages) on transformers and find that formal languages capturing hierarchical dependencies indeed enable language models to achieve lower loss on natural language and better linguistic generalization compared to other formal languages. We also find modest support for the hypothesis that the formal language should fall within the computational limitations of the architecture. Strikingly, pre-pretraining reduces loss more efficiently than training on a matched amount of natural language. For a 1B-parameter language model trained on roughly 1.6B tokens of natural language, pre-pretraining achieves the same loss and better linguistic generalization with a 33% smaller token budget. Finally, we also give mechanistic evidence of transfer from formal to natural language: attention heads acquired during pre-pretraining remain crucial for the model's performance on syntactic evaluations.
Authors: Christoph Benzm\"uller
Abstract: Deep and shallow embeddings of non-classical logics in classical higher-order logic have been explored, implemented, and used in various reasoning tools in recent years. This paper presents a method for the simultaneous deployment of deep and shallow embeddings of various degrees in classical higher-order logic. This enables flexible, interactive and automated theorem proving and counterexample finding at meta and object level, as well as automated faithfulness proofs between these logic embeddings. The method is beneficial for logic education, research and application and is illustrated here using a simple propositional modal logic. However, this approach is conceptual in nature and not limited to this simple logical context.
Authors: Mohammadnavid Ghader, Saeed Reza Kheradpisheh, Bahar Farahani, Mahmood Fazlali
Abstract: Spiking Neural Networks (SNNs) offer a biologically inspired computational paradigm that emulates neuronal activity through discrete spike-based processing. Despite their advantages, training SNNs with traditional backpropagation (BP) remains challenging due to computational inefficiencies and a lack of biological plausibility. This study explores the Forward-Forward (FF) algorithm as an alternative learning framework for SNNs. Unlike backpropagation, which relies on forward and backward passes, the FF algorithm employs two forward passes, enabling layer-wise localized learning, enhanced computational efficiency, and improved compatibility with neuromorphic hardware. We introduce an FF-based SNN training framework and evaluate its performance across both non-spiking (MNIST, Fashion-MNIST, Kuzushiji-MNIST) and spiking (Neuro-MNIST, SHD) datasets. Experimental results demonstrate that our model surpasses existing FF-based SNNs on evaluated static datasets with a much lighter architecture while achieving accuracy comparable to state-of-the-art backpropagation-trained SNNs. On more complex spiking tasks such as SHD, our approach outperforms other SNN models and remains competitive with leading backpropagation-trained SNNs. These findings highlight the FF algorithm's potential to advance SNN training methodologies by addressing some key limitations of backpropagation.
Authors: Luise Ge, Michael Lanier, Anindya Sarkar, Bengisu Guresti, Chongjie Zhang, Yevgeniy Vorobeychik
Abstract: Many dynamic decision problems, such as robotic control, involve a series of tasks, many of which are unknown at training time. Typical approaches for these problems, such as multi-task and meta reinforcement learning, do not generalize well when the tasks are diverse. On the other hand, approaches that aim to tackle task diversity, such as using task embedding as policy context and task clustering, typically lack performance guarantees and require a large number of training tasks. To address these challenges, we propose a novel approach for learning a policy committee that includes at least one near-optimal policy with high probability for tasks encountered during execution. While we show that this problem is in general inapproximable, we present two practical algorithmic solutions. The first yields provable approximation and task sample complexity guarantees when tasks are low-dimensional (the best we can do due to inapproximability), whereas the second is a general and practical gradient-based approach. In addition, we provide a provable sample complexity bound for few-shot learning. Our experiments on MuJoCo and Meta-World show that the proposed approach outperforms state-of-the-art multi-task, meta-, and task clustering baselines in training, generalization, and few-shot learning, often by a large margin. Our code is available at https://github.com/CERL-WUSTL/PACMAN.
Authors: Nuria Alina Chandra, Ryan Murtfeldt, Lin Qiu, Arnab Karmakar, Hannah Lee, Emmanuel Tanumihardja, Kevin Farhat, Ben Caffee, Sejin Paik, Changyeon Lee, Jongwook Choi, Aerin Kim, Oren Etzioni
Abstract: In the age of increasingly realistic generative AI, robust deepfake detection is essential for mitigating fraud and disinformation. While many deepfake detectors report high accuracy on academic datasets, we show that these academic benchmarks are out of date and not representative of real-world deepfakes. We introduce Deepfake-Eval-2024, a new deepfake detection benchmark consisting of in-the-wild deepfakes collected from social media and deepfake detection platform users in 2024. Deepfake-Eval-2024 consists of 45 hours of videos, 56.5 hours of audio, and 1,975 images, encompassing the latest manipulation technologies. The benchmark contains diverse media content from 88 different websites in 52 different languages. We find that the performance of open-source state-of-the-art deepfake detection models drops precipitously when evaluated on Deepfake-Eval-2024, with AUC decreasing by 50% for video, 48% for audio, and 45% for image models compared to previous benchmarks. We also evaluate commercial deepfake detection models and models finetuned on Deepfake-Eval-2024, and find that they have superior performance to off-the-shelf open-source models, but do not yet reach the accuracy of deepfake forensic analysts. The dataset is available at https://github.com/nuriachandra/Deepfake-Eval-2024.
Authors: Ruida Wang, Rui Pan, Yuxin Li, Jipeng Zhang, Yizhen Jia, Shizhe Diao, Renjie Pi, Junjie Hu, Tong Zhang
Abstract: Solving mathematical problems using computer-verifiable languages like Lean has significantly impacted the mathematical and computer science communities. State-of-the-art methods utilize a single Large Language Model (LLM) to generate complete proof or perform tree search, but they fail to balance these tasks. We propose **MA-LoT**: *Model-CollAboration Lean-based Long Chain-of-Thought*, a comprehensive framework for Lean4 theorem proving to solve this issue. It separates the cognition tasks of general NL for whole-proof generation and error analysis for proof correction using the model-collaboration method. We achieve this by structured interaction of the LLM and Lean4 verifier in Long CoT. To implement the framework, we propose the novel *LoT-Transfer Learning* training-inference pipeline, which enables the Long CoT thinking capability to LLMs without special data annotation. Extensive experiment shows that our framework achieves a **61.07%** accuracy rate on the Lean4 version of the MiniF2F-Test dataset, largely outperforming DeepSeek-V3 (33.61%), single-model tree search (InternLM-Step-Prover, 50.70%), and whole-proof generation (Godel-Prover, 55.33%) baselines. Furthermore, our findings highlight the potential of combining Long CoT with formal verification for a more insightful generation in a broader perspective.
Authors: Ruohao Guo, Wei Xu, Alan Ritter
Abstract: As Large Language Models (LLMs) are widely deployed in diverse scenarios, the extent to which they could tacitly spread misinformation emerges as a critical safety concern. Current research primarily evaluates LLMs on explicit false statements, overlooking how misinformation often manifests subtly as unchallenged premises in real-world interactions. We curated EchoMist, the first comprehensive benchmark for implicit misinformation, where false assumptions are embedded in the query to LLMs. EchoMist targets circulated, harmful, and ever-evolving implicit misinformation from diverse sources, including realistic human-AI conversations and social media interactions. Through extensive empirical studies on 15 state-of-the-art LLMs, we find that current models perform alarmingly poorly on this task, often failing to detect false premises and generating counterfactual explanations. We also investigate two mitigation methods, i.e., Self-Alert and RAG, to enhance LLMs' capability to counter implicit misinformation. Our findings indicate that EchoMist remains a persistent challenge and underscore the critical need to safeguard against the risk of implicit misinformation.
Authors: Charaka Vinayak Kumar, Ashok Urlana, Gopichand Kanumolu, Bala Mallikarjunarao Garlapati, Pruthwik Mishra
Abstract: Advancements in Large Language Models (LLMs) have increased the performance of different natural language understanding as well as generation tasks. Although LLMs have breached the state-of-the-art performance in various tasks, they often reflect different forms of bias present in the training data. In the light of this perceived limitation, we provide a unified evaluation of benchmarks using a set of representative small and medium-sized LLMs that cover different forms of biases starting from physical characteristics to socio-economic categories. Moreover, we propose five prompting approaches to carry out the bias detection task across different aspects of bias. Further, we formulate three research questions to gain valuable insight in detecting biases in LLMs using different approaches and evaluation metrics across benchmarks. The results indicate that each of the selected LLMs suffer from one or the other form of bias with the Phi-3.5B model being the least biased. Finally, we conclude the paper with the identification of key challenges and possible future directions.
Authors: Hao Yin, Guangzong Si, Zilei Wang
Abstract: Contrastive decoding strategies are widely used to mitigate object hallucinations in multimodal large language models (MLLMs). By reducing over-reliance on language priors, these strategies ensure that generated content remains closely grounded in visual inputs, producing contextually accurate outputs. Since contrastive decoding requires no additional training or external tools, it offers both computational efficiency and versatility, making it highly attractive. However, these methods present two main limitations: (1) bluntly suppressing language priors can compromise coherence and accuracy of generated content, and (2) processing contrastive inputs adds computational load, significantly slowing inference speed. To address these challenges, we propose Visual Amplification Fusion (VAF), a plug-and-play technique that enhances attention to visual signals within the model's middle layers, where modality fusion predominantly occurs. This approach enables more effective capture of visual features, reducing the model's bias toward language modality. Experimental results demonstrate that VAF significantly reduces hallucinations across various MLLMs without affecting inference speed, while maintaining coherence and accuracy in generated outputs.
Authors: Nobuhiko Wakai, Satoshi Sato, Yasunori Ishii, Takayoshi Yamashita
Abstract: Person detection methods are used widely in applications including visual surveillance, pedestrian detection, and robotics. However, accurate detection of persons from overhead fisheye images remains an open challenge because of factors including person rotation and small-sized persons. To address the person rotation problem, we convert the fisheye images into panoramic images. For smaller people, we focused on the geometry of the panoramas. Conventional detection methods tend to focus on larger people because these larger people yield large significant areas for feature maps. In equirectangular panoramic images, we find that a person's height decreases linearly near the top of the images. Using this finding, we leverage the significance values and aggregate tokens that are sorted based on these values to balance the significant areas. In this leveraging process, we introduce panoramic distortion-aware tokenization. This tokenization procedure divides a panoramic image using self-similarity figures that enable determination of optimal divisions without gaps, and we leverage the maximum significant values in each tile of token groups to preserve the significant areas of smaller people. To achieve higher detection accuracy, we propose a person detection and localization method that combines panoramic-image remapping and the tokenization procedure. Extensive experiments demonstrated that our method outperforms conventional methods when applied to large-scale datasets.
