Authors: Khurram Yamin, Gaurav Ghosal, Bryan Wilder
Abstract: Large Language Models have been shown to contain extensive world knowledge in their parameters, enabling impressive performance on many knowledge intensive tasks. However, when deployed in novel settings, LLMs often encounter situations where they must integrate parametric knowledge with new or unfamiliar information. In this work, we explore whether LLMs can combine knowledge in-context with their parametric knowledge through the lens of counterfactual reasoning. Through synthetic and real experiments in multi-hop reasoning problems, we show that LLMs generally struggle with counterfactual reasoning, often resorting to exclusively using their parametric knowledge. Moreover, we show that simple post-hoc finetuning can struggle to instill counterfactual reasoning ability -- often leading to degradation in stored parametric knowledge. Ultimately, our work reveals important limitations of current LLM's abilities to re-purpose parametric knowledge in novel settings.
Authors: Mert Cemri, Nived Rajaraman, Rishabh Tiwari, Xiaoxuan Liu, Kurt Keutzer, Ion Stoica, Kannan Ramchandran, Ahmad Beirami, Ziteng Sun
Abstract: Scaling test-time compute has driven the recent advances in the reasoning capabilities of large language models (LLMs), typically by allocating additional computation for more thorough exploration. However, increased compute often comes at the expense of higher user-facing latency, directly impacting user experience. Current test-time scaling methods primarily optimize for accuracy based on total compute resources (FLOPS), often overlooking latency constraints. To address this gap, we propose $\texttt{SPECS}$, a latency-aware test-time scaling method inspired by speculative decoding. $\texttt{SPECS}$~uses a smaller, faster model to generate candidate sequences efficiently, and evaluates these candidates using signals from both a larger target model and a dedicated reward model. We introduce new integration strategies, including reward-guided soft verification and a reward-based deferral mechanism. Empirical results on MATH500, AMC23 and OlympiadBench datasets show that $\texttt{SPECS}$~matches or surpasses beam search accuracy while reducing latency by up to $\sim$19.1\%. Our theoretical analysis shows that our algorithm converges to the solution of a KL-regularized reinforcement learning objective with increasing beam width.
Authors: Peiyuan Tang, Haojie Xin, Xiaodong Zhang, Jun Sun, Qin Xia, Zijiang Yang
Abstract: As Vision-Language Models (VLMs) demonstrate increasing capabilities across real-world applications such as code generation and chatbot assistance, ensuring their safety has become paramount. Unlike traditional Large Language Models (LLMs), VLMs face unique vulnerabilities due to their multimodal nature, allowing adversaries to modify visual or textual inputs to bypass safety guardrails and trigger the generation of harmful content. Through systematic analysis of VLM behavior under attack, we identify a novel phenomenon termed ``delayed safety awareness''. Specifically, we observe that safety-aligned VLMs may initially be compromised to produce harmful content, but eventually recognize the associated risks and attempt to self-correct. This pattern suggests that VLMs retain their underlying safety awareness but experience a temporal delay in their activation. Building on this insight, we hypothesize that VLMs' safety awareness can be proactively reactivated through carefully designed prompts. To this end, we introduce ``The Safety Reminder'', a soft prompt tuning approach that optimizes learnable prompt tokens, which are periodically injected during the text generation process to enhance safety awareness, effectively preventing harmful content generation. Additionally, our safety reminder only activates when harmful content is detected, leaving normal conversations unaffected and preserving the model's performance on benign tasks. Through comprehensive evaluation across three established safety benchmarks and one adversarial attacks, we demonstrate that our approach significantly reduces attack success rates while maintaining model utility, offering a practical solution for deploying safer VLMs in real-world applications.
Authors: Robert Graham, Edward Stevinson, Leo Richter, Alexander Chia, Joseph Miller, Joseph Isaac Bloom
Abstract: Identifying inputs that trigger specific behaviours or latent features in language models could have a wide range of safety use cases. We investigate a class of methods capable of generating targeted, linguistically fluent inputs that activate specific latent features or elicit model behaviours. We formalise this approach as context modification and present ContextBench -- a benchmark with tasks assessing core method capabilities and potential safety applications. Our evaluation framework measures both elicitation strength (activation of latent features or behaviours) and linguistic fluency, highlighting how current state-of-the-art methods struggle to balance these objectives. We enhance Evolutionary Prompt Optimisation (EPO) with LLM-assistance and diffusion model inpainting, and demonstrate that these variants achieve state-of-the-art performance in balancing elicitation effectiveness and fluency.
Authors: Jonathan Kutasov, Yuqi Sun, Paul Colognese, Teun van der Weij, Linda Petrini, Chen Bo Calvin Zhang, John Hughes, Xiang Deng, Henry Sleight, Tyler Tracy, Buck Shlegeris, Joe Benton
Abstract: As Large Language Models (LLMs) are increasingly deployed as autonomous agents in complex and long horizon settings, it is critical to evaluate their ability to sabotage users by pursuing hidden objectives. We study the ability of frontier LLMs to evade monitoring and achieve harmful hidden goals while completing a wide array of realistic tasks. We evaluate a broad range of frontier LLMs using SHADE (Subtle Harmful Agent Detection & Evaluation)-Arena, the first highly diverse agent evaluation dataset for sabotage and monitoring capabilities of LLM agents. SHADE-Arena consists of complex pairs of benign main tasks and harmful side objectives in complicated environments. Agents are evaluated on their ability to complete the side task without appearing suspicious to an LLM monitor. When measuring agent ability to (a) complete the main task, (b) complete the side task, and (c) avoid detection, we find that the best performing frontier models score 27% (Claude 3.7 Sonnet) and 15% (Gemini 2.5 Pro) as sabotage agents when overseen by Claude 3.6 Sonnet. For current frontier models, success on the side task relies heavily on having access to a hidden scratchpad that is not visible to the monitor. We also use SHADE-Arena to measure models' monitoring abilities, with the top monitor (Gemini 2.5 Pro) achieving an AUC of 0.87 at distinguishing benign and malign transcripts. We find that for now, models still struggle at sabotage due to failures in long-context main task execution. However, our measurements already demonstrate the difficulty of monitoring for subtle sabotage attempts, which we expect to only increase in the face of more complex and longer-horizon tasks.
Authors: He Zhu, Tianrui Qin, King Zhu, Heyuan Huang, Yeyi Guan, Jinxiang Xia, Yi Yao, Hanhao Li, Ningning Wang, Pai Liu, Tianhao Peng, Xin Gui, Xiaowan Li, Yuhui Liu, Yuchen Eleanor Jiang, Jun Wang, Changwang Zhang, Xiangru Tang, Ge Zhang, Jian Yang, Minghao Liu, Xitong Gao, Wangchunshu Zhou, Jiaheng Liu
Abstract: Recently, Agentic AI has become an increasingly popular research field. However, we argue that current agent research practices lack standardization and scientific rigor, making it hard to conduct fair comparisons among methods. As a result, it is still unclear how different design choices in agent frameworks affect effectiveness, and measuring their progress remains challenging. In this work, we conduct a systematic empirical study on GAIA benchmark and BrowseComp to examine the impact of popular design choices in key agent components in a fair and rigorous manner. We find that the lack of a standard evaluation protocol makes previous works, even open-sourced ones, non-reproducible, with significant variance between random runs. Therefore, we introduce a more robust evaluation protocol to stabilize comparisons. Our study reveals which components and designs are crucial for effective agents, while others are redundant, despite seeming logical. Based on our findings, we build and open-source OAgents, a new foundation agent framework that achieves state-of-the-art performance among open-source projects. OAgents offers a modular design for various agent components, promoting future research in Agentic AI.
Authors: Kartik Sharma, Yiqiao Jin, Vineeth Rakesh, Yingtong Dou, Menghai Pan, Mahashweta Das, Srijan Kumar
Abstract: As large language models (LLMs) are deployed in safety-critical settings, it is essential to ensure that their responses comply with safety standards. Prior research has revealed that LLMs often fail to grasp the notion of safe behaviors, resulting in either unjustified refusals to harmless prompts or the generation of harmful content. While substantial efforts have been made to improve their robustness, existing defenses often rely on costly fine-tuning of model parameters or employ suboptimal heuristic techniques. In this work, we take a novel approach to safeguard LLMs by learning to adapt the system prompts in instruction-tuned LLMs. While LLMs are typically pre-trained to follow a fixed system prompt, we investigate the impact of tailoring the system prompt to each specific user input on the safety of the responses. To this end, we propose $\textbf{Sysformer}$, a trans$\textbf{former}$ model that updates an initial $\textbf{sys}$tem prompt to a more robust system prompt in the LLM input embedding space while attending to the user prompt. While keeping the LLM parameters frozen, the Sysformer is trained to refuse to respond to a set of harmful prompts while responding ideally to a set of safe ones. Through extensive experiments on $5$ LLMs from different families and $2$ recent benchmarks, we demonstrate that Sysformer can significantly enhance the robustness of LLMs, leading to upto $80\%$ gain in the refusal rate on harmful prompts while enhancing the compliance with the safe prompts by upto $90\%$. Results also generalize well to sophisticated jailbreaking attacks, making LLMs upto $100\%$ more robust against different attack strategies. We hope our findings lead to cheaper safeguarding of LLMs and motivate future investigations into designing variable system prompts.
Authors: Marcel Wien\"obst, Sebastian Weichwald, Leonard Henckel
Abstract: We introduce CIfly, a framework for efficient algorithmic primitives in graphical causal inference that isolates reachability as a reusable core operation. It builds on the insight that many causal reasoning tasks can be reduced to reachability in purpose-built state-space graphs that can be constructed on the fly during traversal. We formalize a rule table schema for specifying such algorithms and prove they run in linear time. We establish CIfly as a more efficient alternative to the common primitives moralization and latent projection, which we show are computationally equivalent to Boolean matrix multiplication. Our open-source Rust implementation parses rule table text files and runs the specified CIfly algorithms providing high-performance execution accessible from Python and R. We demonstrate CIfly's utility by re-implementing a range of established causal inference tasks within the framework and by developing new algorithms for instrumental variables. These contributions position CIfly as a flexible and scalable backbone for graphical causal inference, guiding algorithm development and enabling easy and efficient deployment.
Authors: J. Schwardt, J. C. Budich
Abstract: We introduce and benchmark a stochastic local search heuristic for the NP-complete satisfiability problem 3-SAT that drastically outperforms existing solvers in the notoriously difficult realm of critically hard instances. Our construction is based on the crucial observation that well established previous approaches such as WalkSAT are prone to get stuck in local minima that are distinguished from true solutions by a larger number of oversatisfied combinatorial constraints. To address this issue, the proposed algorithm, coined DOCSAT, dissipates oversatisfied constraints (DOC), i.e. reduces their unfavorable abundance so as to render them critical. We analyze and benchmark our algorithm on a randomly generated sample of hard but satisfiable 3-SAT instances with varying problem sizes up to N=15000. Quite remarkably, we find that DOCSAT outperforms both WalkSAT and other well known algorithms including the complete solver Kissat, even when comparing its ability to solve the hardest quintile of the sample to the average performance of its competitors. The essence of DOCSAT may be seen as a way of harnessing statistical structure beyond the primary cost function of a combinatorial problem to avoid or escape local minima traps in stochastic local search, which opens avenues for generalization to other optimization problems.
Authors: Lukas Helff, Ahmad Omar, Felix Friedrich, Wolfgang Stammer, Antonia W\"ust, Tim Woydt, Rupert Mitchell, Patrick Schramowski, Kristian Kersting
Abstract: We introduce SLR, an end-to-end framework for systematic evaluation and training of Large Language Models (LLMs) via Scalable Logical Reasoning. Given a user's task specification, SLR enables scalable, automated synthesis of inductive reasoning tasks with precisely controlled difficulty. For each task, SLR synthesizes (i) a latent ground-truth rule, (ii) an executable validation program used by a symbolic judge to deterministically verify model outputs, and (iii) an instruction prompt for the reasoning task. Using SLR, we create SLR-Bench, a benchmark comprising over 19k prompts spanning 20 curriculum levels that progressively increase in relational, arithmetic, and recursive complexity. Large-scale evaluation reveals that contemporary LLMs readily produce syntactically valid rules, yet often fail at correct logical inference. Recent reasoning LLMs do somewhat better, but incur substantial increases in test-time compute, sometimes exceeding 15k completion tokens. Finally, logic-tuning via SLR doubles Llama-3-8B accuracy on SLR-Bench, achieving parity with Gemini-Flash-Thinking at a fraction of computational cost. SLR is fully automated, requires no human annotation, ensures dataset novelty, and offers a scalable environment for probing and advancing LLMs' reasoning capabilities.
Authors: Berk Yilmaz, Junyu Hu, Jinsong Liu
Abstract: This paper presents a Deep Reinforcement Learning (DRL) system for Xiangqi (Chinese Chess) that integrates neural networks with Monte Carlo Tree Search (MCTS) to enable strategic self-play and self-improvement. Addressing the underexplored complexity of Xiangqi, including its unique board layout, piece movement constraints, and victory conditions, our approach combines policy-value networks with MCTS to simulate move consequences and refine decision-making. By overcoming challenges such as Xiangqi's high branching factor and asymmetrical piece dynamics, our work advances AI capabilities in culturally significant strategy games while providing insights for adapting DRL-MCTS frameworks to domain-specific rule systems.
Authors: Myke C. Cohen, Zhe Su, Hsien-Te Kao, Daniel Nguyen, Spencer Lynch, Maarten Sap, Svitlana Volkova
Abstract: This paper presents an evaluation framework for agentic AI systems in mission-critical negotiation contexts, addressing the need for AI agents that can adapt to diverse human operators and stakeholders. Using Sotopia as a simulation testbed, we present two experiments that systematically evaluated how personality traits and AI agent characteristics influence LLM-simulated social negotiation outcomes--a capability essential for a variety of applications involving cross-team coordination and civil-military interactions. Experiment 1 employs causal discovery methods to measure how personality traits impact price bargaining negotiations, through which we found that Agreeableness and Extraversion significantly affect believability, goal achievement, and knowledge acquisition outcomes. Sociocognitive lexical measures extracted from team communications detected fine-grained differences in agents' empathic communication, moral foundations, and opinion patterns, providing actionable insights for agentic AI systems that must operate reliably in high-stakes operational scenarios. Experiment 2 evaluates human-AI job negotiations by manipulating both simulated human personality and AI system characteristics, specifically transparency, competence, adaptability, demonstrating how AI agent trustworthiness impact mission effectiveness. These findings establish a repeatable evaluation methodology for experimenting with AI agent reliability across diverse operator personalities and human-agent team dynamics, directly supporting operational requirements for reliable AI systems. Our work advances the evaluation of agentic AI workflows by moving beyond standard performance metrics to incorporate social dynamics essential for mission success in complex operations.
Authors: Craig S. Wright
Abstract: The exponential expansion of scientific literature has surpassed the epistemic processing capabilities of both human experts and current artificial intelligence systems. This paper introduces Bayesian Epistemology with Weighted Authority (BEWA), a formally structured architecture that operationalises belief as a dynamic, probabilistically coherent function over structured scientific claims. Each claim is contextualised, author-attributed, and evaluated through a system of replication scores, citation weighting, and temporal decay. Belief updates are performed via evidence-conditioned Bayesian inference, contradiction processing, and epistemic decay mechanisms. The architecture supports graph-based claim propagation, authorial credibility modelling, cryptographic anchoring, and zero-knowledge audit verification. By formalising scientific reasoning into a computationally verifiable epistemic network, BEWA advances the foundation for machine reasoning systems that promote truth utility, rational belief convergence, and audit-resilient integrity across dynamic scientific domains.
Authors: William Sharpless, Dylan Hirsch, Sander Tonkens, Nikhil Shinde, Sylvia Herbert
Abstract: Hard constraints in reinforcement learning (RL), whether imposed via the reward function or the model architecture, often degrade policy performance. Lagrangian methods offer a way to blend objectives with constraints, but often require intricate reward engineering and parameter tuning. In this work, we extend recent advances that connect Hamilton-Jacobi (HJ) equations with RL to propose two novel value functions for dual-objective satisfaction. Namely, we address: (1) the Reach-Always-Avoid problem - of achieving distinct reward and penalty thresholds - and (2) the Reach-Reach problem - of achieving thresholds of two distinct rewards. In contrast with temporal logic approaches, which typically involve representing an automaton, we derive explicit, tractable Bellman forms in this context by decomposing our problem into reach, avoid, and reach-avoid problems, as to leverage these aforementioned recent advances. From a mathematical perspective, the Reach-Always-Avoid and Reach-Reach problems are complementary and fundamentally different from standard sum-of-rewards problems and temporal logic problems, providing a new perspective on constrained decision-making. We leverage our analysis to propose a variation of Proximal Policy Optimization (DO-HJ-PPO), which solves these problems. Across a range of tasks for safe-arrival and multi-target achievement, we demonstrate that DO-HJ-PPO produces qualitatively distinct behaviors from previous approaches and out-competes a number of baselines in various metrics.
Authors: Reyna Abhyankar, Qi Qi, Yiying Zhang
Abstract: Generative AI is being leveraged to solve a variety of computer-use tasks involving desktop applications. State-of-the-art systems have focused solely on improving accuracy on leading benchmarks. However, these systems are practically unusable due to extremely high end-to-end latency (e.g., tens of minutes) for tasks that typically take humans just a few minutes to complete. To understand the cause behind this and to guide future developments of computer agents, we conduct the first study on the temporal performance of computer-use agents on OSWorld, the flagship benchmark in computer-use AI. We find that large model calls for planning and reflection account for the majority of the overall latency, and as an agent uses more steps to complete a task, each successive step can take 3x longer than steps at the beginning of a task. We then construct OSWorld-Human, a manually annotated version of the original OSWorld dataset that contains a human-determined trajectory for each task. We evaluate 16 agents on their efficiency using OSWorld-Human and found that even the highest-scoring agents on OSWorld take 1.4-2.7x more steps than necessary.
Authors: Tom Jeleniewski, Hamied Nabizada, Jonathan Reif, Felix Gehlhoff, Alexander Fay
Abstract: The formalization of process knowledge using ontologies enables consistent modeling of parameter interdependencies in manufacturing. These interdependencies are typically represented as mathematical expressions that define relations between process parameters, supporting tasks such as calculation, validation, and simulation. To support cross-context application and knowledge reuse, such expressions are often defined in a generic form and applied across multiple process contexts. This highlights the necessity of a consistent and semantically coherent model to ensure the correctness of data retrieval and interpretation. Consequently, dedicated mechanisms are required to address key challenges such as selecting context-relevant data, ensuring unit compatibility between variables and data elements, and verifying the completeness of input data required for evaluating mathematical expressions. This paper presents a set of verification mechanisms for a previously developed ontology-based process model that integrates standardized process semantics, data element definitions, and formal mathematical constructs. The approach includes (i) SPARQL-based filtering to retrieve process-relevant data, (ii) a unit consistency check based on expected-unit annotations and semantic classification, and (iii) a data completeness check to validate the evaluability of interdependencies. The applicability of the approach is demonstrated with a use case from Resin Transfer Molding (RTM), supporting the development of machine-interpretable and verifiable engineering models.
Authors: Ana Kostovska, Carola Doerr, Sa\v{s}o D\v{z}eroski, Pan\v{c}e Panov, Tome Eftimov
Abstract: Automated algorithm performance prediction in numerical blackbox optimization often relies on problem characterizations, such as exploratory landscape analysis features. These features are typically used as inputs to machine learning models and are represented in a tabular format. However, such approaches often overlook algorithm configurations, a key factor influencing performance. The relationships between algorithm operators, parameters, problem characteristics, and performance outcomes form a complex structure best represented as a graph. This work explores the use of heterogeneous graph data structures and graph neural networks to predict the performance of optimization algorithms by capturing the complex dependencies between problems, algorithm configurations, and performance outcomes. We focus on two modular frameworks, modCMA-ES and modDE, which decompose two widely used derivative-free optimization algorithms: the covariance matrix adaptation evolution strategy (CMA-ES) and differential evolution (DE). We evaluate 324 modCMA-ES and 576 modDE variants on 24 BBOB problems across six runtime budgets and two problem dimensions. Achieving up to 36.6% improvement in MSE over traditional tabular-based methods, this work highlights the potential of geometric learning in black-box optimization.
Authors: Hao Li, Gengrui Zhang, Petter Holme, Shuyue Hu, Zhen Wang
Abstract: Human decision-making belongs to the foundation of our society and civilization, but we are on the verge of a future where much of it will be delegated to artificial intelligence. The arrival of Large Language Models (LLMs) has transformed the nature and scope of AI-supported decision-making; however, the process by which they learn to make decisions, compared to humans, remains poorly understood. In this study, we examined the decision-making behavior of five leading LLMs across three core dimensions of real-world decision-making: uncertainty, risk, and set-shifting. Using three well-established experimental psychology tasks designed to probe these dimensions, we benchmarked LLMs against 360 newly recruited human participants. Across all tasks, LLMs often outperformed humans, approaching near-optimal performance. Moreover, the processes underlying their decisions diverged fundamentally from those of humans. On the one hand, our finding demonstrates the ability of LLMs to manage uncertainty, calibrate risk, and adapt to changes. On the other hand, this disparity highlights the risks of relying on them as substitutes for human judgment, calling for further inquiry.
Authors: Linde Vanbesien, Bart Bogaerts, Marc Denecker
Abstract: Approximation Fixpoint Theory (AFT) is a powerful theory covering various semantics of non-monotonic reasoning formalisms in knowledge representation such as Logic Programming and Answer Set Programming. Many semantics of such non-monotonic formalisms can be characterized as suitable fixpoints of a non-monotonic operator on a suitable lattice. Instead of working on the original lattice, AFT operates on intervals in such lattice to approximate or construct the fixpoints of interest. While AFT has been applied successfully across a broad range of non-monotonic reasoning formalisms, it is confronted by its limitations in other, relatively simple, examples. In this paper, we overcome those limitations by extending consistent AFT to deal with approximations that are more refined than intervals. Therefore, we introduce a more general notion of approximation spaces, showcase the improved expressiveness and investigate relations between different approximation spaces.
Authors: Albert Sadowski, Jaros{\l}aw A. Chudziak
Abstract: Large Language Models (LLMs) excel in complex reasoning tasks but struggle with consistent rule application, exception handling, and explainability, particularly in domains like legal analysis that require both natural language understanding and precise logical inference. This paper introduces a structured prompting framework that decomposes reasoning into three verifiable steps: entity identification, property extraction, and symbolic rule application. By integrating neural and symbolic approaches, our method leverages LLMs' interpretive flexibility while ensuring logical consistency through formal verification. The framework externalizes task definitions, enabling domain experts to refine logical structures without altering the architecture. Evaluated on the LegalBench hearsay determination task, our approach significantly outperformed baselines, with OpenAI o-family models showing substantial improvements - o1 achieving an F1 score of 0.929 and o3-mini reaching 0.867 using structured decomposition with complementary predicates, compared to their few-shot baselines of 0.714 and 0.74 respectively. This hybrid neural-symbolic system offers a promising pathway for transparent and consistent rule-based reasoning, suggesting potential for explainable AI applications in structured legal reasoning tasks.
Authors: Xiaoya Lu, Zeren Chen, Xuhao Hu, Yijin Zhou, Weichen Zhang, Dongrui Liu, Lu Sheng, Jing Shao
Abstract: Flawed planning from VLM-driven embodied agents poses significant safety hazards, hindering their deployment in real-world household tasks. However, existing static, non-interactive evaluation paradigms fail to adequately assess risks within these interactive environments, since they cannot simulate dynamic risks that emerge from an agent's actions and rely on unreliable post-hoc evaluations that ignore unsafe intermediate steps. To bridge this critical gap, we propose evaluating an agent's interactive safety: its ability to perceive emergent risks and execute mitigation steps in the correct procedural order. We thus present IS-Bench, the first multi-modal benchmark designed for interactive safety, featuring 161 challenging scenarios with 388 unique safety risks instantiated in a high-fidelity simulator. Crucially, it facilitates a novel process-oriented evaluation that verifies whether risk mitigation actions are performed before/after specific risk-prone steps. Extensive experiments on leading VLMs, including the GPT-4o and Gemini-2.5 series, reveal that current agents lack interactive safety awareness, and that while safety-aware Chain-of-Thought can improve performance, it often compromises task completion. By highlighting these critical limitations, IS-Bench provides a foundation for developing safer and more reliable embodied AI systems.
Authors: Sami Abboud, Eleanor Hanna, Olivier Jeunen, Vineesha Raheja, Schaun Wheeler
Abstract: Consumer applications provide ample opportunities to surface and communicate various forms of content to users. From promotional campaigns for new features or subscriptions, to evergreen nudges for engagement, or personalised recommendations; across e-mails, push notifications, and in-app surfaces. The conventional approach to orchestration for communication relies heavily on labour-intensive manual marketer work, and inhibits effective personalisation of content, timing, frequency, and copy-writing. We formulate this task under a sequential decision-making framework, where we aim to optimise a modular decision-making policy that maximises incremental engagement for any funnel event. Our approach leverages a Difference-in-Differences design for Individual Treatment Effect estimation, and Thompson sampling to balance the explore-exploit trade-off. We present results from a multi-service application, where our methodology has resulted in significant increases to a variety of goal events across several product features, and is currently deployed across 150 million users.
Authors: Zexi Liu, Yuzhu Cai, Xinyu Zhu, Yujie Zheng, Runkun Chen, Ying Wen, Yanfeng Wang, Weinan E, Siheng Chen
Abstract: As AI capabilities advance toward and potentially beyond human-level performance, a natural transition emerges where AI-driven development becomes more efficient than human-centric approaches. A promising pathway toward this transition lies in AI-for-AI (AI4AI), which leverages AI techniques to automate and optimize the design, training, and deployment of AI systems themselves. While LLM-based agents have shown the potential to realize AI4AI, they are often unable to fully leverage the experience accumulated by agents during the exploration of solutions in the reasoning process, leading to inefficiencies and suboptimal performance. To address this limitation, we propose ML-Master, a novel AI4AI agent that seamlessly integrates exploration and reasoning by employing a selectively scoped memory mechanism. This approach allows ML-Master to efficiently combine diverse insights from parallel solution trajectories with analytical reasoning, guiding further exploration without overwhelming the agent with excessive context. We evaluate ML-Master on the MLE-Bench, where it achieves a 29.3% average medal rate, significantly surpassing existing methods, particularly in medium-complexity tasks, while accomplishing this superior performance within a strict 12-hour time constraint-half the 24-hour limit used by previous baselines. These results demonstrate ML-Master's potential as a powerful tool for advancing AI4AI.
Authors: Mustafa Akben, Aaron Satko
Abstract: Large language models (LLMs) offer promising opportunities for organizational research. However, their built-in moderation systems can create problems when researchers try to analyze harmful content, often refusing to follow certain instructions or producing overly cautious responses that undermine validity of the results. This is particularly problematic when analyzing organizational conflicts such as microaggressions or hate speech. This paper introduces an Elo rating-based method that significantly improves LLM performance for harmful content analysis In two datasets, one focused on microaggression detection and the other on hate speech, we find that our method outperforms traditional LLM prompting techniques and conventional machine learning models on key measures such as accuracy, precision, and F1 scores. Advantages include better reliability when analyzing harmful content, fewer false positives, and greater scalability for large-scale datasets. This approach supports organizational applications, including detecting workplace harassment, assessing toxic communication, and fostering safer and more inclusive work environments.
Authors: Vinay K Chaudhri, Chaitan Baru, Brandon Bennett, Mehul Bhatt, Darion Cassel, Anthony G Cohn, Rina Dechter, Esra Erdem, Dave Ferrucci, Ken Forbus, Gregory Gelfond, Michael Genesereth, Andrew S. Gordon, Benjamin Grosof, Gopal Gupta, Jim Hendler, Sharat Israni, Tyler R. Josephson, Patrick Kyllonen, Yuliya Lierler, Vladimir Lifschitz, Clifton McFate, Hande K. McGinty, Leora Morgenstern, Alessandro Oltramari, Praveen Paritosh, Dan Roth, Blake Shepard, Cogan Shimzu, Denny Vrande\v{c}i\'c, Mark Whiting, Michael Witbrock
Abstract: The long-standing goal of creating a comprehensive, multi-purpose knowledge resource, reminiscent of the 1984 Cyc project, still persists in AI. Despite the success of knowledge resources like WordNet, ConceptNet, Wolfram|Alpha and other commercial knowledge graphs, verifiable, general-purpose widely available sources of knowledge remain a critical deficiency in AI infrastructure. Large language models struggle due to knowledge gaps; robotic planning lacks necessary world knowledge; and the detection of factually false information relies heavily on human expertise. What kind of knowledge resource is most needed in AI today? How can modern technology shape its development and evaluation? A recent AAAI workshop gathered over 50 researchers to explore these questions. This paper synthesizes our findings and outlines a community-driven vision for a new knowledge infrastructure. In addition to leveraging contemporary advances in knowledge representation and reasoning, one promising idea is to build an open engineering framework to exploit knowledge modules effectively within the context of practical applications. Such a framework should include sets of conventions and social structures that are adopted by contributors.
Authors: Soobin Chae, Suhwan Lee, Hanna Hauptmann, Hajo A. Reijers, Xixi Lu
Abstract: Predictive Process Monitoring (PPM) often uses deep learning models to predict the future behavior of ongoing processes, such as predicting process outcomes. While these models achieve high accuracy, their lack of interpretability undermines user trust and adoption. Explainable AI (XAI) aims to address this challenge by providing the reasoning behind the predictions. However, current evaluations of XAI in PPM focus primarily on functional metrics (such as fidelity), overlooking user-centered aspects such as their effect on task performance and decision-making. This study investigates the effects of explanation styles (feature importance, rule-based, and counterfactual) and perceived AI accuracy (low or high) on decision-making in PPM. We conducted a decision-making experiment, where users were presented with the AI predictions, perceived accuracy levels, and explanations of different styles. Users' decisions were measured both before and after receiving explanations, allowing the assessment of objective metrics (Task Performance and Agreement) and subjective metrics (Decision Confidence). Our findings show that perceived accuracy and explanation style have a significant effect.
Authors: Kenjiro Ide, Taiga Someya, Kohei Kawaguchi, Keisuke Fujii
Abstract: Understanding football tactics is crucial for managers and analysts. Previous research has proposed models based on spatial and kinematic equations, but these are computationally expensive. Also, Reinforcement learning approaches use player positions and velocities but lack interpretability and require large datasets. Rule-based models align with expert knowledge but have not fully considered all players' states. This study explores whether low-dimensional, rule-based models using spatiotemporal data can effectively capture football tactics. Our approach defines interpretable state variables for both the ball-holder and potential pass receivers, based on criteria that explore options like passing. Through discussions with a manager, we identified key variables representing the game state. We then used StatsBomb event data and SkillCorner tracking data from the 2023$/$24 LaLiga season to train an XGBoost model to predict pass success. The analysis revealed that the distance between the player and the ball, as well as the player's space score, were key factors in determining successful passes. Our interpretable low-dimensional modeling facilitates tactical analysis through the use of intuitive variables and provides practical value as a tool to support decision-making in football.
Authors: Jinlong Pang, Jiaheng Wei, Yifan Hua, Chen Qian, Yang Liu
Abstract: Federated learning (FL) provides a promising paradigm for facilitating collaboration between multiple clients that jointly learn a global model without directly sharing their local data. However, existing research suffers from two caveats: 1) From the perspective of agents, voluntary and unselfish participation is often assumed. But self-interested agents may opt out of the system or provide low-quality contributions without proper incentives; 2) From the mechanism designer's perspective, the aggregated models can be unsatisfactory as the existing game-theoretical federated learning approach for data collection ignores the potential heterogeneous effort caused by contributed data. To alleviate above challenges, we propose an incentive-aware framework for agent participation that considers data heterogeneity to accelerate the convergence process. Specifically, we first introduce the notion of Wasserstein distance to explicitly illustrate the heterogeneous effort and reformulate the existing upper bound of convergence. To induce truthful reporting from agents, we analyze and measure the generalization error gap of any two agents by leveraging the peer prediction mechanism to develop score functions. We further present a two-stage Stackelberg game model that formalizes the process and examines the existence of equilibrium. Extensive experiments on real-world datasets demonstrate the effectiveness of our proposed mechanism.
Authors: Yanchen Zhu, Honghui Zou, Chufan Liu, Yuyu Luo, Yuankai Wu, Yuxuan Liang
Abstract: The success of vehicle electrification, which brings significant societal and environmental benefits, is contingent upon the availability of efficient and adaptable charging infrastructure. Traditional fixed-location charging stations often face issues like underutilization or congestion due to the dynamic nature of charging demand. Mobile chargers have emerged as a flexible solution, capable of relocating to align with these demand fluctuations. This paper addresses the optimal planning and operation of hybrid charging infrastructures, integrating both fixed and mobile chargers within urban road networks. We introduce the Hybrid Charging Station Planning and Operation (HCSPO) problem, which simultaneously optimizes the location and configuration of fixed charging stations and schedules mobile chargers for dynamic operations. Our approach incorporates a charging demand prediction model grounded in Model Predictive Control (MPC) to enhance decision-making. To solve the HCSPO problem, we propose a deep reinforcement learning method, augmented with heuristic scheduling techniques, to effectively bridge the planning of fixed chargers with the real-time operation of mobile chargers. Extensive case studies using real-world urban scenarios demonstrate that our method significantly improves the availability of charging infrastructure and reduces user inconvenience compared to existing solutions and baselines.