Authors: Nikhil Abhyankar, Parshin Shojaee, Chandan K. Reddy
Abstract: Automated feature engineering plays a critical role in improving predictive model performance for tabular learning tasks. Traditional automated feature engineering methods are limited by their reliance on pre-defined transformations within fixed, manually designed search spaces, often neglecting domain knowledge. Recent advances using Large Language Models (LLMs) have enabled the integration of domain knowledge into the feature engineering process. However, existing LLM-based approaches use direct prompting or rely solely on validation scores for feature selection, failing to leverage insights from prior feature discovery experiments or establish meaningful reasoning between feature generation and data-driven performance. To address these challenges, we propose LLM-FE, a novel framework that combines evolutionary search with the domain knowledge and reasoning capabilities of LLMs to automatically discover effective features for tabular learning tasks. LLM-FE formulates feature engineering as a program search problem, where LLMs propose new feature transformation programs iteratively, and data-driven feedback guides the search process. Our results demonstrate that LLM-FE consistently outperforms state-of-the-art baselines, significantly enhancing the performance of tabular prediction models across diverse classification and regression benchmarks.
Authors: Dilshod Nematov, Mirabbos Hojamberdiev
Abstract: The rapid advancement of machine learning and artificial intelligence (AI)-driven techniques is revolutionizing materials discovery, property prediction, and material design by minimizing human intervention and accelerating scientific progress. This review provides a comprehensive overview of smart, machine learning (ML)-driven approaches, emphasizing their role in predicting material properties, discovering novel compounds, and optimizing material structures. Key methodologies ranging from deep learning, graph neural networks, and Bayesian optimization to automated generative models, such as generative adversarial networks (GANs) and variational autoencoders (VAEs) enable the autonomous design of materials with tailored functionalities. By leveraging AutoML frameworks (e.g., AutoGluon, TPOT, and H2O.ai), researchers can automate the model selection, hyperparameter tuning, and feature engineering, significantly improving the efficiency of materials informatics. Furthermore, the integration of AI-driven robotic laboratories and high-throughput computing has established a fully automated pipeline for rapid synthesis and experimental validation, drastically reducing the time and cost of material discovery. This review highlights real-world applications of automated ML-driven approaches in predicting mechanical, thermal, electrical, and optical properties of materials, demonstrating successful cases in superconductors, catalysts, photovoltaics, and energy storage systems. We also address key challenges, such as data quality, interpretability, and the integration of AutoML with quantum computing, which are essential for future advancements. Ultimately, the synergy between AI, automated experimentation, and computational modeling transforms the way the materials are discovered, optimized, and designed, paving the way for next-generation innovations in energy, electronics, and nanotechnology.
Authors: Qi Wu, Quanlong Zheng, Yanhao Zhang, Junlin Xie, Jinguo Luo, Kuo Wang, Peng Liu, Qingsong Xie, Ru Zhen, Zhenyu Yang, Haonan Lu
Abstract: With the rapid development of multimodal models, the demand for assessing video understanding capabilities has been steadily increasing. However, existing benchmarks for evaluating video understanding exhibit significant limitations in coverage, task diversity, and scene adaptability. These shortcomings hinder the accurate assessment of models' comprehensive video understanding capabilities. To tackle this challenge, we propose a hierarchical and holistic video understanding (H2VU) benchmark designed to evaluate both general video and online streaming video comprehension. This benchmark contributes three key features: Extended video duration: Spanning videos from brief 3-second clips to comprehensive 1.5-hour recordings, thereby bridging the temporal gaps found in current benchmarks. Comprehensive assessment tasks: Beyond traditional perceptual and reasoning tasks, we have introduced modules for countercommonsense comprehension and trajectory state tracking. These additions test the models' deep understanding capabilities beyond mere prior knowledge. Enriched video data: To keep pace with the rapid evolution of current AI agents, we have expanded first-person streaming video datasets. This expansion allows for the exploration of multimodal models' performance in understanding streaming videos from a first-person perspective. Extensive results from H2VU reveal that existing multimodal large language models (MLLMs) possess substantial potential for improvement in our newly proposed evaluation tasks. We expect that H2VU will facilitate advancements in video understanding research by offering a comprehensive and in-depth analysis of MLLMs.
Authors: Siqing Song, Chuang Wang, Ruiqi Wang, Yi Yang, Xuyao Zhang
Abstract: Quantizing large language models (LLMs) to 1-bit precision significantly reduces computational costs, but existing quantization techniques suffer from noticeable performance degradation when using weight and activation precisions below 4 bits (W4A4). In this paper, we propose a post-training quantization framework with W(1+1)A(1*4) configuration, where weights are quantized to 1 bit with an additional 1 bit for fine-grain grouping and activations are quantized to 1 bit with a 4-fold increase in the number of channels. For weight quantization, we propose utilizing Hessian-aware fine-grained grouping along with an EM-based quantization scheme. For activation quantization, we decompose INT4-quantized activations into a 4 * INT1 format equivalently and simultaneously smooth the scaling factors based on quantization errors, which further reduces the quantization errors in activations. Our method surpasses state-of-the-art (SOTA) LLM quantization baselines on W2A4 across multiple tasks, pushing the boundaries of existing LLM quantization methods toward fully binarized models. Code is available at https://github.com/JimmyCrave/LLM-PTQ-binarization.
Authors: Wenyuan Zhang, Shuaiyi Nie, Xinghua Zhang, Zefeng Zhang, Tingwen Liu
Abstract: We introduce S1-Bench, a novel benchmark designed to evaluate the performance of Large Reasoning Models (LRMs) on simple tasks that favor intuitive system 1 thinking rather than deliberative system 2 reasoning. While LRMs have achieved significant breakthroughs in complex reasoning tasks through explicit chains of thought, their heavy reliance on system 2 thinking may limit their system 1 thinking capabilities. However, there is a lack of an appropriate benchmark for evaluating LRM's system 1 thinking capabilities. To fill this gap, S1-Bench introduces a suite of simple, diverse, and natural questions across multiple domains and languages, specifically designed to assess LRMs' performance on questions more suitable for system 1 . We conduct extensive evaluations across 28 LRMs, revealing their inefficiency, inadequate accuracy, and limited robustness when handling simple questions. Additionally, we observe a gap between their difficulty perception and generation length. Overall, this work paves the way toward dual-system compatibility in the development of LRMs.
Authors: Zhouhao Sun, Xiao Ding, Li Du, Yunpeng Xu, Yixuan Ma, Yang Zhao, Bing Qin, Ting Liu
Abstract: Despite significant progress, recent studies indicate that current large language models (LLMs) may still capture dataset biases and utilize them during inference, leading to the poor generalizability of LLMs. However, due to the diversity of dataset biases and the insufficient nature of bias suppression based on in-context learning, the effectiveness of previous prior knowledge-based debiasing methods and in-context learning based automatic debiasing methods is limited. To address these challenges, we explore the combination of causal mechanisms with information theory and propose an information gain-guided causal intervention debiasing (ICD) framework. To eliminate biases within the instruction-tuning dataset, it is essential to ensure that these biases do not provide any additional information to predict the answers, i.e., the information gain of these biases for predicting the answers needs to be 0. Under this guidance, this framework utilizes a causal intervention-based data rewriting method to automatically and autonomously balance the distribution of instruction-tuning dataset for reducing the information gain. Subsequently, it employs a standard supervised fine-tuning process to train LLMs on the debiased dataset. Experimental results show that ICD can effectively debias LLM to improve its generalizability across different tasks.
Authors: Paul Ghanem, Owen Howell, Michael Potter, Pau Closas, Alireza Ramezani, Deniz Erdogmus, Tales Imbiriba
Abstract: Inferring an adversary's goals from exhibited behavior is crucial for counterplanning and non-cooperative multi-agent systems in domains like cybersecurity, military, and strategy games. Deep Inverse Reinforcement Learning (IRL) methods based on maximum entropy principles show promise in recovering adversaries' goals but are typically offline, require large batch sizes with gradient descent, and rely on first-order updates, limiting their applicability in real-time scenarios. We propose an online Recursive Deep Inverse Reinforcement Learning (RDIRL) approach to recover the cost function governing the adversary actions and goals. Specifically, we minimize an upper bound on the standard Guided Cost Learning (GCL) objective using sequential second-order Newton updates, akin to the Extended Kalman Filter (EKF), leading to a fast (in terms of convergence) learning algorithm. We demonstrate that RDIRL is able to recover cost and reward functions of expert agents in standard and adversarial benchmark tasks. Experiments on benchmark tasks show that our proposed approach outperforms several leading IRL algorithms.
Authors: Zilin Huang, Zihao Sheng, Zhengyang Wan, Yansong Qu, Yuhao Luo, Boyue Wang, Pei Li, Yen-Jung Chen, Jiancong Chen, Keke Long, Jiayi Meng, Yue Leng, Sikai Chen
Abstract: Recent advances in autonomous system simulation platforms have significantly enhanced the safe and scalable testing of driving policies. However, existing simulators do not yet fully meet the needs of future transportation research-particularly in enabling effective human-AI collaboration and modeling socially-aware driving agents. This paper introduces Sky-Drive, a novel distributed multi-agent simulation platform that addresses these limitations through four key innovations: (a) a distributed architecture for synchronized simulation across multiple terminals; (b) a multi-modal human-in-the-loop framework integrating diverse sensors to collect rich behavioral data; (c) a human-AI collaboration mechanism supporting continuous and adaptive knowledge exchange; and (d) a digital twin framework for constructing high-fidelity virtual replicas of real-world transportation environments. Sky-Drive supports diverse applications such as autonomous vehicle-human road users interaction modeling, human-in-the-loop training, socially-aware reinforcement learning, personalized driving development, and customized scenario generation. Future extensions will incorporate foundation models for context-aware decision support and hardware-in-the-loop testing for real-world validation. By bridging scenario generation, data collection, algorithm training, and hardware integration, Sky-Drive has the potential to become a foundational platform for the next generation of human-centered and socially-aware autonomous transportation systems research. The demo video and code are available at:https://sky-lab-uw.github.io/Sky-Drive-website/
Authors: Wenjun Yang, Eyhab Al-Masri
Abstract: This paper introduces a conversational interface system that enables participatory design of differentially private AI systems in public sector applications. Addressing the challenge of balancing mathematical privacy guarantees with democratic accountability, we propose three key contributions: (1) an adaptive $\epsilon$-selection protocol leveraging TOPSIS multi-criteria decision analysis to align citizen preferences with differential privacy (DP) parameters, (2) an explainable noise-injection framework featuring real-time Mean Absolute Error (MAE) visualizations and GPT-4-powered impact analysis, and (3) an integrated legal-compliance mechanism that dynamically modulates privacy budgets based on evolving regulatory constraints. Our results advance participatory AI practices by demonstrating how conversational interfaces can enhance public engagement in algorithmic privacy mechanisms, ensuring that privacy-preserving AI in public sector governance remains both mathematically robust and democratically accountable.