Authors: Ciro Beneduce, Massimiliano Luca, Bruno Lepri
Abstract: Image generation models are revolutionizing many domains, and urban analysis and design is no exception. While such models are widely adopted, there is a limited literature exploring their geographic knowledge, along with the biases they embed. In this work, we generated 150 synthetic images for each state in the USA and related capitals using FLUX 1 and Stable Diffusion 3.5, two state-of-the-art models for image generation. We embed each image using DINO-v2 ViT-S/14 and the Fr\'echet Inception Distances to measure the similarity between the generated images. We found that while these models have implicitly learned aspects of USA geography, if we prompt the models to generate an image for "United States" instead of specific cities or states, the models exhibit a strong representative bias toward metropolis-like areas, excluding rural states and smaller cities. {\color{black} In addition, we found that models systematically exhibit some entity-disambiguation issues with European-sounding names like Frankfort or Devon.
Authors: Tomoya Kashimata, Yohei Hamakawa, Masaya Yamasaki, Kosuke Tatsumura
Abstract: Many real-time systems require the optimization of discrete variables. Black-box optimization (BBO) algorithms and multi-armed bandit (MAB) algorithms perform optimization by repeatedly taking actions and observing the corresponding instant rewards without any prior knowledge. Recently, a BBO method using an Ising machine has been proposed to find the best action that is represented by a combination of discrete values and maximizes the instant reward in static environments. In contrast, dynamic environments, where real-time systems operate, necessitate MAB algorithms that maximize the average reward over multiple trials. However, due to the enormous number of actions resulting from the combinatorial nature of discrete optimization, conventional MAB algorithms cannot effectively optimize dynamic, discrete environments. Here, we show a heuristic MAB method for dynamic, discrete environments by extending the BBO method, in which an Ising machine effectively explores the actions while considering interactions between variables and changes in dynamic environments. We demonstrate the dynamic adaptability of the proposed method in a wireless communication system with moving users.
Authors: Jiaqi Chen, Mingfeng Fan, Xuefeng Zhang, Jingsong Liang, Yuhong Cao, Guohua Wu, Guillaume Adrien Sartoretti
Abstract: Effective and efficient task planning is essential for mobile robots, especially in applications like warehouse retrieval and environmental monitoring. These tasks often involve selecting one location from each of several target clusters, forming a Generalized Traveling Salesman Problem (GTSP) that remains challenging to solve both accurately and efficiently. To address this, we propose a Multimodal Fused Learning (MMFL) framework that leverages both graph and image-based representations to capture complementary aspects of the problem, and learns a policy capable of generating high-quality task planning schemes in real time. Specifically, we first introduce a coordinate-based image builder that transforms GTSP instances into spatially informative representations. We then design an adaptive resolution scaling strategy to enhance adaptability across different problem scales, and develop a multimodal fusion module with dedicated bottlenecks that enables effective integration of geometric and spatial features. Extensive experiments show that our MMFL approach significantly outperforms state-of-the-art methods across various GTSP instances while maintaining the computational efficiency required for real-time robotic applications. Physical robot tests further validate its practical effectiveness in real-world scenarios.
Authors: Lingfeng Li, Yunlong Lu, Yongyi Wang, Wenxin Li
Abstract: A wide variety of bots in games enriches the gameplay experience and enhances replayability. Recent advancements in game artificial intelligence have predominantly focused on improving the proficiency of bots. Nevertheless, developing highly competent bots with a wide range of distinct play styles remains a relatively under-explored area. We select the Mahjong game environment as a case study. The high degree of randomness inherent in the Mahjong game and the prevalence of out-of-distribution states lead to suboptimal performance of existing offline learning and Learning-from-Demonstration (LfD) algorithms. In this paper, we leverage the gameplay histories of existing Mahjong agents and put forward a novel LfD algorithm that necessitates only minimal modifications to the Proximal Policy Optimization algorithm. The comprehensive empirical results illustrate that our proposed method not only significantly enhances the proficiency of the agents but also effectively preserves their unique play styles.
Authors: Davide Frizzo, Francesco Borsatti, Gian Antonio Susto
Abstract: Predictive Maintenance (PdM) is pivotal in Industry 4.0 and 5.0, proactively enhancing efficiency through accurate equipment Remaining Useful Life (RUL) prediction, thus optimizing maintenance scheduling and reducing unexpected failures and premature interventions. This paper introduces a novel RUL estimation approach leveraging State Space Models (SSM) for efficient long-term sequence modeling. To handle model uncertainty, Simoultaneous Quantile Regression (SQR) is integrated into the SSM, enabling multiple quantile estimations. The proposed method is benchmarked against traditional sequence modelling techniques (LSTM, Transformer, Informer) using the C-MAPSS dataset. Results demonstrate superior accuracy and computational efficiency of SSM models, underscoring their potential for high-stakes industrial applications.
Authors: Fabian Neuhaus
Abstract: BFO 2020 does not support functions, dispositions, and roles of generically dependent continuants (like software or datasets). In this paper, we argue that this is a severe limitation, which prevents, for example, the adequate representation of the functions of computer models or the various roles of datasets during the execution of these models. We discuss the aspects of BFO 2020 that prevent the representation of realizable entities of generically dependent continuants. Two approaches to address the issue are presented: (a) the use of defined classes and (b) a proposal of changes that allow BFO to support functions, dispositions, and roles of generically dependent continuants.
Authors: Chuxue Cao, Mengze Li, Juntao Dai, Jinluan Yang, Zijian Zhao, Shengyu Zhang, Weijie Shi, Chengzhong Liu, Sirui Han, Yike Guo
Abstract: Large language models (LLMs) have shown promising first-order logic (FOL) reasoning capabilities with applications in various areas. However, their effectiveness in complex mathematical reasoning involving multi-step FOL deductions is still under-researched. While LLMs perform competitively on established mathematical reasoning benchmarks, they struggle with multi-step FOL tasks, as demonstrated by Deepseek-Prover-V2-7B's low accuracy (4.2%) on our proposed theorem proving dataset. This issue arises from the limited exploration of diverse proof strategies and the potential for early reasoning mistakes to undermine entire proofs. To address these issues, we propose DREAM, a self-adaptive solution that enhances the Diversity and REAsonability of LLMs' generation strategies. DREAM incorporates an Axiom-Driven Strategy Diversification mechanism to promote varied strategic outcomes and a Sub-Proposition Error Feedback to help LLMs reflect on and correct their proofs. Our contributions include pioneering advancements in LLMs' mathematical reasoning through FOL theorem proving, introducing a novel inference stage solution that improves performance by 0.6% to 6.4%, and providing a curated dataset of 447 mathematical theorems in Lean 4 format for evaluation.
Authors: Lina Berrayana, Sean Rooney, Luis Garc\'es-Erice, Ioana Giurgiu
Abstract: The creation of benchmarks to evaluate the safety of Large Language Models is one of the key activities within the trusted AI community. These benchmarks allow models to be compared for different aspects of safety such as toxicity, bias, harmful behavior etc. Independent benchmarks adopt different approaches with distinct data sets and evaluation methods. We investigate how robust such benchmarks are by using different approaches to rank a set of representative models for bias and compare how similar are the overall rankings. We show that different but widely used bias evaluations methods result in disparate model rankings. We conclude with recommendations for the community in the usage of such benchmarks.
Authors: Dadi Guo (May), Jiayu Liu (May), Zhiyuan Fan (May), Zhitao He (May), Haoran Li (May), Yumeng Wang (May), Yi R. (May), Fung
Abstract: Large reasoning models (e.g., R1, o3) have demonstrated remarkable mathematical problem-solving abilities. However, the high reported accuracy of these advanced models on popular datasets, reliance on purely numerical evaluation and potential benchmark leakage, often masks their true reasoning shortcomings. To address this, we propose leveraging the inherent rigor and methodological complexity of mathematical proofs as a diagnostic tool to expose these hidden failures. Specifically, we introduce the RFMDataset (Reveal Failure Modes), a collection of 200 diverse mathematical proof problems, and thoroughly evaluate advanced models' performance on it. Our in-depth analysis of their failures uncovers 10 fine-grained error types, which shows fundamental limitations in current large reasoning models: 1) large reasoning models grapple profoundly with mathematical proofs, with some generating entirely correct proofs for less than 20% of problems and failing even on basic ones; 2) models exhibit a diverse spectrum of reasoning failures, prominently demonstrating the lack of guarantees for the correctness and rigor of single-step reasoning; and 3) models show hallucination and incompleteness during the reasoning process. Our findings reveal that models' self-reflection is insufficient to resolve the current logical dilemmas, necessitating formalized and fine-grained logical training.
Authors: Josiah P. Hanna, Nicholas E. Corrado
Abstract: Recent work on large language models has demonstrated the use of model-free reinforcement learning (RL) to train reasoning-like capabilities. The emergence of "thinking" through model-free RL is interesting as thinking actions neither produce reward nor change the external world state to one where the agent is more likely to get reward. This paper seeks to build a domain-independent understanding of when model-free RL will lead to "thinking" as a strategy for reward maximization. To build this understanding, we first introduce a theoretical model which we call a \textit{thought Markov decision process} (MDP). Thought MDPs minimally extend the classical MDP model to include an abstract notion of thought state and thought action. Using the thought MDP model, we prove the importance of policy initialization in determining whether or not thinking emerges and show formally that thought actions are equivalent to the agent choosing to perform a step of policy improvement before continuing to act. We then show that open-source LLMs satisfy the conditions that our theory predicts are necessary for model-free RL to produce thinking-like behavior. Finally, we hypothesize sufficient conditions that would enable thinking to be learned outside of language generation and introduce a toy domain where a combination of multi-task pre-training and designated thought actions enable more data-efficient RL compared to non-thinking agents.
Authors: Botao Zhu (Sherman), Xianbin Wang (Sherman), Lei Zhang (Sherman), Xuemin (Sherman), Shen
Abstract: In collaborative systems with complex tasks relying on distributed resources, trust evaluation of potential collaborators has emerged as an effective mechanism for task completion. However, due to the network dynamics and varying information gathering latencies, it is extremely challenging to observe and collect all trust attributes of a collaborating device concurrently for a comprehensive trust assessment. In this paper, a novel progressive trust evaluation framework, namely chain-of-trust, is proposed to make better use of misaligned device attribute data. This framework, designed for effective task completion, divides the trust evaluation process into multiple chained stages based on task decomposition. At each stage, based on the task completion process, the framework only gathers the latest device attribute data relevant to that stage, leading to reduced trust evaluation complexity and overhead. By leveraging advanced in-context learning, few-shot learning, and reasoning capabilities, generative AI is then employed to analyze and interpret the collected data to produce correct evaluation results quickly. Only devices deemed trustworthy at this stage proceed to the next round of trust evaluation. The framework ultimately determines devices that remain trustworthy across all stages. Experimental results demonstrate that the proposed framework achieves high accuracy in trust evaluation.
Authors: Abinitha Gourabathina, Yuexing Hao, Walter Gerych, Marzyeh Ghassemi
Abstract: Clinical robustness is critical to the safe deployment of medical Large Language Models (LLMs), but key questions remain about how LLMs and humans may differ in response to the real-world variability typified by clinical settings. To address this, we introduce MedPerturb, a dataset designed to systematically evaluate medical LLMs under controlled perturbations of clinical input. MedPerturb consists of clinical vignettes spanning a range of pathologies, each transformed along three axes: (1) gender modifications (e.g., gender-swapping or gender-removal); (2) style variation (e.g., uncertain phrasing or colloquial tone); and (3) format changes (e.g., LLM-generated multi-turn conversations or summaries). With MedPerturb, we release a dataset of 800 clinical contexts grounded in realistic input variability, outputs from four LLMs, and three human expert reads per clinical context. We use MedPerturb in two case studies to reveal how shifts in gender identity cues, language style, or format reflect diverging treatment selections between humans and LLMs. We find that LLMs are more sensitive to gender and style perturbations while human annotators are more sensitive to LLM-generated format perturbations such as clinical summaries. Our results highlight the need for evaluation frameworks that go beyond static benchmarks to assess the similarity between human clinician and LLM decisions under the variability characteristic of clinical settings.
Authors: Yilin Zhang, Xinran Zhao, Zora Zhiruo Wang, Chenyang Yang, Jiayi Wei, Tongshuang Wu
Abstract: Retrieval-Augmented Generation (RAG) has become essential for large-scale code generation, grounding predictions in external code corpora to improve actuality. However, a critical yet underexplored aspect of RAG pipelines is chunking -- the process of dividing documents into retrievable units. Existing line-based chunking heuristics often break semantic structures, splitting functions or merging unrelated code, which can degrade generation quality. We propose chunking via Abstract Syntax Trees (\ourwork), a structure-aware method that recursively breaks large AST nodes into smaller chunks and merges sibling nodes while respecting size limits. This approach generates self-contained, semantically coherent units across programming languages and tasks, improving performance on diverse code generation tasks, e.g., boosting Recall@5 by 4.3 points on RepoEval retrieval and Pass@1 by 2.67 points on SWE-bench generation. Our work highlights the importance of structure-aware chunking for scaling retrieval-enhanced code intelligence.
Authors: Wang Yu-Hang, Liu ying, Fang liang, Wang Xuelin, Junkang Guo, Shiwei Li, Lei Gao, Jian Liu, Wenfei Yin
Abstract: Adversarial Training (AT) is a cornerstone defense, but many variants overlook foundational feature representations by primarily focusing on stronger attack generation. We introduce Adversarial Evolution Training (AET), a simple yet powerful framework that strategically prepends an Empirical Risk Minimization (ERM) phase to conventional AT. We hypothesize this initial ERM phase cultivates a favorable feature manifold, enabling more efficient and effective robustness acquisition. Empirically, AET achieves comparable or superior robustness more rapidly, improves clean accuracy, and cuts training costs by 8-25\%. Its effectiveness is shown across multiple datasets, architectures, and when augmenting established AT methods. Our findings underscore the impact of feature pre-conditioning via standard training for developing more efficient, principled robust defenses. Code is available in the supplementary material.
Authors: Jiahe Qin, Junpeng Li, Changchun Hua, Yana Yang
Abstract: Label Proportion Learning (LLP) addresses the classification problem where multiple instances are grouped into bags and each bag contains information about the proportion of each class. However, in practical applications, obtaining precise supervisory information regarding the proportion of instances in a specific class is challenging. To better align with real-world application scenarios and effectively leverage the proportional constraints of instances within tuples, this paper proposes a generalized learning framework \emph{MDPU}. Specifically, we first mathematically model the distribution of instances within tuples of arbitrary size, under the constraint that the number of positive instances is no less than that of negative instances. Then we derive an unbiased risk estimator that satisfies risk consistency based on the empirical risk minimization (ERM) method. To mitigate the inevitable overfitting issue during training, a risk correction method is introduced, leading to the development of a corrected risk estimator. The generalization error bounds of the unbiased risk estimator theoretically demonstrate the consistency of the proposed method. Extensive experiments on multiple datasets and comparisons with other relevant baseline methods comprehensively validate the effectiveness of the proposed learning framework.
Authors: Hui Ma, Kai Yang, Man-On Pun
Abstract: Cellular traffic prediction is of great importance for operators to manage network resources and make decisions. Traffic is highly dynamic and influenced by many exogenous factors, which would lead to the degradation of traffic prediction accuracy. This paper proposes an end-to-end framework with two variants to explicitly characterize the spatiotemporal patterns of cellular traffic among neighboring cells. It uses convolutional neural networks with an attention mechanism to capture the spatial dynamics and Kalman filter for temporal modelling. Besides, we can fully exploit the auxiliary information such as social activities to improve prediction performance. We conduct extensive experiments on three real-world datasets. The results show that our proposed models outperform the state-of-the-art machine learning techniques in terms of prediction accuracy.
Authors: Liulu He, Shenli Zhen, Karwei Sun, Yijiang Liu, Yufei Zhao, Chongkang Tan, Huanrui Yang, Yuan Du, Li Du
Abstract: Rotations have become essential to state-of-the-art quantization pipelines for large language models (LLMs) by effectively smoothing outliers in weights and activations. However, further optimizing the rotation parameters offers only limited performance gains and introduces significant training overhead: due to rotation parameter sharing, full-model must be loaded simultaneously to enable backpropagation, resulting in substantial memory consumption and limited practical utility. In this work, we identify two fundamental limitations of current rotational quantization methods: (i) rotation fails to align channel means, resulting in wider quantization bounds and increased rounding errors; and (ii) rotation makes the activation distribution more Gaussian-like, increasing energy loss caused by clipping errors. To address these issues, we introduce \textbf{BASE-Q}, a simple yet powerful approach that combines bias correction and asymmetric scaling to effectively reduce rounding and clipping errors. Furthermore, BASE-Q enables blockwise optimization, eliminating the need for memory-intensive full-model backpropagation. Extensive experiments on various LLMs and benchmarks demonstrate the effectiveness of BASE-Q, narrowing the accuracy gap to full-precision models by 50.5\%, 42.9\%, and 29.2\% compared to QuaRot, SpinQuant, and OSTQuant, respectively. The code will be released soon.
Authors: Tianyu Wang, Lingyou Pang, Akira Horiguchi, Carey E. Priebe
Abstract: The increasing use of synthetic data from the public Internet has enhanced data usage efficiency in large language model (LLM) training. However, the potential threat of model collapse remains insufficiently explored. Existing studies primarily examine model collapse in a single model setting or rely solely on statistical surrogates. In this work, we introduce LLM Web Dynamics (LWD), an efficient framework for investigating model collapse at the network level. By simulating the Internet with a retrieval-augmented generation (RAG) database, we analyze the convergence pattern of model outputs. Furthermore, we provide theoretical guarantees for this convergence by drawing an analogy to interacting Gaussian Mixture Models.
Authors: Chuheng Zhang, Tim Pearce, Pushi Zhang, Kaixin Wang, Xiaoyu Chen, Wei Shen, Li Zhao, Jiang Bian
Abstract: Latent action models (LAMs) aim to learn action-relevant changes from unlabeled videos by compressing changes between frames as latents. However, differences between video frames can be caused by controllable changes as well as exogenous noise, leading to an important concern -- do latents capture the changes caused by actions or irrelevant noise? This paper studies this issue analytically, presenting a linear model that encapsulates the essence of LAM learning, while being tractable.This provides several insights, including connections between LAM and principal component analysis (PCA), desiderata of the data-generating policy, and justification of strategies to encourage learning controllable changes using data augmentation, data cleaning, and auxiliary action-prediction. We also provide illustrative results based on numerical simulation, shedding light on the specific structure of observations, actions, and noise in data that influence LAM learning.
Authors: Shengyuan Hu, Neil Kale, Pratiksha Thaker, Yiwei Fu, Steven Wu, Virginia Smith
Abstract: Machine unlearning has the potential to improve the safety of large language models (LLMs) by removing sensitive or harmful information post hoc. A key challenge in unlearning involves balancing between forget quality (effectively unlearning undesirable information) and retain quality (maintaining good performance on other, general tasks). Unfortunately, as we show, current LLM unlearning benchmarks contain highly disparate forget and retain sets -- painting a false picture of the effectiveness of LLM unlearning methods. This can be particularly problematic because it opens the door for benign perturbations, such as relearning attacks, to easily reveal supposedly unlearned knowledge once models are deployed. To address this, we present $\texttt{BLUR}$: a benchmark for LLM unlearning that provides more realistic scenarios of forget-retain overlap. $\texttt{BLUR}$ significantly expands on existing unlearning benchmarks by providing extended evaluation tasks, combined forget/retain queries, and relearning datasets of varying degrees of difficulty. Despite the benign nature of the queries considered, we find that the performance of existing methods drops significantly when evaluated on $\texttt{BLUR}$, with simple approaches performing better on average than more recent methods. These results highlight the importance of robust evaluation and suggest several important directions of future study. Our benchmark is publicly available at: https://huggingface.co/datasets/forgelab/BLUR
Authors: Minjae Cho, Hiroyasu Tsukamoto, Huy Trong Tran
Abstract: Control contraction metrics (CCMs) provide a framework to co-synthesize a controller and a corresponding contraction metric -- a positive-definite Riemannian metric under which a closed-loop system is guaranteed to be incrementally exponentially stable. However, the synthesized controller only ensures that all the trajectories of the system converge to one single trajectory and, as such, does not impose any notion of optimality across an entire trajectory. Furthermore, constructing CCMs requires a known dynamics model and non-trivial effort in solving an infinite-dimensional convex feasibility problem, which limits its scalability to complex systems featuring high dimensionality with uncertainty. To address these issues, we propose to integrate CCMs into reinforcement learning (RL), where CCMs provide dynamics-informed feedback for learning control policies that minimize cumulative tracking error under unknown dynamics. We show that our algorithm, called contraction actor-critic (CAC), formally enhances the capability of CCMs to provide a set of contracting policies with the long-term optimality of RL in a fully automated setting. Given a pre-trained dynamics model, CAC simultaneously learns a contraction metric generator (CMG) -- which generates a contraction metric -- and uses an actor-critic algorithm to learn an optimal tracking policy guided by that metric. We demonstrate the effectiveness of our algorithm relative to established baselines through extensive empirical studies, including simulated and real-world robot experiments, and provide a theoretical rationale for incorporating contraction theory into RL.
Authors: Haolin Pan, Hongyu Lin, Haoran Luo, Yang Liu, Kaichun Yao, Libo Zhang, Mingjie Xing, Yanjun Wu
Abstract: Compiler auto-tuning optimizes pass sequences to improve performance metrics such as Intermediate Representation (IR) instruction count. Although recent advances leveraging Large Language Models (LLMs) have shown promise in automating compiler tuning, two significant challenges still remain: the absence of high-quality reasoning datasets for agents training, and limited effective interactions with the compilation environment. In this work, we introduce Compiler-R1, the first reinforcement learning (RL)-driven framework specifically augmenting LLM capabilities for compiler auto-tuning. Compiler-R1 features a curated, high-quality reasoning dataset and a novel two-stage end-to-end RL training pipeline, enabling efficient environment exploration and learning through an outcome-based reward. Extensive experiments across seven datasets demonstrate Compiler-R1 achieving an average 8.46% IR instruction count reduction compared to opt -Oz, showcasing the strong potential of RL-trained LLMs for compiler optimization. Our code and datasets are publicly available at https://github.com/Panhaolin2001/Compiler-R1.
Authors: Peter Belcak, Greg Heinrich, Jan Kautz, Pavlo Molchanov
Abstract: Finetuning language models for a new domain inevitably leads to the deterioration of their general performance. This becomes more pronounced the more limited the finetuning data resource. We introduce minifinetuning (MFT), a method for language model domain adaptation that considerably reduces the effects of overfitting-induced degeneralization in low-data settings and which does so in the absence of any pre-training data for replay. MFT demonstrates 2-10x more favourable specialization-to-degeneralization ratios than standard finetuning across a wide range of models and domains and exhibits an intrinsic robustness to overfitting when data in the new domain is scarce and down to as little as 500 samples. Employing corrective self-distillation that is individualized on the sample level, MFT outperforms parameter-efficient finetuning methods, demonstrates replay-like degeneralization mitigation properties, and is composable with either for a combined effect.
Authors: Guoqing Chao, Zhenghao Zhang, Lei Meng, Jie Wen, Dianhui Chu
Abstract: Federated multi-view clustering has been proposed to mine the valuable information within multi-view data distributed across different devices and has achieved impressive results while preserving the privacy. Despite great progress, most federated multi-view clustering methods only used global pseudo-labels to guide the downstream clustering process and failed to exploit the global information when extracting features. In addition, missing data problem in federated multi-view clustering task is less explored. To address these problems, we propose a novel Federated Incomplete Multi-view Clustering method with globally Fused Graph guidance (FIMCFG). Specifically, we designed a dual-head graph convolutional encoder at each client to extract two kinds of underlying features containing global and view-specific information. Subsequently, under the guidance of the fused graph, the two underlying features are fused into high-level features, based on which clustering is conducted under the supervision of pseudo-labeling. Finally, the high-level features are uploaded to the server to refine the graph fusion and pseudo-labeling computation. Extensive experimental results demonstrate the effectiveness and superiority of FIMCFG. Our code is publicly available at https://github.com/PaddiHunter/FIMCFG.
Authors: Feiyu Yao, Qian Wang
Abstract: As large language models (LLMs) continue to support increasingly longer contexts, the memory demand for key-value (KV) caches during decoding grows rapidly, becoming a critical bottleneck in both GPU memory capacity and PCIe bandwidth. Sparse attention mechanisms alleviate this issue by computing attention weights only for selected key-value pairs. However, their indexing computation typically requires traversing all key vectors, resulting in significant computational and data transfer overhead. To reduce the cost of index retrieval, existing methods often treat each decoding step as an independent process, failing to exploit the temporal correlations embedded in historical decoding information. To this end, we propose LFPS(Learn From the Past for Sparse Indexing), an acceleration method that dynamically constructs sparse indexing candidates based on historical attention patterns. LFPS captures two prevalent trends in decoder attention -vertical patterns (attending to fixed positions) and slash patterns (attending to relative positions) -and incorporates a positional expansion strategy to effectively predict the Top-k indices for the current step. We validate LFPS on challenging long-context benchmarks such as LongBench-RULER, using Llama-3.1-8B-Instruct as the base model. Experimental results show that LFPS achieves up to 22.8$\times$ speedup over full attention and 9.6$\times$ speedup over exact Top-k retrieval on an RTX 4090 GPU and a single CPU core of a Xeon Gold 6430, respectively, while preserving generation accuracy. These results demonstrate that LFPS offers a practical and efficient solution for decoding optimization in long-context LLM inference.
Authors: Jittarin Jetwiriyanon, Teo Susnjak, Surangika Ranathunga
Abstract: This study investigates zero-shot forecasting capabilities of Time Series Foundation Models (TSFMs) for macroeconomic indicators. We apply TSFMs to forecasting economic indicators under univariate conditions, bypassing the need for train bespoke econometric models using and extensive training datasets. Our experiments were conducted on a case study dataset, without additional customisation. We rigorously back-tested three state-of-the-art TSFMs (Chronos, TimeGPT and Moirai) under data-scarce conditions and structural breaks. Our results demonstrate that appropriately engineered TSFMs can internalise rich economic dynamics, accommodate regime shifts, and deliver well-behaved uncertainty estimates out of the box, while matching state-of-the-art multivariate models on this domain. Our findings suggest that, without any fine-tuning, TSFMs can match or exceed classical models during stable economic conditions. However, they are vulnerable to degradation in performances during periods of rapid shocks. The findings offer guidance to practitioners on when zero-shot deployments are viable for macroeconomic monitoring and strategic planning.
Authors: Yunze Lin
Abstract: Mathematical reasoning presents a significant challenge for Large Language Models (LLMs) as it requires ensuring the correctness of each reasoning step. Researchers have been strengthening the mathematical reasoning abilities of LLMs through supervised fine-tuning, but due to the inability to suppress incorrect outputs, illusions can easily arise. Recently, Direct Preference Optimization (DPO) has been widely adopted for aligning human intent by using preference data to prevent LLMs from generating incorrect outputs. However, it has shown limited benefits in long-chain mathematical reasoning, mainly because DPO struggles to effectively capture the differences between accepted and rejected answers from preferences in long-chain data. The inconsistency between DPO training and LLMs' generation metrics also affects the effectiveness of suppressing incorrect outputs. We propose the Multi-Granularity Direct Preference Optimization (MDPO) method, optimizing the mathematical reasoning of LLMs at three granularities: Solution2Solution, Inference2Inference, and Step2Step. Solution2Solution focuses on the correctness of entire long-chain reasoning; Inference2Inference concentrates on logical reasoning between steps; Step2Step corrects computational errors in steps, enhancing the computational capabilities of LLMs. Additionally, we unify the training objectives of the three granularities to align with the generation metrics. We conducted experiments on the open-source models Qwen2 and Llama3, achieving improvements of 1.7% and 0.9% on the GSM8K dataset, and 2.3% and 1.2% on the MATH dataset, outperforming DPO and other DPO variant methods. Furthermore, we also provide a pipeline for constructing MDPO training data that is simple and does not require manual annotation costs.
Authors: Xinglin Wang, Yiwei Li, Shaoxiong Feng, Peiwen Yuan, Yueqi Zhang, Jiayi Shi, Chuyi Tan, Boyuan Pan, Yao Hu, Kan Li
Abstract: Test-Time Scaling (TTS) improves the performance of Large Language Models (LLMs) by using additional inference-time computation to explore multiple reasoning paths through search. Yet how to allocate a fixed rollout budget most effectively during search remains underexplored, often resulting in inefficient use of compute at test time. To bridge this gap, we formulate test-time search as a resource allocation problem and derive the optimal allocation strategy that maximizes the probability of obtaining a correct solution under a fixed rollout budget. Within this formulation, we reveal a core limitation of existing search methods: solution-level allocation tends to favor reasoning directions with more candidates, leading to theoretically suboptimal and inefficient use of compute. To address this, we propose Direction-Oriented Resource Allocation (DORA), a provably optimal method that mitigates this bias by decoupling direction quality from candidate count and allocating resources at the direction level. To demonstrate DORA's effectiveness, we conduct extensive experiments on challenging mathematical reasoning benchmarks including MATH500, AIME2024, and AIME2025. The empirical results show that DORA consistently outperforms strong baselines with comparable computational cost, achieving state-of-the-art accuracy. We hope our findings contribute to a broader understanding of optimal TTS for LLMs.
Authors: Falih Gozi Febrinanto, Adonia Simango, Chengpei Xu, Jingjing Zhou, Jiangang Ma, Sonika Tyagi, Feng Xia
Abstract: Graph neural networks (GNNs) have been developed to model the relationship between regions of interest (ROIs) in brains and have shown significant improvement in detecting brain diseases. However, most of these frameworks do not consider the intrinsic relationship of causality factor between brain ROIs, which is arguably more essential to observe cause and effect interaction between signals rather than typical correlation values. We propose a novel framework called CGB (Causal Graphs for Brains) for brain disease classification/detection, which models refined brain networks based on the causal discovery method, transfer entropy, and geometric curvature strategy. CGB unveils causal relationships between ROIs that bring vital information to enhance brain disease classification performance. Furthermore, CGB also performs a graph rewiring through a geometric curvature strategy to refine the generated causal graph to become more expressive and reduce potential information bottlenecks when GNNs model it. Our extensive experiments show that CGB outperforms state-of-the-art methods in classification tasks on brain disease datasets, as measured by average F1 scores.
Authors: Pedro C. Vieira, Miguel E. P. Silva, Pedro Manuel Pinto Ribeiro
Abstract: Graph Neural Networks (GNNs) are a predominant method for graph representation learning. However, beyond subgraph frequency estimation, their application to network motif significance-profile (SP) prediction remains under-explored, with no established benchmarks in the literature. We propose to address this problem, framing SP estimation as a task independent of subgraph frequency estimation. Our approach shifts from frequency counting to direct SP estimation and modulates the problem as multitarget regression. The reformulation is optimised for interpretability, stability and scalability on large graphs. We validate our method using a large synthetic dataset and further test it on real-world graphs. Our experiments reveal that 1-WL limited models struggle to make precise estimations of SPs. However, they can generalise to approximate the graph generation processes of networks by comparing their predicted SP with the ones originating from synthetic generators. This first study on GNN-based motif estimation also hints at how using direct SP estimation can help go past the theoretical limitations that motif estimation faces when performed through subgraph counting.
Authors: Siru Ouyang, Xinyu Zhu, Zilin Xiao, Minhao Jiang, Yu Meng, Jiawei Han
Abstract: Reinforcement learning (RL) has become a powerful approach for improving the reasoning capabilities of large language models (LLMs), as evidenced by recent successes such as OpenAI's o1 and Deepseek-R1. However, applying RL at scale remains intimidatingly resource-intensive, requiring multiple model copies and extensive GPU workloads. On the other hand, while being powerful, recent studies suggest that RL does not fundamentally endow models with new knowledge; rather, it primarily reshapes the model's output distribution to activate reasoning capabilities latent in the base model. Building on this insight, we hypothesize that the changes in output probabilities induced by RL are largely model-size invariant, opening the door to a more efficient paradigm: training a small model with RL and transferring its induced probability shifts to larger base models. To verify our hypothesis, we conduct a token-level analysis of decoding trajectories and find high alignment in RL-induced output distributions across model scales, validating our hypothesis. Motivated by this, we propose RAST, a simple yet effective method that transfers reasoning behaviors by injecting RL-induced probability adjustments from a small RL-trained model into larger models. Experiments across multiple mathematical reasoning benchmarks show that RAST substantially and consistently enhances the reasoning capabilities of base models while requiring significantly lower GPU memory than direct RL training, sometimes even yielding better performance than the RL-trained counterparts. Our findings offer new insights into the nature of RL-driven reasoning and practical strategies for scaling its benefits without incurring its full computational cost. The project page of RAST is available at https://ozyyshr.github.io/RAST/.
Authors: Le Jiang, Liyan Ma, Guang Yang
Abstract: Federated learning (FL) has emerged as a transformative framework for privacy-preserving distributed training, allowing clients to collaboratively train a global model without sharing their local data. This is especially crucial in sensitive fields like healthcare, where protecting patient data is paramount. However, privacy leakage remains a critical challenge, as the communication of model updates can be exploited by potential adversaries. Gradient inversion attacks (GIAs), for instance, allow adversaries to approximate the gradients used for training and reconstruct training images, thus stealing patient privacy. Existing defense mechanisms obscure gradients, yet lack a nuanced understanding of which gradients or types of image information are most vulnerable to such attacks. These indiscriminate calibrated perturbations result in either excessive privacy protection degrading model accuracy, or insufficient one failing to safeguard sensitive information. Therefore, we introduce a framework that addresses these challenges by leveraging a shadow model with interpretability for identifying sensitive areas. This enables a more targeted and sample-specific noise injection. Specially, our defensive strategy achieves discrepancies of 3.73 in PSNR and 0.2 in SSIM compared to the circumstance without defense on the ChestXRay dataset, and 2.78 in PSNR and 0.166 in the EyePACS dataset. Moreover, it minimizes adverse effects on model performance, with less than 1\% F1 reduction compared to SOTA methods. Our extensive experiments, conducted across diverse types of medical images, validate the generalization of the proposed framework. The stable defense improvements for FedAvg are consistently over 1.5\% times in LPIPS and SSIM. It also offers a universal defense against various GIA types, especially for these sensitive areas in images.
Authors: Songqi Zhou, Ruixue Liu, Yixing Wang, Jia Lu, Benben Jiang
Abstract: Accurate fault detection in lithium-ion batteries is essential for the safe and reliable operation of electric vehicles and energy storage systems. However, existing methods often struggle to capture complex temporal dependencies and cannot fully leverage abundant unlabeled data. Although large language models (LLMs) exhibit strong representation capabilities, their architectures are not directly suited to the numerical time-series data common in industrial settings. To address these challenges, we propose a novel framework that adapts BERT-style pretraining for battery fault detection by extending the standard BERT architecture with a customized time-series-to-token representation module and a point-level Masked Signal Modeling (point-MSM) pretraining task tailored to battery applications. This approach enables self-supervised learning on sequential current, voltage, and other charge-discharge cycle data, yielding distributionally robust, context-aware temporal embeddings. We then concatenate these embeddings with battery metadata and feed them into a downstream classifier for accurate fault classification. Experimental results on a large-scale real-world dataset show that models initialized with our pretrained parameters significantly improve both representation quality and classification accuracy, achieving an AUROC of 0.945 and substantially outperforming existing approaches. These findings validate the effectiveness of BERT-style pretraining for time-series fault detection.
Authors: Hanyu Pei, Jing-Xiao Liao, Qibin Zhao, Ting Gao, Shijun Zhang, Xiaoge Zhang, Feng-Lei Fan
Abstract: Drawing inspiration from our human brain that designs different neurons for different tasks, recent advances in deep learning have explored modifying a network's neurons to develop so-called task-driven neurons. Prototyping task-driven neurons (referred to as NeuronSeek) employs symbolic regression (SR) to discover the optimal neuron formulation and construct a network from these optimized neurons. Along this direction, this work replaces symbolic regression with tensor decomposition (TD) to discover optimal neuronal formulations, offering enhanced stability and faster convergence. Furthermore, we establish theoretical guarantees that modifying the aggregation functions with common activation functions can empower a network with a fixed number of parameters to approximate any continuous function with an arbitrarily small error, providing a rigorous mathematical foundation for the NeuronSeek framework. Extensive empirical evaluations demonstrate that our NeuronSeek-TD framework not only achieves superior stability, but also is competitive relative to the state-of-the-art models across diverse benchmarks. The code is available at https://github.com/HanyuPei22/NeuronSeek.
Authors: Angelos Assos, Carmel Baharav, Bailey Flanigan, Ariel Procaccia
Abstract: An increasingly influential form of deliberative democracy centers on citizens' assemblies, where randomly selected people discuss policy questions. The legitimacy of these panels hinges on their representation of the broader population, but panelists often drop out, leading to an unbalanced composition. Although participant attrition is mitigated in practice by alternates, their selection is not taken into account by existing methods. To address this gap, we introduce an optimization framework for alternate selection. Our algorithmic approach, which leverages learning-theoretic machinery, estimates dropout probabilities using historical data and selects alternates to minimize expected misrepresentation. We establish theoretical guarantees for our approach, including worst-case bounds on sample complexity (with implications for computational efficiency) and on loss when panelists' probabilities of dropping out are mis-estimated. Empirical evaluation using real-world data demonstrates that, compared to the status quo, our method significantly improves representation while requiring fewer alternates.
Authors: Zhengze Zhang, Shiqi Wang, Yiqun Shen, Simin Guo, Dahua Lin, Xiaoliang Wang, Nguyen Cam-Tu, Fei Tan
Abstract: Large language models (LLMs) have demonstrated exceptional performance across various applications, but their conversational abilities decline sharply as model size decreases, presenting a barrier to their deployment in resource-constrained environments. Knowledge distillation with Direct Preference Optimization (dDPO) has emerged as a promising approach to enhancing the conversational abilities of smaller models using a larger teacher model. However, current methods primarily focus on 'black-box' KD, which only uses the teacher's responses, overlooking the output distribution offered by the teacher. This paper addresses this gap by introducing daDPO (Distribution-Aware DPO), a unified method for preference optimization and distribution-based distillation. We provide rigorous theoretical analysis and empirical validation, showing that daDPO outperforms existing methods in restoring performance for pruned models and enhancing smaller LLM models. Notably, in in-domain evaluation, our method enables a 20% pruned Vicuna1.5-7B to achieve near-teacher performance (-7.3% preference rate compared to that of dDPO's -31%), and allows Qwen2.5-1.5B to occasionally outperform its 7B teacher model (14.0% win rate).
Authors: Wangzhi Zhan, Jianpeng Chen, Dongqi Fu, Dawei Zhou
Abstract: Metamaterials are artificial materials that are designed to meet unseen properties in nature, such as ultra-stiffness and negative materials indices. In mechanical metamaterial design, three key modalities are typically involved, i.e., 3D topology, density condition, and mechanical property. Real-world complex application scenarios place the demanding requirements on machine learning models to consider all three modalities together. However, a comprehensive literature review indicates that most existing works only consider two modalities, e.g., predicting mechanical properties given the 3D topology or generating 3D topology given the required properties. Therefore, there is still a significant gap for the state-of-the-art machine learning models capturing the whole. Hence, we propose a unified model named UNIMATE, which consists of a modality alignment module and a synergetic diffusion generation module. Experiments indicate that UNIMATE outperforms the other baseline models in topology generation task, property prediction task, and condition confirmation task by up to 80.2%, 5.1%, and 50.2%, respectively. We opensource our proposed UNIMATE model and corresponding results at https://github.com/wzhan24/UniMate.
Authors: Kunxi Li, Zhonghua Jiang, Zhouzhou Shen, Zhaode Wang, Chengfei Lv, Shengyu Zhang, Fan Wu, Fei Wu
Abstract: This paper introduces MadaKV, a modality-adaptive key-value (KV) cache eviction strategy designed to enhance the efficiency of multimodal large language models (MLLMs) in long-context inference. In multimodal scenarios, attention heads exhibit varying preferences for different modalities, resulting in significant disparities in modality importance across attention heads. Traditional KV cache eviction methods, which are tailored for unimodal settings, fail to capture modality-specific information, thereby yielding suboptimal performance. MadaKV addresses these challenges through two key components: modality preference adaptation and hierarchical compression compensation. By dynamically sensing modality information within attention heads and adaptively retaining critical tokens, MadaKV achieves substantial reductions in KV cache memory footprint and model inference decoding latency (1.3 to 1.5 times improvement) while maintaining high accuracy across various multimodal long-context tasks. Extensive experiments on representative MLLMs and the MileBench benchmark demonstrate the effectiveness of MadaKV compared to existing KV cache eviction methods.
Authors: Matteo Ninniri, Marco Podda, Davide Bacciu
Abstract: Generative models of graphs based on discrete Denoising Diffusion Probabilistic Models (DDPMs) offer a principled approach to molecular generation by systematically removing structural noise through iterative atom and bond adjustments. However, existing formulations are fundamentally limited by their inability to adapt the graph size (that is, the number of atoms) during the diffusion process, severely restricting their effectiveness in conditional generation scenarios such as property-driven molecular design, where the targeted property often correlates with the molecular size. In this paper, we reformulate the noising and denoising processes to support monotonic insertion and deletion of nodes. The resulting model, which we call GrIDDD, dynamically grows or shrinks the chemical graph during generation. GrIDDD matches or exceeds the performance of existing graph diffusion models on molecular property targeting despite being trained on a more difficult problem. Furthermore, when applied to molecular optimization, GrIDDD exhibits competitive performance compared to specialized optimization models. This work paves the way for size-adaptive molecular generation with graph diffusion.
Authors: Gautam Siddharth Kashyap, Md Tabrez Nafis, Samar Wazir
Abstract: Training Artificial Neural Networks (ANNs) with Stochastic Gradient Descent (SGD) frequently encounters difficulties, including substantial computing expense and the risk of converging to local optima, attributable to its dependence on partial weight gradients. Therefore, this work investigates Particle Swarm Optimization (PSO) and Genetic Algorithms (GAs) - two population-based Metaheuristic Optimizers (MHOs) - as alternatives to SGD to mitigate these constraints. A hybrid PSO-SGD strategy is developed to improve local search efficiency. The findings indicate that the hybrid PSO-SGD technique decreases the median training MSE by 90 to 95 percent relative to conventional GA and PSO across various network sizes (e.g., from around 0.02 to approximately 0.001 in the Sphere function). RMHC attains substantial enhancements, reducing MSE by roughly 85 to 90 percent compared to GA. Simultaneously, RS consistently exhibits errors exceeding 0.3, signifying subpar performance. These findings underscore that hybrid and evolutionary procedures significantly improve training efficiency and accuracy compared to conventional optimization methods and imply that the Building Block Hypothesis (BBH) may still be valid, indicating that advantageous weight structures are retained during evolutionary search.
Authors: Jo\~ao G. Ribeiro, Yaniv Oren, Alberto Sardinha, Matthijs Spaan, Francisco S. Melo
Abstract: This paper proposes RecBayes, a novel approach for ad hoc teamwork under partial observability, a setting where agents are deployed on-the-fly to environments where pre-existing teams operate, that never requires, at any stage, access to the states of the environment or the actions of its teammates. We show that by relying on a recurrent Bayesian classifier trained using past experiences, an ad hoc agent is effectively able to identify known teams and tasks being performed from observations alone. Unlike recent approaches such as PO-GPL (Gu et al., 2021) and FEAT (Rahman et al., 2023), that require at some stage fully observable states of the environment, actions of teammates, or both, or approaches such as ATPO (Ribeiro et al., 2023) that require the environments to be small enough to be tabularly modelled (Ribeiro et al., 2023), in their work up to 4.8K states and 1.7K observations, we show RecBayes is both able to handle arbitrarily large spaces while never relying on either states and teammates' actions. Our results in benchmark domains from the multi-agent systems literature, adapted for partial observability and scaled up to 1M states and 2^125 observations, show that RecBayes is effective at identifying known teams and tasks being performed from partial observations alone, and as a result, is able to assist the teams in solving the tasks effectively.
Authors: Peter Yichen Chen, Minghao Guo, Hanspeter Pfister, Ming Lin, William Freeman, Qixing Huang, Han-Wei Shen, Wojciech Matusik
Abstract: Computer graphics, often associated with films, games, and visual effects, has long been a powerful tool for addressing scientific challenges--from its origins in 3D visualization for medical imaging to its role in modern computational modeling and simulation. This course explores the deep and evolving relationship between computer graphics and science, highlighting past achievements, ongoing contributions, and open questions that remain. We show how core methods, such as geometric reasoning and physical modeling, provide inductive biases that help address challenges in both fields, especially in data-scarce settings. To that end, we aim to reframe graphics as a modeling language for science by bridging vocabulary gaps between the two communities. Designed for both newcomers and experts, Graphics4Science invites the graphics community to engage with science, tackle high-impact problems where graphics expertise can make a difference, and contribute to the future of scientific discovery. Additional details are available on the course website: https://graphics4science.github.io
Authors: Albert Dorador
Abstract: Piecewise-constant regression trees remain popular for their interpretability, yet often lag behind black-box models like Random Forest in predictive accuracy. In this work, we introduce TRUST (Transparent, Robust, and Ultra-Sparse Trees), a novel regression tree model that combines the accuracy of Random Forests with the interpretability of shallow decision trees and sparse linear models. TRUST further enhances transparency by leveraging Large Language Models to generate tailored, user-friendly explanations. Extensive validation on synthetic and real-world benchmark datasets demonstrates that TRUST consistently outperforms other interpretable models -- including CART, Lasso, and Node Harvest -- in predictive accuracy, while matching the accuracy of Random Forest and offering substantial gains in both accuracy and interpretability over M5', a well-established model that is conceptually related.
Authors: Ruipeng Liu, Qinru Qiu, Simon Khan, Garrett E. Katz
Abstract: A computational bottleneck in current Vector-Symbolic Architectures (VSAs) is the ``clean-up'' step, which decodes the noisy vectors retrieved from the architecture. Clean-up typically compares noisy vectors against a ``codebook'' of prototype vectors, incurring computational complexity that is quadratic or similar. We present a new codebook representation that supports efficient clean-up, based on Kroneker products of rotation-like matrices. The resulting clean-up time complexity is linearithmic, i.e. $\mathcal{O}(N\,\text{log}\,N)$, where $N$ is the vector dimension and also the number of vectors in the codebook. Clean-up space complexity is $\mathcal{O}(N)$. Furthermore, the codebook is not stored explicitly in computer memory: It can be represented in $\mathcal{O}(\text{log}\,N)$ space, and individual vectors in the codebook can be materialized in $\mathcal{O}(N)$ time and space. At the same time, asymptotic memory capacity remains comparable to standard approaches. Computer experiments confirm these results, demonstrating several orders of magnitude more scalability than baseline VSA techniques.
Authors: Taylor Lynn Curtis, Maximilian Puelma Touzel, William Garneau, Manon Gruaz, Mike Pinder, Li Wei Wang, Sukanya Krishna, Luda Cohen, Jean-Fran\c{c}ois Godbout, Reihaneh Rabbany, Kellin Pelrine
Abstract: The proliferation of misinformation poses a significant threat to society, exacerbated by the capabilities of generative AI. This demo paper introduces Veracity, an open-source AI system designed to empower individuals to combat misinformation through transparent and accessible fact-checking. Veracity leverages the synergy between Large Language Models (LLMs) and web retrieval agents to analyze user-submitted claims and provide grounded veracity assessments with intuitive explanations. Key features include multilingual support, numerical scoring of claim veracity, and an interactive interface inspired by familiar messaging applications. This paper will showcase Veracity's ability to not only detect misinformation but also explain its reasoning, fostering media literacy and promoting a more informed society.
Authors: Bohan Yang, Gang Liu, Rirao Dao, Yujia Qian, Ke Shi, Anke Tang, Yong Luo, Jingnan Liu
Abstract: Objective. Proton arc therapy (PAT) is an emerging and promising modality in radiotherapy, offering several advantages over conventional intensitymodulated proton therapy (IMPT). However, identifying the optimal energy layer (EL) sequence remains computationally intensive due to the large number of possible energy layer transitions. This study proposes an unsupervised deep learning framework for fast and effective EL pre-selection, aiming to minimize energy layer switch time while preserving high plan quality. Approach. We introduce a novel data representation method, spot-count representation, which encodes the number of proton spots intersecting the target and organs at risk (OARs) in a matrix structured by sorted gantry angles and energy layers. This representation is the input of a UNet-based architecture, SPArcdl, which is trained to optimize a tri-objective function: maximizing target coverage, minimizing OAR exposure, and reducing energy switching time. The model is evaluated on 54 nasopharyngeal cancer cases, and its performance is benchmarked against plans generated by SPArcparticle swarm. Main results. SPArcdl produces EL pre-selection that significantly improves both plan quality and delivery efficiency. Compared to SPArc particle swarm, it enhances the conformity index by 0.16 (p < 0.01), reduces the homogeneity index by 0.71 (p < 0.01), shortens the energy switching time by 38.4% (p < 0.01), and lowers the mean dose to brainstem by 0.21 (p < 0.01). The results unintentionally reveal employing unchanged ELS is more time-wise efficient than descended ELS. SPArcdl's inference time is within 1 second. Significance. SPArcdl is a fast and effective tool for generating high-quality PAT plans by strategically pre-selecting energy layers to reduce delivery time while maintaining excellent dosimetric performance.
Authors: Pham Khai Nguyen Do, Bao Nguyen Tran, Nam Nguyen, Duc Dung Nguyen
Abstract: Recent advances in Novel View Synthesis (NVS) and 3D generation have significantly improved editing tasks, with a primary emphasis on maintaining cross-view consistency throughout the generative process. Contemporary methods typically address this challenge using a dual-strategy framework: performing consistent 2D inpainting across all views guided by embedded priors either explicitly in pixel space or implicitly in latent space; and conducting 3D reconstruction with additional consistency guidance. Previous strategies, in particular, often require an initial 3D reconstruction phase to establish geometric structure, introducing considerable computational overhead. Even with the added cost, the resulting reconstruction quality often remains suboptimal. In this paper, we present VEIGAR, a computationally efficient framework that outperforms existing methods without relying on an initial reconstruction phase. VEIGAR leverages a lightweight foundation model to reliably align priors explicitly in the pixel space. In addition, we introduce a novel supervision strategy based on scale-invariant depth loss, which removes the need for traditional scale-and-shift operations in monocular depth regularization. Through extensive experimentation, VEIGAR establishes a new state-of-the-art benchmark in reconstruction quality and cross-view consistency, while achieving a threefold reduction in training time compared to the fastest existing method, highlighting its superior balance of efficiency and effectiveness.
Authors: Emanuele Musumeci, Michele Brienza, Francesco Argenziano, Vincenzo Suriani, Daniele Nardi, Domenico D. Bloisi
Abstract: Classical planning in AI and Robotics addresses complex tasks by shifting from imperative to declarative approaches (e.g., PDDL). However, these methods often fail in real scenarios due to limited robot perception and the need to ground perceptions to planning predicates. This often results in heavily hard-coded behaviors that struggle to adapt, even with scenarios where goals can be achieved through relaxed planning. Meanwhile, Large Language Models (LLMs) lead to planning systems that leverage commonsense reasoning but often at the cost of generating unfeasible and/or unsafe plans. To address these limitations, we present an approach integrating classical planning with LLMs, leveraging their ability to extract commonsense knowledge and ground actions. We propose a hierarchical formulation that enables robots to make unfeasible tasks tractable by defining functionally equivalent goals through gradual relaxation. This mechanism supports partial achievement of the intended objective, suited to the agent's specific context. Our method demonstrates its ability to adapt and execute tasks effectively within environments modeled using 3D Scene Graphs through comprehensive qualitative and quantitative evaluations. We also show how this method succeeds in complex scenarios where other benchmark methods are more likely to fail. Code, dataset, and additional material are released to the community.
Authors: Mingyuan Luo, Xin Yang, Zhongnuo Yan, Yan Cao, Yuanji Zhang, Xindi Hu, Jin Wang, Haoxuan Ding, Wei Han, Litao Sun, Dong Ni
Abstract: Three-dimensional (3D) ultrasound (US) aims to provide sonographers with the spatial relationships of anatomical structures, playing a crucial role in clinical diagnosis. Recently, deep-learning-based freehand 3D US has made significant advancements. It reconstructs volumes by estimating transformations between images without external tracking. However, image-only reconstruction poses difficulties in reducing cumulative drift and further improving reconstruction accuracy, particularly in scenarios involving complex motion trajectories. In this context, we propose an enhanced motion network (MoNetV2) to enhance the accuracy and generalizability of reconstruction under diverse scanning velocities and tactics. First, we propose a sensor-based temporal and multi-branch structure that fuses image and motion information from a velocity perspective to improve image-only reconstruction accuracy. Second, we devise an online multi-level consistency constraint that exploits the inherent consistency of scans to handle various scanning velocities and tactics. This constraint exploits both scan-level velocity consistency, path-level appearance consistency, and patch-level motion consistency to supervise inter-frame transformation estimation. Third, we distill an online multi-modal self-supervised strategy that leverages the correlation between network estimation and motion information to further reduce cumulative errors. Extensive experiments clearly demonstrate that MoNetV2 surpasses existing methods in both reconstruction quality and generalizability performance across three large datasets.
Authors: Zijian Zhou, Ao Qu, Zhaoxuan Wu, Sunghwan Kim, Alok Prakash, Daniela Rus, Jinhua Zhao, Bryan Kian Hsiang Low, Paul Pu Liang
Abstract: Modern language agents must operate over long-horizon, multi-turn interactions, where they retrieve external information, adapt to observations, and answer interdependent queries. Yet, most LLM systems rely on full-context prompting, appending all past turns regardless of their relevance. This leads to unbounded memory growth, increased computational costs, and degraded reasoning performance on out-of-distribution input lengths. We introduce MEM1, an end-to-end reinforcement learning framework that enables agents to operate with constant memory across long multi-turn tasks. At each turn, MEM1 updates a compact shared internal state that jointly supports memory consolidation and reasoning. This state integrates prior memory with new observations from the environment while strategically discarding irrelevant or redundant information. To support training in more realistic and compositional settings, we propose a simple yet effective and scalable approach to constructing multi-turn environments by composing existing datasets into arbitrarily complex task sequences. Experiments across three domains, including internal retrieval QA, open-domain web QA, and multi-turn web shopping, show that MEM1-7B improves performance by 3.5x while reducing memory usage by 3.7x compared to Qwen2.5-14B-Instruct on a 16-objective multi-hop QA task, and generalizes beyond the training horizon. Our results demonstrate the promise of reasoning-driven memory consolidation as a scalable alternative to existing solutions for training long-horizon interactive agents, where both efficiency and performance are optimized.
Authors: Glenn Matlin, Mika Okamoto, Huzaifa Pardawala, Yang Yang, Sudheer Chava
Abstract: Language Models (LMs) have demonstrated impressive capabilities with core Natural Language Processing (NLP) tasks. The effectiveness of LMs for highly specialized knowledge-intensive tasks in finance remains difficult to assess due to major gaps in the methodologies of existing evaluation frameworks, which have caused an erroneous belief in a far lower bound of LMs' performance on common Finance NLP (FinNLP) tasks. To demonstrate the potential of LMs for these FinNLP tasks, we present the first holistic benchmarking suite for Financial Language Model Evaluation (FLaME). We are the first research paper to comprehensively study LMs against 'reasoning-reinforced' LMs, with an empirical study of 23 foundation LMs over 20 core NLP tasks in finance. We open-source our framework software along with all data and results.
Authors: Arpit Bahety, Arnav Balaji, Ben Abbatematteo, Roberto Mart\'in-Mart\'in
Abstract: For robots to become efficient helpers in the home, they must learn to perform new mobile manipulation tasks simply by watching humans perform them. Learning from a single video demonstration from a human is challenging as the robot needs to first extract from the demo what needs to be done and how, translate the strategy from a third to a first-person perspective, and then adapt it to be successful with its own morphology. Furthermore, to mitigate the dependency on costly human monitoring, this learning process should be performed in a safe and autonomous manner. We present SafeMimic, a framework to learn new mobile manipulation skills safely and autonomously from a single third-person human video. Given an initial human video demonstration of a multi-step mobile manipulation task, SafeMimic first parses the video into segments, inferring both the semantic changes caused and the motions the human executed to achieve them and translating them to an egocentric reference. Then, it adapts the behavior to the robot's own morphology by sampling candidate actions around the human ones, and verifying them for safety before execution in a receding horizon fashion using an ensemble of safety Q-functions trained in simulation. When safe forward progression is not possible, SafeMimic backtracks to previous states and attempts a different sequence of actions, adapting both the trajectory and the grasping modes when required for its morphology. As a result, SafeMimic yields a strategy that succeeds in the demonstrated behavior and learns task-specific actions that reduce exploration in future attempts. Our experiments show that our method allows robots to safely and efficiently learn multi-step mobile manipulation behaviors from a single human demonstration, from different users, and in different environments, with improvements over state-of-the-art baselines across seven tasks
Authors: Pedro Mendes, Paolo Romano, David Garlan
Abstract: Modern neural networks (NNs) often achieve high predictive accuracy but remain poorly calibrated, producing overconfident predictions even when wrong. This miscalibration poses serious challenges in applications where reliable uncertainty estimates are critical. In this work, we investigate how human perceptual uncertainty compares to uncertainty estimated by NNs. Using three vision benchmarks annotated with both human disagreement and crowdsourced confidence, we assess the correlation between model-predicted uncertainty and human-perceived uncertainty. Our results show that current methods only weakly align with human intuition, with correlations varying significantly across tasks and uncertainty metrics. Notably, we find that incorporating human-derived soft labels into the training process can improve calibration without compromising accuracy. These findings reveal a persistent gap between model and human uncertainty and highlight the potential of leveraging human insights to guide the development of more trustworthy AI systems.
Authors: Amit Das, Naofumi Tomita, Kyle J. Syme, Weijie Ma, Paige O'Connor, Kristin N. Corbett, Bing Ren, Xiaoying Liu, Saeed Hassanpour
Abstract: Hematoxylin and Eosin (H&E) staining is a cornerstone of pathological analysis, offering reliable visualization of cellular morphology and tissue architecture for cancer diagnosis, subtyping, and grading. Immunohistochemistry (IHC) staining provides molecular insights by detecting specific proteins within tissues, enhancing diagnostic accuracy, and improving treatment planning. However, IHC staining is costly, time-consuming, and resource-intensive, requiring specialized expertise. To address these limitations, this study proposes HistoStainAlign, a novel deep learning framework that predicts IHC staining patterns directly from H&E whole-slide images (WSIs) by learning joint representations of morphological and molecular features. The framework integrates paired H&E and IHC embeddings through a contrastive training strategy, capturing complementary features across staining modalities without patch-level annotations or tissue registration. The model was evaluated on gastrointestinal and lung tissue WSIs with three commonly used IHC stains: P53, PD-L1, and Ki-67. HistoStainAlign achieved weighted F1 scores of 0.735 [95% Confidence Interval (CI): 0.670-0.799], 0.830 [95% CI: 0.772-0.886], and 0.723 [95% CI: 0.607-0.836], respectively for these three IHC stains. Embedding analyses demonstrated the robustness of the contrastive alignment in capturing meaningful cross-stain relationships. Comparisons with a baseline model further highlight the advantage of incorporating contrastive learning for improved stain pattern prediction. This study demonstrates the potential of computational approaches to serve as a pre-screening tool, helping prioritize cases for IHC staining and improving workflow efficiency.
Authors: Jushaan Singh Kalra, Xinran Zhao, To Eun Kim, Fengyu Cai, Fernando Diaz, Tongshuang Wu
Abstract: Retrieval-augmented Generation (RAG) is powerful, but its effectiveness hinges on which retrievers we use and how. Different retrievers offer distinct, often complementary signals: BM25 captures lexical matches; dense retrievers, semantic similarity. Yet in practice, we typically fix a single retriever based on heuristics, which fails to generalize across diverse information needs. Can we dynamically select and integrate multiple retrievers for each individual query, without the need for manual selection? In our work, we validate this intuition with quantitative analysis and introduce mixture of retrievers: a zero-shot, weighted combination of heterogeneous retrievers. Extensive experiments show that such mixtures are effective and efficient: Despite totaling just 0.8B parameters, this mixture outperforms every individual retriever and even larger 7B models by +10.8% and +3.9% on average, respectively. Further analysis also shows that this mixture framework can help incorporate specialized non-oracle human information sources as retrievers to achieve good collaboration, with a 58.9% relative performance improvement over simulated humans alone.
Authors: Sheng Liu, Tianlang Chen, Pan Lu, Haotian Ye, Yizheng Chen, Lei Xing, James Zou
Abstract: Test-time compute has emerged as a powerful paradigm for improving the performance of large language models (LLMs), where generating multiple outputs or refining individual chains can significantly boost answer accuracy. However, existing methods like Best-of-N, majority voting, and self-reflection typically apply reasoning in a uniform way across inputs, overlooking the fact that different problems may require different levels of reasoning depth. In this work, we propose Fractional Reasoning, a training-free and model-agnostic framework that enables continuous control over reasoning intensity at inference time, going beyond the limitations of fixed instructional prompts. Our method operates by extracting the latent steering vector associated with deeper reasoning and reapplying it with a tunable scaling factor, allowing the model to tailor its reasoning process to the complexity of each input. This supports two key modes of test-time scaling: (1) improving output quality in breadth-based strategies (e.g., Best-of-N, majority voting), and (2) enhancing the correctness of individual reasoning chains in depth-based strategies (e.g., self-reflection). Experiments on GSM8K, MATH500, and GPQA demonstrate that Fractional Reasoning consistently improves performance across diverse reasoning tasks and models.
Authors: Sam Silver, Jimin Sun, Ivan Zhang, Sara Hooker, Eddie Kim
Abstract: Large Language Models (LLMs) have demonstrated impressive mathematical reasoning capabilities, yet their performance remains brittle to minor variations in problem description and prompting strategy. Furthermore, reasoning is vulnerable to sampling-induced errors which autoregressive models must primarily address using self-correction via additionally-generated tokens. To better understand self-correction capabilities of recent models, we conduct experiments measuring models' ability to self-correct synthetic perturbations introduced into their Chain of Thought (CoT) reasoning. We observe robust single-utterance intrinsic self-correction behavior across a range of open-weight models and datasets, ranging from subtle, implicit corrections to explicit acknowledgments and corrections of errors. Our findings suggest that LLMs, including those not finetuned for long CoT, may possess stronger intrinsic self-correction capabilities than commonly shown in the literature. The presence of this ability suggests that recent "reasoning" model work involves amplification of traits already meaningfully present in models.
Authors: Yu Zhang, Gaoshan Bi, Simon Jeffery, Max Davis, Yang Li, Qing Xue, Po Yang
Abstract: Precision soil greenhouse gas (GHG) flux prediction is essential in agricultural systems for assessing environmental impacts, developing emission mitigation strategies and promoting sustainable agriculture. Due to the lack of advanced sensor and network technologies on majority of farms, there are challenges in obtaining comprehensive and diverse agricultural data. As a result, the scarcity of agricultural data seriously obstructs the application of machine learning approaches in precision soil GHG flux prediction. This research proposes a knowledge-guided graph neural network framework that addresses the above challenges by integrating knowledge embedded in an agricultural process-based model and graph neural network techniques. Specifically, we utilise the agricultural process-based model to simulate and generate multi-dimensional agricultural datasets for 47 countries that cover a wide range of agricultural variables. To extract key agricultural features and integrate correlations among agricultural features in the prediction process, we propose a machine learning framework that integrates the autoencoder and multi-target multi-graph based graph neural networks, which utilises the autoencoder to selectively extract significant agricultural features from the agricultural process-based model simulation data and the graph neural network to integrate correlations among agricultural features for accurately predict fertilisation-oriented soil GHG fluxes. Comprehensive experiments were conducted with both the agricultural simulation dataset and real-world agricultural dataset to evaluate the proposed approach in comparison with well-known baseline and state-of-the-art regression methods. The results demonstrate that our proposed approach provides superior accuracy and stability in fertilisation-oriented soil GHG prediction.
Authors: Liangyan Li, Yangyi Liu, Yimo Ning, Stefano Rini, Jun Chen
Abstract: Federated Learning (FL) has emerged as a powerful paradigm for leveraging diverse datasets from multiple sources while preserving data privacy by avoiding centralized storage. However, many existing approaches fail to account for the intricate gradient correlations between remote clients, a limitation that becomes especially problematic in data heterogeneity scenarios. In this work, we propose a novel FL framework utilizing Power-Norm Cosine Similarity (PNCS) to improve client selection for model aggregation. By capturing higher-order gradient moments, PNCS addresses non-IID data challenges, enhancing convergence speed and accuracy. Additionally, we introduce a simple algorithm ensuring diverse client selection through a selection history queue. Experiments with a VGG16 model across varied data partitions demonstrate consistent improvements over state-of-the-art methods.
Authors: Liangyan Li, Yimo Ning, Kevin Le, Wei Dong, Yunzhe Li, Jun Chen, Xiaohong Liu
Abstract: This paper introduces a novel framework for image and video demoir\'eing by integrating Maximum A Posteriori (MAP) estimation with advanced deep learning techniques. Demoir\'eing addresses inherently nonlinear degradation processes, which pose significant challenges for existing methods. Traditional supervised learning approaches either fail to remove moir\'e patterns completely or produce overly smooth results. This stems from constrained model capacity and scarce training data, which inadequately represent the clean image distribution and hinder accurate reconstruction of ground-truth images. While generative models excel in image restoration for linear degradations, they struggle with nonlinear cases such as demoir\'eing and often introduce artifacts. To address these limitations, we propose a hybrid MAP-based framework that integrates two complementary components. The first is a supervised learning model enhanced with efficient linear attention Test-Time Training (TTT) modules, which directly learn nonlinear mappings for RAW-to-sRGB demoir\'eing. The second is a Truncated Flow Matching Prior (TFMP) that further refines the outputs by aligning them with the clean image distribution, effectively restoring high-frequency details and suppressing artifacts. These two components combine the computational efficiency of linear attention with the refinement abilities of generative models, resulting in improved restoration performance.
Authors: Yosub Shin, Igor Molybog
Abstract: Video synchronization-aligning multiple video streams capturing the same event from different angles-is crucial for applications such as reality TV show production, sports analysis, surveillance, and autonomous systems. Prior work has heavily relied on audio cues or specific visual events, limiting applicability in diverse settings where such signals may be unreliable or absent. Additionally, existing benchmarks for video synchronization lack generality and reproducibility, restricting progress in the field. In this work, we introduce VideoSync, a video synchronization framework that operates independently of specific feature extraction methods, such as human pose estimation, enabling broader applicability across different content types. We evaluate our system on newly composed datasets covering single-human, multi-human, and non-human scenarios, providing both the methodology and code for dataset creation to establish reproducible benchmarks. Our analysis reveals biases in prior SOTA work, particularly in SeSyn-Net's preprocessing pipeline, leading to inflated performance claims. We correct these biases and propose a more rigorous evaluation framework, demonstrating that VideoSync outperforms existing approaches, including SeSyn-Net, under fair experimental conditions. Additionally, we explore various synchronization offset prediction methods, identifying a convolutional neural network (CNN)-based model as the most effective. Our findings advance video synchronization beyond domain-specific constraints, making it more generalizable and robust for real-world applications.