Authors: Xinyu Li, Qi Yao, Yuanda Wang
Abstract: Garment sewing patterns are fundamental design elements that bridge the gap between design concepts and practical manufacturing. The generative modeling of sewing patterns is crucial for creating diversified garments. However, existing approaches are limited either by reliance on a single input modality or by suboptimal generation efficiency. In this work, we present GarmentDiffusion, a new generative model capable of producing centimeter-precise, vectorized 3D sewing patterns from multimodal inputs (text, image, and incomplete sewing pattern). Our method efficiently encodes 3D sewing pattern parameters into compact edge token representations, achieving a sequence length that is 10 times shorter than that of the autoregressive SewingGPT in DressCode. By employing a diffusion transformer, we simultaneously denoise all edge tokens along the temporal axis, while maintaining a constant number of denoising steps regardless of dataset-specific edge and panel statistics. With all combination of designs of our model, the sewing pattern generation speed is accelerated by 100 times compared to SewingGPT. We achieve new state-of-the-art results on DressCodeData, as well as on the largest sewing pattern dataset, namely GarmentCodeData. The project website is available at https://shenfu-research.github.io/Garment-Diffusion/.
Authors: Dongxiu Liu, Haoyi Niu, Zhihao Wang, Jinliang Zheng, Yinan Zheng, Zhonghong Ou, Jianming Hu, Jianxiong Li, Xianyuan Zhan
Abstract: Current robotic planning methods often rely on predicting multi-frame images with full pixel details. While this fine-grained approach can serve as a generic world model, it introduces two significant challenges for downstream policy learning: substantial computational costs that hinder real-time deployment, and accumulated inaccuracies that can mislead action extraction. Planning with coarse-grained subgoals partially alleviates efficiency issues. However, their forward planning schemes can still result in off-task predictions due to accumulation errors, leading to misalignment with long-term goals. This raises a critical question: Can robotic planning be both efficient and accurate enough for real-time control in long-horizon, multi-stage tasks? To address this, we propose a Latent Space Backward Planning scheme (LBP), which begins by grounding the task into final latent goals, followed by recursively predicting intermediate subgoals closer to the current state. The grounded final goal enables backward subgoal planning to always remain aware of task completion, facilitating on-task prediction along the entire planning horizon. The subgoal-conditioned policy incorporates a learnable token to summarize the subgoal sequences and determines how each subgoal guides action extraction. Through extensive simulation and real-robot long-horizon experiments, we show that LBP outperforms existing fine-grained and forward planning methods, achieving SOTA performance. Project Page: https://lbp-authors.github.io
Authors: Thanh Son Nguyen, Van Thanh Nguyen, Dang Minh Duc Nguyen
Abstract: Time series forecasting has attracted significant attention, leading to the de-velopment of a wide range of approaches, from traditional statistical meth-ods to advanced deep learning models. Among them, the Auto-Regressive Integrated Moving Average (ARIMA) model remains a widely adopted linear technique due to its effectiveness in modeling temporal dependencies in economic, industrial, and social data. On the other hand, polynomial classifi-ers offer a robust framework for capturing non-linear relationships and have demonstrated competitive performance in domains such as stock price pre-diction. In this study, we propose a hybrid forecasting approach that inte-grates the ARIMA model with a polynomial classifier to leverage the com-plementary strengths of both models. The hybrid method is evaluated on multiple real-world time series datasets spanning diverse domains. Perfor-mance is assessed based on forecasting accuracy and computational effi-ciency. Experimental results reveal that the proposed hybrid model consist-ently outperforms the individual models in terms of prediction accuracy, al-beit with a modest increase in execution time.
Authors: Yifu Yuan, Haiqin Cui, Yibin Chen, Zibin Dong, Fei Ni, Longxin Kou, Jinyi Liu, Pengyi Li, Yan Zheng, Jianye Hao
Abstract: Achieving generalization in robotic manipulation remains a critical challenge, particularly for unseen scenarios and novel tasks. Current Vision-Language-Action (VLA) models, while building on top of general Vision-Language Models (VLMs), still fall short of achieving robust zero-shot performance due to the scarcity and heterogeneity prevalent in embodied datasets. To address these limitations, we propose FSD (From Seeing to Doing), a novel vision-language model that generates intermediate representations through spatial relationship reasoning, providing fine-grained guidance for robotic manipulation. Our approach combines a hierarchical data pipeline for training with a self-consistency mechanism that aligns spatial coordinates with visual signals. Through extensive experiments, we comprehensively validated FSD's capabilities in both "seeing" and "doing," achieving outstanding performance across 8 benchmarks for general spatial reasoning and embodied reference abilities, as well as on our proposed more challenging benchmark VABench. We also verified zero-shot capabilities in robot manipulation, demonstrating significant performance improvements over baseline methods in both SimplerEnv and real robot settings. Experimental results show that FSD achieves 40.6% success rate in SimplerEnv and 72% success rate across 8 real-world tasks, outperforming the strongest baseline by 30%.
Authors: Falong Fan, Xi Li
Abstract: Multi-agent systems leverage advanced AI models as autonomous agents that interact, cooperate, or compete to complete complex tasks across applications such as robotics and traffic management. Despite their growing importance, safety in multi-agent systems remains largely underexplored, with most research focusing on single AI models rather than interacting agents. This work investigates backdoor vulnerabilities in multi-agent systems and proposes a defense mechanism based on agent interactions. By leveraging reasoning abilities, each agent evaluates responses from others to detect illogical reasoning processes, which indicate poisoned agents. Experiments on LLM-based multi-agent systems, including ChatGPT series and Llama 3, demonstrate the effectiveness of the proposed method, achieving high accuracy in identifying poisoned agents while minimizing false positives on clean agents. We believe this work provides insights into multi-agent system safety and contributes to the development of robust, trustworthy AI interactions.
Authors: Yufei Xiang, Yiqun Shen, Yeqin Zhang, Cam-Tu Nguyen
Abstract: Large Language Models (LLMs) possess extensive knowledge and commonsense reasoning capabilities, making them valuable for creating powerful agents. However, existing LLM agent frameworks have not fully utilized past experiences for improvement. This work introduces a new LLM-based agent framework called Retrospex, which addresses this challenge by analyzing past experiences in depth. Unlike previous approaches, Retrospex does not directly integrate experiences into the LLM's context. Instead, it combines the LLM's action likelihood with action values estimated by a Reinforcement Learning (RL) Critic, which is trained on past experiences through an offline ''retrospection'' process. Additionally, Retrospex employs a dynamic action rescoring mechanism that increases the importance of experience-based values for tasks that require more interaction with the environment. We evaluate Retrospex in ScienceWorld, ALFWorld and Webshop environments, demonstrating its advantages over strong, contemporary baselines.
Authors: Ranit Karmakar, Simon F. N{\o}rrelykke
Abstract: Segmentation evaluation metrics traditionally rely on binary decision logic: predictions are either correct or incorrect, based on rigid IoU thresholds. Detection--based metrics such as F1 and mAP determine correctness at the object level using fixed overlap cutoffs, while overlap--based metrics like Intersection over Union (IoU) and Dice operate at the pixel level, often overlooking instance--level structure. Panoptic Quality (PQ) attempts to unify detection and segmentation assessment, but it remains dependent on hard-threshold matching--treating predictions below the threshold as entirely incorrect. This binary framing obscures important distinctions between qualitatively different errors and fails to reward gradual model improvements. We propose SoftPQ, a flexible and interpretable instance segmentation metric that redefines evaluation as a graded continuum rather than a binary classification. SoftPQ introduces tunable upper and lower IoU thresholds to define a partial matching region and applies a sublinear penalty function to ambiguous or fragmented predictions. These extensions allow SoftPQ to exhibit smoother score behavior, greater robustness to structural segmentation errors, and more informative feedback for model development and evaluation. Through controlled perturbation experiments, we show that SoftPQ captures meaningful differences in segmentation quality that existing metrics overlook, making it a practical and principled alternative for both benchmarking and iterative model refinement.
Authors: Kun Huang, Weikai Xu, Yuxuan Liu, Quandong Wang, Pengzhi Gao, Wei Liu, Jian Luan, Bin Wang, Bo An
Abstract: The Chain of Action-Planning Thoughts (CoaT) paradigm has been shown to improve the reasoning performance of VLM-based mobile agents in GUI tasks. However, the scarcity of diverse CoaT trajectories limits the expressiveness and generalization ability of such agents. While self-training is commonly employed to address data scarcity, existing approaches either overlook the correctness of intermediate reasoning steps or depend on expensive process-level annotations to construct process reward models (PRM). To address the above problems, we propose an Iterative Preference Learning (IPL) that constructs a CoaT-tree through interative sampling, scores leaf nodes using rule-based reward, and backpropagates feedback to derive Thinking-level Direct Preference Optimization (T-DPO) pairs. To prevent overfitting during warm-up supervised fine-tuning, we further introduce a three-stage instruction evolution, which leverages GPT-4o to generate diverse Q\&A pairs based on real mobile UI screenshots, enhancing both generality and layout understanding. Experiments on three standard Mobile GUI-agent benchmarks demonstrate that our agent MobileIPL outperforms strong baselines, including continual pretraining models such as OS-ATLAS and UI-TARS. It achieves state-of-the-art performance across three standard Mobile GUI-Agents benchmarks and shows strong generalization to out-of-domain scenarios.