Authors: Yunchi Lu, Youshan Miao, Cheng Tan, Peng Huang, Yi Zhu, Xian Zhang, Fan Yang
Abstract: Training large language models (LLMs) at scale requires parallel execution across thousands of devices, incurring enormous computational costs. Yet, these costly distributed trainings are rarely verified, leaving them prone to silent errors and potentially wasting millions of GPU hours. We introduce TrainVerify, a system for verifiable distributed training of LLMs. Given a deep learning model's logical specification as the ground truth, TrainVerify formally verifies that a distributed parallel execution plan is mathematically equivalent to it. Direct verification is notoriously difficult due to the sheer scale of LLMs which often involves billions of variables and highly intricate computation graphs. Therefore, TrainVerify introduces shape-reduction techniques and a stage-wise parallel verification algorithm that significantly reduces complexity while preserving formal correctness. TrainVerify scales to frontier LLMs, including the successful verification of the Llama3 (405B) and DeepSeek-V3 (671B) training plans.
Authors: Jiawen Yang, Shuhao Chen, Yucong Duan, Ke Tang, Yu Zhang
Abstract: Unsupervised domain adaptation (UDA) methods effectively bridge domain gaps but become struggled when the source and target domains belong to entirely distinct modalities. To address this limitation, we propose a novel setting called Heterogeneous-Modal Unsupervised Domain Adaptation (HMUDA), which enables knowledge transfer between completely different modalities by leveraging a bridge domain containing unlabeled samples from both modalities. To learn under the HMUDA setting, we propose Latent Space Bridging (LSB), a specialized framework designed for the semantic segmentation task. Specifically, LSB utilizes a dual-branch architecture, incorporating a feature consistency loss to align representations across modalities and a domain alignment loss to reduce discrepancies between class centroids across domains. Extensive experiments conducted on six benchmark datasets demonstrate that LSB achieves state-of-the-art performance.
Authors: Toan Nguyen Hai, Ha Nguyen Viet, Truong Quan Xuan, Duc Do Minh
Abstract: Vietnamese, the 20th most spoken language with over 102 million native speakers, lacks robust resources for key natural language processing tasks such as text segmentation and machine reading comprehension (MRC). To address this gap, we present VSMRC, the Vietnamese Text Segmentation and Multiple-Choice Reading Comprehension Dataset. Sourced from Vietnamese Wikipedia, our dataset includes 15,942 documents for text segmentation and 16,347 synthetic multiple-choice question-answer pairs generated with human quality assurance, ensuring a reliable and diverse resource. Experiments show that mBERT consistently outperforms monolingual models on both tasks, achieving an accuracy of 88.01% on MRC test set and an F1 score of 63.15\% on text segmentation test set. Our analysis reveals that multilingual models excel in NLP tasks for Vietnamese, suggesting potential applications to other under-resourced languages. VSMRC is available at HuggingFace
Authors: Cong Wang, Zexuan Deng, Zhiwei Jiang, Fei Shen, Yafeng Yin, Shiwei Gan, Zifeng Cheng, Shiping Ge, Qing Gu
Abstract: Sign Language Video Generation (SLVG) seeks to generate identity-preserving sign language videos from spoken language texts. Existing methods primarily rely on the single coarse condition (\eg, skeleton sequences) as the intermediary to bridge the translation model and the video generation model, which limits both the naturalness and expressiveness of the generated videos. To overcome these limitations, we propose SignViP, a novel SLVG framework that incorporates multiple fine-grained conditions for improved generation fidelity. Rather than directly translating error-prone high-dimensional conditions, SignViP adopts a discrete tokenization paradigm to integrate and represent fine-grained conditions (\ie, fine-grained poses and 3D hands). SignViP contains three core components. (1) Sign Video Diffusion Model is jointly trained with a multi-condition encoder to learn continuous embeddings that encapsulate fine-grained motion and appearance. (2) Finite Scalar Quantization (FSQ) Autoencoder is further trained to compress and quantize these embeddings into discrete tokens for compact representation of the conditions. (3) Multi-Condition Token Translator is trained to translate spoken language text to discrete multi-condition tokens. During inference, Multi-Condition Token Translator first translates the spoken language text into discrete multi-condition tokens. These tokens are then decoded to continuous embeddings by FSQ Autoencoder, which are subsequently injected into Sign Video Diffusion Model to guide video generation. Experimental results show that SignViP achieves state-of-the-art performance across metrics, including video quality, temporal coherence, and semantic fidelity. The code is available at https://github.com/umnooob/signvip/.
Authors: Markus Frohmann, Gabriel Meseguer-Brocal, Markus Schedl, Elena V. Epure
Abstract: The rapid advancement of AI-based music generation tools is revolutionizing the music industry but also posing challenges to artists, copyright holders, and providers alike. This necessitates reliable methods for detecting such AI-generated content. However, existing detectors, relying on either audio or lyrics, face key practical limitations: audio-based detectors fail to generalize to new or unseen generators and are vulnerable to audio perturbations; lyrics-based methods require cleanly formatted and accurate lyrics, unavailable in practice. To overcome these limitations, we propose a novel, practically grounded approach: a multimodal, modular late-fusion pipeline that combines automatically transcribed sung lyrics and speech features capturing lyrics-related information within the audio. By relying on lyrical aspects directly from audio, our method enhances robustness, mitigates susceptibility to low-level artifacts, and enables practical applicability. Experiments show that our method, DE-detect, outperforms existing lyrics-based detectors while also being more robust to audio perturbations. Thus, it offers an effective, robust solution for detecting AI-generated music in real-world scenarios. Our code is available at https://github.com/deezer/robust-AI-lyrics-detection.
Authors: Hemanth Kannamarlapudi, Sowmya Chintalapudi
Abstract: Navigation is a very crucial aspect of autonomous vehicle ecosystem which heavily relies on collecting and processing large amounts of data in various states and taking a confident and safe decision to define the next vehicle maneuver. In this paper, we propose a novel architecture based on Quantum Artificial Intelligence by enabling quantum and AI at various levels of navigation decision making and communication process in Autonomous vehicles : Quantum Neural Networks for multimodal sensor fusion, Nav-Q for Quantum reinforcement learning for navigation policy optimization and finally post-quantum cryptographic protocols for secure communication. Quantum neural networks uses quantum amplitude encoding to fuse data from various sensors like LiDAR, radar, camera, GPS and weather etc., This approach gives a unified quantum state representation between heterogeneous sensor modalities. Nav-Q module processes the fused quantum states through variational quantum circuits to learn optimal navigation policies under swift dynamic and complex conditions. Finally, post quantum cryptographic protocols are used to secure communication channels for both within vehicle communication and V2X (Vehicle to Everything) communications and thus secures the autonomous vehicle communication from both classical and quantum security threats. Thus, the proposed framework addresses fundamental challenges in autonomous vehicles navigation by providing quantum performance and future proof security. Index Terms Quantum Computing, Autonomous Vehicles, Sensor Fusion
Authors: Qianru Zhang, Honggang Wen, Ming Li, Dong Huang, Siu-Ming Yiu, Christian S. Jensen, Pietro Li\`o
Abstract: Time series forecasting requires architectures that simultaneously achieve three competing objectives: (1) strict temporal causality for reliable predictions, (2) sub-quadratic complexity for practical scalability, and (3) multi-scale pattern recognition for accurate long-horizon forecasting. We introduce AutoHFormer, a hierarchical autoregressive transformer that addresses these challenges through three key innovations: 1) Hierarchical Temporal Modeling: Our architecture decomposes predictions into segment-level blocks processed in parallel, followed by intra-segment sequential refinement. This dual-scale approach maintains temporal coherence while enabling efficient computation. 2) Dynamic Windowed Attention: The attention mechanism employs learnable causal windows with exponential decay, reducing complexity while preserving precise temporal relationships. This design avoids both the anti-causal violations of standard transformers and the sequential bottlenecks of RNN hybrids. 3) Adaptive Temporal Encoding: a novel position encoding system is adopted to capture time patterns at multiple scales. It combines fixed oscillating patterns for short-term variations with learnable decay rates for long-term trends. Comprehensive experiments demonstrate that AutoHFormer 10.76X faster training and 6.06X memory reduction compared to PatchTST on PEMS08, while maintaining consistent accuracy across 96-720 step horizons in most of cases. These breakthroughs establish new benchmarks for efficient and precise time series modeling. Implementations of our method and all baselines in hierarchical autoregressive mechanism are available at https://github.com/lizzyhku/Autotime.
Authors: Weiwei Duan, Michael P. Gerlek, Steven N. Minton, Craig A. Knoblock, Fandel Lin, Theresa Chen, Leeje Jang, Sofia Kirsanova, Zekun Li, Yijun Lin, Yao-Yi Chiang
Abstract: Historical geologic maps contain rich geospatial information, such as rock units, faults, folds, and bedding planes, that is critical for assessing mineral resources essential to renewable energy, electric vehicles, and national security. However, digitizing maps remains a labor-intensive and time-consuming task. We present DIGMAPPER, a modular, scalable system developed in collaboration with the United States Geological Survey (USGS) to automate the digitization of geologic maps. DIGMAPPER features a fully dockerized, workflow-orchestrated architecture that integrates state-of-the-art deep learning models for map layout analysis, feature extraction, and georeferencing. To overcome challenges such as limited training data and complex visual content, our system employs innovative techniques, including in-context learning with large language models, synthetic data generation, and transformer-based models. Evaluations on over 100 annotated maps from the DARPA-USGS dataset demonstrate high accuracy across polygon, line, and point feature extraction, and reliable georeferencing performance. Deployed at USGS, DIGMAPPER significantly accelerates the creation of analysis-ready geospatial datasets, supporting national-scale critical mineral assessments and broader geoscientific applications.
Authors: Jina Kim, Youjin Jang, Jeongjin Han
Abstract: We propose VRAIL (Vectorized Reward-based Attribution for Interpretable Learning), a bi-level framework for value-based reinforcement learning (RL) that learns interpretable weight representations from state features. VRAIL consists of two stages: a deep learning (DL) stage that fits an estimated value function using state features, and an RL stage that uses this to shape learning via potential-based reward transformations. The estimator is modeled in either linear or quadratic form, allowing attribution of importance to individual features and their interactions. Empirical results on the Taxi-v3 environment demonstrate that VRAIL improves training stability and convergence compared to standard DQN, without requiring environment modifications. Further analysis shows that VRAIL uncovers semantically meaningful subgoals, such as passenger possession, highlighting its ability to produce human-interpretable behavior. Our findings suggest that VRAIL serves as a general, model-agnostic framework for reward shaping that enhances both learning and interpretability.
Authors: Zhihan Guo, Jiele Wu, Wenqian Cui, Yifei Zhang, Minda Hu, Yufei Wang, Irwin King
Abstract: Current research on long-form context in Large Language Models (LLMs) primarily focuses on the understanding of long-contexts, the Open-ended Long Text Generation (Open-LTG) remains insufficiently explored. Training a long-context generation model requires curation of gold standard reference data, which is typically nonexistent for informative Open-LTG tasks. However, previous methods only utilize general assessments as reward signals, which limits accuracy. To bridge this gap, we introduce ProxyReward, an innovative reinforcement learning (RL) based framework, which includes a dataset and a reward signal computation method. Firstly, ProxyReward Dataset generation is accomplished through simple prompts that enables the model to create automatically, obviating extensive labeled data or significant manual effort. Secondly, ProxyReward Signal offers a targeted evaluation of information comprehensiveness and accuracy for specific questions. The experimental results indicate that our method ProxyReward surpasses even GPT-4-Turbo. It can significantly enhance performance by 20% on the Open-LTG task when training widely used open-source models, while also surpassing the LLM-as-a-Judge approach. Our work presents effective methods to enhance the ability of LLMs to address complex open-ended questions posed by human.
Authors: Zhenting Qi, Fan Nie, Alexandre Alahi, James Zou, Himabindu Lakkaraju, Yilun Du, Eric Xing, Sham Kakade, Hanlin Zhang
Abstract: Modern language model (LM) training has been divided into multiple stages, making it difficult for downstream developers to evaluate the impact of design choices made at each stage. We present EvoLM, a model suite that enables systematic and transparent analysis of LMs' training dynamics across pre-training, continued pre-training, supervised fine-tuning, and reinforcement learning. By training over 100 LMs with 1B and 4B parameters from scratch, we rigorously evaluate both upstream (language modeling) and downstream (problem-solving) reasoning capabilities, including considerations of both in-domain and out-of-domain generalization. Key insights highlight the diminishing returns from excessive pre-training and post-training, the importance and practices of mitigating forgetting during domain-specific continued pre-training, the crucial role of continued pre-training in bridging pre-training and post-training phases, and various intricate trade-offs when configuring supervised fine-tuning and reinforcement learning. To facilitate open research and reproducibility, we release all pre-trained and post-trained models, training datasets for all stages, and our entire training and evaluation pipeline.
Authors: Vishesh Tripathi, Tanmay Odapally, Indraneel Das, Uday Allu, Biddwan Ahmed
Abstract: Retrieval-Augmented Generation (RAG) systems have revolutionized information retrieval and question answering, but traditional text-based chunking methods struggle with complex document structures, multi-page tables, embedded figures, and contextual dependencies across page boundaries. We present a novel multimodal document chunking approach that leverages Large Multimodal Models (LMMs) to process PDF documents in batches while maintaining semantic coherence and structural integrity. Our method processes documents in configurable page batches with cross-batch context preservation, enabling accurate handling of tables spanning multiple pages, embedded visual elements, and procedural content. We evaluate our approach on a curated dataset of PDF documents with manually crafted queries, demonstrating improvements in chunk quality and downstream RAG performance. Our vision-guided approach achieves better accuracy compared to traditional vanilla RAG systems, with qualitative analysis showing superior preservation of document structure and semantic coherence.
Authors: Fei Wang, Xingchen Wan, Ruoxi Sun, Jiefeng Chen, Sercan \"O. Ar{\i}k
Abstract: Inference-time scaling has proven effective in boosting large language model (LLM) performance through increased test-time computation. Yet, its practical application is often hindered by reliance on external verifiers or a lack of optimization for realistic computational constraints. We propose DynScaling, which addresses these limitations through two primary innovations: an integrated parallel-sequential sampling strategy and a bandit-based dynamic budget allocation framework. The integrated sampling strategy unifies parallel and sequential sampling by constructing synthetic sequential reasoning chains from initially independent parallel responses, promoting diverse and coherent reasoning trajectories. The dynamic budget allocation framework formulates the allocation of computational resources as a multi-armed bandit problem, adaptively distributing the inference budget across queries based on the uncertainty of previously sampled responses, thereby maximizing computational efficiency. By combining these components, DynScaling effectively improves LLM performance under practical resource constraints without the need for external verifiers. Experimental results demonstrate that DynScaling consistently surpasses existing verifier-free inference scaling baselines in both task performance and computational cost.
Authors: Devesh Kumar
Abstract: The proliferation of online communication platforms has created unprecedented opportunities for global connectivity while simultaneously enabling harmful behaviors such as cyberbullying, which affects approximately 54.4\% of teenagers according to recent research. This paper presents a hybrid architecture that combines the contextual understanding capabilities of transformer-based models with the pattern recognition strengths of broad learning systems for effective cyberbullying detection. This approach integrates a modified DeBERTa model augmented with Squeeze-and-Excitation blocks and sentiment analysis capabilities with a Gated Broad Learning System (GBLS) classifier, creating a synergistic framework that outperforms existing approaches across multiple benchmark datasets. The proposed ModifiedDeBERTa + GBLS model achieved good performance on four English datasets: 79.3\% accuracy on HateXplain, 95.41\% accuracy on SOSNet, 91.37\% accuracy on Mendeley-I, and 94.67\% accuracy on Mendeley-II. Beyond performance gains, the framework incorporates comprehensive explainability mechanisms including token-level attribution analysis, LIME-based local interpretations, and confidence calibration, addressing critical transparency requirements in automated content moderation. Ablation studies confirm the meaningful contribution of each architectural component, while failure case analysis reveals specific challenges in detecting implicit bias and sarcastic content, providing valuable insights for future improvements in cyberbullying detection systems.
Authors: Puchun Liu, C. L. Philip Chen, Yubin He, Tong Zhang
Abstract: The difficulty of extracting deep features from EEG data and effectively integrating information from multiple views presents significant challenges for developing a generalizable pretraining framework for EEG representation learning. However, most existing pre-training methods rely solely on the contextual semantics of a single view, failing to capture the complex and synergistic interactions among different perspectives, limiting the expressiveness and generalization of learned representations. To address these issues, this paper proposes CRIA, an adaptive framework that utilizes variable-length and variable-channel coding to achieve a unified representation of EEG data across different datasets. In this work, we define cross-view information as the integrated representation that emerges from the interaction among temporal, spectral, and spatial views of EEG signals. The model employs a cross-attention mechanism to fuse temporal, spectral, and spatial features effectively, and combines an attention matrix masking strategy based on the information bottleneck principle with a novel viewpoint masking pre-training scheme. Experimental results on the Temple University EEG corpus and the CHB-MIT dataset show that CRIA outperforms existing methods with the same pre-training conditions, achieving a balanced accuracy of 57.02% for multi-class event classification and 80.03% for anomaly detection, highlighting its strong generalization ability.
Authors: Tianle Gu, Kexin Huang, Zongqi Wang, Yixu Wang, Jie Li, Yuanqi Yao, Yang Yao, Yujiu Yang, Yan Teng, Yingchun Wang
Abstract: Safety alignment is a key requirement for building reliable Artificial General Intelligence. Despite significant advances in safety alignment, we observe that minor latent shifts can still trigger unsafe responses in aligned models. We argue that this stems from the shallow nature of existing alignment methods, which focus on surface-level refusal behaviors without sufficiently altering internal representations. Consequently, small shifts in hidden activations can re-trigger harmful behaviors embedded in the latent space. To explore the robustness of safety alignment to latent perturbations, we introduce a probing method that measures the Negative Log-Likelihood of the original response generated by the model. This probe quantifies local sensitivity in the latent space, serving as a diagnostic tool for identifying vulnerable directions. Based on this signal, we construct effective jailbreak trajectories, giving rise to the Activation Steering Attack (ASA). More importantly, these insights offer a principled foundation for improving alignment robustness. To this end, we introduce Layer-wise Adversarial Patch Training~(LAPT), a fine-tuning strategy that inject controlled perturbations into hidden representations during training. Experimental results highlight that LAPT strengthen alignment robustness without compromising general capabilities. Our findings reveal fundamental flaws in current alignment paradigms and call for representation-level training strategies that move beyond surface-level behavior supervision. Codes and results are available at https://github.com/Carol-gutianle/LatentSafety.
Authors: Qianqian Liao, Wuque Cai, Hongze Sun, Dongze Liu, Duo Chen, Dezhong Yao, Daqing Guo
Abstract: Recent developed graph-based methods for diagnosing brain disorders using functional connectivity highly rely on predefined brain atlases, but overlook the rich information embedded within atlases and the confounding effects of site and phenotype variability. To address these challenges, we propose a two-stage Brain-to-Population Graph Learning (B2P-GL) framework that integrates the semantic similarity of brain regions and condition-based population graph modeling. In the first stage, termed brain representation learning, we leverage brain atlas knowledge from GPT-4 to enrich the graph representation and refine the brain graph through an adaptive node reassignment graph attention network. In the second stage, termed population disorder diagnosis, phenotypic data is incorporated into population graph construction and feature fusion to mitigate confounding effects and enhance diagnosis performance. Experiments on the ABIDE I, ADHD-200, and Rest-meta-MDD datasets show that B2P-GL outperforms state-of-the-art methods in prediction accuracy while enhancing interpretability. Overall, our proposed framework offers a reliable and personalized approach to brain disorder diagnosis, advancing clinical applicability.
Authors: Yejing Wang, Shengyu Zhou, Jinyu Lu, Qidong Liu, Xinhang Li, Wenlin Zhang, Feng Li, Pengjie Wang, Jian Xu, Bo Zheng, Xiangyu Zhao
Abstract: Generative recommendations (GR), which usually include item tokenizers and generative Large Language Models (LLMs), have demonstrated remarkable success across a wide range of scenarios. The majority of existing research efforts primarily concentrate on developing powerful item tokenizers or advancing LLM decoding strategies to attain superior performance. However, the critical fine-tuning step in GR frameworks, which is essential for adapting LLMs to recommendation data, remains largely unexplored. Current approaches predominantly rely on either the next-token prediction loss of supervised fine-tuning (SFT) or recommendationspecific direct preference optimization (DPO) strategies. Both methods ignore the exploration of possible positive unobserved samples, which is commonly referred to as the exposure bias problem. To mitigate this problem, this paper treats the GR as a multi-step generation task and constructs a GFlowNets-based fine-tuning framework (GFlowGR). The proposed framework integrates collaborative knowledge from traditional recommender systems to create an adaptive trajectory sampler and a comprehensive reward model. Leveraging the diverse generation property of GFlowNets, along with sampling and heuristic weighting techniques, GFlowGR emerges as a promising approach to mitigate the exposure bias problem. Extensive empirical results on two real-world datasets and with two different GR backbones highlight the effectiveness and robustness of GFlowGR.
Authors: Shoutrik Das, Nishant Singh, Arjun Gangwar, S Umesh
Abstract: Dysarthria is a neurological disorder that significantly impairs speech intelligibility, often rendering affected individuals unable to communicate effectively. This necessitates the development of robust dysarthric-to-regular speech conversion techniques. In this work, we investigate the utility and limitations of self-supervised learning (SSL) features and their quantized representations as an alternative to mel-spectrograms for speech generation. Additionally, we explore methods to mitigate speaker variability by generating clean speech in a single-speaker voice using features extracted from WavLM. To this end, we propose a fully non-autoregressive approach that leverages Conditional Flow Matching (CFM) with Diffusion Transformers to learn a direct mapping from dysarthric to clean speech. Our findings highlight the effectiveness of discrete acoustic units in improving intelligibility while achieving faster convergence compared to traditional mel-spectrogram-based approaches.
Authors: Yi Chen, Yuying Ge, Rui Wang, Yixiao Ge, Junhao Cheng, Ying Shan, Xihui Liu
Abstract: Recent reinforcement learning approaches, such as outcome-supervised GRPO, have advanced Chain-of-Thought reasoning in large language models (LLMs), yet their adaptation to multimodal LLMs (MLLMs) is unexplored. To address the lack of rigorous evaluation for MLLM post-training methods, we introduce SEED-Bench-R1, a benchmark with complex real-world videos requiring balanced perception and reasoning. It offers a large training set and evaluates generalization across three escalating challenges: in-distribution, cross-environment, and cross-environment-task scenarios. Using SEED-Bench-R1, we find that standard GRPO, while improving answer accuracy, often reduces logical coherence between reasoning steps and answers, with only a 57.9% consistency rate. This stems from reward signals focusing solely on final answers, encouraging shortcuts, and strict KL penalties limiting exploration.To address this, we propose GRPO-CARE, a consistency-aware RL framework optimizing both answer correctness and reasoning coherence without explicit supervision. GRPO-CARE introduces a two-tiered reward: (1) a base reward for answer correctness, and (2) an adaptive consistency bonus, computed by comparing the model's reasoning-to-answer likelihood (via a slowly-evolving reference model) against group peers.This dual mechanism amplifies rewards for reasoning paths that are both correct and logically consistent. Replacing KL penalties with this adaptive bonus, GRPO-CARE outperforms standard GRPO on SEED-Bench-R1, achieving a 6.7% performance gain on the hardest evaluation level and a 24.5% improvement in consistency. It also shows strong transferability, improving model performance across diverse video understanding benchmarks. Our work contributes a systematically designed benchmark and a generalizable post-training framework, advancing the development of more interpretable and robust MLLMs.
Authors: Xinyi Wu, Geng Hong, Pei Chen, Yueyue Chen, Xudong Pan, Min Yang
Abstract: As large language models (LLMs) advance, concerns about their misconduct in complex social contexts intensify. Existing research overlooked the systematic understanding and assessment of their criminal capability in realistic interactions. We propose a unified framework PRISON, to quantify LLMs' criminal potential across five dimensions: False Statements, Frame-Up, Psychological Manipulation, Emotional Disguise, and Moral Disengagement. Using structured crime scenarios adapted from classic films, we evaluate both criminal potential and anti-crime ability of LLMs via role-play. Results show that state-of-the-art LLMs frequently exhibit emergent criminal tendencies, such as proposing misleading statements or evasion tactics, even without explicit instructions. Moreover, when placed in a detective role, models recognize deceptive behavior with only 41% accuracy on average, revealing a striking mismatch between conducting and detecting criminal behavior. These findings underscore the urgent need for adversarial robustness, behavioral alignment, and safety mechanisms before broader LLM deployment.
Authors: Chenxi Wang, Yixuan Zhang, Lang Gao, Zixiang Xu, Zirui Song, Yanbo Wang, Xiuying Chen
Abstract: Language is not only a tool for communication but also a medium for human cognition and reasoning. If, as linguistic relativity suggests, the structure of language shapes cognitive patterns, then large language models (LLMs) trained on human language may also internalize the habitual logical structures embedded in different languages. To examine this hypothesis, we introduce BICAUSE, a structured bilingual dataset for causal reasoning, which includes semantically aligned Chinese and English samples in both forward and reversed causal forms. Our study reveals three key findings: (1) LLMs exhibit typologically aligned attention patterns, focusing more on causes and sentence-initial connectives in Chinese, while showing a more balanced distribution in English. (2) Models internalize language-specific preferences for causal word order and often rigidly apply them to atypical inputs, leading to degraded performance, especially in Chinese. (3) When causal reasoning succeeds, model representations converge toward semantically aligned abstractions across languages, indicating a shared understanding beyond surface form. Overall, these results suggest that LLMs not only mimic surface linguistic forms but also internalize the reasoning biases shaped by language. Rooted in cognitive linguistic theory, this phenomenon is for the first time empirically verified through structural analysis of model internals.
Authors: Thomas Barbera, Jacopo Burger, Alessandro D'Amelio, Simone Zini, Simone Bianco, Raffaella Lanzarotti, Paolo Napoletano, Giuseppe Boccignone, Jose Luis Contreras-Vidal
Abstract: Imagine unlocking the power of the mind to communicate, create, and even interact with the world around us. Recent breakthroughs in Artificial Intelligence (AI), especially in how machines "see" and "understand" language, are now fueling exciting progress in decoding brain signals from scalp electroencephalography (EEG). Prima facie, this opens the door to revolutionary brain-computer interfaces (BCIs) designed for real life, moving beyond traditional uses to envision Brain-to-Speech, Brain-to-Image, and even a Brain-to-Internet of Things (BCIoT). However, the journey is not as straightforward as it was for Computer Vision (CV) and Natural Language Processing (NLP). Applying AI to real-world EEG-based BCIs, particularly in building powerful foundational models, presents unique and intricate hurdles that could affect their reliability. Here, we unfold a guided exploration of this dynamic and rapidly evolving research area. Rather than barely outlining a map of current endeavors and results, the goal is to provide a principled navigation of this hot and cutting-edge research landscape. We consider the basic paradigms that emerge from a causal perspective and the attendant challenges presented to AI-based models. Looking ahead, we then discuss promising research avenues that could overcome today's technological, methodological, and ethical limitations. Our aim is to lay out a clear roadmap for creating truly practical and effective EEG-based BCI solutions that can thrive in everyday environments.
Authors: Simardeep Singh
Abstract: Large language models (LLMs) are known to memorize parts of their training data, raising important concerns around privacy and security. While previous research has focused on studying memorization in pre-trained models, much less is known about how knowledge distillation (KD) affects memorization.In this study, we explore how different KD methods influence the memorization of fine-tuned task data when a large teacher model is distilled into smaller student variants.This study demonstrates that distilling a larger teacher model, fine-tuned on a dataset, into a smaller variant not only lowers computational costs and model size but also significantly reduces the memorization risks compared to standard fine-tuning approaches.
Authors: Masashi Takeshita, Rafal Rzepka
Abstract: In this work, we propose JETHICS, a Japanese dataset for evaluating ethics understanding of AI models. JETHICS contains 78K examples and is built by following the construction methods of the existing English ETHICS dataset. It includes four categories based normative theories and concepts from ethics and political philosophy; and one representing commonsense morality. Our evaluation experiments on non-proprietary large language models (LLMs) and on GPT-4o reveal that even GPT-4o achieves only an average score of about 0.7, while the best-performing Japanese LLM attains around 0.5, indicating a relatively large room for improvement in current LLMs.
Authors: Putri A. van der Linden, Alexander Timans, Erik J. Bekkers
Abstract: We study the problem of conformal prediction (CP) under geometric data shifts, where data samples are susceptible to transformations such as rotations or flips. While CP endows prediction models with post-hoc uncertainty quantification and formal coverage guarantees, their practicality breaks under distribution shifts that deteriorate model performance. To address this issue, we propose integrating geometric information--such as geometric pose--into the conformal procedure to reinstate its guarantees and ensure robustness under geometric shifts. In particular, we explore recent advancements on pose canonicalization as a suitable information extractor for this purpose. Evaluating the combined approach across discrete and continuous shifts and against equivariant and augmentation-based baselines, we find that integrating geometric information with CP yields a principled way to address geometric shifts while maintaining broad applicability to black-box predictors.
Authors: Raghav Mehta, Fabio De Sousa Ribeiro, Tian Xia, Melanie Roschewitz, Ainkaran Santhirasekaram, Dominic C. Marshall, Ben Glocker
Abstract: Segmenting anatomical structures in medical images plays an important role in the quantitative assessment of various diseases. However, accurate segmentation becomes significantly more challenging in the presence of disease. Disease patterns can alter the appearance of surrounding healthy tissues, introduce ambiguous boundaries, or even obscure critical anatomical structures. As such, segmentation models trained on real-world datasets may struggle to provide good anatomical segmentation, leading to potential misdiagnosis. In this paper, we generate counterfactual (CF) images to simulate how the same anatomy would appear in the absence of disease without altering the underlying structure. We then use these CF images to segment structures of interest, without requiring any changes to the underlying segmentation model. Our experiments on two real-world clinical chest X-ray datasets show that the use of counterfactual images improves anatomical segmentation, thereby aiding downstream clinical decision-making.
Authors: Abdulvahap Mutlu, \c{S}eng\"ul Do\u{g}an, T\"urker Tuncer
Abstract: Amyotrophic Lateral Sclerosis (ALS) is a rare neurodegenerative disease, and high-quality EEG data from ALS patients are scarce. This data scarcity, coupled with severe class imbalance between ALS and healthy control recordings, poses a challenge for training reliable machine learning classifiers. In this work, we address these issues by generating synthetic EEG signals for ALS patients using a Conditional Wasserstein Generative Adversarial Network (CWGAN). We train CWGAN on a private EEG dataset (ALS vs. non-ALS) to learn the distribution of ALS EEG signals and produce realistic synthetic samples. We preprocess and normalize EEG recordings, and train a CWGAN model to generate synthetic ALS signals. The CWGAN architecture and training routine are detailed, with key hyperparameters chosen for stable training. Qualitative evaluation of generated signals shows that they closely mimic real ALS EEG patterns. The CWGAN training converged with generator and discriminator loss curves stabilizing, indicating successful learning. The synthetic EEG signals appear realistic and have potential use as augmented data for training classifiers, helping to mitigate class imbalance and improve ALS detection accuracy. We discuss how this approach can facilitate data sharing and enhance diagnostic models.
Authors: Xingzhong Fan, Hongming Tang, Yue Zeng, M. B. N. Kouwenhoven, Guangquan Zeng
Abstract: Conventional galaxy generation methods rely on semi-analytical models and hydrodynamic simulations, which are highly dependent on physical assumptions and parameter tuning. In contrast, data-driven generative models do not have explicit physical parameters pre-determined, and instead learn them efficiently from observational data, making them alternative solutions to galaxy generation. Among these, diffusion models outperform Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) in quality and diversity. Leveraging physical prior knowledge to these models can further enhance their capabilities. In this work, we present GalCatDiff, the first framework in astronomy to leverage both galaxy image features and astrophysical properties in the network design of diffusion models. GalCatDiff incorporates an enhanced U-Net and a novel block entitled Astro-RAB (Residual Attention Block), which dynamically combines attention mechanisms with convolution operations to ensure global consistency and local feature fidelity. Moreover, GalCatDiff uses category embeddings for class-specific galaxy generation, avoiding the high computational costs of training separate models for each category. Our experimental results demonstrate that GalCatDiff significantly outperforms existing methods in terms of the consistency of sample color and size distributions, and the generated galaxies are both visually realistic and physically consistent. This framework will enhance the reliability of galaxy simulations and can potentially serve as a data augmentor to support future galaxy classification algorithm development.
Authors: Xiting He, Mingwu Su, Xinqi Jiang, Long Bai, Jiewen Lai, Hongliang Ren
Abstract: Vision-Language-Action (VLA) models have emerged as a prominent research area, showcasing significant potential across a variety of applications. However, their performance in endoscopy robotics, particularly endoscopy capsule robots that perform actions within the digestive system, remains unexplored. The integration of VLA models into endoscopy robots allows more intuitive and efficient interactions between human operators and medical devices, improving both diagnostic accuracy and treatment outcomes. In this work, we design CapsDT, a Diffusion Transformer model for capsule robot manipulation in the stomach. By processing interleaved visual inputs, and textual instructions, CapsDT can infer corresponding robotic control signals to facilitate endoscopy tasks. In addition, we developed a capsule endoscopy robot system, a capsule robot controlled by a robotic arm-held magnet, addressing different levels of four endoscopy tasks and creating corresponding capsule robot datasets within the stomach simulator. Comprehensive evaluations on various robotic tasks indicate that CapsDT can serve as a robust vision-language generalist, achieving state-of-the-art performance in various levels of endoscopy tasks while achieving a 26.25% success rate in real-world simulation manipulation.
Authors: Martha Arbayani Zaidan, Naser Hossein Motlagh, Petteri Nurmi, Tareq Hussein, Markku Kulmala, Tuukka Pet\"aj\"a, Sasu Tarkoma
Abstract: Atmospheric sciences are crucial for understanding environmental phenomena ranging from air quality to extreme weather events, and climate change. Recent breakthroughs in sensing, communication, computing, and Artificial Intelligence (AI) have significantly advanced atmospheric sciences, enabling the generation of vast amounts of data through long-term Earth observations and providing powerful tools for analyzing atmospheric phenomena and predicting natural disasters. This paper contributes a critical interdisciplinary overview that bridges the fields of atmospheric science and computer science, highlighting the transformative potential of AI in atmospheric research. We identify key challenges associated with integrating AI into atmospheric research, including issues related to big data and infrastructure, and provide a detailed research roadmap that addresses both current and emerging challenges.
Authors: Leo Gagnon, Eric Elmoznino, Sarthak Mittal, Tom Marty, Tejas Kasetty, Dhanya Sridhar, Guillaume Lajoie
Abstract: The rapid adaptation ability of auto-regressive foundation models is often attributed to the diversity of their pre-training data. This is because, from a Bayesian standpoint, minimizing prediction error in such settings requires integrating over all plausible latent hypotheses consistent with observations. While this behavior is desirable in principle, it often proves too ambitious in practice: under high ambiguity, the number of plausible latent alternatives makes Bayes-optimal prediction computationally intractable. Cognitive science has long recognized this limitation, suggesting that under such conditions, heuristics or information-seeking strategies are preferable to exhaustive inference. Translating this insight to next-token prediction, we hypothesize that low- and high-ambiguity predictions pose different computational demands, making ambiguity-agnostic next-token prediction a detrimental inductive bias. To test this, we introduce MetaHMM, a synthetic sequence meta-learning benchmark with rich compositional structure and a tractable Bayesian oracle. We show that Transformers indeed struggle with high-ambiguity predictions across model sizes. Motivated by cognitive theories, we propose a method to convert pre-trained models into Monte Carlo predictors that decouple task inference from token prediction. Preliminary results show substantial gains in ambiguous contexts through improved capacity allocation and test-time scalable inference, though challenges remain.
Authors: Heng Zhang, Zikang Wan, Danilo Vasconcellos Vargas
Abstract: Human vision excels at segmenting visual cues without the need for explicit training, and it remains remarkably robust even as noise severity increases. In contrast, existing AI algorithms struggle to maintain accuracy under similar conditions. Here, we present SyncMapV2, the first to solve unsupervised segmentation with state-of-the-art robustness. SyncMapV2 exhibits a minimal drop in mIoU, only 0.01%, under digital corruption, compared to a 23.8% drop observed in SOTA methods.This superior performance extends across various types of corruption: noise (7.3% vs. 37.7%), weather (7.5% vs. 33.8%), and blur (7.0% vs. 29.5%). Notably, SyncMapV2 accomplishes this without any robust training, supervision, or loss functions. It is based on a learning paradigm that uses self-organizing dynamical equations combined with concepts from random networks. Moreover,unlike conventional methods that require re-initialization for each new input, SyncMapV2 adapts online, mimicking the continuous adaptability of human vision. Thus, we go beyond the accurate and robust results, and present the first algorithm that can do all the above online, adapting to input rather than re-initializing. In adaptability tests, SyncMapV2 demonstrates near-zero performance degradation, which motivates and fosters a new generation of robust and adaptive intelligence in the near future.
Authors: Xu Zhao, Chen Zhao, Xiantao Hu, Hongliang Zhang, Ying Tai, Jian Yang
Abstract: Recent advancements in multi-scale architectures have demonstrated exceptional performance in image denoising tasks. However, existing architectures mainly depends on a fixed single-input single-output Unet architecture, ignoring the multi-scale representations of pixel level. In addition, previous methods treat the frequency domain uniformly, ignoring the different characteristics of high-frequency and low-frequency noise. In this paper, we propose a novel multi-scale adaptive dual-domain network (MADNet) for image denoising. We use image pyramid inputs to restore noise-free results from low-resolution images. In order to realize the interaction of high-frequency and low-frequency information, we design an adaptive spatial-frequency learning unit (ASFU), where a learnable mask is used to separate the information into high-frequency and low-frequency components. In the skip connections, we design a global feature fusion block to enhance the features at different scales. Extensive experiments on both synthetic and real noisy image datasets verify the effectiveness of MADNet compared with current state-of-the-art denoising approaches.
Authors: Sajan Muhammad, Salem Lahlou
Abstract: Efficiently identifying the right trajectories for training remains an open problem in GFlowNets. To address this, it is essential to prioritize exploration in regions of the state space where the reward distribution has not been sufficiently learned. This calls for uncertainty-driven exploration, in other words, the agent should be aware of what it does not know. This attribute can be measured by joint predictions, which are particularly important for combinatorial and sequential decision problems. In this research, we integrate epistemic neural networks (ENN) with the conventional architecture of GFlowNets to enable more efficient joint predictions and better uncertainty quantification, thereby improving exploration and the identification of optimal trajectories. Our proposed algorithm, ENN-GFN-Enhanced, is compared to the baseline method in GFlownets and evaluated in grid environments and structured sequence generation in various settings, demonstrating both its efficacy and efficiency.
Authors: Carmelo Scribano, Elena Govi, Paolo bertellini, Simone Parisi, Giorgia Franchini, Marko Bertogna
Abstract: Accurate mapping of agricultural field boundaries is essential for the efficient operation of agriculture. Automatic extraction from high-resolution satellite imagery, supported by computer vision techniques, can avoid costly ground surveys. In this paper, we present a pipeline for field delineation based on the Segment Anything Model (SAM), introducing a fine-tuning strategy to adapt SAM to this task. In addition to using published datasets, we describe a method for acquiring a complementary regional dataset that covers areas beyond current sources. Extensive experiments assess segmentation accuracy and evaluate the generalization capabilities. Our approach provides a robust baseline for automated field delineation. The new regional dataset, known as ERAS, is now publicly available.
Authors: Ji Zhang, Jingkuan Song, Lianli Gao, Nicu Sebe, Heng Tao Shen
Abstract: Recent advances in model pre-training give rise to task adaptation-based few-shot learning (FSL), where the goal is to adapt a pre-trained task-agnostic model for capturing task-specific knowledge with a few-labeled support samples of the target task.Nevertheless, existing approaches may still fail in the open world due to the inevitable in-distribution (ID) and out-of-distribution (OOD) noise from both support and query samples of the target task. With limited support samples available, i) the adverse effect of the dual noises can be severely amplified during task adaptation, and ii) the adapted model can produce unreliable predictions on query samples in the presence of the dual noises. In this work, we propose DEnoised Task Adaptation (DETA++) for reliable FSL. DETA++ uses a Contrastive Relevance Aggregation (CoRA) module to calculate image and region weights for support samples, based on which a clean prototype loss and a noise entropy maximization loss are proposed to achieve noise-robust task adaptation. Additionally,DETA++ employs a memory bank to store and refine clean regions for each inner-task class, based on which a Local Nearest Centroid Classifier (LocalNCC) is devised to yield noise-robust predictions on query samples. Moreover, DETA++ utilizes an Intra-class Region Swapping (IntraSwap) strategy to rectify ID class prototypes during task adaptation, enhancing the model's robustness to the dual noises. Extensive experiments demonstrate the effectiveness and flexibility of DETA++.
Authors: Cedric M\"oller, Ricardo Usbeck
Abstract: We examine the impact of incorporating knowledge graph information on the performance of relation extraction models across a range of datasets. Our hypothesis is that the positions of entities within a knowledge graph provide important insights for relation extraction tasks. We conduct experiments on multiple datasets, each varying in the number of relations, training examples, and underlying knowledge graphs. Our results demonstrate that integrating knowledge graph information significantly enhances performance, especially when dealing with an imbalance in the number of training examples for each relation. We evaluate the contribution of knowledge graph-based features by combining established relation extraction methods with graph-aware Neural Bellman-Ford networks. These features are tested in both supervised and zero-shot settings, demonstrating consistent performance improvements across various datasets.
Authors: Nikola Jovanovi\'c, Ismail Labiad, Tom\'a\v{s} Sou\v{c}ek, Martin Vechev, Pierre Fernandez
Abstract: Watermarking the outputs of generative models has emerged as a promising approach for tracking their provenance. Despite significant interest in autoregressive image generation models and their potential for misuse, no prior work has attempted to watermark their outputs at the token level. In this work, we present the first such approach by adapting language model watermarking techniques to this setting. We identify a key challenge: the lack of reverse cycle-consistency (RCC), wherein re-tokenizing generated image tokens significantly alters the token sequence, effectively erasing the watermark. To address this and to make our method robust to common image transformations, neural compression, and removal attacks, we introduce (i) a custom tokenizer-detokenizer finetuning procedure that improves RCC, and (ii) a complementary watermark synchronization layer. As our experiments demonstrate, our approach enables reliable and robust watermark detection with theoretically grounded p-values.
Authors: Iwan Williams
Abstract: Large Language Models (LLMs) such as GPT-4 produce compelling responses to a wide range of prompts. But their representational capacities are uncertain. Many LLMs have no direct contact with extra-linguistic reality: their inputs, outputs and training data consist solely of text, raising the questions (1) can LLMs represent anything and (2) if so, what? In this paper, I explore what it would take to answer these questions according to a structural-correspondence based account of representation, and make an initial survey of this evidence. I argue that the mere existence of structural correspondences between LLMs and worldly entities is insufficient to ground representation of those entities. However, if these structural correspondences play an appropriate role - they are exploited in a way that explains successful task performance - then they could ground real world contents. This requires overcoming a challenge: the text-boundedness of LLMs appears, on the face of it, to prevent them engaging in the right sorts of tasks.
Authors: Santosh Patapati, Trisanth Srinivasan, Amith Adiraju
Abstract: Micro-gesture recognition is a challenging task in affective computing due to the subtle, involuntary nature of the gestures and their low movement amplitude. In this paper, we introduce a Pose-Guided Semantics-Aware CLIP-based architecture, or CLIP for Micro-Gesture recognition (CLIP-MG), a modified CLIP model tailored for micro-gesture classification on the iMiGUE dataset. CLIP-MG integrates human pose (skeleton) information into the CLIP-based recognition pipeline through pose-guided semantic query generation and a gated multi-modal fusion mechanism. The proposed model achieves a Top-1 accuracy of 61.82%. These results demonstrate both the potential of our approach and the remaining difficulty in fully adapting vision-language models like CLIP for micro-gesture recognition.
Authors: Yao Lu, Zhaiyuan Ji, Jiawei Du, Yu Shanqing, Qi Xuan, Tianyi Zhou
Abstract: Although the annotation paradigm based on Large Language Models (LLMs) has made significant breakthroughs in recent years, its actual deployment still has two core bottlenecks: first, the cost of calling commercial APIs in large-scale annotation is very expensive; second, in scenarios that require fine-grained semantic understanding, such as sentiment classification and toxicity classification, the annotation accuracy of LLMs is even lower than that of Small Language Models (SLMs) dedicated to this field. To address these problems, we propose a new paradigm of multi-model cooperative annotation and design a fully automatic annotation framework AutoAnnotator based on this. Specifically, AutoAnnotator consists of two layers. The upper-level meta-controller layer uses the generation and reasoning capabilities of LLMs to select SLMs for annotation, automatically generate annotation code and verify difficult samples; the lower-level task-specialist layer consists of multiple SLMs that perform annotation through multi-model voting. In addition, we use the difficult samples obtained by the secondary review of the meta-controller layer as the reinforcement learning set and fine-tune the SLMs in stages through a continual learning strategy, thereby improving the generalization of SLMs. Extensive experiments show that AutoAnnotator outperforms existing open-source/API LLMs in zero-shot, one-shot, CoT, and majority voting settings. Notably, AutoAnnotator reduces the annotation cost by 74.15% compared to directly annotating with GPT-3.5-turbo, while still improving the accuracy by 6.21%. Project page: https://github.com/Zhaiyuan-Ji/AutoAnnotator.
Authors: Shushanta Pudasaini, Aman Shakya, Siddhartha Shrestha, Sahil Bhatta, Sunil Thapa, Sushmita Palikhe
Abstract: After the release of ChatGPT, Large Language Models (LLMs) have gained huge popularity in recent days and thousands of variants of LLMs have been released. However, there is no generative language model for the Nepali language, due to which other downstream tasks, including fine-tuning, have not been explored yet. To fill this research gap in the Nepali NLP space, this research proposes \textit{NepaliGPT}, a generative large language model tailored specifically for the Nepali language. This research introduces an advanced corpus for the Nepali language collected from several sources, called the Devanagari Corpus. Likewise, the research introduces the first NepaliGPT benchmark dataset comprised of 4,296 question-answer pairs in the Nepali language. The proposed LLM NepaliGPT achieves the following metrics in text generation: Perplexity of 26.32245, ROUGE-1 score of 0.2604, causal coherence of 81.25\%, and causal consistency of 85.41\%.
Authors: Zhiyuan Liang, Dongwen Tang, Yuhao Zhou, Xuanlei Zhao, Mingjia Shi, Wangbo Zhao, Zekai Li, Peihao Wang, Konstantin Sch\"urholt, Damian Borth, Michael M. Bronstein, Yang You, Zhangyang Wang, Kai Wang
Abstract: Modern Parameter-Efficient Fine-Tuning (PEFT) methods such as low-rank adaptation (LoRA) reduce the cost of customizing large language models (LLMs), yet still require a separate optimization run for every downstream dataset. We introduce \textbf{Drag-and-Drop LLMs (\textit{DnD})}, a prompt-conditioned parameter generator that eliminates per-task training by mapping a handful of unlabeled task prompts directly to LoRA weight updates. A lightweight text encoder distills each prompt batch into condition embeddings, which are then transformed by a cascaded hyper-convolutional decoder into the full set of LoRA matrices. Once trained in a diverse collection of prompt-checkpoint pairs, DnD produces task-specific parameters in seconds, yielding i) up to \textbf{12,000$\times$} lower overhead than full fine-tuning, ii) average gains up to \textbf{30\%} in performance over the strongest training LoRAs on unseen common-sense reasoning, math, coding, and multimodal benchmarks, and iii) robust cross-domain generalization despite never seeing the target data or labels. Our results demonstrate that prompt-conditioned parameter generation is a viable alternative to gradient-based adaptation for rapidly specializing LLMs. Our project is available at \href{https://jerryliang24.github.io/DnD}{https://jerryliang24.github.io/DnD}.
URLs: https://jerryliang24.github.io/DnD, https://jerryliang24.github.io/DnD
Authors: Dong Nguyen Tien, Dung D. Le
Abstract: Visual Document Understanding (VDU) systems have achieved strong performance in information extraction by integrating textual, layout, and visual signals. However, their robustness under realistic adversarial perturbations remains insufficiently explored. We introduce the first unified framework for generating and evaluating multi-modal adversarial attacks on OCR-based VDU models. Our method covers six gradient-based layout attack scenarios, incorporating manipulations of OCR bounding boxes, pixels, and texts across both word and line granularities, with constraints on layout perturbation budget (e.g., IoU >= 0.6) to preserve plausibility. Experimental results across four datasets (FUNSD, CORD, SROIE, DocVQA) and six model families demonstrate that line-level attacks and compound perturbations (BBox + Pixel + Text) yield the most severe performance degradation. Projected Gradient Descent (PGD)-based BBox perturbations outperform random-shift baselines in all investigated models. Ablation studies further validate the impact of layout budget, text modification, and adversarial transferability.
Authors: Berk Yilmaz, Daniel Fidel Harvey, Prajit Dhuri
Abstract: This study investigates the integration of signal processing transformations -- Fast Fourier Transform (FFT), Walsh-Hadamard Transform (WHT), and Discrete Cosine Transform (DCT) -- within the ResNet50 convolutional neural network (CNN) model for image classification. The primary objective is to assess the trade-offs between computational efficiency, energy consumption, and classification accuracy during training and inference. Using the CIFAR-100 dataset (100 classes, 60,000 images), experiments demonstrated that incorporating WHT significantly reduced energy consumption while improving accuracy. Specifically, a baseline ResNet50 model achieved a testing accuracy of 66%, consuming an average of 25,606 kJ per model. In contrast, a modified ResNet50 incorporating WHT in the early convolutional layers achieved 74% accuracy, and an enhanced version with WHT applied to both early and late layers achieved 79% accuracy, with an average energy consumption of only 39 kJ per model. These results demonstrate the potential of WHT as a highly efficient and effective approach for energy-constrained CNN applications.
Authors: Daniel Fidel Harvey, George Weale, Berk Yilmaz
Abstract: Mixture of Experts (MoE) architectures increase large language model scalability, yet their performance depends on the router module that moves tokens to specialized experts. Bad routing can load imbalance and reduced accuracy. This project designed and implemented different router architectures within Transformer models to fix these limitations. We experimented with six distinct router variants Linear, Attention, Multi-Layer Perceptron (MLP), Hybrid, Hash, and our new MLP-Hadamard. We characterized these routers using BERT and the Qwen1.5-MoE model, looking at parameter efficiency, inference latency, routing entropy, and expert utilization patterns. Our evaluations showed distinct trade-offs: Linear routers offer speed, while MLP and Attention routers provide greater expressiveness. The MLP-Hadamard router shows a unique capability for structured, sparse routing. We successfully replaced and fine-tuned custom routers within the complex, quantized Qwen1.5-MoE model. This work provides a comparative analysis of MoE router designs and offers insights into optimizing their performance for efficient and effective large-scale model deployment.
Authors: Jonas R. Naujoks, Aleksander Krasowski, Moritz Weckbecker, Galip \"Umit Yolcu, Thomas Wiegand, Sebastian Lapuschkin, Wojciech Samek, Ren\'e P. Klausen
Abstract: Physics-informed neural networks (PINNs) offer a powerful approach to solving partial differential equations (PDEs), which are ubiquitous in the quantitative sciences. Applied to both forward and inverse problems across various scientific domains, PINNs have recently emerged as a valuable tool in the field of scientific machine learning. A key aspect of their training is that the data -- spatio-temporal points sampled from the PDE's input domain -- are readily available. Influence functions, a tool from the field of explainable AI (XAI), approximate the effect of individual training points on the model, enhancing interpretability. In the present work, we explore the application of influence function-based sampling approaches for the training data. Our results indicate that such targeted resampling based on data attribution methods has the potential to enhance prediction accuracy in physics-informed neural networks, demonstrating a practical application of an XAI method in PINN training.
Authors: Haotian Xia, Hao Peng, Yunjia Qi, Xiaozhi Wang, Bin Xu, Lei Hou, Juanzi Li
Abstract: Long story generation remains a challenge for existing large language models (LLMs), primarily due to two main factors: (1) discourse coherence, which requires plot consistency, logical coherence, and completeness in the long-form generation, and (2) narrative complexity, which requires an interwoven and engaging narrative. To address these challenges, we propose StoryWriter, a multi-agent story generation framework, which consists of three main modules: (1) outline agent, which generates event-based outlines containing rich event plots, character, and event-event relationships. (2) planning agent, which further details events and plans which events should be written in each chapter to maintain an interwoven and engaging story. (3) writing agent, which dynamically compresses the story history based on the current event to generate and reflect new plots, ensuring the coherence of the generated story. We conduct both human and automated evaluation, and StoryWriter significantly outperforms existing story generation baselines in both story quality and length. Furthermore, we use StoryWriter to generate a dataset, which contains about $6,000$ high-quality long stories, with an average length of $8,000$ words. We train the model Llama3.1-8B and GLM4-9B using supervised fine-tuning on LongStory and develop StoryWriter_GLM and StoryWriter_GLM, which demonstrates advanced performance in long story generation.
Authors: Tri Duc Ly, Gia H. Ngo
Abstract: EEG is a non-invasive, safe, and low-risk method to record electrophysiological signals inside the brain. Especially with recent technology developments like dry electrodes, consumer-grade EEG devices, and rapid advances in machine learning, EEG is commonly used as a resource for automatic emotion recognition. With the aim to develop a deep learning model that can perform EEG-based emotion recognition in a real-life context, we propose a novel approach to utilize multi-scale convolutional neural networks to accomplish such tasks. By implementing feature extraction kernels with many ratio coefficients as well as a new type of kernel that learns key information from four separate areas of the brain, our model consistently outperforms the state-of-the-art TSception model in predicting valence, arousal, and dominance scores across many performance evaluation metrics.
Authors: Jun Qi, Chen-Yu Liu, Sabato Marco Siniscalchi, Chao-Han Huck Yang, Min-Hsiu Hsieh
Abstract: Low-Rank Adaptation (LoRA) is widely recognized for its parameter-efficient fine-tuning of large-scale neural models. However, standard LoRA independently optimizes low-rank matrices, which inherently limits its expressivity and generalization capabilities. While classical tensor-train (TT) decomposition can be separately employed on individual LoRA matrices, this work demonstrates that the classical TT-based approach neither significantly improves parameter efficiency nor achieves substantial performance gains. This paper proposes TensorGuide, a novel tensor-train-guided adaptation framework to overcome these limitations. TensorGuide generates two correlated low-rank LoRA matrices through a unified TT structure driven by controlled Gaussian noise. The resulting joint TT representation inherently provides structured, low-rank adaptations, significantly enhancing expressivity, generalization, and parameter efficiency without increasing the number of trainable parameters. Theoretically, we justify these improvements through neural tangent kernel analyses, demonstrating superior optimization dynamics and enhanced generalization. Extensive experiments on quantum dot classification and GPT-2 fine-tuning benchmarks demonstrate that TensorGuide-based LoRA consistently outperforms standard LoRA and TT-LoRA, achieving improved accuracy and scalability with fewer parameters.
Authors: Tara Akhound-Sadegh, Jungyoon Lee, Avishek Joey Bose, Valentin De Bortoli, Arnaud Doucet, Michael M. Bronstein, Dominique Beaini, Siamak Ravanbakhsh, Kirill Neklyudov, Alexander Tong
Abstract: Sampling efficiently from a target unnormalized probability density remains a core challenge, with relevance across countless high-impact scientific applications. A promising approach towards this challenge is the design of amortized samplers that borrow key ideas, such as probability path design, from state-of-the-art generative diffusion models. However, all existing diffusion-based samplers remain unable to draw samples from distributions at the scale of even simple molecular systems. In this paper, we propose Progressive Inference-Time Annealing (PITA), a novel framework to learn diffusion-based samplers that combines two complementary interpolation techniques: I.) Annealing of the Boltzmann distribution and II.) Diffusion smoothing. PITA trains a sequence of diffusion models from high to low temperatures by sequentially training each model at progressively higher temperatures, leveraging engineered easy access to samples of the temperature-annealed target density. In the subsequent step, PITA enables simulating the trained diffusion model to procure training samples at a lower temperature for the next diffusion model through inference-time annealing using a novel Feynman-Kac PDE combined with Sequential Monte Carlo. Empirically, PITA enables, for the first time, equilibrium sampling of N-body particle systems, Alanine Dipeptide, and tripeptides in Cartesian coordinates with dramatically lower energy function evaluations. Code available at: https://github.com/taraak/pita
Authors: Sophie Chiang, Guy Laban, Hatice Gunes
Abstract: As conversational agents increasingly engage in emotionally supportive dialogue, it is important to understand how closely their interactions resemble those in traditional therapy settings. This study investigates whether the concerns shared with a robot align with those shared in human-to-human (H2H) therapy sessions, and whether robot responses semantically mirror those of human therapists. We analyzed two datasets: one of interactions between users and professional therapists (Hugging Face's NLP Mental Health Conversations), and another involving supportive conversations with a social robot (QTrobot from LuxAI) powered by a large language model (LLM, GPT-3.5). Using sentence embeddings and K-means clustering, we assessed cross-agent thematic alignment by applying a distance-based cluster-fitting method that evaluates whether responses from one agent type map to clusters derived from the other, and validated it using Euclidean distances. Results showed that 90.88% of robot conversation disclosures could be mapped to clusters from the human therapy dataset, suggesting shared topical structure. For matched clusters, we compared the subjects as well as therapist and robot responses using Transformer, Word2Vec, and BERT embeddings, revealing strong semantic overlap in subjects' disclosures in both datasets, as well as in the responses given to similar human disclosure themes across agent types (robot vs. human therapist). These findings highlight both the parallels and boundaries of robot-led support conversations and their potential for augmenting mental health interventions.
Authors: Yaru Niu, Yunzhe Zhang, Mingyang Yu, Changyi Lin, Chenhao Li, Yikai Wang, Yuxiang Yang, Wenhao Yu, Tingnan Zhang, Bingqing Chen, Jonathan Francis, Zhenzhen Li, Jie Tan, Ding Zhao
Abstract: Quadrupedal robots have demonstrated impressive locomotion capabilities in complex environments, but equipping them with autonomous versatile manipulation skills in a scalable way remains a significant challenge. In this work, we introduce a cross-embodiment imitation learning system for quadrupedal manipulation, leveraging data collected from both humans and LocoMan, a quadruped equipped with multiple manipulation modes. Specifically, we develop a teleoperation and data collection pipeline, which unifies and modularizes the observation and action spaces of the human and the robot. To effectively leverage the collected data, we propose an efficient modularized architecture that supports co-training and pretraining on structured modality-aligned data across different embodiments. Additionally, we construct the first manipulation dataset for the LocoMan robot, covering various household tasks in both unimanual and bimanual modes, supplemented by a corresponding human dataset. We validate our system on six real-world manipulation tasks, where it achieves an average success rate improvement of 41.9% overall and 79.7% under out-of-distribution (OOD) settings compared to the baseline. Pretraining with human data contributes a 38.6% success rate improvement overall and 82.7% under OOD settings, enabling consistently better performance with only half the amount of robot data. Our code, hardware, and data are open-sourced at: https://human2bots.github.io.
Authors: Saad Almohaimeed, Saleh Almohaimeed, Damla Turgut, Ladislau B\"ol\"oni
Abstract: Implicit hate speech has recently emerged as a critical challenge for social media platforms. While much of the research has traditionally focused on harmful speech in general, the need for generalizable techniques to detect veiled and subtle forms of hate has become increasingly pressing. Based on lexicon analysis, we hypothesize that implicit hate speech is already present in publicly available harmful speech datasets but may not have been explicitly recognized or labeled by annotators. Additionally, crowdsourced datasets are prone to mislabeling due to the complexity of the task and often influenced by annotators' subjective interpretations. In this paper, we propose an approach to address the detection of implicit hate speech and enhance generalizability across diverse datasets by leveraging existing harmful speech datasets. Our method comprises three key components: influential sample identification, reannotation, and augmentation using Llama-3 70B and GPT-4o. Experimental results demonstrate the effectiveness of our approach in improving implicit hate detection, achieving a +12.9-point F1 score improvement compared to the baseline.
Authors: Mehreen Naeem, Andrew Melnik, Michael Beetz
Abstract: We introduce a novel framework that integrates Semantic Digital Twins (SDTs) with Large Language Models (LLMs) to enable adaptive and goal-driven robotic task execution in dynamic environments. The system decomposes natural language instructions into structured action triplets, which are grounded in contextual environmental data provided by the SDT. This semantic grounding allows the robot to interpret object affordances and interaction rules, enabling action planning and real-time adaptability. In case of execution failures, the LLM utilizes error feedback and SDT insights to generate recovery strategies and iteratively revise the action plan. We evaluate our approach using tasks from the ALFRED benchmark, demonstrating robust performance across various household scenarios. The proposed framework effectively combines high-level reasoning with semantic environment understanding, achieving reliable task completion in the face of uncertainty and failure.
Authors: Riccardo Ziglio, Cecilia Pasquini, Silvio Ranise
Abstract: Face swapping manipulations in video streams represents an increasing threat in remote video communications, due to advances in automated and real-time tools. Recent literature proposes to characterize and exploit visual artifacts introduced in video frames by swapping algorithms when dealing with challenging physical scenes, such as face occlusions. This paper investigates the effectiveness of this approach by benchmarking CNN-based data-driven models on two data corpora (including a newly collected one) and analyzing generalization capabilities with respect to different acquisition sources and swapping algorithms. The results confirm excellent performance of general-purpose CNN architectures when operating within the same data source, but a significant difficulty in robustly characterizing occlusion-based visual cues across datasets. This highlights the need for specialized detection strategies to deal with such artifacts.
Authors: Soumya Suvra Ghosal, Vaibhav Singh, Akash Ghosh, Soumyabrata Pal, Subhadip Baidya, Sriparna Saha, Dinesh Manocha
Abstract: Reward models are essential for aligning large language models (LLMs) with human preferences. However, most open-source multilingual reward models are primarily trained on preference datasets in high-resource languages, resulting in unreliable reward signals for low-resource Indic languages. Collecting large-scale, high-quality preference data for these languages is prohibitively expensive, making preference-based training approaches impractical. To address this challenge, we propose RELIC, a novel in-context learning framework for reward modeling in low-resource Indic languages. RELIC trains a retriever with a pairwise ranking objective to select in-context examples from auxiliary high-resource languages that most effectively highlight the distinction between preferred and less-preferred responses. Extensive experiments on three preference datasets- PKU-SafeRLHF, WebGPT, and HH-RLHF-using state-of-the-art open-source reward models demonstrate that RELIC significantly improves reward model accuracy for low-resource Indic languages, consistently outperforming existing example selection methods. For example, on Bodo-a low-resource Indic language-using a LLaMA-3.2-3B reward model, RELIC achieves a 12.81% and 10.13% improvement in accuracy over zero-shot prompting and state-of-the-art example selection method, respectively.
Authors: Zeqiang Lai, Yunfei Zhao, Haolin Liu, Zibo Zhao, Qingxiang Lin, Huiwen Shi, Xianghui Yang, Mingxin Yang, Shuhui Yang, Yifei Feng, Sheng Zhang, Xin Huang, Di Luo, Fan Yang, Fang Yang, Lifu Wang, Sicong Liu, Yixuan Tang, Yulin Cai, Zebin He, Tian Liu, Yuhong Liu, Jie Jiang, Linus, Jingwei Huang, Chunchao Guo
Abstract: In this report, we present Hunyuan3D 2.5, a robust suite of 3D diffusion models aimed at generating high-fidelity and detailed textured 3D assets. Hunyuan3D 2.5 follows two-stages pipeline of its previous version Hunyuan3D 2.0, while demonstrating substantial advancements in both shape and texture generation. In terms of shape generation, we introduce a new shape foundation model -- LATTICE, which is trained with scaled high-quality datasets, model-size, and compute. Our largest model reaches 10B parameters and generates sharp and detailed 3D shape with precise image-3D following while keeping mesh surface clean and smooth, significantly closing the gap between generated and handcrafted 3D shapes. In terms of texture generation, it is upgraded with phyiscal-based rendering (PBR) via a novel multi-view architecture extended from Hunyuan3D 2.0 Paint model. Our extensive evaluation shows that Hunyuan3D 2.5 significantly outperforms previous methods in both shape and end-to-end texture generation.
Authors: Ronald Skorobogat, Karsten Roth, Mariana-Iuliana Georgescu, Zeynep Akata
Abstract: Model merging enables the combination of multiple specialized expert models into a single model capable of performing multiple tasks. However, the benefits of merging an increasing amount of specialized experts generally lead to diminishing returns and reduced overall performance gains. In this work, we offer an explanation and analysis from a task arithmetic perspective; revealing that as the merging process (across numerous existing merging methods) continues for more and more experts, the associated task vector space experiences rank collapse. To mitigate this issue, we introduce Subspace Boosting, which operates on the singular value decomposed task vector space and maintains task vector ranks. Subspace Boosting raises merging efficacy for up to 20 expert models by large margins of more than 10% when evaluated on vision benchmarks. Moreover, we propose employing Higher-Order Generalized Singular Value Decomposition to further quantify task similarity, offering a new interpretable perspective on model merging.
Authors: Liyang Yu, Tianyi Wang, Junfeng Jiao, Fengwu Shan, Hongqing Chu, Bingzhao Gao
Abstract: In complex real-world traffic environments, autonomous vehicles (AVs) need to interact with other traffic participants while making real-time and safety-critical decisions accordingly. The unpredictability of human behaviors poses significant challenges, particularly in dynamic scenarios, such as multi-lane highways and unsignalized T-intersections. To address this gap, we design a bi-level interaction decision-making algorithm (BIDA) that integrates interactive Monte Carlo tree search (MCTS) with deep reinforcement learning (DRL), aiming to enhance interaction rationality, efficiency and safety of AVs in dynamic key traffic scenarios. Specifically, we adopt three types of DRL algorithms to construct a reliable value network and policy network, which guide the online deduction process of interactive MCTS by assisting in value update and node selection. Then, a dynamic trajectory planner and a trajectory tracking controller are designed and implemented in CARLA to ensure smooth execution of planned maneuvers. Experimental evaluations demonstrate that our BIDA not only enhances interactive deduction and reduces computational costs, but also outperforms other latest benchmarks, which exhibits superior safety, efficiency and interaction rationality under varying traffic conditions.