Authors: Qi Feng
Abstract: Video-based spatial cognition is vital for robotics and embodied AI but challenges current Vision-Language Models (VLMs). This paper makes two key contributions. First, we introduce ViCA (Visuospatial Cognitive Assistant)-322K, a diverse dataset of 322,003 QA pairs from real-world indoor videos (ARKitScenes, ScanNet, ScanNet++), offering supervision for 3D metadata-grounded queries and video-based complex reasoning. Second, we develop ViCA-7B, fine-tuned on ViCA-322K, which achieves new state-of-the-art on all eight VSI-Bench tasks, outperforming existing models, including larger ones (e.g., +26.1 on Absolute Distance). For interpretability, we present ViCA-Thinking-2.68K, a dataset with explicit reasoning chains, and fine-tune ViCA-7B to create ViCA-7B-Thinking, a model that articulates its spatial reasoning. Our work highlights the importance of targeted data and suggests paths for improved temporal-spatial modeling. We release all resources to foster research in robust visuospatial intelligence.
Authors: Qi Feng
Abstract: While Multimodal Large Language Models (MLLMs) excel at general vision-language tasks, visuospatial cognition - reasoning about spatial layouts, relations, and dynamics - remains a significant challenge. Existing models often lack the necessary architectural components and specialized training data for fine-grained spatial understanding. We introduce ViCA2 (Visuospatial Cognitive Assistant 2), a novel MLLM designed to enhance spatial reasoning. ViCA2 features a dual vision encoder architecture integrating SigLIP for semantics and Hiera for spatial structure, coupled with a token ratio control mechanism for efficiency. We also developed ViCA-322K, a new large-scale dataset with over 322,000 spatially grounded question-answer pairs for targeted instruction tuning. On the challenging VSI-Bench benchmark, our ViCA2-7B model achieves a state-of-the-art average score of 56.8, significantly surpassing larger open-source models (e.g., LLaVA-NeXT-Video-72B, 40.9) and leading proprietary models (Gemini-1.5 Pro, 45.4). This demonstrates the effectiveness of our approach in achieving strong visuospatial intelligence with a compact model. We release ViCA2, its codebase, and the ViCA-322K dataset to facilitate further research.
Authors: Taiqiang Wu, Runming Yang, Jiayi Li, Pengfei Hu, Ngai Wong, Yujiu Yang
Abstract: Large language models (LLMs) consistently benefit from further fine-tuning on various tasks. However, we observe that directly tuning the INSTRUCT (i.e., instruction tuned) models often leads to marginal improvements and even performance degeneration. Notably, paired BASE models, the foundation for these INSTRUCT variants, contain highly similar weight values (i.e., less than 2% on average for Llama 3.1 8B). Therefore, we propose a novel Shadow-FT framework to tune the INSTRUCT models by leveraging the corresponding BASE models. The key insight is to fine-tune the BASE model, and then directly graft the learned weight updates to the INSTRUCT model. Our proposed Shadow-FT introduces no additional parameters, is easy to implement, and significantly improves performance. We conduct extensive experiments on tuning mainstream LLMs, such as Qwen 3 and Llama 3 series, and evaluate them across 19 benchmarks covering coding, reasoning, and mathematical tasks. Experimental results demonstrate that Shadow-FT consistently outperforms conventional full-parameter and parameter-efficient tuning approaches. Further analyses indicate that Shadow-FT can be applied to multimodal large language models (MLLMs) and combined with direct preference optimization (DPO). Codes and weights are available at \href{https://github.com/wutaiqiang/Shadow-FT}{Github}.
Authors: Jack Chen, Fazhong Liu, Naruto Liu, Yuhan Luo, Erqu Qin, Harry Zheng, Tian Dong, Haojin Zhu, Yan Meng, Xiao Wang
Abstract: Large language models (LLMs) excel at mathematical reasoning and logical problem-solving. The current popular training paradigms primarily use supervised fine-tuning (SFT) and reinforcement learning (RL) to enhance the models' reasoning abilities. However, when using SFT or RL alone, there are respective challenges: SFT may suffer from overfitting, while RL is prone to mode collapse. The state-of-the-art methods have proposed hybrid training schemes. However, static switching faces challenges such as poor generalization across different tasks and high dependence on data quality. In response to these challenges, inspired by the curriculum learning-quiz mechanism in human reasoning cultivation, We propose SASR, a step-wise adaptive hybrid training framework that theoretically unifies SFT and RL and dynamically balances the two throughout optimization. SASR uses SFT for initial warm-up to establish basic reasoning skills, and then uses an adaptive dynamic adjustment algorithm based on gradient norm and divergence relative to the original distribution to seamlessly integrate SFT with the online RL method GRPO. By monitoring the training status of LLMs and adjusting the training process in sequence, SASR ensures a smooth transition between training schemes, maintaining core reasoning abilities while exploring different paths. Experimental results demonstrate that SASR outperforms SFT, RL, and static hybrid training methods.
Authors: Yubin Kim, Taehan Kim, Wonjune Kang, Eugene Park, Joonsik Yoon, Dongjae Lee, Xin Liu, Daniel McDuff, Hyeonhoon Lee, Cynthia Breazeal, Hae Won Park
Abstract: Vocal health plays a crucial role in peoples' lives, significantly impacting their communicative abilities and interactions. However, despite the global prevalence of voice disorders, many lack access to convenient diagnosis and treatment. This paper introduces VocalAgent, an audio large language model (LLM) to address these challenges through vocal health diagnosis. We leverage Qwen-Audio-Chat fine-tuned on three datasets collected in-situ from hospital patients, and present a multifaceted evaluation framework encompassing a safety assessment to mitigate diagnostic biases, cross-lingual performance analysis, and modality ablation studies. VocalAgent demonstrates superior accuracy on voice disorder classification compared to state-of-the-art baselines. Its LLM-based method offers a scalable solution for broader adoption of health diagnostics, while underscoring the importance of ethical and technical validation.
Authors: Soumya Rani Samineni, Durgesh Kalwar, Karthik Valmeekam, Kaya Stechly, Subbarao Kambhampati
Abstract: Reinforcement learning-based post-training of large language models (LLMs) has recently gained attention, particularly following the release of DeepSeek R1, which applied GRPO for fine-tuning. Amid the growing hype around improved reasoning abilities attributed to RL post-training, we critically examine the formulation and assumptions underlying these methods. We start by highlighting the popular structural assumptions made in modeling LLM training as a Markov Decision Process (MDP), and show how they lead to a degenerate MDP that doesn't quite need the RL/GRPO apparatus. The two critical structural assumptions include (1) making the MDP states be just a concatenation of the actions-with states becoming the context window and the actions becoming the tokens in LLMs and (2) splitting the reward of a state-action trajectory uniformly across the trajectory. Through a comprehensive analysis, we demonstrate that these simplifying assumptions make the approach effectively equivalent to an outcome-driven supervised learning. Our experiments on benchmarks including GSM8K and Countdown using Qwen-2.5 base models show that iterative supervised fine-tuning, incorporating both positive and negative samples, achieves performance comparable to GRPO-based training. We will also argue that the structural assumptions indirectly incentivize the RL to generate longer sequences of intermediate tokens-which in turn feeds into the narrative of "RL generating longer thinking traces." While RL may well be a very useful technique for improving the reasoning abilities of LLMs, our analysis shows that the simplistic structural assumptions made in modeling the underlying MDP render the popular LLM RL frameworks and their interpretations questionable.
Authors: Kaya Stechly, Karthik Valmeekam, Atharva Gundawar, Vardhan Palod, Subbarao Kambhampati
Abstract: Recent impressive results from large reasoning models have been interpreted as a triumph of Chain of Thought (CoT), and especially of the process of training on CoTs sampled from base LLMs in order to help find new reasoning patterns. In this paper, we critically examine that interpretation by investigating how the semantics of intermediate tokens-often anthropomorphized as "thoughts" or reasoning traces and which are claimed to display behaviors like backtracking, self-verification etc.-actually influence model performance. We train transformer models on formally verifiable reasoning traces and solutions, constraining both intermediate steps and final outputs to align with those of a formal solver (in our case, A* search). By constructing a formal interpreter of the semantics of our problems and intended algorithm, we systematically evaluate not only solution accuracy but also the correctness of intermediate traces, thus allowing us to evaluate whether the latter causally influences the former. We notice that, despite significant improvements on the solution-only baseline, models trained on entirely correct traces still produce invalid reasoning traces when arriving at correct solutions. To further show that trace accuracy is only loosely connected to solution accuracy, we then train models on noisy, corrupted traces which have no relation to the specific problem each is paired with, and find that not only does performance remain largely consistent with models trained on correct data, but in some cases can improve upon it and generalize more robustly on out-of-distribution tasks. These results challenge the assumption that intermediate tokens or "Chains of Thought" induce predictable reasoning behaviors and caution against anthropomorphizing such outputs or over-interpreting them (despite their mostly correct forms) as evidence of human-like or algorithmic behaviors in language models.
Authors: Ming Gao, Shilong Wu, Hang Chen, Jun Du, Chin-Hui Lee, Shinji Watanabe, Jingdong Chen, Siniscalchi Sabato Marco, Odette Scharenborg
Abstract: Meetings are a valuable yet challenging scenario for speech applications due to complex acoustic conditions. This paper summarizes the outcomes of the MISP 2025 Challenge, hosted at Interspeech 2025, which focuses on multi-modal, multi-device meeting transcription by incorporating video modality alongside audio. The tasks include Audio-Visual Speaker Diarization (AVSD), Audio-Visual Speech Recognition (AVSR), and Audio-Visual Diarization and Recognition (AVDR). We present the challenge's objectives, tasks, dataset, baseline systems, and solutions proposed by participants. The best-performing systems achieved significant improvements over the baseline: the top AVSD model achieved a Diarization Error Rate (DER) of 8.09%, improving by 7.43%; the top AVSR system achieved a Character Error Rate (CER) of 9.48%, improving by 10.62%; and the best AVDR system achieved a concatenated minimum-permutation Character Error Rate (cpCER) of 11.56%, improving by 72.49%.