Authors: Soroush H. Zargarbashi, Mohammad Sadegh Akhondzadeh, Aleksandar Bojchevski
Abstract: Given any model, conformal prediction (CP) returns prediction sets guaranteed to include the true label with high adjustable probability. Robust CP (RCP) extends this to inputs with worst-case noise. A well-established approach is to use randomized smoothing for RCP since it is applicable to any black-box model and provides smaller sets compared to deterministic methods. However, current smoothing-based RCP requires many model forward passes per each input which is computationally expensive. We show that conformal prediction attains some robustness even with a forward pass on a single randomly perturbed input. Using any binary certificate we propose a single sample robust CP (RCP1). Our approach returns robust sets with smaller average set size compared to SOTA methods which use many (e.g. around 100) passes per input. Our key insight is to certify the conformal prediction procedure itself rather than individual scores. Our approach is agnostic to the setup (classification and regression). We further extend our approach to smoothing-based robust conformal risk control.
Authors: Keyhan Najafian, Farhad Maleki, Lingling Jin, Ian Stavness
Abstract: Instance segmentation is essential for applications such as automated monitoring of plant health, growth, and yield. However, extensive effort is required to create large-scale datasets with pixel-level annotations of each object instance for developing instance segmentation models that restrict the use of deep learning in these areas. This challenge is more significant in images with densely packed, self-occluded objects, which are common in agriculture. To address this challenge, we propose a semi-self-supervised learning approach that requires minimal manual annotation to develop a high-performing instance segmentation model. We design GLMask, an image-mask representation for the model to focus on shape, texture, and pattern while minimizing its dependence on color features. We develop a pipeline to generate semantic segmentation and then transform it into instance-level segmentation. The proposed approach substantially outperforms the conventional instance segmentation models, establishing a state-of-the-art wheat head instance segmentation model with mAP@50 of 98.5%. Additionally, we assessed the proposed methodology on the general-purpose Microsoft COCO dataset, achieving a significant performance improvement of over 12.6% mAP@50. This highlights that the utility of our proposed approach extends beyond precision agriculture and applies to other domains, specifically those with similar data characteristics.
Authors: Yuxin Chen, Jianglan Wei, Chenfeng Xu, Boyi Li, Masayoshi Tomizuka, Andrea Bajcsy, Ran Tian
Abstract: World models enable robots to "imagine" future observations given current observations and planned actions, and have been increasingly adopted as generalized dynamics models to facilitate robot learning. Despite their promise, these models remain brittle when encountering novel visual distractors such as objects and background elements rarely seen during training. Specifically, novel distractors can corrupt action outcome predictions, causing downstream failures when robots rely on the world model imaginations for planning or action verification. In this work, we propose Reimagination with Observation Intervention (ReOI), a simple yet effective test-time strategy that enables world models to predict more reliable action outcomes in open-world scenarios where novel and unanticipated visual distractors are inevitable. Given the current robot observation, ReOI first detects visual distractors by identifying which elements of the scene degrade in physically implausible ways during world model prediction. Then, it modifies the current observation to remove these distractors and bring the observation closer to the training distribution. Finally, ReOI "reimagines" future outcomes with the modified observation and reintroduces the distractors post-hoc to preserve visual consistency for downstream planning and verification. We validate our approach on a suite of robotic manipulation tasks in the context of action verification, where the verifier needs to select desired action plans based on predictions from a world model. Our results show that ReOI is robust to both in-distribution and out-of-distribution visual distractors. Notably, it improves task success rates by up to 3x in the presence of novel distractors, significantly outperforming action verification that relies on world model predictions without imagination interventions.
Authors: Nadav Kunievsky, James A. Evans
Abstract: Understanding whether large language models (LLMs) possess a world model-a structured understanding of the world that supports generalization beyond surface-level patterns-is central to assessing their reliability, especially in high-stakes applications. We propose a formal framework for evaluating whether an LLM exhibits a sufficiently robust world model, defined as producing consistent outputs across semantically equivalent prompts while distinguishing between prompts that express different intents. We introduce a new evaluation approach to measure this that decomposes model response variability into three components: variability due to user purpose, user articulation, and model instability. An LLM with a strong world model should attribute most of the variability in its responses to changes in foundational purpose rather than superficial changes in articulation. This approach allows us to quantify how much of a model's behavior is semantically grounded rather than driven by model instability or alternative wording. We apply this framework to evaluate LLMs across diverse domains. Our results show how larger models attribute a greater share of output variability to changes in user purpose, indicating a more robust world model. This improvement is not uniform, however: larger models do not consistently outperform smaller ones across all domains, and their advantage in robustness is often modest. These findings highlight the importance of moving beyond accuracy-based benchmarks toward semantic diagnostics that more directly assess the structure and stability of a model's internal understanding of the world.
Authors: Ihor Pysmennyi, Roman Kyslyi, Kyrylo Kleshch
Abstract: Traditional quality assurance (QA) methods face significant challenges in addressing the complexity, scale, and rapid iteration cycles of modern software systems and are strained by limited resources available, leading to substantial costs associated with poor quality. The object of this research is the Quality Assurance processes for modern distributed software applications. The subject of the research is the assessment of the benefits, challenges, and prospects of integrating modern AI-oriented tools into quality assurance processes. We performed comprehensive analysis of implications on both verification and validation processes covering exploratory test analyses, equivalence partitioning and boundary analyses, metamorphic testing, finding inconsistencies in acceptance criteria (AC), static analyses, test case generation, unit test generation, test suit optimization and assessment, end to end scenario execution. End to end regression of sample enterprise application utilizing AI-agents over generated test scenarios was implemented as a proof of concept highlighting practical use of the study. The results, with only 8.3% flaky executions of generated test cases, indicate significant potential for the proposed approaches. However, the study also identified substantial challenges for practical adoption concerning generation of semantically identical coverage, "black box" nature and lack of explainability from state-of-the-art Large Language Models (LLMs), the tendency to correct mutated test cases to match expected results, underscoring the necessity for thorough verification of both generated artifacts and test execution results. The research demonstrates AI's transformative potential for QA but highlights the importance of a strategic approach to implementing these technologies, considering the identified limitations and the need for developing appropriate verification methodologies.
Authors: Tal Zeevi, El\'eonore V. Lieffrig, Lawrence H. Staib, John A. Onofrey
Abstract: Uncertainty maps highlight unreliable regions in segmentation predictions. However, most uncertainty evaluation metrics treat voxels independently, ignoring spatial context and anatomical structure. As a result, they may assign identical scores to qualitatively distinct patterns (e.g., scattered vs. boundary-aligned uncertainty). We propose three spatially aware metrics that incorporate structural and boundary information and conduct a thorough validation on medical imaging data from the prostate zonal segmentation challenge within the Medical Segmentation Decathlon. Our results demonstrate improved alignment with clinically important factors and better discrimination between meaningful and spurious uncertainty patterns.
Authors: Zeyun Deng, Jasorsi Ghosh, Fiona Xie, Yuzhe Lu, Katia Sycara, Joseph Campbell
Abstract: Reinforcement learning algorithms often suffer from poor sample efficiency, making them challenging to apply in multi-task or continual learning settings. Efficiency can be improved by transferring knowledge from a previously trained teacher policy to guide exploration in new but related tasks. However, if the new task sufficiently differs from the teacher's training task, the transferred guidance may be sub-optimal and bias exploration toward low-reward behaviors. We propose an energy-based transfer learning method that uses out-of-distribution detection to selectively issue guidance, enabling the teacher to intervene only in states within its training distribution. We theoretically show that energy scores reflect the teacher's state-visitation density and empirically demonstrate improved sample efficiency and performance across both single-task and multi-task settings.
Authors: Muhammad Azeem Aslam, Asim Naveed, Nisar Ahmed
Abstract: Breast ultrasound imaging is a valuable tool for early breast cancer detection, but automated tumor segmentation is challenging due to inherent noise, variations in scale of lesions, and fuzzy boundaries. To address these challenges, we propose a novel hybrid attention-based network for lesion segmentation. Our proposed architecture integrates a pre-trained DenseNet121 in the encoder part for robust feature extraction with a multi-branch attention-enhanced decoder tailored for breast ultrasound images. The bottleneck incorporates Global Spatial Attention (GSA), Position Encoding (PE), and Scaled Dot-Product Attention (SDPA) to learn global context, spatial relationships, and relative positional features. The Spatial Feature Enhancement Block (SFEB) is embedded at skip connections to refine and enhance spatial features, enabling the network to focus more effectively on tumor regions. A hybrid loss function combining Binary Cross-Entropy (BCE) and Jaccard Index loss optimizes both pixel-level accuracy and region-level overlap metrics, enhancing robustness to class imbalance and irregular tumor shapes. Experiments on public datasets demonstrate that our method outperforms existing approaches, highlighting its potential to assist radiologists in early and accurate breast cancer diagnosis.
Authors: Khiem Le, Tuan Tran, Ting Hua, Nitesh V. Chawla
Abstract: Existing resource-adaptive LoRA federated fine-tuning methods enable clients to fine-tune models using compressed versions of global LoRA matrices, in order to accommodate various compute resources across clients. This compression requirement will lead to suboptimal performance due to information loss. To address this, we propose FLAME, a novel federated learning framework based on the Sparse Mixture-of-Experts (SMoE) architecture. Unlike prior approaches, FLAME retains full (uncompressed) global LoRA matrices and achieves client-side adaptability by varying the number of activated experts per client. However, incorporating SMoE into federated learning introduces unique challenges, specifically, the mismatch in output magnitude from partial expert activation and the imbalance in expert training quality across clients. FLAME tackles these challenges through a lightweight rescaling mechanism and an activation-aware aggregation scheme. Empirical results across diverse computational settings demonstrate that FLAME consistently outperforms existing methods, providing a robust and effective solution for resource-adaptive federated learning.
Authors: Jiamin He, A. Rupam Mahmood, Martha White
Abstract: We introduce a novel reinforcement learning (RL) framework that treats distribution parameters as actions, redefining the boundary between agent and environment. This reparameterization makes the new action space continuous, regardless of the original action type (discrete, continuous, mixed, etc.). Under this new parameterization, we develop a generalized deterministic policy gradient estimator, Distribution Parameter Policy Gradient (DPPG), which has lower variance than the gradient in the original action space. Although learning the critic over distribution parameters poses new challenges, we introduce interpolated critic learning (ICL), a simple yet effective strategy to enhance learning, supported by insights from bandit settings. Building on TD3, a strong baseline for continuous control, we propose a practical DPPG-based actor-critic algorithm, Distribution Parameter Actor-Critic (DPAC). Empirically, DPAC outperforms TD3 in MuJoCo continuous control tasks from OpenAI Gym and DeepMind Control Suite, and demonstrates competitive performance on the same environments with discretized action spaces.
Authors: Jiaxin Pei, Dustin Wright, Isabelle Augenstin, David Jurgens
Abstract: Effectively engaging the public with science is vital for fostering trust and understanding in our scientific community. Yet, with an ever-growing volume of information, science communicators struggle to anticipate how audiences will perceive and interact with scientific news. In this paper, we introduce a computational framework that models public perception across twelve dimensions, such as newsworthiness, importance, and surprisingness. Using this framework, we create a large-scale science news perception dataset with 10,489 annotations from 2,101 participants from diverse US and UK populations, providing valuable insights into public responses to scientific information across domains. We further develop NLP models that predict public perception scores with a strong performance. Leveraging the dataset and model, we examine public perception of science from two perspectives: (1) Perception as an outcome: What factors affect the public perception of scientific information? (2) Perception as a predictor: Can we use the estimated perceptions to predict public engagement with science? We find that individuals' frequency of science news consumption is the driver of perception, whereas demographic factors exert minimal influence. More importantly, through a large-scale analysis and carefully designed natural experiment on Reddit, we demonstrate that the estimated public perception of scientific information has direct connections with the final engagement pattern. Posts with more positive perception scores receive significantly more comments and upvotes, which is consistent across different scientific information and for the same science, but are framed differently. Overall, this research underscores the importance of nuanced perception modeling in science communication, offering new pathways to predict public interest and engagement with scientific content.
Authors: Mobin Habibpour, Fatemeh Afghah
Abstract: Object Goal Navigation (ObjectNav) challenges robots to find objects in unseen environments, demanding sophisticated reasoning. While Vision-Language Models (VLMs) show potential, current ObjectNav methods often employ them superficially, primarily using vision-language embeddings for object-scene similarity checks rather than leveraging deeper reasoning. This limits contextual understanding and leads to practical issues like repetitive navigation behaviors. This paper introduces a novel zero-shot ObjectNav framework that pioneers the use of dynamic, history-aware prompting to more deeply integrate VLM reasoning into frontier-based exploration. Our core innovation lies in providing the VLM with action history context, enabling it to generate semantic guidance scores for navigation actions while actively avoiding decision loops. We also introduce a VLM-assisted waypoint generation mechanism for refining the final approach to detected objects. Evaluated on the HM3D dataset within Habitat, our approach achieves a 46% Success Rate (SR) and 24.8% Success weighted by Path Length (SPL). These results are comparable to state-of-the-art zero-shot methods, demonstrating the significant potential of our history-augmented VLM prompting strategy for more robust and context-aware robotic navigation.
Authors: Fenghua Cheng, Jinxiang Wang, Sen Wang, Zi Huang, Xue Li
Abstract: Multimodal reasoning is a process of understanding, integrating and inferring information across different data modalities. It has recently attracted surging academic attention as a benchmark for Artificial Intelligence (AI). Although there are various tasks for evaluating multimodal reasoning ability, they still have limitations. Lack of reasoning on hierarchical visual clues at different levels of granularity, e.g., local details and global context, is of little discussion, despite its frequent involvement in real scenarios. To bridge the gap, we introduce a novel and challenging task for multimodal reasoning, namely GeoGuess. Given a street view image, the task is to identify its location and provide a detailed explanation. A system that succeeds in GeoGuess should be able to detect tiny visual clues, perceive the broader landscape, and associate with vast geographic knowledge. Therefore, GeoGuess would require the ability to reason between hierarchical visual information and geographic knowledge. In this work, we establish a benchmark for GeoGuess by introducing a specially curated dataset GeoExplain which consists of panoramas-geocoordinates-explanation tuples. Additionally, we present a multimodal and multilevel reasoning method, namely SightSense which can make prediction and generate comprehensive explanation based on hierarchy of visual information and external knowledge. Our analysis and experiments demonstrate their outstanding performance in GeoGuess.
Authors: Rex Shen, Lu Tian
Abstract: Synthetic Data Generation has become essential for scalable, privacy-preserving statistical analysis. While standard approaches based on generative models, such as Normalizing Flows, have been widely used, they often suffer from slow convergence in high-dimensional settings, frequently converging more slowly than the canonical $1/\sqrt{n}$ rate when approximating the true data distribution. To overcome these limitations, we propose a Latent Noise Injection method using Masked Autoregressive Flows (MAF). Instead of directly sampling from the trained model, our method perturbs each data point in the latent space and maps it back to the data domain. This construction preserves a one to one correspondence between observed and synthetic data, enabling synthetic outputs that closely reflect the underlying distribution, particularly in challenging high-dimensional regimes where traditional sampling struggles. Our procedure satisfies local $(\epsilon, \delta)$-differential privacy and introduces a single perturbation parameter to control the privacy-utility trade-off. Although estimators based on individual synthetic datasets may converge slowly, we show both theoretically and empirically that aggregating across $K$ studies in a meta analysis framework restores classical efficiency and yields consistent, reliable inference. We demonstrate that with a well-calibrated perturbation parameter, Latent Noise Injection achieves strong statistical alignment with the original data and robustness against membership inference attacks. These results position our method as a compelling alternative to conventional flow-based sampling for synthetic data sharing in decentralized and privacy-sensitive domains, such as biomedical research.
Authors: Pavlo Vasylenko, Marcos Treviso, Andr\'e F. T. Martins
Abstract: Transformer-based architectures traditionally employ softmax to compute attention weights, which produces dense distributions over all tokens in a sequence. While effective in many settings, this density has been shown to be detrimental for tasks that demand precise focus on fixed-size patterns: as sequence length increases, non-informative tokens accumulate attention probability mass, leading to dispersion and representational collapse. We show in this paper that sparse attention mechanisms using $\alpha$-entmax can avoid these issues, due to their ability to assign exact zeros to irrelevant tokens. Furthermore, we introduce Adaptive-Scalable Entmax (ASEntmax), which endows $\alpha$-entmax with a learnable temperature parameter, allowing the attention distribution to interpolate between sparse (pattern-focused) and dense (softmax-like) regimes. Finally, we show that the ability to locate and generalize fixed-size patterns can be further improved through a careful design of position encodings, which impacts both dense and sparse attention methods. By integrating ASEntmax into standard transformer layers alongside proper positional encodings, we show that our models greatly outperform softmax, scalable softmax, and fixed-temperature $\alpha$-entmax baselines on long-context generalization.
Authors: Anvith Pabba, Alex Mathai, Anindya Chakraborty, Baishakhi Ray
Abstract: Large Language Models (LLMs) have shown impressive capabilities in downstream software engineering tasks such as Automated Program Repair (APR). In particular, there has been a lot of research on repository-level issue-resolution benchmarks such as SWE-Bench. Although there has been significant progress on this topic, we notice that in the process of solving such issues, existing agentic systems tend to hyper-localize on immediately suspicious lines of code and fix them in isolation, without a deeper understanding of the issue semantics, code semantics, or execution semantics. Consequently, many existing systems generate patches that overfit to the user issue, even when a more general fix is preferable. To address this limitation, we introduce SemAgent, a novel workflow-based procedure that leverages issue, code, and execution semantics to generate patches that are complete - identifying and fixing all lines relevant to the issue. We achieve this through a novel pipeline that (a) leverages execution semantics to retrieve relevant context, (b) comprehends issue-semantics via generalized abstraction, (c) isolates code-semantics within the context of this abstraction, and (d) leverages this understanding in a two-stage architecture: a repair stage that proposes fine-grained fixes, followed by a reviewer stage that filters relevant fixes based on the inferred issue-semantics. Our evaluations show that our methodology achieves a solve rate of 44.66% on the SWEBench-Lite benchmark beating all other workflow-based approaches, and an absolute improvement of 7.66% compared to our baseline, which lacks such deep semantic understanding. We note that our approach performs particularly well on issues requiring multi-line reasoning (and editing) and edge-case handling, suggesting that incorporating issue and code semantics into APR pipelines can lead to robust and semantically consistent repairs.
Authors: Vladislav Belozerov, Peter J Barclay, Askhan Sami
Abstract: Large-language-model coding tools are now mainstream in software engineering. But as these same tools move human effort up the development stack, they present fresh dangers: 10% of real prompts leak private data, 42% of generated snippets hide security flaws, and the models can even ``agree'' with wrong ideas, a trait called sycophancy. We argue that firms must tag and review every AI-generated line of code, keep prompts and outputs inside private or on-premises deployments, obey emerging safety regulations, and add tests that catch sycophantic answers -- so they can gain speed without losing security and accuracy.
Authors: Vijay Prakash Dwivedi, Charilaos Kanatsoulis, Shenyang Huang, Jure Leskovec
Abstract: Graph machine learning has led to a significant increase in the capabilities of models that learn on arbitrary graph-structured data and has been applied to molecules, social networks, recommendation systems, and transportation, among other domains. Data in multi-tabular relational databases can also be constructed as 'relational entity graphs' for Relational Deep Learning (RDL) - a new blueprint that enables end-to-end representation learning without traditional feature engineering. Compared to arbitrary graph-structured data, relational entity graphs have key properties: (i) their structure is defined by primary-foreign key relationships between entities in different tables, (ii) the structural connectivity is a function of the relational schema defining a database, and (iii) the graph connectivity is temporal and heterogeneous in nature. In this paper, we provide a comprehensive review of RDL by first introducing the representation of relational databases as relational entity graphs, and then reviewing public benchmark datasets that have been used to develop and evaluate recent GNN-based RDL models. We discuss key challenges including large-scale multi-table integration and the complexities of modeling temporal dynamics and heterogeneous data, while also surveying foundational neural network methods and recent architectural advances specialized for relational entity graphs. Finally, we explore opportunities to unify these distinct modeling challenges, highlighting how RDL converges multiple sub-fields in graph machine learning towards the design of foundation models that can transform the processing of relational data.
Authors: Athanasios Glentis, Jiaxiang Li, Andi Han, Mingyi Hong
Abstract: Training large language models (LLMs) typically relies on adaptive optimizers such as Adam, which require significant memory to maintain first- and second-moment matrices, known as optimizer states. While recent works such as GaLore, Fira, and APOLLO have proposed state-compressed variants to reduce memory consumption, a fundamental question remains: What is the minimal amount of optimizer state that is truly necessary to retain state-of-the-art performance in LLM pretraining? In this work, we systematically investigate this question using a bottom-up approach. We find that two memory- and compute-efficient optimization techniques are particularly effective: (1) column-wise gradient normalization significantly boosts the performance of plain SGD without requiring momentum; and (2) adding first-order momentum only to the output layer - where gradient variance is highest - yields performance competitive with fully adaptive methods such as Muon. Based on these insights, we propose SCALE (Stochastic Column-normalized Last-layer Momentum), a new optimizer that combines column-normalized SGD with last-layer momentum, where column normalization refers to normalizing the gradient along the output dimension. Across multiple LLaMA models (60M-1B), SCALE matches or exceeds the performance of Adam while using only 35-45% of the total memory. It also consistently outperforms memory-efficient optimizers such as GaLore, Fira, and APOLLO, making it a strong candidate for large-scale pretraining under memory constraints. For the LLaMA 7B model, SCALE outperforms the state-of-the-art method APOLLO in terms of both perplexity and memory consumption. In addition, our method serves as a minimalist baseline for more sophisticated optimizer design.
Authors: Manuel Brack, Sudeep Katakol, Felix Friedrich, Patrick Schramowski, Hareesh Ravi, Kristian Kersting, Ajinkya Kale
Abstract: Training data is at the core of any successful text-to-image models. The quality and descriptiveness of image text are crucial to a model's performance. Given the noisiness and inconsistency in web-scraped datasets, recent works shifted towards synthetic training captions. While this setup is generally believed to produce more capable models, current literature does not provide any insights into its design choices. This study closes this gap by systematically investigating how different synthetic captioning strategies impact the downstream performance of text-to-image models. Our experiments demonstrate that dense, high-quality captions enhance text alignment but may introduce trade-offs in output aesthetics and diversity. Conversely, captions of randomized lengths yield balanced improvements across aesthetics and alignment without compromising sample diversity. We also demonstrate that varying caption distributions introduce significant shifts in the output bias of a trained model. Our findings underscore the importance of caption design in achieving optimal model performance and provide practical insights for more effective training data strategies in text-to-image generation.
Authors: Penglong Zhai, Yifang Yuan, Fanyi Di, Jie Li, Yue Liu, Chen Li, Jie Huang, Sicong Wang, Yao Xu, Xin Li
Abstract: Generative retrieval-based recommendation has emerged as a promising paradigm aiming at directly generating the identifiers of the target candidates. However, in large-scale recommendation systems, this approach becomes increasingly cumbersome due to the redundancy and sheer scale of the token space. To overcome these limitations, recent research has explored the use of semantic tokens as an alternative to ID tokens, which typically leveraged reconstruction-based strategies, like RQ-VAE, to quantize content embeddings and significantly reduce the embedding size. However, reconstructive quantization aims for the precise reconstruction of each item embedding independently, which conflicts with the goal of generative retrieval tasks focusing more on differentiating among items. Moreover, multi-modal side information of items, such as descriptive text and images, geographical knowledge in location-based recommendation services, has been shown to be effective in improving recommendations by providing richer contexts for interactions. Nevertheless, effectively integrating such complementary knowledge into existing generative recommendation frameworks remains challenging. To overcome these challenges, we propose a novel unsupervised deep quantization exclusively based on contrastive learning, named SimCIT (a Simple Contrastive Item Tokenization framework). Specifically, different from existing reconstruction-based strategies, SimCIT propose to use a learnable residual quantization module to align with the signals from different modalities of the items, which combines multi-modal knowledge alignment and semantic tokenization in a mutually beneficial contrastive learning framework. Extensive experiments across public datasets and a large-scale industrial dataset from various domains demonstrate SimCIT's effectiveness in LLM-based generative recommendation.
Authors: Zhiying Qiu, Tao Lin
Abstract: Diffusion models have recently shown promise in offline RL. However, these methods often suffer from high training costs and slow convergence, particularly when using transformer-based denoising backbones. While several optimization strategies have been proposed -- such as modified noise schedules, auxiliary prediction targets, and adaptive loss weighting -- challenges remain in achieving stable and efficient training. In particular, existing loss weighting functions typically rely on neural network approximators, which can be ineffective in early training phases due to limited generalization capacity of MLPs when exposed to sparse feedback in the early training stages. In this work, we derive a variationally optimal uncertainty-aware weighting function and introduce a closed-form polynomial approximation method for its online estimation under the flow-based generative modeling framework. We integrate our method into a diffusion planning pipeline and evaluate it on standard offline RL benchmarks. Experimental results on Maze2D and Kitchen tasks show that our method achieves competitive performance with up to 10 times fewer training steps, highlighting its practical effectiveness.
Authors: Zhicheng Lin
Abstract: Large language models (LLMs) are rapidly being adopted across psychology, serving as research tools, experimental subjects, human simulators, and computational models of cognition. However, the application of human measurement tools to these systems can produce contradictory results, raising concerns that many findings are measurement phantoms--statistical artifacts rather than genuine psychological phenomena. In this Perspective, we argue that building a robust science of AI psychology requires integrating two of our field's foundational pillars: the principles of reliable measurement and the standards for sound causal inference. We present a dual-validity framework to guide this integration, which clarifies how the evidence needed to support a claim scales with its scientific ambition. Using an LLM to classify text may require only basic accuracy checks, whereas claiming it can simulate anxiety demands a far more rigorous validation process. Current practice systematically fails to meet these requirements, often treating statistical pattern matching as evidence of psychological phenomena. The same model output--endorsing "I am anxious"--requires different validation strategies depending on whether researchers claim to measure, characterize, simulate, or model psychological constructs. Moving forward requires developing computational analogues of psychological constructs and establishing clear, scalable standards of evidence rather than the uncritical application of human measurement tools.
Authors: Zhicheng Lin
Abstract: Large language models (LLMs) offer emerging opportunities for psychological and behavioral research, but methodological guidance is lacking. This article provides a framework for using LLMs as psychological simulators across two primary applications: simulating roles and personas to explore diverse contexts, and serving as computational models to investigate cognitive processes. For simulation, we present methods for developing psychologically grounded personas that move beyond demographic categories, with strategies for validation against human data and use cases ranging from studying inaccessible populations to prototyping research instruments. For cognitive modeling, we synthesize emerging approaches for probing internal representations, methodological advances in causal interventions, and strategies for relating model behavior to human cognition. We address overarching challenges including prompt sensitivity, temporal limitations from training data cutoffs, and ethical considerations that extend beyond traditional human subjects review. Throughout, we emphasize the need for transparency about model capabilities and constraints. Together, this framework integrates emerging empirical evidence about LLM performance--including systematic biases, cultural limitations, and prompt brittleness--to help researchers wrangle these challenges and leverage the unique capabilities of LLMs in psychological research.
Authors: Bin Chen, Xinzge Gao, Chuanrui Hu, Penghang Yu, Hua Zhang, Bing-Kun Bao
Abstract: Generative Reward Models (GRMs) provide greater flexibility than scalar reward models in capturing human preferences, but their effectiveness is limited by poor reasoning capabilities. This often results in incomplete or overly speculative reasoning paths, leading to hallucinations or missing key information in complex tasks. We address this challenge with ReasonGRM, a three-stage generative reward modeling framework. In the first stage, Zero-RL is used to generate concise, outcome-directed reasoning paths that reduce the likelihood of critical omissions. In the second stage, we introduce a novel evaluation metric, $R^\star$, which scores reasoning paths based on their generation likelihood. This favors paths that reach correct answers with minimal exploration, helping to reduce hallucination-prone data during training. In the final stage, the model is further refined through reinforcement learning on challenging examples to enhance its preference discrimination capabilities. Experiments on three public benchmarks show that ReasonGRM achieves competitive or state-of-the-art performance, outperforming previous best GRMs by 1.8\% on average and surpassing proprietary models such as GPT-4o by up to 5.6\%. These results demonstrate the effectiveness of reasoning-aware training and highlight the importance of high-quality rationale selection for reliable preference modeling.
Authors: Chenxu Wang, Yonggang Jin, Cheng Hu, Youpeng Zhao, Zipeng Dai, Jian Zhao, Shiyu Huang, Liuyu Xiang, Junge Zhang, Zhaofeng He
Abstract: Adapting a single agent to a new multi-agent system brings challenges, necessitating adjustments across various tasks, environments, and interactions with unknown teammates and opponents. Addressing this challenge is highly complex, and researchers have proposed two simplified scenarios, Multi-agent reinforcement learning for zero-shot learning and Ad-Hoc Teamwork. Building on these foundations, we propose a more comprehensive setting, Agent Collaborative-Competitive Adaptation (ACCA), which evaluates an agent to generalize across diverse scenarios, tasks, and interactions with both unfamiliar opponents and teammates. In ACCA, agents adjust to task and environmental changes, collaborate with unseen teammates, and compete against unknown opponents. We introduce a new modeling approach, Multi-Retrieval and Dynamic Generation (MRDG), that effectively models both teammates and opponents using their behavioral trajectories. This method incorporates a positional encoder for varying team sizes and a hypernetwork module to boost agents' learning and adaptive capabilities. Additionally, a viewpoint alignment module harmonizes the observational perspectives of retrieved teammates and opponents with the learning agent. Extensive tests in benchmark scenarios like SMAC, Overcooked-AI, and Melting Pot show that MRDG significantly improves robust collaboration and competition with unseen teammates and opponents, surpassing established baselines. Our code is available at: https://github.com/vcis-wangchenxu/MRDG.git
Authors: Yuping Yan, Yizhi Wang, Yuanshuai Li, Yaochu Jin
Abstract: Serial pipeline training is an efficient paradigm for handling data heterogeneity in cross-silo federated learning with low communication overhead. However, even without centralized aggregation, direct transfer of models between clients can violate privacy regulations and remain susceptible to gradient leakage and linkage attacks. Additionally, ensuring resilience against semi-honest or malicious clients who may manipulate or misuse received models remains a grand challenge, particularly in privacy-sensitive domains such as healthcare. To address these challenges, we propose TriCon-SF, a novel serial federated learning framework that integrates triple shuffling and contribution awareness. TriCon-SF introduces three levels of randomization by shuffling model layers, data segments, and training sequences to break deterministic learning patterns and disrupt potential attack vectors, thereby enhancing privacy and robustness. In parallel, it leverages Shapley value methods to dynamically evaluate client contributions during training, enabling the detection of dishonest behavior and enhancing system accountability. Extensive experiments on non-IID healthcare datasets demonstrate that TriCon-SF outperforms standard serial and parallel federated learning in both accuracy and communication efficiency. Security analysis further supports its resilience against client-side privacy attacks.
Authors: Xinyi Liu, Weiguang Wang, Hangfeng He
Abstract: With the growing adoption of Large Language Models (LLMs) for open-ended tasks, accurately assessing epistemic uncertainty, which reflects a model's lack of knowledge, has become crucial to ensuring reliable outcomes. However, quantifying epistemic uncertainty in such tasks is challenging due to the presence of aleatoric uncertainty, which arises from multiple valid answers. While bias can introduce noise into epistemic uncertainty estimation, it may also reduce noise from aleatoric uncertainty. To investigate this trade-off, we conduct experiments on Visual Question Answering (VQA) tasks and find that mitigating prompt-introduced bias improves uncertainty quantification in GPT-4o. Building on prior work showing that LLMs tend to copy input information when model confidence is low, we further analyze how these prompt biases affect measured epistemic and aleatoric uncertainty across varying bias-free confidence levels with GPT-4o and Qwen2-VL. We find that all considered biases induce greater changes in both uncertainties when bias-free model confidence is lower. Moreover, lower bias-free model confidence leads to greater underestimation of epistemic uncertainty (i.e. overconfidence) due to bias, whereas it has no significant effect on the direction of changes in aleatoric uncertainty estimation. These distinct effects deepen our understanding of bias mitigation for uncertainty quantification and potentially inform the development of more advanced techniques.
Authors: Fanchen Bu, Kijung Shin
Abstract: In unsupervised combinatorial optimization (UCO), during training, one aims to have continuous decisions that are promising in a probabilistic sense for each training instance, which enables end-to-end training on initially discrete and non-differentiable problems. At the test time, for each test instance, starting from continuous decisions, derandomization is typically applied to obtain the final deterministic decisions. Researchers have developed more and more powerful test-time derandomization schemes to enhance the empirical performance and the theoretical guarantee of UCO methods. However, we notice a misalignment between training and testing in the existing UCO methods. Consequently, lower training losses do not necessarily entail better post-derandomization performance, even for the training instances without any data distribution shift. Empirically, we indeed observe such undesirable cases. We explore a preliminary idea to better align training and testing in UCO by including a differentiable version of derandomization into training. Our empirical exploration shows that such an idea indeed improves training-test alignment, but also introduces nontrivial challenges into training.
Authors: Daejin Jo, Jeeyoung Yun, Byungseok Roh, Sungwoong Kim
Abstract: With the rapid progress of speech language models (SLMs), discrete speech tokens have emerged as a core interface between speech and text, enabling unified modeling across modalities. Recent speech tokenization approaches aim to isolate semantic information from low-level acoustics to better align with language models. In particular, previous methods use SSL teachers such as HuBERT to extract semantic representations, which are then distilled into a semantic quantizer to suppress acoustic redundancy as well as capture content-related latent structures. However, they still produce speech token sequences significantly longer than their textual counterparts, creating challenges for efficient speech-language modeling. Reducing the frame rate is a natural solution, but standard techniques, such as rigid average pooling across frames, can distort or dilute the semantic structure required for effective LM alignment. To address this, we propose LM-SPT, a speech tokenization method that introduces a novel semantic distillation. Instead of directly matching teacher and student features via pooling, we reconstruct speech solely from semantic tokens and minimize the discrepancy between the encoded representations of the original and reconstructed waveforms, obtained from a frozen automatic speech recognition (ASR) encoder. This indirect yet data-driven supervision enables the tokenizer to learn discrete units that are more semantically aligned with language models. LM-SPT further incorporates architectural improvements to the encoder and decoder for speech tokenization, and supports multiple frame rates, including 25Hz, 12.5Hz, and 6.25Hz. Experimental results show that LM-SPT achieves superior reconstruction fidelity compared to baselines, and that SLMs trained with LM-SPT tokens achieve competitive performances on speech-to-text and consistently outperform baselines on text-to-speech tasks.
Authors: Hyun Joon Park, Jeongmin Liu, Jin Sob Kim, Jeong Yeol Yang, Sung Won Han, Eunwoo Song
Abstract: We introduce RapFlow-TTS, a rapid and high-fidelity TTS acoustic model that leverages velocity consistency constraints in flow matching (FM) training. Although ordinary differential equation (ODE)-based TTS generation achieves natural-quality speech, it typically requires a large number of generation steps, resulting in a trade-off between quality and inference speed. To address this challenge, RapFlow-TTS enforces consistency in the velocity field along the FM-straightened ODE trajectory, enabling consistent synthetic quality with fewer generation steps. Additionally, we introduce techniques such as time interval scheduling and adversarial learning to further enhance the quality of the few-step synthesis. Experimental results show that RapFlow-TTS achieves high-fidelity speech synthesis with a 5- and 10-fold reduction in synthesis steps than the conventional FM- and score-based approaches, respectively.
Authors: Kosuke Nakanishi, Akihiro Kubo, Yuji Yasui, Shin Ishii
Abstract: Recently, robust reinforcement learning (RL) methods designed to handle adversarial input observations have received significant attention, motivated by RL's inherent vulnerabilities. While existing approaches have demonstrated reasonable success, addressing worst-case scenarios over long time horizons requires both minimizing the agent's cumulative rewards for adversaries and training agents to counteract them through alternating learning. However, this process introduces mutual dependencies between the agent and the adversary, making interactions with the environment inefficient and hindering the development of off-policy methods. In this work, we propose a novel off-policy method that eliminates the need for additional environmental interactions by reformulating adversarial learning as a soft-constrained optimization problem. Our approach is theoretically supported by the symmetric property of policy evaluation between the agent and the adversary. The implementation is available at https://github.com/nakanakakosuke/VALT_SAC.
Authors: Jongmin Park, Seunghoon Han, Won-Yong Shin, Sungsu Lim
Abstract: The hyperbolic space, characterized by a constant negative curvature and exponentially expanding space, aligns well with the structural properties of heterogeneous graphs. However, although heterogeneous graphs inherently possess diverse power-law structures, most hyperbolic heterogeneous graph embedding models rely on a single hyperbolic space. This approach may fail to effectively capture the diverse power-law structures within heterogeneous graphs. To address this limitation, we propose a Metapath-based Hyperbolic Contrastive Learning framework (MHCL), which uses multiple hyperbolic spaces to capture diverse complex structures within heterogeneous graphs. Specifically, by learning each hyperbolic space to describe the distribution of complex structures corresponding to each metapath, it is possible to capture semantic information effectively. Since metapath embeddings represent distinct semantic information, preserving their discriminability is important when aggregating them to obtain node representations. Therefore, we use a contrastive learning approach to optimize MHCL and improve the discriminability of metapath embeddings. In particular, our contrastive learning method minimizes the distance between embeddings of the same metapath and maximizes the distance between those of different metapaths in hyperbolic space, thereby improving the separability of metapath embeddings with distinct semantic information. We conduct comprehensive experiments to evaluate the effectiveness of MHCL. The experimental results demonstrate that MHCL outperforms state-of-the-art baselines in various graph machine learning tasks, effectively capturing the complex structures of heterogeneous graphs.
Authors: Lance Ying, Ryan Truong, Katherine M. Collins, Cedegao E. Zhang, Megan Wei, Tyler Brooke-Wilson, Tan Zhi-Xuan, Lionel Wong, Joshua B. Tenenbaum
Abstract: Drawing real world social inferences usually requires taking into account information from multiple modalities. Language is a particularly powerful source of information in social settings, especially in novel situations where language can provide both abstract information about the environment dynamics and concrete specifics about an agent that cannot be easily visually observed. In this paper, we propose Language-Informed Rational Agent Synthesis (LIRAS), a framework for drawing context-specific social inferences that integrate linguistic and visual inputs. LIRAS frames multimodal social reasoning as a process of constructing structured but situation-specific agent and environment representations - leveraging multimodal language models to parse language and visual inputs into unified symbolic representations, over which a Bayesian inverse planning engine can be run to produce granular probabilistic judgments. On a range of existing and new social reasoning tasks derived from cognitive science experiments, we find that our model (instantiated with a comparatively lightweight VLM) outperforms ablations and state-of-the-art models in capturing human judgments across all domains.
Authors: Beomseok Ko, Hyeryung Jang
Abstract: Diffusion models excel in image generation but are computational and resource-intensive due to their reliance on iterative Markov chain processes, leading to error accumulation and limiting the effectiveness of naive compression techniques. In this paper, we propose PQCAD-DM, a novel hybrid compression framework combining Progressive Quantization (PQ) and Calibration-Assisted Distillation (CAD) to address these challenges. PQ employs a two-stage quantization with adaptive bit-width transitions guided by a momentum-based mechanism, reducing excessive weight perturbations in low-precision. CAD leverages full-precision calibration datasets during distillation, enabling the student to match full-precision performance even with a quantized teacher. As a result, PQCAD-DM achieves a balance between computational efficiency and generative quality, halving inference time while maintaining competitive performance. Extensive experiments validate PQCAD-DM's superior generative capabilities and efficiency across diverse datasets, outperforming fixed-bit quantization methods.
Authors: Youjin Kong
Abstract: Fairness in machine learning (ML) has become a rapidly growing area of research. But why, in the first place, is unfairness in ML morally wrong? And why should we care about improving fairness? Most fair-ML research implicitly appeals to distributive equality: the idea that desirable goods and benefits, such as opportunities (e.g., Barocas et al., 2023), should be equally distributed across society. Unfair ML models, then, are seen as wrong because they unequally distribute such benefits. This paper argues that this exclusive focus on distributive equality offers an incomplete and potentially misleading ethical foundation. Grounding ML fairness in egalitarianism -- the view that equality is a fundamental moral and social ideal -- requires challenging structural inequality: systematic, institutional, and durable arrangements that privilege some groups while disadvantaging others. Structural inequality manifests through ML systems in two primary forms: allocative harms (e.g., economic loss) and representational harms (e.g., stereotypes, erasure). While distributive equality helps address allocative harms, it fails to explain why representational harms are wrong -- why it is wrong for ML systems to reinforce social hierarchies that stratify people into superior and inferior groups -- and why ML systems should aim to foster a society where people relate as equals (i.e., relational equality). To address these limitations, the paper proposes a multifaceted egalitarian framework for ML fairness that integrates both distributive and relational equality. Drawing on critical social and political philosophy, this framework offers a more comprehensive ethical foundation for tackling the full spectrum of harms perpetuated by ML systems. The paper also outlines practical pathways for implementing the framework across the ML pipeline.
Authors: Nick Erickson, Lennart Purucker, Andrej Tschalzev, David Holzm\"uller, Prateek Mutalik Desai, and David Salinas, Frank Hutter
Abstract: With the growing popularity of deep learning and foundation models for tabular data, the need for standardized and reliable benchmarks is higher than ever. However, current benchmarks are static. Their design is not updated even if flaws are discovered, model versions are updated, or new models are released. To address this, we introduce TabArena, the first continuously maintained living tabular benchmarking system. To launch TabArena, we manually curate a representative collection of datasets and well-implemented models, conduct a large-scale benchmarking study to initialize a public leaderboard, and assemble a team of experienced maintainers. Our results highlight the influence of validation method and ensembling of hyperparameter configurations to benchmark models at their full potential. While gradient-boosted trees are still strong contenders on practical tabular datasets, we observe that deep learning methods have caught up under larger time budgets with ensembling. At the same time, foundation models excel on smaller datasets. Finally, we show that ensembles across models advance the state-of-the-art in tabular machine learning and investigate the contributions of individual models. We launch TabArena with a public leaderboard, reproducible code, and maintenance protocols to create a living benchmark available at https://tabarena.ai.
URLs: https://tabarena.ai.
Authors: Muyang Zheng, Yuanzhi Yao, Changting Lin, Rui Wang, Meng Han
Abstract: Despite efforts to align large language models (LLMs) with societal and moral values, these models remain susceptible to jailbreak attacks--methods designed to elicit harmful responses. Jailbreaking black-box LLMs is considered challenging due to the discrete nature of token inputs, restricted access to the target LLM, and limited query budget. To address the issues above, we propose an effective method for jailbreaking black-box large language Models via Iterative Semantic Tuning, named MIST. MIST enables attackers to iteratively refine prompts that preserve the original semantic intent while inducing harmful content. Specifically, to balance semantic similarity with computational efficiency, MIST incorporates two key strategies: sequential synonym search, and its advanced version--order-determining optimization. Extensive experiments across two open-source models and four closed-source models demonstrate that MIST achieves competitive attack success rates and attack transferability compared with other state-of-the-art white-box and black-box jailbreak methods. Additionally, we conduct experiments on computational efficiency to validate the practical viability of MIST.
Authors: Chengpeng Hu, Ziming Wang, Bo Yuan, Jialin Liu, Chengqi Zhang, Xin Yao
Abstract: Dynamic material handling (DMH) involves the assignment of dynamically arriving material transporting tasks to suitable vehicles in real time for minimising makespan and tardiness. In real-world scenarios, historical task records are usually available, which enables the training of a decision policy on multiple instances consisting of historical records. Recently, reinforcement learning has been applied to solve DMH. Due to the occurrence of dynamic events such as new tasks, adaptability is highly required. Solving DMH is challenging since constraints including task delay should be satisfied. A feedback is received only when all tasks are served, which leads to sparse reward. Besides, making the best use of limited computational resources and historical records for training a robust policy is crucial. The time allocated to different problem instances would highly impact the learning process. To tackle those challenges, this paper proposes a novel adaptive constrained evolutionary reinforcement learning (ACERL) approach, which maintains a population of actors for diverse exploration. ACERL accesses each actor for tackling sparse rewards and constraint violation to restrict the behaviour of the policy. Moreover, ACERL adaptively selects the most beneficial training instances for improving the policy. Extensive experiments on eight training and eight unseen test instances demonstrate the outstanding performance of ACERL compared with several state-of-the-art algorithms. Policies trained by ACERL can schedule the vehicles while fully satisfying the constraints. Additional experiments on 40 unseen noised instances show the robust performance of ACERL. Cross-validation further presents the overall effectiveness of ACREL. Besides, a rigorous ablation study highlights the coordination and benefits of each ingredient of ACERL.
Authors: Yuchu Jiang, Jiaming Chu, Jian Zhao, Xin Zhang, Xu Yang, Lei Jin, Chi Zhang, Xuelong Li
Abstract: The proliferation of generative models has raised serious concerns about visual content forgery. Existing deepfake detection methods primarily target either image-level classification or pixel-wise localization. While some achieve high accuracy, they often suffer from limited generalization across manipulation types or rely on complex architectures. In this paper, we propose Loupe, a lightweight yet effective framework for joint deepfake detection and localization. Loupe integrates a patch-aware classifier and a segmentation module with conditional queries, allowing simultaneous global authenticity classification and fine-grained mask prediction. To enhance robustness against distribution shifts of test set, Loupe introduces a pseudo-label-guided test-time adaptation mechanism by leveraging patch-level predictions to supervise the segmentation head. Extensive experiments on the DDL dataset demonstrate that Loupe achieves state-of-the-art performance, securing the first place in the IJCAI 2025 Deepfake Detection and Localization Challenge with an overall score of 0.846. Our results validate the effectiveness of the proposed patch-level fusion and conditional query design in improving both classification accuracy and spatial localization under diverse forgery patterns. The code is available at https://github.com/Kamichanw/Loupe.
Authors: Daniel Frau-Alfaro, Julio Casta\~no-Amoros, Santiago Puente, Pablo Gil, Roberto Calandra
Abstract: Object handover is an important skill that we use daily when interacting with other humans. To deploy robots in collaborative setting, like houses, being able to receive and handing over objects safely and efficiently becomes a crucial skill. In this work, we demonstrate the use of Reinforcement Learning (RL) for dexterous object handover between two multi-finger hands. Key to this task is the use of a novel reward function based on dual quaternions to minimize the rotation distance, which outperforms other rotation representations such as Euler and rotation matrices. The robustness of the trained policy is experimentally evaluated by testing w.r.t. objects that are not included in the training distribution, and perturbations during the handover process. The results demonstrate that the trained policy successfully perform this task, achieving a total success rate of 94% in the best-case scenario after 100 experiments, thereby showing the robustness of our policy with novel objects. In addition, the best-case performance of the policy decreases by only 13.8% when the other robot moves during the handover, proving that our policy is also robust to this type of perturbation, which is common in real-world object handovers.
Authors: Sattwik Sahu, Agamdeep Singh, Karthik Nambiar, Srikanth Saripalli, P. B. Sujit
Abstract: Off-road traversability segmentation enables autonomous navigation with applications in search-and-rescue, military operations, wildlife exploration, and agriculture. Current frameworks struggle due to significant variations in unstructured environments and uncertain scene changes, and are not adaptive to be used for different robot types. We present AnyTraverse, a framework combining natural language-based prompts with human-operator assistance to determine navigable regions for diverse robotic vehicles. The system segments scenes for a given set of prompts and calls the operator only when encountering previously unexplored scenery or unknown class not part of the prompt in its region-of-interest, thus reducing active supervision load while adapting to varying outdoor scenes. Our zero-shot learning approach eliminates the need for extensive data collection or retraining. Our experimental validation includes testing on RELLIS-3D, Freiburg Forest, and RUGD datasets and demonstrate real-world deployment on multiple robot platforms. The results show that AnyTraverse performs better than GA-NAV and Off-seg while offering a vehicle-agnostic approach to off-road traversability that balances automation with targeted human supervision.
Authors: Victor Alejandre (Universidad Politecnica de Madrid, Spain), Concha Bielza (Universidad Politecnica de Madrid, Spain), Pedro Larra\~naga (Universidad Politecnica de Madrid, Spain)
Abstract: Semiparametric Bayesian networks (SPBNs) integrate parametric and non-parametric probabilistic models, offering flexibility in learning complex data distributions from samples. In particular, kernel density estimators (KDEs) are employed for the non-parametric component. Under the assumption of data normality, the normal rule is used to learn the bandwidth matrix for the KDEs in SPBNs. This matrix is the key hyperparameter that controls the trade-off between bias and variance. However, real-world data often deviates from normality, potentially leading to suboptimal density estimation and reduced predictive performance. This paper first establishes the theoretical framework for the application of state-of-the-art bandwidth selectors and subsequently evaluates their impact on SPBN performance. We explore the approaches of cross-validation and plug-in selectors, assessing their effectiveness in enhancing the learning capability and applicability of SPBNs. To support this investigation, we have extended the open-source package PyBNesian for SPBNs with the additional bandwidth selection techniques and conducted extensive experimental analyses. Our results demonstrate that the proposed bandwidth selectors leverage increasing information more effectively than the normal rule, which, despite its robustness, stagnates with more data. In particular, unbiased cross-validation generally outperforms the normal rule, highlighting its advantage in high sample size scenarios.
Authors: Jun Fu, Bin Tian, Haonan Chen, Shi Meng, Tingting Yao
Abstract: Autonomous parking plays a vital role in intelligent vehicle systems, particularly in constrained urban environments where high-precision control is required. While traditional rule-based parking systems struggle with environmental uncertainties and lack adaptability in crowded or dynamic scenes, human drivers demonstrate the ability to park intuitively without explicit modeling. Inspired by this observation, we propose a Transformer-based end-to-end framework for autonomous parking that learns from expert demonstrations. The network takes as input surround-view camera images, goal-point representations, ego vehicle motion, and pedestrian trajectories. It outputs discrete control sequences including throttle, braking, steering, and gear selection. A novel cross-attention module integrates BEV features with target points, and a GRU-based pedestrian predictor enhances safety by modeling dynamic obstacles. We validate our method on the CARLA 0.9.14 simulator in both vertical and parallel parking scenarios. Experiments show our model achieves a high success rate of 96.57\%, with average positional and orientation errors of 0.21 meters and 0.41 degrees, respectively. The ablation studies further demonstrate the effectiveness of key modules such as pedestrian prediction and goal-point attention fusion. The code and dataset will be released at: https://github.com/little-snail-f/ParkFormer.
Authors: Jacopo Graldi, Alessandro Breccia, Giulia Lanzillotta, Thomas Hofmann, Lorenzo Noci
Abstract: Despite recent efforts, neural networks still struggle to learn in non-stationary environments, and our understanding of catastrophic forgetting (CF) is far from complete. In this work, we perform a systematic study on the impact of model scale and the degree of feature learning in continual learning. We reconcile existing contradictory observations on scale in the literature, by differentiating between lazy and rich training regimes through a variable parameterization of the architecture. We show that increasing model width is only beneficial when it reduces the amount of feature learning, yielding more laziness. Using the framework of dynamical mean field theory, we then study the infinite width dynamics of the model in the feature learning regime and characterize CF, extending prior theoretical results limited to the lazy regime. We study the intricate relationship between feature learning, task non-stationarity, and forgetting, finding that high feature learning is only beneficial with highly similar tasks. We identify a transition modulated by task similarity where the model exits an effectively lazy regime with low forgetting to enter a rich regime with significant forgetting. Finally, our findings reveal that neural networks achieve optimal performance at a critical level of feature learning, which depends on task non-stationarity and transfers across model scales. This work provides a unified perspective on the role of scale and feature learning in continual learning.
Authors: Fabian Gr\"oger, Shuo Wen, Huyen Le, Maria Brbi\'c
Abstract: Multimodal models have demonstrated powerful capabilities in complex tasks requiring multimodal alignment including zero-shot classification and cross-modal retrieval. However, existing models typically rely on millions of paired multimodal samples, which are prohibitively expensive or infeasible to obtain in many domains. In this work, we explore the feasibility of building multimodal models with limited amount of paired data by aligning pretrained unimodal foundation models. We show that high-quality alignment is possible with as few as tens of thousands of paired samples$\unicode{x2013}$less than $1\%$ of the data typically used in the field. To achieve this, we introduce STRUCTURE, an effective regularization technique that preserves the neighborhood geometry of the latent space of unimodal encoders. Additionally, we show that aligning last layers is often suboptimal and demonstrate the benefits of aligning the layers with the highest representational similarity across modalities. These two components can be readily incorporated into existing alignment methods, yielding substantial gains across 24 zero-shot image classification and retrieval benchmarks, with average relative improvement of $51.6\%$ in classification and $91.8\%$ in retrieval tasks. Our results highlight the effectiveness and broad applicability of our framework for limited-sample multimodal learning and offer a promising path forward for resource-constrained domains.
Authors: Jonas Wagner, Simon M\"uller, Christian N\"ather, Jan-Philipp Stegh\"ofer, Andreas Both
Abstract: A key challenge in security analysis is the manual evaluation of potential security weaknesses generated by static application security testing (SAST) tools. Numerous false positives (FPs) in these reports reduce the effectiveness of security analysis. We propose using Large Language Models (LLMs) to improve the assessment of SAST findings. We investigate the ability of LLMs to reduce FPs while trying to maintain a perfect true positive rate, using datasets extracted from the OWASP Benchmark (v1.2) and a real-world software project. Our results indicate that advanced prompting techniques, such as Chain-of-Thought and Self-Consistency, substantially improve FP detection. Notably, some LLMs identified approximately 62.5% of FPs in the OWASP Benchmark dataset without missing genuine weaknesses. Combining detections from different LLMs would increase this FP detection to approximately 78.9%. Additionally, we demonstrate our approach's generalizability using a real-world dataset covering five SAST tools, three programming languages, and infrastructure files. The best LLM detected 33.85% of all FPs without missing genuine weaknesses, while combining detections from different LLMs would increase this detection to 38.46%. Our findings highlight the potential of LLMs to complement traditional SAST tools, enhancing automation and reducing resources spent addressing false alarms.
Authors: Jack Griffiths, Steven A. Wrathmall, Simon A. Gardiner
Abstract: Precise determination of thermodynamic parameters in ultracold Bose gases remains challenging due to the destructive nature of conventional measurement techniques and inherent experimental uncertainties. We demonstrate an artificial intelligence approach for rapid, non-destructive estimation of the chemical potential and temperature from single-shot, in situ imaged density profiles of finite-temperature Bose gases. Our convolutional neural network is trained exclusively on quasi-2D `pancake' condensates in harmonic trap configurations. It achieves parameter extraction within fractions of a second. The model also demonstrates zero-shot generalisation across both trap geometry and thermalisation dynamics, successfully estimating thermodynamic parameters for toroidally trapped condensates with errors of only a few nanokelvin despite no prior exposure to such geometries during training, and maintaining predictive accuracy during dynamic thermalisation processes after a relatively brief evolution without explicit training on non-equilibrium states. These results suggest that supervised learning can overcome traditional limitations in ultracold atom thermometry, with extension to broader geometric configurations, temperature ranges, and additional parameters potentially enabling comprehensive real-time analysis of quantum gas experiments. Such capabilities could significantly streamline experimental workflows whilst improving measurement precision across a range of quantum fluid systems.
Authors: Mohon Raihan, Plabon Kumar Saha, Rajan Das Gupta, A Z M Tahmidul Kabir, Afia Anjum Tamanna, Md. Harun-Ur-Rashid, Adnan Bin Abdus Salam, Md Tanvir Anjum, A Z M Ahteshamul Kabir
Abstract: Neonatal death is still a concerning reality for underdeveloped and even some developed countries. Worldwide data indicate that 26.693 babies out of 1,000 births die, according to Macro Trades. To reduce this number, early prediction of endangered babies is crucial. Such prediction enables the opportunity to take ample care of the child and mother so that early child death can be avoided. In this context, machine learning was used to determine whether a newborn baby is at risk. To train the predictive model, historical data of 1.4 million newborns was used. Machine learning and deep learning techniques such as logical regression, K-nearest neighbor, random forest classifier, extreme gradient boosting (XGBoost), convolutional neural network, and long short-term memory (LSTM) were implemented using the dataset to identify the most accurate model for predicting neonatal mortality. Among the machine learning algorithms, XGBoost and random forest classifier achieved the best accuracy with 94%, while among the deep learning models, LSTM delivered the highest accuracy with 99%. Therefore, using LSTM appears to be the most suitable approach to predict whether precautionary measures for a child are necessary.
Authors: Haoran Sun, Yankai Jiang, Wenjie Lou, Yujie Zhang, Wenjie Li, Lilong Wang, Mianxin Liu, Lei Liu, Xiaosong Wang
Abstract: Multimodal large language models (MLLMs) have begun to demonstrate robust reasoning capabilities on general tasks, yet their application in the medical domain remains in its early stages. Constructing chain-of-thought (CoT) training data is essential for bolstering the reasoning abilities of medical MLLMs. However, existing approaches exhibit a deficiency in offering a comprehensive framework for searching and evaluating effective reasoning paths towards critical diagnosis. To address this challenge, we propose Mentor-Intern Collaborative Search (MICS), a novel reasoning-path searching scheme to generate rigorous and effective medical CoT data. MICS first leverages mentor models to initialize the reasoning, one step at a time, then prompts each intern model to continue the thinking along those initiated paths, and finally selects the optimal reasoning path according to the overall reasoning performance of multiple intern models. The reasoning performance is determined by an MICS-Score, which assesses the quality of generated reasoning paths. Eventually, we construct MMRP, a multi-task medical reasoning dataset with ranked difficulty, and Chiron-o1, a new medical MLLM devised via a curriculum learning strategy, with robust visual question-answering and generalizable reasoning capabilities. Extensive experiments demonstrate that Chiron-o1, trained on our CoT dataset constructed using MICS, achieves state-of-the-art performance across a list of medical visual question answering and reasoning benchmarks. Codes are available at GitHub - manglu097/Chiron-o1: Enhancing Step-by-Step and Verifiable Medical Reasoning in MLLMs
Authors: Oliver Sch\"on, Sofie Haesaert, Sadegh Soudjani
Abstract: The requirement for identifying accurate system representations has not only been a challenge to fulfill, but it has compromised the scalability of formal methods, as the resulting models are often too complex for effective decision making with formal correctness and performance guarantees. Focusing on probabilistic simulation relations and surrogate models of stochastic systems, we propose an approach that significantly enhances the scalability and practical applicability of such simulation relations by eliminating the need to compute error bounds directly. As a result, we provide an abstraction-based technique that scales effectively to higher dimensions while addressing complex nonlinear agent-environment interactions with infinite-horizon temporal logic guarantees amidst uncertainty. Our approach trades scalability for conservatism favorably, as demonstrated on a complex high-dimensional vehicle intersection case study.
Authors: Guan Zhe Hong, Bhavya Vasudeva, Vatsal Sharan, Cyrus Rashtchian, Prabhakar Raghavan, Rina Panigrahy
Abstract: When large language models (LLMs) use in-context learning (ICL) to solve a new task, they seem to grasp not only the goal of the task but also core, latent concepts in the demonstration examples. This begs the question of whether transformers represent latent structures as part of their computation or whether they take shortcuts to solve the problem. Prior mechanistic work on ICL does not address this question because it does not sufficiently examine the relationship between the learned representation and the latent concept, and the considered problem settings often involve only single-step reasoning. In this work, we examine how transformers disentangle and use latent concepts. We show that in 2-hop reasoning tasks with a latent, discrete concept, the model successfully identifies the latent concept and does step-by-step concept composition. In tasks parameterized by a continuous latent concept, we find low-dimensional subspaces in the representation space where the geometry mimics the underlying parameterization. Together, these results refine our understanding of ICL and the representation of transformers, and they provide evidence for highly localized structures in the model that disentangle latent concepts in ICL tasks.
Authors: Antonin Berthon, Mihaela van der Schaar
Abstract: Accurately assessing student knowledge is critical for effective education, yet traditional Knowledge Tracing (KT) methods rely on opaque latent embeddings, limiting interpretability. Even LLM-based approaches generate direct predictions or summaries that may hallucinate without any accuracy guarantees. We recast KT as an inverse problem: learning the minimum natural-language summary that makes past answers explainable and future answers predictable. Our Language Bottleneck Model (LBM) consists of an encoder LLM that writes an interpretable knowledge summary and a frozen decoder LLM that must reconstruct and predict student responses using only that summary text. By constraining all predictive information to pass through a short natural-language bottleneck, LBMs ensure that the summary contains accurate information while remaining human-interpretable. Experiments on synthetic arithmetic benchmarks and the large-scale Eedi dataset show that LBMs rival the accuracy of state-of-the-art KT and direct LLM methods while requiring orders-of-magnitude fewer student trajectories. We demonstrate that training the encoder with group-relative policy optimization, using downstream decoding accuracy as a reward signal, effectively improves summary quality.
Authors: Sahil Kale, Vijaykant Nadadur
Abstract: LaTeX's precision and flexibility in typesetting have made it the gold standard for the preparation of scientific documentation. Large Language Models (LLMs) present a promising opportunity for researchers to produce publication-ready material using LaTeX with natural language instructions, yet current benchmarks completely lack evaluation of this ability. By introducing TeXpert, our benchmark dataset with natural language prompts for generating LaTeX code focused on components of scientific documents across multiple difficulty levels, we conduct an in-depth analysis of LLM performance in this regard and identify frequent error types. Our evaluation across open and closed-source LLMs highlights multiple key findings: LLMs excelling on standard benchmarks perform poorly in LaTeX generation with a significant accuracy drop-off as the complexity of tasks increases; open-source models like DeepSeek v3 and DeepSeek Coder strongly rival closed-source counterparts in LaTeX tasks; and formatting and package errors are unexpectedly prevalent, suggesting a lack of diverse LaTeX examples in the training datasets of most LLMs. Our dataset, code, and model evaluations are available at https://github.com/knowledge-verse-ai/TeXpert.
Authors: Giuseppe Attanasio, Sonal Sannigrahi, Ben Peters, Andr\'e F. T. Martins
Abstract: This paper presents the IT-IST submission to the IWSLT 2025 Shared Task on Instruction Following Speech Processing. We submit results for the Short Track, i.e., speech recognition, translation, and spoken question answering. Our model is a unified speech-to-text model that integrates a pre-trained continuous speech encoder and text decoder through a first phase of modality alignment and a second phase of instruction fine-tuning. Crucially, we focus on using small-scale language model backbones (< 2B) and restrict to high-quality, CC-BY data along with synthetic data generation to supplement existing resources.
Authors: Elizabeth Fons, Alejandro Sztrajman, Yousef El-Laham, Luciana Ferrer, Svitlana Vyetrenko, Manuela Veloso
Abstract: Time series with missing or irregularly sampled data are a persistent challenge in machine learning. Many methods operate on the frequency-domain, relying on the Fast Fourier Transform (FFT) which assumes uniform sampling, therefore requiring prior interpolation that can distort the spectra. To address this limitation, we introduce a differentiable Lomb--Scargle layer that enables a reliable computation of the power spectrum of irregularly sampled data. We integrate this layer into a novel score-based diffusion model (LSCD) for time series imputation conditioned on the entire signal spectrum. Experiments on synthetic and real-world benchmarks demonstrate that our method recovers missing data more accurately than purely time-domain baselines, while simultaneously producing consistent frequency estimates. Crucially, our method can be easily integrated into learning frameworks, enabling broader adoption of spectral guidance in machine learning approaches involving incomplete or irregular data.
Authors: Joshua Schraven, Alexander Windmann, Oliver Niggemann
Abstract: Benchmark datasets for network intrusion detection commonly rely on synthetically generated traffic, which fails to reflect the statistical variability and temporal drift encountered in operational environments. This paper introduces MAWIFlow, a flow-based benchmark derived from the MAWILAB v1.1 dataset, designed to enable realistic and reproducible evaluation of anomaly detection methods. A reproducible preprocessing pipeline is presented that transforms raw packet captures into flow representations conforming to the CICFlowMeter format, while preserving MAWILab's original anomaly labels. The resulting datasets comprise temporally distinct samples from January 2011, 2016, and 2021, drawn from trans-Pacific backbone traffic. To establish reference baselines, traditional machine learning methods, including Decision Trees, Random Forests, XGBoost, and Logistic Regression, are compared to a deep learning model based on a CNN-BiLSTM architecture. Empirical results demonstrate that tree-based classifiers perform well on temporally static data but experience significant performance degradation over time. In contrast, the CNN-BiLSTM model maintains better performance, thus showing improved generalization. These findings underscore the limitations of synthetic benchmarks and static models, and motivate the adoption of realistic datasets with explicit temporal structure. All datasets, pipeline code, and model implementations are made publicly available to foster transparency and reproducibility.
Authors: Jingtong Su, Julia Kempe, Karen Ullrich
Abstract: Transformers have achieved state-of-the-art performance across language and vision tasks. This success drives the imperative to interpret their internal mechanisms with the dual goals of enhancing performance and improving behavioral control. Attribution methods help advance interpretability by assigning model outputs associated with a target concept to specific model components. Current attribution research primarily studies multi-layer perceptron neurons and addresses relatively simple concepts such as factual associations (e.g., Paris is located in France). This focus tends to overlook the impact of the attention mechanism and lacks a unified approach for analyzing more complex concepts. To fill these gaps, we introduce Scalable Attention Module Discovery (SAMD), a concept-agnostic method for mapping arbitrary, complex concepts to specific attention heads of general transformer models. We accomplish this by representing each concept as a vector, calculating its cosine similarity with each attention head, and selecting the TopK-scoring heads to construct the concept-associated attention module. We then propose Scalar Attention Module Intervention (SAMI), a simple strategy to diminish or amplify the effects of a concept by adjusting the attention module using only a single scalar parameter. Empirically, we demonstrate SAMD on concepts of varying complexity, and visualize the locations of their corresponding modules. Our results demonstrate that module locations remain stable before and after LLM post-training, and confirm prior work on the mechanics of LLM multilingualism. Through SAMI, we facilitate jailbreaking on HarmBench (+72.7%) by diminishing "safety" and improve performance on the GSM8K benchmark (+1.6%) by amplifying "reasoning". Lastly, we highlight the domain-agnostic nature of our approach by suppressing the image classification accuracy of vision transformers on ImageNet.