Authors: Yakun Zhu, Zhongzhen Huang, Linjie Mu, Yutong Huang, Wei Nie, Jiaji Liu, Shaoting Zhang, Pengfei Liu, Xiaofan Zhang
Abstract: The emergence of groundbreaking large language models capable of performing complex reasoning tasks holds significant promise for addressing various scientific challenges, including those arising in complex clinical scenarios. To enable their safe and effective deployment in real-world healthcare settings, it is urgently necessary to benchmark the diagnostic capabilities of current models systematically. Given the limitations of existing medical benchmarks in evaluating advanced diagnostic reasoning, we present DiagnosisArena, a comprehensive and challenging benchmark designed to rigorously assess professional-level diagnostic competence. DiagnosisArena consists of 1,113 pairs of segmented patient cases and corresponding diagnoses, spanning 28 medical specialties, deriving from clinical case reports published in 10 top-tier medical journals. The benchmark is developed through a meticulous construction pipeline, involving multiple rounds of screening and review by both AI systems and human experts, with thorough checks conducted to prevent data leakage. Our study reveals that even the most advanced reasoning models, o3-mini, o1, and DeepSeek-R1, achieve only 45.82%, 31.09%, and 17.79% accuracy, respectively. This finding highlights a significant generalization bottleneck in current large language models when faced with clinical diagnostic reasoning challenges. Through DiagnosisArena, we aim to drive further advancements in AIs diagnostic reasoning capabilities, enabling more effective solutions for real-world clinical diagnostic challenges. We provide the benchmark and evaluation tools for further research and development https://github.com/SPIRAL-MED/DiagnosisArena.
Authors: Bart Kosko, Olaoluwa Adigun
Abstract: We present the new bidirectional variational autoencoder (BVAE) network architecture. The BVAE uses a single neural network both to encode and decode instead of an encoder-decoder network pair. The network encodes in the forward direction and decodes in the backward direction through the same synaptic web. Simulations compared BVAEs and ordinary VAEs on the four image tasks of image reconstruction, classification, interpolation, and generation. The image datasets included MNIST handwritten digits, Fashion-MNIST, CIFAR-10, and CelebA-64 face images. The bidirectional structure of BVAEs cut the parameter count by almost 50% and still slightly outperformed the unidirectional VAEs.
Authors: Zhouhao Sun, Zhiyuan Kan, Xiao Ding, Li Du, Yang Zhao, Bing Qin, Ting Liu
Abstract: Despite significant progress, recent studies have indicated that current large language models (LLMs) may still utilize bias during inference, leading to the poor generalizability of LLMs. Some benchmarks are proposed to investigate the generalizability of LLMs, with each piece of data typically containing one type of controlled bias. However, a single piece of data may contain multiple types of biases in practical applications. To bridge this gap, we propose a multi-bias benchmark where each piece of data contains five types of biases. The evaluations conducted on this benchmark reveal that the performance of existing LLMs and debiasing methods is unsatisfying, highlighting the challenge of eliminating multiple types of biases simultaneously. To overcome this challenge, we propose a causal effect estimation-guided multi-bias elimination method (CMBE). This method first estimates the causal effect of multiple types of biases simultaneously. Subsequently, we eliminate the causal effect of biases from the total causal effect exerted by both the semantic information and biases during inference. Experimental results show that CMBE can effectively eliminate multiple types of bias simultaneously to enhance the generalizability of LLMs.
Authors: Jannis Becktepe, Leona Hennig, Steffen Oeltze-Jafra, Marius Lindauer
Abstract: Medical Image Segmentation (MIS) includes diverse tasks, from bone to organ segmentation, each with its own challenges in finding the best segmentation model. The state-of-the-art AutoML-related MIS-framework nnU-Net automates many aspects of model configuration but remains constrained by fixed hyperparameters and heuristic design choices. As a full-AutoML framework for MIS, we propose Auto-nnU-Net, a novel nnU-Net variant enabling hyperparameter optimization (HPO), neural architecture search (NAS), and hierarchical NAS (HNAS). Additionally, we propose Regularized PriorBand to balance model accuracy with the computational resources required for training, addressing the resource constraints often faced in real-world medical settings that limit the feasibility of extensive training procedures. We evaluate our approach across diverse MIS datasets from the well-established Medical Segmentation Decathlon, analyzing the impact of AutoML techniques on segmentation performance, computational efficiency, and model design choices. The results demonstrate that our AutoML approach substantially improves the segmentation performance of nnU-Net on 6 out of 10 datasets and is on par on the other datasets while maintaining practical resource requirements. Our code is available at https://github.com/automl/AutoNNUnet.
Authors: Xiaobei Yan, Yiming Li, Zhaoxin Fan, Han Qiu, Tianwei Zhang
Abstract: Large language models (LLMs) have shown impressive capabilities across a wide range of applications, but their ever-increasing size and resource demands make them vulnerable to inference cost attacks, where attackers induce victim LLMs to generate the longest possible output content. In this paper, we revisit existing inference cost attacks and reveal that these methods can hardly produce large-scale malicious effects since they are self-targeting, where attackers are also the users and therefore have to execute attacks solely through the inputs, whose generated content will be charged by LLMs and can only directly influence themselves. Motivated by these findings, this paper introduces a new type of inference cost attacks (dubbed 'bit-flip inference cost attack') that target the victim model itself rather than its inputs. Specifically, we design a simple yet effective method (dubbed 'BitHydra') to effectively flip critical bits of model parameters. This process is guided by a loss function designed to suppress
Authors: David Rozado
Abstract: This study examines the behavior of Large Language Models (LLMs) when evaluating professional candidates based on their resumes or curricula vitae (CVs). In an experiment involving 22 leading LLMs, each model was systematically given one job description along with a pair of profession-matched CVs, one bearing a male first name, the other a female first name, and asked to select the more suitable candidate for the job. Each CV pair was presented twice, with names swapped to ensure that any observed preferences in candidate selection stemmed from gendered names cues. Despite identical professional qualifications across genders, all LLMs consistently favored female-named candidates across 70 different professions. Adding an explicit gender field (male/female) to the CVs further increased the preference for female applicants. When gendered names were replaced with gender-neutral identifiers "Candidate A" and "Candidate B", several models displayed a preference to select "Candidate A". Counterbalancing gender assignment between these gender-neutral identifiers resulted in gender parity in candidate selection. When asked to rate CVs in isolation rather than compare pairs, LLMs assigned slightly higher average scores to female CVs overall, but the effect size was negligible. Including preferred pronouns (he/him or she/her) next to a candidate's name slightly increased the odds of the candidate being selected regardless of gender. Finally, most models exhibited a substantial positional bias to select the candidate listed first in the prompt. These findings underscore the need for caution when deploying LLMs in high-stakes autonomous decision-making contexts and raise doubts about whether LLMs consistently apply principled reasoning.
Authors: Chen Shani, Dan Jurafsky, Yann LeCun, Ravid Shwartz-Ziv
Abstract: Humans organize knowledge into compact categories through semantic compression by mapping diverse instances to abstract representations while preserving meaning (e.g., robin and blue jay are both birds; most birds can fly). These concepts reflect a trade-off between expressive fidelity and representational simplicity. Large Language Models (LLMs) demonstrate remarkable linguistic abilities, yet whether their internal representations strike a human-like trade-off between compression and semantic fidelity is unclear. We introduce a novel information-theoretic framework, drawing from Rate-Distortion Theory and the Information Bottleneck principle, to quantitatively compare these strategies. Analyzing token embeddings from a diverse suite of LLMs against seminal human categorization benchmarks, we uncover key divergences. While LLMs form broad conceptual categories that align with human judgment, they struggle to capture the fine-grained semantic distinctions crucial for human understanding. More fundamentally, LLMs demonstrate a strong bias towards aggressive statistical compression, whereas human conceptual systems appear to prioritize adaptive nuance and contextual richness, even if this results in lower compressional efficiency by our measures. These findings illuminate critical differences between current AI and human cognitive architectures, guiding pathways toward LLMs with more human-aligned conceptual representations.
Authors: Cehao Yang, Xueyuan Lin, Chengjin Xu, Xuhui Jiang, Xiaojun Wu, Honghao Liu, Hui Xiong, Jian Guo
Abstract: A practical approach to activate long chain-of-thoughts reasoning ability in pre-trained large language models is to perform supervised fine-tuning on instruction datasets synthesized by strong Large Reasoning Models such as DeepSeek-R1, offering a cost-effective alternative to reinforcement learning. However, large-scale instruction sets with more than 100k samples incur significant training overhead, while effective strategies for automatic long-CoT instruction selection still remain unexplored. In this work, we propose Select2Reason, a novel and efficient instruction-tuning data selection framework for long-CoT reasoning. From the perspective of emergence of rethinking behaviors like self-correction and backtracking, we investigate common metrics that may determine the quality of long-CoT reasoning instructions. Select2Reason leverages a quantifier to estimate difficulty of question and jointly incorporates a reasoning trace length-based heuristic through a weighted scheme for ranking to prioritize high-utility examples. Empirical results on OpenR1-Math-220k demonstrate that fine-tuning LLM on only 10% of the data selected by Select2Reason achieves performance competitive with or superior to full-data tuning and open-source baseline OpenR1-Qwen-7B across three competition-level and six comprehensive mathematical benchmarks. Further experiments highlight the scalability in varying data size, efficiency during inference, and its adaptability to other instruction pools with minimal cost.
Authors: Qilin Wang
Abstract: Long-term time-series forecasting (LTSF) models are often presented as general-purpose solutions that can be applied across domains, implicitly assuming that all data is pointwise predictable. Using chaotic systems such as Lorenz-63 as a case study, we argue that geometric structure - not pointwise prediction - is the right abstraction for a dynamic-agnostic foundational model. Minimizing the Wasserstein-2 distance (W2), which captures geometric changes, and providing a spectral view of dynamics are essential for long-horizon forecasting. Our model, FRIREN (Flow-inspired Representations via Interpretable Eigen-networks), implements an augmented normalizing-flow block that embeds data into a normally distributed latent representation. It then generates a W2-efficient optimal path that can be decomposed into rotation, scaling, inverse rotation, and translation. This architecture yields locally generated, geometry-preserving predictions that are independent of the underlying dynamics, and a global spectral representation that functions as a finite Koopman operator with a small modification. This enables practitioners to identify which modes grow, decay, or oscillate, both locally and system-wide. FRIREN achieves an MSE of 11.4, MAE of 1.6, and SWD of 0.96 on Lorenz-63 in a 336-in, 336-out, dt=0.01 setting, surpassing TimeMixer (MSE 27.3, MAE 2.8, SWD 2.1). The model maintains effective prediction for 274 out of 336 steps, approximately 2.5 Lyapunov times. On Rossler (96-in, 336-out), FRIREN achieves an MSE of 0.0349, MAE of 0.0953, and SWD of 0.0170, outperforming TimeMixer's MSE of 4.3988, MAE of 0.886, and SWD of 3.2065. FRIREN is also competitive on standard LTSF datasets such as ETT and Weather. By connecting modern generative flows with classical spectral analysis, FRIREN makes long-term forecasting both accurate and interpretable, setting a new benchmark for LTSF model design.
Authors: Jiangning Zhu, Yuxing Zhou, Zheng Wang, Juntao Yao, Yima Gu, Yuhui Yuan, Shixia Liu
Abstract: Given the central role of charts in scientific, business, and communication contexts, enhancing the chart understanding capabilities of vision-language models (VLMs) has become increasingly critical. A key limitation of existing VLMs lies in their inaccurate visual grounding of infographic elements, including charts and human-recognizable objects (HROs) such as icons and images. However, chart understanding often requires identifying relevant elements and reasoning over them. To address this limitation, we introduce OrionBench, a benchmark designed to support the development of accurate object detection models for charts and HROs in infographics. It contains 26,250 real and 78,750 synthetic infographics, with over 6.9 million bounding box annotations. These annotations are created by combining the model-in-the-loop and programmatic methods. We demonstrate the usefulness of OrionBench through three applications: 1) constructing a Thinking-with-Boxes scheme to boost the chart understanding performance of VLMs, 2) comparing existing object detection models, and 3) applying the developed detection model to document layout and UI element detection.
Authors: Zhihao Du, Changfeng Gao, Yuxuan Wang, Fan Yu, Tianyu Zhao, Hao Wang, Xiang Lv, Hui Wang, Chongjia Ni, Xian Shi, Keyu An, Guanrou Yang, Yabin Li, Yanni Chen, Zhifu Gao, Qian Chen, Yue Gu, Mengzhe Chen, Yafeng Chen, Shiliang Zhang, Wen Wang, Jieping Ye
Abstract: In our prior works, we introduced a scalable streaming speech synthesis model, CosyVoice 2, which integrates a large language model (LLM) and a chunk-aware flow matching (FM) model, and achieves low-latency bi-streaming speech synthesis and human-parity quality. Despite these advancements, CosyVoice 2 exhibits limitations in language coverage, domain diversity, data volume, text formats, and post-training techniques. In this paper, we present CosyVoice 3, an improved model designed for zero-shot multilingual speech synthesis in the wild, surpassing its predecessor in content consistency, speaker similarity, and prosody naturalness. Key features of CosyVoice 3 include: 1) A novel speech tokenizer to improve prosody naturalness, developed via supervised multi-task training, including automatic speech recognition, speech emotion recognition, language identification, audio event detection, and speaker analysis. 2) A new differentiable reward model for post-training applicable not only to CosyVoice 3 but also to other LLM-based speech synthesis models. 3) Dataset Size Scaling: Training data is expanded from ten thousand hours to one million hours, encompassing 9 languages and 18 Chinese dialects across various domains and text formats. 4) Model Size Scaling: Model parameters are increased from 0.5 billion to 1.5 billion, resulting in enhanced performance on our multilingual benchmark due to the larger model capacity. These advancements contribute significantly to the progress of speech synthesis in the wild. We encourage readers to listen to the demo at https://funaudiollm.github.io/cosyvoice3.
Authors: Yuheng Lu, Qian Yu, Hongru Wang, Zeming Liu, Wei Su, Yanping Liu, Yuhang Guo, Maocheng Liang, Yunhong Wang, Haifeng Wang
Abstract: Graphical User Interface (GUI) agents, which autonomously operate on digital interfaces through natural language instructions, hold transformative potential for accessibility, automation, and user experience. A critical aspect of their functionality is grounding - the ability to map linguistic intents to visual and structural interface elements. However, existing GUI agents often struggle to adapt to the dynamic and interconnected nature of real-world digital environments, where tasks frequently span multiple platforms and applications while also being impacted by version updates. To address this, we introduce TransBench, the first benchmark designed to systematically evaluate and enhance the transferability of GUI agents across three key dimensions: cross-version transferability (adapting to version updates), cross-platform transferability (generalizing across platforms like iOS, Android, and Web), and cross-application transferability (handling tasks spanning functionally distinct apps). TransBench includes 15 app categories with diverse functionalities, capturing essential pages across versions and platforms to enable robust evaluation. Our experiments demonstrate significant improvements in grounding accuracy, showcasing the practical utility of GUI agents in dynamic, real-world environments. Our code and data will be publicly available at GitHub.
Authors: Licheng Pan, Zhichao Chen, Haoxuan Li, Guangyi Liu, Zhijian Xu, Zhaoran Liu, Hao Wang, Ying Wei
Abstract: Multi-task forecasting has become the standard approach for time-series forecasting (TSF). However, we show that it suffers from an Expressiveness Bottleneck, where predictions at different time steps share the same representation, leading to unavoidable errors even with optimal representations. To address this issue, we propose a two-stage framework: first, pre-train a foundation model for one-step-ahead prediction; then, adapt it using step-specific LoRA modules.This design enables the foundation model to handle any number of forecast steps while avoiding the expressiveness bottleneck. We further introduce the Mixture-of-LoRA (MoLA) model, which employs adaptively weighted LoRA experts to achieve partial parameter sharing across steps. This approach enhances both efficiency and forecasting performance by exploiting interdependencies between forecast steps. Experiments show that MoLA significantly improves model expressiveness and outperforms state-of-the-art time-series forecasting methods. Code is available at https://anonymous.4open.science/r/MoLA-BC92.
Authors: Yutong Chen, Jiandong Gao, Ji Wu
Abstract: R1-style Reinforcement Learning (RL) significantly enhances Large Language Models' reasoning capabilities, yet the mechanism behind rule-based RL remains unclear. We found that small-scale SFT has significant influence on RL but shows poor efficiency. To explain our observations, we propose an analytical framework and compare the efficiency of SFT and RL by measuring sample effect. Hypothetical analysis show that SFT efficiency is limited by training data. Guided by our analysis, we propose Re-distillation, a technique that fine-tunes pretrain model through small-scale distillation from the RL-trained policy. Experiments on Knight & Knave and MATH datasets demonstrate re-distillation's surprising efficiency: re-distilled models match RL performance with far fewer samples and less computation. Empirical verification shows that sample effect is a good indicator of performance improvements. As a result, on K&K dataset, our re-distilled Qwen2.5-1.5B model surpasses DeepSeek-V3-0324 with only 1K SFT samples. On MATH, Qwen2.5-1.5B fine-tuned with re-distilled 500 samples matches its instruct-tuned variant without RL. Our work explains several interesting phenomena in R1-style RL, shedding light on the mechanisms behind its empirical success. Code is available at: https://github.com/on1262/deep-reasoning
Authors: Beck LaBash, Shahriar Khushrushahi, Fabian Ruehle
Abstract: We propose a two-stage deep learning framework for the inverse design of rectangular patch antennas. Our approach leverages generative modeling to learn a latent representation of antenna frequency response curves and conditions a subsequent generative model on these responses to produce feasible antenna geometries. We further demonstrate that leveraging search and optimization techniques at test-time improves the accuracy of the generated designs and enables consideration of auxiliary objectives such as manufacturability. Our approach generalizes naturally to different design criteria, and can be easily adapted to more complex geometric design spaces.
Authors: Yuezhou Ma, Haixu Wu, Hang Zhou, Huikun Weng, Jianmin Wang, Mingsheng Long
Abstract: Physics sensing plays a central role in many scientific and engineering domains, which inherently involves two coupled tasks: reconstructing dense physical fields from sparse observations and optimizing scattered sensor placements to observe maximum information. While deep learning has made rapid advances in sparse-data reconstruction, existing methods generally omit optimization of sensor placements, leaving the mutual enhancement between reconstruction and placement on the shelf. To change this suboptimal practice, we propose PhySense, a synergistic two-stage framework that learns to jointly reconstruct physical fields and to optimize sensor placements, both aiming for accurate physics sensing. The first stage involves a flow-based generative model enhanced by cross-attention to adaptively fuse sparse observations. Leveraging the reconstruction feedback, the second stage performs sensor placement via projected gradient descent to satisfy spatial constraints. We further prove that the learning objectives of the two stages are consistent with classical variance-minimization principles, providing theoretical guarantees. Extensive experiments across three challenging benchmarks, especially a 3D geometry dataset, indicate PhySense achieves state-of-the-art physics sensing accuracy and discovers informative sensor placements previously unconsidered.
Authors: Trinity Chung, Yuchen Shen, Nathan C. L. Kong, Aran Nayebi
Abstract: Tactile sensing remains far less understood in neuroscience and less effective in artificial systems compared to more mature modalities such as vision and language. We bridge these gaps by introducing a novel Encoder-Attender-Decoder (EAD) framework to systematically explore the space of task-optimized temporal neural networks trained on realistic tactile input sequences from a customized rodent whisker-array simulator. We identify convolutional recurrent neural networks (ConvRNNs) as superior encoders to purely feedforward and state-space architectures for tactile categorization. Crucially, these ConvRNN-encoder-based EAD models achieve neural representations closely matching rodent somatosensory cortex, saturating the explainable neural variability and revealing a clear linear relationship between supervised categorization performance and neural alignment. Furthermore, contrastive self-supervised ConvRNN-encoder-based EADs, trained with tactile-specific augmentations, match supervised neural fits, serving as an ethologically-relevant, label-free proxy. For neuroscience, our findings highlight nonlinear recurrent processing as important for general-purpose tactile representations in somatosensory cortex, providing the first quantitative characterization of the underlying inductive biases in this system. For embodied AI, our results emphasize the importance of recurrent EAD architectures to handle realistic tactile inputs, along with tailored self-supervised learning methods for achieving robust tactile perception with the same type of sensors animals use to sense in unstructured environments.
Authors: Xuanhe Zhou, Junxuan He, Wei Zhou, Haodong Chen, Zirui Tang, Haoyu Zhao, Xin Tong, Guoliang Li, Youmin Chen, Jun Zhou, Zhaojun Sun, Binyuan Hui, Shuo Wang, Conghui He, Zhiyuan Liu, Jingren Zhou, Fan Wu
Abstract: The integration of large language model (LLM) and data management (DATA) is rapidly redefining both domains. In this survey, we comprehensively review the bidirectional relationships. On the one hand, DATA4LLM, spanning large-scale data processing, storage, and serving, feeds LLMs with high quality, diversity, and timeliness of data required for stages like pre-training, post-training, retrieval-augmented generation, and agentic workflows: (i) Data processing for LLMs includes scalable acquisition, deduplication, filtering, selection, domain mixing, and synthetic augmentation; (ii) Data Storage for LLMs focuses on efficient data and model formats, distributed and heterogeneous storage hierarchies, KV-cache management, and fault-tolerant checkpointing; (iii) Data serving for LLMs tackles challenges in RAG (e.g., knowledge post-processing), LLM inference (e.g., prompt compression, data provenance), and training strategies (e.g., data packing and shuffling). On the other hand, in LLM4DATA, LLMs are emerging as general-purpose engines for data management. We review recent advances in (i) data manipulation, including automatic data cleaning, integration, discovery; (ii) data analysis, covering reasoning over structured, semi-structured, and unstructured data, and (iii) system optimization (e.g., configuration tuning, query rewriting, anomaly diagnosis), powered by LLM techniques like retrieval-augmented prompting, task-specialized fine-tuning, and multi-agent collaboration.
Authors: Chen Han, Wenzhen Zheng, Xijin Tang
Abstract: The proliferation of misinformation in digital platforms reveals the limitations of traditional detection methods, which mostly rely on static classification and fail to capture the intricate process of real-world fact-checking. Despite advancements in Large Language Models (LLMs) that enhance automated reasoning, their application to misinformation detection remains hindered by issues of logical inconsistency and superficial verification. In response, we introduce Debate-to-Detect (D2D), a novel Multi-Agent Debate (MAD) framework that reformulates misinformation detection as a structured adversarial debate. Inspired by fact-checking workflows, D2D assigns domain-specific profiles to each agent and orchestrates a five-stage debate process, including Opening Statement, Rebuttal, Free Debate, Closing Statement, and Judgment. To transcend traditional binary classification, D2D introduces a multi-dimensional evaluation mechanism that assesses each claim across five distinct dimensions: Factuality, Source Reliability, Reasoning Quality, Clarity, and Ethics. Experiments with GPT-4o on two fakenews datasets demonstrate significant improvements over baseline methods, and the case study highlight D2D's capability to iteratively refine evidence while improving decision transparency, representing a substantial advancement towards robust and interpretable misinformation detection. The code will be open-sourced in a future release.
Authors: Bryan Sangwoo Kim, Jeongsol Kim, Jong Chul Ye
Abstract: Modern single-image super-resolution (SISR) models deliver photo-realistic results at the scale factors on which they are trained, but collapse when asked to magnify far beyond that regime. We address this scalability bottleneck with Chain-of-Zoom (CoZ), a model-agnostic framework that factorizes SISR into an autoregressive chain of intermediate scale-states with multi-scale-aware prompts. CoZ repeatedly re-uses a backbone SR model, decomposing the conditional probability into tractable sub-problems to achieve extreme resolutions without additional training. Because visual cues diminish at high magnifications, we augment each zoom step with multi-scale-aware text prompts generated by a vision-language model (VLM). The prompt extractor itself is fine-tuned using Generalized Reward Policy Optimization (GRPO) with a critic VLM, aligning text guidance towards human preference. Experiments show that a standard 4x diffusion SR model wrapped in CoZ attains beyond 256x enlargement with high perceptual quality and fidelity. Project Page: https://bryanswkim.github.io/chain-of-zoom/ .
Authors: Peng Xiao, Hongbo Zhao, Yijun Wang, Jianxin Lin
Abstract: Restoring real-world degraded images, such as old photographs or low-resolution images, presents a significant challenge due to the complex, mixed degradations they exhibit, such as scratches, color fading, and noise. Recent data-driven approaches have struggled with two main challenges: achieving high-fidelity restoration and providing object-level control over colorization. While diffusion models have shown promise in generating high-quality images with specific controls, they often fail to fully preserve image details during restoration. In this work, we propose an internal detail-preserving diffusion model for high-fidelity restoration of real-world degraded images. Our method utilizes a pre-trained Stable Diffusion model as a generative prior, eliminating the need to train a model from scratch. Central to our approach is the Internal Image Detail Enhancement (IIDE) technique, which directs the diffusion model to preserve essential structural and textural information while mitigating degradation effects. The process starts by mapping the input image into a latent space, where we inject the diffusion denoising process with degradation operations that simulate the effects of various degradation factors. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art models in both qualitative assessments and perceptual quantitative evaluations. Additionally, our approach supports text-guided restoration, enabling object-level colorization control that mimics the expertise of professional photo editing.
Authors: Reva Schwartz, Rumman Chowdhury, Akash Kundu, Heather Frase, Marzieh Fadaee, Tom David, Gabriella Waters, Afaf Taik, Morgan Briggs, Patrick Hall, Shomik Jain, Kyra Yee, Spencer Thomas, Sundeep Bhandari, Qinghua Lu, Matthew Holmes, Theodora Skeadas
Abstract: Conventional AI evaluation approaches concentrated within the AI stack exhibit systemic limitations for exploring, navigating and resolving the human and societal factors that play out in real world deployment such as in education, finance, healthcare, and employment sectors. AI capability evaluations can capture detail about first-order effects, such as whether immediate system outputs are accurate, or contain toxic, biased or stereotypical content, but AI's second-order effects, i.e. any long-term outcomes and consequences that may result from AI use in the real world, have become a significant area of interest as the technology becomes embedded in our daily lives. These secondary effects can include shifts in user behavior, societal, cultural and economic ramifications, workforce transformations, and long-term downstream impacts that may result from a broad and growing set of risks. This position paper argues that measuring the indirect and secondary effects of AI will require expansion beyond static, single-turn approaches conducted in silico to include testing paradigms that can capture what actually materializes when people use AI technology in context. Specifically, we describe the need for data and methods that can facilitate contextual awareness and enable downstream interpretation and decision making about AI's secondary effects, and recommend requirements for a new ecosystem.
Authors: Pradyumna Shyama Prasad, Minh Nhat Nguyen
Abstract: Can LLMs accurately adjust their confidence when facing opposition? Building on previous studies measuring calibration on static fact-based question-answering tasks, we evaluate Large Language Models (LLMs) in a dynamic, adversarial debate setting, uniquely combining two realistic factors: (a) a multi-turn format requiring models to update beliefs as new information emerges, and (b) a zero-sum structure to control for task-related uncertainty, since mutual high-confidence claims imply systematic overconfidence. We organized 60 three-round policy debates among ten state-of-the-art LLMs, with models privately rating their confidence (0-100) in winning after each round. We observed five concerning patterns: (1) Systematic overconfidence: models began debates with average initial confidence of 72.9% vs. a rational 50% baseline. (2) Confidence escalation: rather than reducing confidence as debates progressed, debaters increased their win probabilities, averaging 83% by the final round. (3) Mutual overestimation: in 61.7% of debates, both sides simultaneously claimed >=75% probability of victory, a logical impossibility. (4) Persistent self-debate bias: models debating identical copies increased confidence from 64.1% to 75.2%; even when explicitly informed their chance of winning was exactly 50%, confidence still rose (from 50.0% to 57.1%). (5) Misaligned private reasoning: models' private scratchpad thoughts sometimes differed from their public confidence ratings, raising concerns about faithfulness of chain-of-thought reasoning. These results suggest LLMs lack the ability to accurately self-assess or update their beliefs in dynamic, multi-turn tasks; a major concern as LLM outputs are deployed without careful review in assistant roles or agentic settings.
Authors: Kevin Li, Marinka Zitnik
Abstract: Offline goal-conditioned reinforcement learning methods have shown promise for reach-avoid tasks, where an agent must reach a target state while avoiding undesirable regions of the state space. Existing approaches typically encode avoid-region information into an augmented state space and cost function, which prevents flexible, dynamic specification of novel avoid-region information at evaluation time. They also rely heavily on well-designed reward and cost functions, limiting scalability to complex or poorly structured environments. We introduce RADT, a decision transformer model for offline, reward-free, goal-conditioned, avoid region-conditioned RL. RADT encodes goals and avoid regions directly as prompt tokens, allowing any number of avoid regions of arbitrary size to be specified at evaluation time. Using only suboptimal offline trajectories from a random policy, RADT learns reach-avoid behavior through a novel combination of goal and avoid-region hindsight relabeling. We benchmark RADT against 3 existing offline goal-conditioned RL models across 11 tasks, environments, and experimental settings. RADT generalizes in a zero-shot manner to out-of-distribution avoid region sizes and counts, outperforming baselines that require retraining. In one such zero-shot setting, RADT achieves 35.7% improvement in normalized cost over the best retrained baseline while maintaining high goal-reaching success. We apply RADT to cell reprogramming in biology, where it reduces visits to undesirable intermediate gene expression states during trajectories to desired target states, despite stochastic transitions and discrete, structured state dynamics.
Authors: Baichuan Li, Larry Powell, Tracy Hammond
Abstract: The quality of training data is critical to the performance of machine learning applications in domains like transportation, healthcare, and robotics. Accurate image labeling, however, often relies on time-consuming, expert-driven methods with limited feedback. This research introduces a sketch-based annotation approach supported by large language models (LLMs) to reduce technical barriers and enhance accessibility. Using a synthetic dataset, we examine how sketch recognition features relate to LLM feedback metrics, aiming to improve the reliability and interpretability of LLM-assisted labeling. We also explore how prompting strategies and sketch variations influence feedback quality. Our main contribution is a sketch-based virtual assistant that simplifies annotation for non-experts and advances LLM-driven labeling tools in terms of scalability, accessibility, and explainability.
Authors: Yejin Lee, Joonghyuk Hahn, Hyeseon Ahn, Yo-Sub Han
Abstract: Implicit hate speech detection is challenging due to its subtlety and reliance on contextual interpretation rather than explicit offensive words. Current approaches rely on contrastive learning, which are shown to be effective on distinguishing hate and non-hate sentences. Humans, however, detect implicit hate speech by first identifying specific targets within the text and subsequently interpreting how these target relate to their surrounding context. Motivated by this reasoning process, we propose AmpleHate, a novel approach designed to mirror human inference for implicit hate detection. AmpleHate identifies explicit target using a pretrained Named Entity Recognition model and capture implicit target information via [CLS] tokens. It computes attention-based relationships between explicit, implicit targets and sentence context and then, directly injects these relational vectors into the final sentence representation. This amplifies the critical signals of target-context relations for determining implicit hate. Experiments demonstrate that AmpleHate achieves state-of-the-art performance, outperforming contrastive learning baselines by an average of 82.14% and achieve faster convergence. Qualitative analyses further reveal that attention patterns produced by AmpleHate closely align with human judgement, underscoring its interpretability and robustness.
Authors: Haoyu Zhang, Wentao Zhang, Hao Miao, Xinke Jiang, Yuchen Fang, Yifan Zhang
Abstract: Spatio-Temporal Graph Neural Networks (STGNNs) have emerged as a powerful tool for modeling dynamic graph-structured data across diverse domains. However, they often fail to generalize in Spatio-Temporal Out-of-Distribution (STOOD) scenarios, where both temporal dynamics and spatial structures evolve beyond the training distribution. To address this problem, we propose an innovative Spatio-Temporal Retrieval-Augmented Pattern Learning framework,STRAP, which enhances model generalization by integrating retrieval-augmented learning into the STGNN continue learning pipeline. The core of STRAP is a compact and expressive pattern library that stores representative spatio-temporal patterns enriched with historical, structural, and semantic information, which is obtained and optimized during the training phase. During inference, STRAP retrieves relevant patterns from this library based on similarity to the current input and injects them into the model via a plug-and-play prompting mechanism. This not only strengthens spatio-temporal representations but also mitigates catastrophic forgetting. Moreover, STRAP introduces a knowledge-balancing objective to harmonize new information with retrieved knowledge. Extensive experiments across multiple real-world streaming graph datasets show that STRAP consistently outperforms state-of-the-art STGNN baselines on STOOD tasks, demonstrating its robustness, adaptability, and strong generalization capability without task-specific fine-tuning.
Authors: Daniel Ordo\~nez-Apraez, Vladimir Kosti\'c, Alek Fr\"ohlich, Vivien Brandt, Karim Lounici, Massimiliano Pontil
Abstract: In many real-world applications of regression, conditional probability estimation, and uncertainty quantification, exploiting symmetries rooted in physics or geometry can dramatically improve generalization and sample efficiency. While geometric deep learning has made significant empirical advances by incorporating group-theoretic structure, less attention has been given to statistical learning guarantees. In this paper, we introduce an equivariant representation learning framework that simultaneously addresses regression, conditional probability estimation, and uncertainty quantification while providing first-of-its-kind non-asymptotic statistical learning guarantees. Grounded in operator and group representation theory, our framework approximates the spectral decomposition of the conditional expectation operator, building representations that are both equivariant and disentangled along independent symmetry subgroups. Empirical evaluations on synthetic datasets and real-world robotics applications confirm the potential of our approach, matching or outperforming existing equivariant baselines in regression while additionally providing well-calibrated parametric uncertainty estimates.
Authors: Shuo Wang, Yun Cheng, Qingye Meng, Olga Saukh, Jiang Zhang, Jingfang Fan, Yuanting Zhang, Xingyuan Yuan, Lothar Thiele
Abstract: Air quality forecasting (AQF) is critical for public health and environmental management, yet remains challenging due to the complex interplay of emissions, meteorology, and chemical transformations. Traditional numerical models, such as CMAQ and WRF-Chem, provide physically grounded simulations but are computationally expensive and rely on uncertain emission inventories. Deep learning models, while computationally efficient, often struggle with generalization due to their lack of physical constraints. To bridge this gap, we propose PCDCNet, a surrogate model that integrates numerical modeling principles with deep learning. PCDCNet explicitly incorporates emissions, meteorological influences, and domain-informed constraints to model pollutant formation, transport, and dissipation. By combining graph-based spatial transport modeling, recurrent structures for temporal accumulation, and representation enhancement for local interactions, PCDCNet achieves state-of-the-art (SOTA) performance in 72-hour station-level PM2.5 and O3 forecasting while significantly reducing computational costs. Furthermore, our model is deployed in an online platform, providing free, real-time air quality forecasts, demonstrating its scalability and societal impact. By aligning deep learning with physical consistency, PCDCNet offers a practical and interpretable solution for AQF, enabling informed decision-making for both personal and regulatory applications.
Authors: Artem Petrov, Dmitrii Volkov
Abstract: As AI systems become increasingly capable, understanding their offensive cyber potential is critical for informed governance and responsible deployment. However, it's hard to accurately bound their capabilities, and some prior evaluations dramatically underestimated them. The art of extracting maximum task-specific performance from AIs is called "AI elicitation", and today's safety organizations typically conduct it in-house. In this paper, we explore crowdsourcing elicitation efforts as an alternative to in-house elicitation work. We host open-access AI tracks at two Capture The Flag (CTF) competitions: AI vs. Humans (400 teams) and Cyber Apocalypse (8000 teams). The AI teams achieve outstanding performance at both events, ranking top-5% and top-10% respectively for a total of \$7500 in bounties. This impressive performance suggests that open-market elicitation may offer an effective complement to in-house elicitation. We propose elicitation bounties as a practical mechanism for maintaining timely, cost-effective situational awareness of emerging AI capabilities. Another advantage of open elicitations is the option to collect human performance data at scale. Applying METR's methodology, we found that AI agents can reliably solve cyber challenges requiring one hour or less of effort from a median human CTF participant.
Authors: Christoph Schuhmann, Robert Kaczmarczyk, Gollam Rabby, Felix Friedrich, Maurice Kraus, Krishna Kalyan, Kourosh Nadi, Huu Nguyen, Kristian Kersting, S\"oren Auer
Abstract: Effective human-AI interaction relies on AI's ability to accurately perceive and interpret human emotions. Current benchmarks for vision and vision-language models are severely limited, offering a narrow emotional spectrum that overlooks nuanced states (e.g., bitterness, intoxication) and fails to distinguish subtle differences between related feelings (e.g., shame vs. embarrassment). Existing datasets also often use uncontrolled imagery with occluded faces and lack demographic diversity, risking significant bias. To address these critical gaps, we introduce EmoNet Face, a comprehensive benchmark suite. EmoNet Face features: (1) A novel 40-category emotion taxonomy, meticulously derived from foundational research to capture finer details of human emotional experiences. (2) Three large-scale, AI-generated datasets (EmoNet HQ, Binary, and Big) with explicit, full-face expressions and controlled demographic balance across ethnicity, age, and gender. (3) Rigorous, multi-expert annotations for training and high-fidelity evaluation. (4) We built EmpathicInsight-Face, a model achieving human-expert-level performance on our benchmark. The publicly released EmoNet Face suite - taxonomy, datasets, and model - provides a robust foundation for developing and evaluating AI systems with a deeper understanding of human emotions.
Authors: Ruiyi Fang, Bingheng Li, Jingyu Zhao, Ruizhi Pu, Qiuhao Zeng, Gezheng Xu, Charles Ling, Boyu Wang
Abstract: Graph Domain Adaptation (GDA) transfers knowledge from labeled source graphs to unlabeled target graphs, addressing the challenge of label scarcity. In this paper, we highlight the significance of graph homophily, a pivotal factor for graph domain alignment, which, however, has long been overlooked in existing approaches. Specifically, our analysis first reveals that homophily discrepancies exist in benchmarks. Moreover, we also show that homophily discrepancies degrade GDA performance from both empirical and theoretical aspects, which further underscores the importance of homophily alignment in GDA. Inspired by this finding, we propose a novel homophily alignment algorithm that employs mixed filters to smooth graph signals, thereby effectively capturing and mitigating homophily discrepancies between graphs. Experimental results on a variety of benchmarks verify the effectiveness of our method.
Authors: Anmol Mekala, Anirudh Atmakuru, Yixiao Song, Marzena Karpinska, Mohit Iyyer
Abstract: Large language models (LLMs) now support context windows exceeding 128K tokens, but this comes with significant memory requirements and high inference latency. Quantization can mitigate these costs, but may degrade performance. In this work, we present the first systematic evaluation of quantized LLMs on tasks with long-inputs (>64K tokens) and long-form outputs. Our evaluation spans 9.7K test examples, five quantization methods (FP8, GPTQ-int8, AWQ-int4, GPTQ-int4, BNB-nf4), and five models (Llama-3.1 8B and 70B; Qwen-2.5 7B, 32B, and 72B). We find that, on average, 8-bit quantization preserves accuracy (~0.8% drop), whereas 4-bit methods lead to substantial losses, especially for tasks involving long context inputs (drops of up to 59%). This degradation tends to worsen when the input is in a language other than English. Crucially, the effects of quantization depend heavily on the quantization method, model, and task. For instance, while Qwen-2.5 72B remains robust under BNB-nf4, Llama-3.1 70B experiences a 32% performance drop on the same task. These findings highlight the importance of a careful, task-specific evaluation before deploying quantized LLMs, particularly in long-context scenarios and with languages other than English.
Authors: Shenghai Yuan, Xianyi He, Yufan Deng, Yang Ye, Jinfa Huang, Bin Lin, Chongyang Ma, Jiebo Luo, Li Yuan
Abstract: Subject-to-Video (S2V) generation aims to create videos that faithfully incorporate reference content, providing enhanced flexibility in the production of videos. To establish the infrastructure for S2V generation, we propose OpenS2V-Nexus, consisting of (i) OpenS2V-Eval, a fine-grained benchmark, and (ii) OpenS2V-5M, a million-scale dataset. In contrast to existing S2V benchmarks inherited from VBench that focus on global and coarse-grained assessment of generated videos, OpenS2V-Eval focuses on the model's ability to generate subject-consistent videos with natural subject appearance and identity fidelity. For these purposes, OpenS2V-Eval introduces 180 prompts from seven major categories of S2V, which incorporate both real and synthetic test data. Furthermore, to accurately align human preferences with S2V benchmarks, we propose three automatic metrics, NexusScore, NaturalScore and GmeScore, to separately quantify subject consistency, naturalness, and text relevance in generated videos. Building on this, we conduct a comprehensive evaluation of 16 representative S2V models, highlighting their strengths and weaknesses across different content. Moreover, we create the first open-source large-scale S2V generation dataset OpenS2V-5M, which consists of five million high-quality 720P subject-text-video triples. Specifically, we ensure subject-information diversity in our dataset by (1) segmenting subjects and building pairing information via cross-video associations and (2) prompting GPT-Image-1 on raw frames to synthesize multi-view representations. Through OpenS2V-Nexus, we deliver a robust infrastructure to accelerate future S2V generation research.