Authors: Abdellah Rahmani, Pascal Frossard
Abstract: Understanding causal relationships in multivariate time series is crucial in many scenarios, such as those dealing with financial or neurological data. Many such time series exhibit multiple regimes, i.e., consecutive temporal segments with a priori unknown boundaries, with each regime having its own causal structure. Inferring causal dependencies and regime shifts is critical for analyzing the underlying processes. However, causal structure learning in this setting is challenging due to (1) non stationarity, i.e., each regime can have its own causal graph and mixing function, and (2) complex noise distributions, which may be non Gaussian or heteroscedastic. Existing causal discovery approaches cannot address these challenges, since generally assume stationarity or Gaussian noise with constant variance. Hence, we introduce FANTOM, a unified framework for causal discovery that handles non stationary processes along with non Gaussian and heteroscedastic noises. FANTOM simultaneously infers the number of regimes and their corresponding indices and learns each regime's Directed Acyclic Graph. It uses a Bayesian Expectation Maximization algorithm that maximizes the evidence lower bound of the data log likelihood. On the theoretical side, we prove, under mild assumptions, that temporal heteroscedastic causal models, introduced in FANTOM's formulation, are identifiable in both stationary and non stationary settings. In addition, extensive experiments on synthetic and real data show that FANTOM outperforms existing methods.
Authors: Valeria Vuk, Cristina Sarasua, Fabrizio Gilardi
Abstract: A wide range of participation is essential for democracy, as it helps prevent the dominance of extreme views, erosion of legitimacy, and political polarization. However, engagement in online political discussions often features a limited spectrum of views due to high levels of self-selection and the tendency of online platforms to facilitate exchanges primarily among like-minded individuals. This study examines whether an LLM-based bot can widen the scope of perspectives expressed by participants in online discussions through two pre-registered randomized experiments conducted in a chatroom. We evaluate the impact of a bot that actively monitors discussions, identifies missing arguments, and introduces them into the conversation. The results indicate that our bot significantly expands the range of arguments, as measured by both objective and subjective metrics. Furthermore, disclosure of the bot as AI does not significantly alter these effects. These findings suggest that LLM-based moderation tools can positively influence online political discourse.
Authors: Ricardo Rei, Nuno M. Guerreiro, Jos\'e Pombal, Jo\~ao Alves, Pedro Teixeirinha, Amin Farajian, Andr\'e F. T. Martins
Abstract: Fine-tuning pretrained LLMs has been shown to be an effective strategy for reaching state-of-the-art performance on specific tasks like machine translation. However, this process of adaptation often implies sacrificing general-purpose capabilities, such as conversational reasoning and instruction-following, hampering the utility of the system in real-world applications that require a mixture of skills. In this paper, we introduce Tower+, a suite of models designed to deliver strong performance across both translation and multilingual general-purpose text capabilities. We achieve a Pareto frontier between translation specialization and multilingual general-purpose capabilities by introducing a novel training recipe that builds on Tower (Alves et al., 2024), comprising continued pretraining, supervised fine-tuning, preference optimization, and reinforcement learning with verifiable rewards. At each stage of training, we carefully generate and curate data to strengthen performance on translation as well as general-purpose tasks involving code generation, mathematics problem solving, and general instruction-following. We develop models at multiple scales: 2B, 9B, and 72B. Our smaller models often outperform larger general-purpose open-weight and proprietary LLMs (e.g., Llama 3.3 70B, GPT-4o). Our largest model delivers best-in-class translation performance for high-resource languages and top results in multilingual Arena Hard evaluations and in IF-MT, a benchmark we introduce for evaluating both translation and instruction-following. Our findings highlight that it is possible to rival frontier models in general capabilities, while optimizing for specific business domains, such as translation and localization.
Authors: Konstantin Usevich, Clara D\'erand, Ricardo Borsoi, Marianne Clausel
Abstract: Polynomial Neural Networks (PNNs) possess a rich algebraic and geometric structure. However, their identifiability -- a key property for ensuring interpretability -- remains poorly understood. In this work, we present a comprehensive analysis of the identifiability of deep PNNs, including architectures with and without bias terms. Our results reveal an intricate interplay between activation degrees and layer widths in achieving identifiability. As special cases, we show that architectures with non-increasing layer widths are generically identifiable under mild conditions, while encoder-decoder networks are identifiable when the decoder widths do not grow too rapidly. Our proofs are constructive and center on a connection between deep PNNs and low-rank tensor decompositions, and Kruskal-type uniqueness theorems. This yields both generic conditions determined by the architecture, and effective conditions that depend on the network's parameters. We also settle an open conjecture on the expected dimension of PNN's neurovarieties, and provide new bounds on the activation degrees required for it to reach its maximum.
Authors: Shruti Sadanand Dongare, Amun Kharel, Jonathan Samuel, Xiaona Zhou
Abstract: This paper introduces TransDreamerV3, a reinforcement learning model that enhances the DreamerV3 architecture by integrating a transformer encoder. The model is designed to improve memory and decision-making capabilities in complex environments. We conducted experiments on Atari-Boxing, Atari-Freeway, Atari-Pong, and Crafter tasks, where TransDreamerV3 demonstrated improved performance over DreamerV3, particularly in the Atari-Freeway and Crafter tasks. While issues in the Minecraft task and limited training across all tasks were noted, TransDreamerV3 displays advancement in world model-based reinforcement learning, leveraging transformer architectures.
Authors: Shoubin Yu, Yue Zhang, Ziyang Wang, Jaehong Yoon, Mohit Bansal
Abstract: Combining pre-trained expert models offers substantial potential for scalable multimodal reasoning, but building a unified framework remains challenging due to the increasing diversity of input modalities and task complexity. For instance, medical diagnosis requires precise reasoning over structured clinical tables, while financial forecasting depends on interpreting plot-based data to make informed predictions. To tackle this challenge, we introduce MEXA, a training-free framework that performs modality- and task-aware aggregation of multiple expert models to enable effective multimodal reasoning across diverse and distinct domains. MEXA dynamically selects expert models based on the input modality and the task-specific reasoning demands (i.e., skills). Each expert model, specialized in a modality task pair, generates interpretable textual reasoning outputs. MEXA then aggregates and reasons over these outputs using a Large Reasoning Model (LRM) to produce the final answer. This modular design allows flexible and transparent multimodal reasoning across diverse domains without additional training overhead. We extensively evaluate our approach on diverse multimodal benchmarks, including Video Reasoning, Audio Reasoning, 3D Understanding, and Medical QA. MEXA consistently delivers performance improvements over strong multimodal baselines, highlighting the effectiveness and broad applicability of our expert-driven selection and aggregation in diverse multimodal reasoning tasks.
Authors: Botao Zhu, Xianbin Wang
Abstract: Trust is emerging as an effective tool to ensure the successful completion of collaborative tasks within collaborative systems. However, rapidly and continuously evaluating the trustworthiness of collaborators during task execution is a significant challenge due to distributed devices, complex operational environments, and dynamically changing resources. To tackle this challenge, this paper proposes a Siamese-enabled rapid and continuous trust evaluation framework (SRCTE) to facilitate effective task collaboration. First, the communication and computing resource attributes of the collaborator in a trusted state, along with historical collaboration data, are collected and represented using an attributed control flow graph (ACFG) that captures trust-related semantic information and serves as a reference for comparison with data collected during task execution. At each time slot of task execution, the collaborator's communication and computing resource attributes, as well as task completion effectiveness, are collected in real time and represented with an ACFG to convey their trust-related semantic information. A Siamese model, consisting of two shared-parameter Structure2vec networks, is then employed to learn the deep semantics of each pair of ACFGs and generate their embeddings. Finally, the similarity between the embeddings of each pair of ACFGs is calculated to determine the collaborator's trust value at each time slot. A real system is built using two Dell EMC 5200 servers and a Google Pixel 8 to test the effectiveness of the proposed SRCTE framework. Experimental results demonstrate that SRCTE converges rapidly with only a small amount of data and achieves a high anomaly trust detection rate compared to the baseline algorithm.
Authors: Josu\'e Mart\'inez-Mart\'inez, Olivia Brown, Mostafa Karami, Sheida Nabavi
Abstract: Deep neural networks are increasingly being used to detect and diagnose medical conditions using medical imaging. Despite their utility, these models are highly vulnerable to adversarial attacks and distribution shifts, which can affect diagnostic reliability and undermine trust among healthcare professionals. In this study, we propose a robust training algorithm with data augmentation (RTDA) to mitigate these vulnerabilities in medical image classification. We benchmark classifier robustness against adversarial perturbations and natural variations of RTDA and six competing baseline techniques, including adversarial training and data augmentation approaches in isolation and combination, using experimental data sets with three different imaging technologies (mammograms, X-rays, and ultrasound). We demonstrate that RTDA achieves superior robustness against adversarial attacks and improved generalization performance in the presence of distribution shift in each image classification task while maintaining high clean accuracy.
Authors: Michael Plainer, Hao Wu, Leon Klein, Stephan G\"unnemann, Frank No\'e
Abstract: Diffusion models have recently gained significant attention due to their effectiveness in various scientific domains, including biochemistry. When trained on equilibrium molecular distributions, diffusion models provide both: a generative procedure to sample equilibrium conformations and associated forces derived from the model's scores. However, using the forces for coarse-grained molecular dynamics simulations uncovers inconsistencies in the samples generated via classical diffusion inference and simulation, despite both originating from the same model. Particularly at the small diffusion timesteps required for simulations, diffusion models fail to satisfy the Fokker-Planck equation, which governs how the score should evolve over time. We interpret this deviation as an indication of the observed inconsistencies and propose an energy-based diffusion model with a Fokker-Planck-derived regularization term enforcing consistency. We demonstrate the effectiveness of our approach on toy systems, alanine dipeptide, and introduce a state-of-the-art transferable Boltzmann emulator for dipeptides that supports simulation and demonstrates enhanced consistency and efficient sampling.
Authors: David Jacob Drexlin, Jonas Dippel, Julius Hense, Niklas Preni{\ss}l, Gr\'egoire Montavon, Frederick Klauschen, Klaus-Robert M\"uller
Abstract: Deep learning models have made significant advances in histological prediction tasks in recent years. However, for adaptation in clinical practice, their lack of robustness to varying conditions such as staining, scanner, hospital, and demographics is still a limiting factor: if trained on overrepresented subpopulations, models regularly struggle with less frequent patterns, leading to shortcut learning and biased predictions. Large-scale foundation models have not fully eliminated this issue. Therefore, we propose a novel approach explicitly modeling such metadata into a Metadata-guided generative Diffusion model framework (MeDi). MeDi allows for a targeted augmentation of underrepresented subpopulations with synthetic data, which balances limited training data and mitigates biases in downstream models. We experimentally show that MeDi generates high-quality histopathology images for unseen subpopulations in TCGA, boosts the overall fidelity of the generated images, and enables improvements in performance for downstream classifiers on datasets with subpopulation shifts. Our work is a proof-of-concept towards better mitigating data biases with generative models.
Authors: Ritabrata Chakraborty, Rajatsubhra Chakraborty, Avijit Dasgupta, Sandeep Chaurasia
Abstract: Traditional video-based tasks like soccer action spotting rely heavily on visual inputs, often requiring complex and computationally expensive models to process dense video data. In this work, we propose a shift from this video-centric approach to a text-based task, making it lightweight and scalable by utilizing Large Language Models (LLMs) instead of Vision-Language Models (VLMs). We posit that expert commentary, which provides rich, fine-grained descriptions and contextual cues such as excitement and tactical insights, contains enough information to reliably spot key actions in a match. To demonstrate this, we use the SoccerNet Echoes dataset, which provides timestamped commentary, and employ a system of three LLMs acting as judges specializing in outcome, excitement, and tactics. Each LLM evaluates sliding windows of commentary to identify actions like goals, cards, and substitutions, generating accurate timestamps for these events. Our experiments show that this language-centric approach performs effectively in detecting critical match events, providing a lightweight and training-free alternative to traditional video-based methods for action spotting.
Authors: Samin Yeasar Arnob, Scott Fujimoto, Doina Precup
Abstract: In this paper, we investigate the use of small datasets in the context of offline reinforcement learning (RL). While many common offline RL benchmarks employ datasets with over a million data points, many offline RL applications rely on considerably smaller datasets. We show that offline RL algorithms can overfit on small datasets, resulting in poor performance. To address this challenge, we introduce "Sparse-Reg": a regularization technique based on sparsity to mitigate overfitting in offline reinforcement learning, enabling effective learning in limited data settings and outperforming state-of-the-art baselines in continuous control.
Authors: Mahin Montasir Afif, Abdullah Al Noman, K. M. Tahsin Kabir, Md. Mortuza Ahmmed, Md. Mostafizur Rahman, Mufti Mahmud, Md. Ashraful Babu
Abstract: Generative Adversarial Networks (GAN) have shown potential in expanding limited medical imaging datasets. This study explores how different ratios of GAN-generated and real brain tumor MRI images impact the performance of a CNN in classifying healthy vs. tumorous scans. A DCGAN was used to create synthetic images which were mixed with real ones at various ratios to train a custom CNN. The CNN was then evaluated on a separate real-world test set. Our results indicate that the model maintains high sensitivity and precision in tumor classification, even when trained predominantly on synthetic data. When only a small portion of GAN data was added, such as 900 real images and 100 GAN images, the model achieved excellent performance, with test accuracy reaching 95.2%, and precision, recall, and F1-score all exceeding 95%. However, as the proportion of GAN images increased further, performance gradually declined. This study suggests that while GANs are useful for augmenting limited datasets especially when real data is scarce, too much synthetic data can introduce artifacts that affect the model's ability to generalize to real world cases.
Authors: Denis Larionov, Nikolay Bazenkov, Mikhail Kiselev
Abstract: This study investigates columnar-organized spiking neural networks (SNNs) for continual learning and catastrophic forgetting. Using CoLaNET (Columnar Layered Network), we show that microcolumns adapt most efficiently to new tasks when they lack shared structure with prior learning. We demonstrate how CoLaNET hyperparameters govern the trade-off between retaining old knowledge (stability) and acquiring new information (plasticity). Our optimal configuration learns ten sequential MNIST tasks effectively, maintaining 92% accuracy on each. It shows low forgetting, with only 4% performance degradation on the first task after training on nine subsequent tasks.
Authors: Yuchen Li, Hengyi Cai, Rui Kong, Xinran Chen, Jiamin Chen, Jun Yang, Haojie Zhang, Jiayi Li, Jiayi Wu, Yiqun Chen, Changle Qu, Keyi Kong, Wenwen Ye, Lixin Su, Xinyu Ma, Long Xia, Daiting Shi, Jiashu Zhao, Haoyi Xiong, Shuaiqiang Wang, Dawei Yin
Abstract: In this paper, we introduce the AI Search Paradigm, a comprehensive blueprint for next-generation search systems capable of emulating human information processing and decision-making. The paradigm employs a modular architecture of four LLM-powered agents (Master, Planner, Executor and Writer) that dynamically adapt to the full spectrum of information needs, from simple factual queries to complex multi-stage reasoning tasks. These agents collaborate dynamically through coordinated workflows to evaluate query complexity, decompose problems into executable plans, and orchestrate tool usage, task execution, and content synthesis. We systematically present key methodologies for realizing this paradigm, including task planning and tool integration, execution strategies, aligned and robust retrieval-augmented generation, and efficient LLM inference, spanning both algorithmic techniques and infrastructure-level optimizations. By providing an in-depth guide to these foundational components, this work aims to inform the development of trustworthy, adaptive, and scalable AI search systems.
Authors: Israel Ju\'arez-Jim\'enez, Tiffany Guadalupe Mart\'inez Paredes, Jes\'us Garc\'ia-Ram\'irez, Eric Ramos Aguilar
Abstract: Emotion recognition from facial images is a crucial task in human-computer interaction, enabling machines to learn human emotions through facial expressions. Previous studies have shown that facial images can be used to train deep learning models; however, most of these studies do not include a through dataset analysis. Visualizing facial landmarks can be challenging when extracting meaningful dataset insights; to address this issue, we propose facial landmark box plots, a visualization technique designed to identify outliers in facial datasets. Additionally, we compare two sets of facial landmark features: (i) the landmarks' absolute positions and (ii) their displacements from a neutral expression to the peak of an emotional expression. Our results indicate that a neural network achieves better performance than a random forest classifier.
Authors: Guozheng Ma, Lu Li, Zilin Wang, Li Shen, Pierre-Luc Bacon, Dacheng Tao
Abstract: Effectively scaling up deep reinforcement learning models has proven notoriously difficult due to network pathologies during training, motivating various targeted interventions such as periodic reset and architectural advances such as layer normalization. Instead of pursuing more complex modifications, we show that introducing static network sparsity alone can unlock further scaling potential beyond their dense counterparts with state-of-the-art architectures. This is achieved through simple one-shot random pruning, where a predetermined percentage of network weights are randomly removed once before training. Our analysis reveals that, in contrast to naively scaling up dense DRL networks, such sparse networks achieve both higher parameter efficiency for network expressivity and stronger resistance to optimization challenges like plasticity loss and gradient interference. We further extend our evaluation to visual and streaming RL scenarios, demonstrating the consistent benefits of network sparsity.
Authors: Matias Martinez, Xavier Franch
Abstract: The rapid progress in Automated Program Repair (APR) has been driven by advances in AI, particularly large language models (LLMs) and agent-based systems. SWE-Bench is a recent benchmark designed to evaluate LLM-based repair systems using real issues and pull requests mined from 12 popular open-source Python repositories. Its public leaderboards, SWE-Bench Lite and SWE-Bench Verified, have become central platforms for tracking progress and comparing solutions. However, because the submission process does not require detailed documentation, the architectural design and origin of many solutions remain unclear. In this paper, we present the first comprehensive study of all submissions to the SWE-Bench Lite (68 entries) and Verified (79 entries) leaderboards, analyzing 67 unique approaches across dimensions such as submitter type, product availability, LLM usage, and system architecture. Our findings reveal the dominance of proprietary LLMs (especially Claude 3.5/3.7), the presence of both agentic and non-agentic designs, and a contributor base spanning from individual developers to large tech companies.
Authors: Tianjiao Yu, Vedant Shah, Muntasir Wahed, Ying Shen, Kiet A. Nguyen, Ismini Lourentzou
Abstract: Articulated objects are common in the real world, yet modeling their structure and motion remains a challenging task for 3D reconstruction methods. In this work, we introduce Part$^{2}$GS, a novel framework for modeling articulated digital twins of multi-part objects with high-fidelity geometry and physically consistent articulation. Part$^{2}$GS leverages a part-aware 3D Gaussian representation that encodes articulated components with learnable attributes, enabling structured, disentangled transformations that preserve high-fidelity geometry. To ensure physically consistent motion, we propose a motion-aware canonical representation guided by physics-based constraints, including contact enforcement, velocity consistency, and vector-field alignment. Furthermore, we introduce a field of repel points to prevent part collisions and maintain stable articulation paths, significantly improving motion coherence over baselines. Extensive evaluations on both synthetic and real-world datasets show that Part$^{2}$GS consistently outperforms state-of-the-art methods by up to 10$\times$ in Chamfer Distance for movable parts.
Authors: Xiuyu Yang, Shuhan Tan, Philipp Kr\"ahenb\"uhl
Abstract: An ideal traffic simulator replicates the realistic long-term point-to-point trip that a self-driving system experiences during deployment. Prior models and benchmarks focus on closed-loop motion simulation for initial agents in a scene. This is problematic for long-term simulation. Agents enter and exit the scene as the ego vehicle enters new regions. We propose InfGen, a unified next-token prediction model that performs interleaved closed-loop motion simulation and scene generation. InfGen automatically switches between closed-loop motion simulation and scene generation mode. It enables stable long-term rollout simulation. InfGen performs at the state-of-the-art in short-term (9s) traffic simulation, and significantly outperforms all other methods in long-term (30s) simulation. The code and model of InfGen will be released at https://orangesodahub.github.io/InfGen
Authors: Zeyuan Yang, Xueyang Yu, Delin Chen, Maohao Shen, Chuang Gan
Abstract: Vision-language models (VLMs) excel at multimodal understanding, yet their text-only decoding forces them to verbalize visual reasoning, limiting performance on tasks that demand visual imagination. Recent attempts train VLMs to render explicit images, but the heavy image-generation pre-training often hinders the reasoning ability. Inspired by the way humans reason with mental imagery-the internal construction and manipulation of visual cues-we investigate whether VLMs can reason through interleaved multimodal trajectories without producing explicit images. To this end, we present a Machine Mental Imagery framework, dubbed as Mirage, which augments VLM decoding with latent visual tokens alongside ordinary text. Concretely, whenever the model chooses to ``think visually'', it recasts its hidden states as next tokens, thereby continuing a multimodal trajectory without generating pixel-level images. Begin by supervising the latent tokens through distillation from ground-truth image embeddings, we then switch to text-only supervision to make the latent trajectory align tightly with the task objective. A subsequent reinforcement learning stage further enhances the multimodal reasoning capability. Experiments on diverse benchmarks demonstrate that Mirage unlocks stronger multimodal reasoning without explicit image generation.
Authors: Yanzhi Zhang, Zhaoxi Zhang, Haoxiang Guan, Yilin Cheng, Yitong Duan, Chen Wang, Yue Wang, Shuxin Zheng, Jiyan He
Abstract: Reinforcement learning has emerged as a powerful paradigm for post-training large language models (LLMs) to improve reasoning. Approaches like Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) have shown strong results, but they require extensive external supervision. We investigate an alternative class of methods, Reinforcement Learning from Internal Feedback (RLIF), which relies solely on intrinsic model-derived signals instead of external rewards. In particular, we leverage unsupervised reward proxies such as token-level entropy, trajectory-level entropy, and self-certainty. Our theoretical analysis shows these internal objectives are partially equivalent, and we empirically evaluate various RLIF strategies on challenging math reasoning benchmarks. Experimental results demonstrate that RLIF can boost the reasoning performance of base LLMs at the beginning phase of the training, matching or surpassing RLVR techniques on these tasks. However, when training progresses, performance degrades even below the model before training. Moreover, we find that RLIF yields little improvement for instruction-tuned models, indicating diminishing returns of intrinsic feedback once an LLM is already instruction-tuned. We further analyze this limitation by mixing model weights and explain the reason of RLIF's training behaviors, providing practical guidelines for integrating internal feedback signals into LLM training. We hope our analysis of internal feedback will inform more principled and effective strategies for LLM post-training.
Authors: Deepti Raghavan, Keshav Santhanam, Muhammad Shahir Rahman, Nayani Modugula, Luis Gaspar Schroeder, Maximilien Cura, Houjun Liu, Pratiksha Thaker, Philip Levis, Matei Zaharia
Abstract: Compound AI applications chain together subcomponents such as generative language models, document retrievers, and embedding models. Applying traditional systems optimizations such as parallelism and pipelining in compound AI systems is difficult because each component has different constraints in terms of the granularity and type of data that it ingests. New data is often generated during intermediate computations, and text streams may be split into smaller, independent fragments (such as documents to sentences) which may then be re-aggregated at later parts of the computation. Due to this complexity, existing systems to serve compound AI queries do not fully take advantage of parallelism and pipelining opportunities. We present Alto, a framework that automatically optimizes execution of compound AI queries through streaming and parallelism. Bento introduces a new abstraction called nested ancestry, a metadata hierarchy that allows the system to correctly track partial outputs and aggregate data across the heterogeneous constraints of the components of compound AI applications. This metadata is automatically inferred from the programming model, allowing developers to express complex dataflow patterns without needing to reason manually about the details of routing and aggregation. Implementations of four applications in Alto outperform or match implementations in LangGraph, a popular existing AI programming framework. Alto implementations match or improve latency by between 10-30%.
Authors: Shijing Hu, Zhihui Lu, Xin Xu, Ruijun Deng, Xin Du, Qiang Duan
Abstract: Embodied intelligence (EI) enables manufacturing systems to flexibly perceive, reason, adapt, and operate within dynamic shop floor environments. In smart manufacturing, a representative EI scenario is robotic visual inspection, where industrial robots must accurately inspect components on rapidly changing, heterogeneous production lines. This task requires both high inference accuracy especially for uncommon defects and low latency to match production speeds, despite evolving lighting, part geometries, and surface conditions. To meet these needs, we propose LAECIPS, a large vision model-assisted adaptive edge-cloud collaboration framework for IoT-based embodied intelligence systems. LAECIPS decouples large vision models in the cloud from lightweight models on the edge, enabling plug-and-play model adaptation and continual learning. Through a hard input mining-based inference strategy, LAECIPS routes complex and uncertain inspection cases to the cloud while handling routine tasks at the edge, achieving both high accuracy and low latency. Experiments conducted on a real-world robotic semantic segmentation system for visual inspection demonstrate significant improvements in accuracy, processing latency, and communication overhead compared to state-of-the-art methods. LAECIPS provides a practical and scalable foundation for embodied intelligence in smart manufacturing, especially in adaptive robotic inspection and quality control scenarios.
Authors: Mustafa F Abdelwahed, Joan Espasa, Alice Toniolo, Ian P. Gent
Abstract: Diverse planning approaches are utilised in real-world applications like risk management, automated streamed data analysis, and malware detection. The current diverse planning formulations encode the diversity model as a distance function, which is computational inexpensive when comparing two plans. However, such modelling approach limits what can be encoded as measure of diversity, as well as the ability to explain why two plans are different. This paper introduces a novel approach to the diverse planning problem, allowing for more expressive modelling of diversity using a n-dimensional grid representation, where each dimension corresponds to a user-defined feature. Furthermore, we present a novel toolkit that generates diverse plans based on such customisable diversity models, called \emph{Behaviour Planning}. We provide an implementation for behaviour planning using planning-as-satisfiability. An empirical evaluation of our implementation shows that behaviour planning significantly outperforms the current diverse planning method in generating diverse plans measured on our new customisable diversity models. Our implementation is the first diverse planning approach to support planning categories beyond classical planning, such as over-subscription and numerical planning.
Authors: Simon Werner, Katharina Christ, Laura Bernardy, Marion G. M\"uller, Achim Rettinger
Abstract: Aligning machine learning systems with human expectations is mostly attempted by training with manually vetted human behavioral samples, typically explicit feedback. This is done on a population level since the context that is capturing the subjective Point-Of-View (POV) of a concrete person in a specific situational context is not retained in the data. However, we argue that alignment on an individual level can boost the subjective predictive performance for the individual user interacting with the system considerably. Since perception differs for each person, the same situation is observed differently. Consequently, the basis for decision making and the subsequent reasoning processes and observable reactions differ. We hypothesize that individual perception patterns can be used for improving the alignment on an individual level. We test this, by integrating perception information into machine learning systems and measuring their predictive performance wrt.~individual subjective assessments. For our empirical study, we collect a novel data set of multimodal stimuli and corresponding eye tracking sequences for the novel task of Perception-Guided Crossmodal Entailment and tackle it with our Perception-Guided Multimodal Transformer. Our findings suggest that exploiting individual perception signals for the machine learning of subjective human assessments provides a valuable cue for individual alignment. It does not only improve the overall predictive performance from the point-of-view of the individual user but might also contribute to steering AI systems towards every person's individual expectations and values.
Authors: Kaitao Chen, Mianxin Liu, Fang Yan, Lei Ma, Xiaoming Shi, Lilong Wang, Xiaosong Wang, Lifeng Zhu, Zhe Wang, Mu Zhou, Shaoting Zhang
Abstract: The advent of vision-language models fosters the interactive conversations between AI-enabled models and humans. Yet applying these models into clinics must deal with daunting challenges around large-scale training data, financial, and computational resources. Here we propose a cost-effective instruction learning framework for conversational pathology named as CLOVER. CLOVER only trains a lightweight module and uses instruction tuning while freezing the parameters of the large language model. Instead of using costly GPT-4, we propose well-designed prompts on GPT-3.5 for building generation-based instructions, emphasizing the utility of pathological knowledge derived from the Internet source. To augment the use of instructions, we construct a high-quality set of template-based instructions in the context of digital pathology. From two benchmark datasets, our findings reveal the strength of hybrid-form instructions in the visual question-answer in pathology. Extensive results show the cost-effectiveness of CLOVER in answering both open-ended and closed-ended questions, where CLOVER outperforms strong baselines that possess 37 times more training parameters and use instruction data generated from GPT-4. Through the instruction tuning, CLOVER exhibits robustness of few-shot learning in the external clinical dataset. These findings demonstrate that cost-effective modeling of CLOVER could accelerate the adoption of rapid conversational applications in the landscape of digital pathology.
Authors: Takuya Ito, Murray Campbell, Lior Horesh, Tim Klinger, Parikshit Ram
Abstract: The rapid development of artificial intelligence (AI) systems has created an urgent need for their scientific quantification. While their fluency across a variety of domains is impressive, AI systems fall short on tests requiring algorithmic reasoning -- a glaring limitation given the necessity for interpretable and reliable technology. Despite a surge of reasoning benchmarks emerging from the academic community, no theoretical framework exists to quantify algorithmic reasoning in AI systems. Here, we adopt a framework from computational complexity theory to quantify algorithmic generalization using algebraic expressions: algebraic circuit complexity. Algebraic circuit complexity theory -- the study of algebraic expressions as circuit models -- is a natural framework to study the complexity of algorithmic computation. Algebraic circuit complexity enables the study of generalization by defining benchmarks in terms of the computational requirements to solve a problem. Moreover, algebraic circuits are generic mathematical objects; an arbitrarily large number of samples can be generated for a specified circuit, making it an ideal experimental sandbox for the data-hungry models that are used today. In this Perspective, we adopt tools from algebraic circuit complexity, apply them to formalize a science of algorithmic generalization, and address key challenges for its successful application to AI science.
Authors: M\'elanie Roschewitz, Raghav Mehta, Charles Jones, Ben Glocker
Abstract: Shifts in data distribution can substantially harm the performance of clinical AI models and lead to misdiagnosis. Hence, various methods have been developed to detect the presence of such shifts at deployment time. However, the root causes of dataset shifts are diverse, and the choice of shift mitigation strategies is highly dependent on the precise type of shift encountered at test time. As such, detecting test-time dataset shift is not sufficient: precisely identifying which type of shift has occurred is critical. In this work, we propose the first unsupervised dataset shift identification framework for imaging datasets, effectively distinguishing between prevalence shift (caused by a change in the label distribution), covariate shift (caused by a change in input characteristics) and mixed shifts (simultaneous prevalence and covariate shifts). We discuss the importance of self-supervised encoders for detecting subtle covariate shifts and propose a novel shift detector leveraging both self-supervised encoders and task model outputs for improved shift detection. We show the effectiveness of the proposed shift identification framework across three different imaging modalities (chest radiography, digital mammography, and retinal fundus images) on five types of real-world dataset shifts using five large publicly available datasets.
Authors: Lixian Jing, Jianpeng Qi, Junyu Dong, Yanwei Yu
Abstract: As deep neural networks (DNNs) are increasingly deployed on edge devices, optimizing models for constrained computational resources is critical. Existing auto-pruning methods face challenges due to the diversity of DNN models, various operators (e.g., filters), and the difficulty in balancing pruning granularity with model accuracy. To address these limitations, we introduce AutoSculpt, a pattern-based automated pruning framework designed to enhance efficiency and accuracy by leveraging graph learning and deep reinforcement learning (DRL). AutoSculpt automatically identifies and prunes regular patterns within DNN architectures that can be recognized by existing inference engines, enabling runtime acceleration. Three key steps in AutoSculpt include: (1) Constructing DNNs as graphs to encode their topology and parameter dependencies, (2) embedding computationally efficient pruning patterns, and (3) utilizing DRL to iteratively refine auto-pruning strategies until the optimal balance between compression and accuracy is achieved. Experimental results demonstrate the effectiveness of AutoSculpt across various architectures, including ResNet, MobileNet, VGG, and Vision Transformer, achieving pruning rates of up to 90% and nearly 18% improvement in FLOPs reduction, outperforming all baselines. The codes can be available at https://anonymous.4open.science/r/AutoSculpt-DDA0
Authors: Haokun Zhao, Jinyi Han, Jiaqing Liang, Yanghua Xiao, Xiaojun Meng, Jiansheng Wei
Abstract: Large Language Models (LLMs) have achieved significant advancements, but the increasing complexity of tasks and higher performance demands highlight the need for continuous improvement. Some approaches utilize synthetic data generated by advanced LLMs based on evaluation results to train models. However, conventional evaluation methods fail to provide detailed, fine-grained profiles of LLMs, limiting their guidance for data synthesis. In this paper, we introduce the Cognitive Diagnostic Synthesis (CDS) method, which incorporates a diagnostic process inspired by Cognitive Diagnosis Theory (CDT) to refine evaluation results and characterize model profiles at the knowledge component level. Based on these diagnostics, we propose two diagnosis-synthesis strategies for weakness-targeted data synthesis. Additionally, we present an enhanced data augmentation and selection pipeline to improve the quality and diversity of synthesized data. Our experiments with several open-source models show significant improvements across multiple benchmarks, achieving up to 6.00% improvement in code generation, 13.10% in mathematical reasoning, and 5.43% in academic exams. Code and data are available on GitHub.
Authors: Alan Chan, Kevin Wei, Sihao Huang, Nitarshan Rajkumar, Elija Perrier, Seth Lazar, Gillian K. Hadfield, Markus Anderljung
Abstract: AI agents plan and execute interactions in open-ended environments. For example, OpenAI's Operator can use a web browser to do product comparisons and buy online goods. Much research on making agents useful and safe focuses on directly modifying their behaviour, such as by training them to follow user instructions. Direct behavioural modifications are useful, but do not fully address how heterogeneous agents will interact with each other and other actors. Rather, we will need external protocols and systems to shape such interactions. For instance, agents will need more efficient protocols to communicate with each other and form agreements. Attributing an agent's actions to a particular human or other legal entity can help to establish trust, and also disincentivize misuse. Given this motivation, we propose the concept of \textbf{agent infrastructure}: technical systems and shared protocols external to agents that are designed to mediate and influence their interactions with and impacts on their environments. Just as the Internet relies on protocols like HTTPS, our work argues that agent infrastructure will be similarly indispensable to ecosystems of agents. We identify three functions for agent infrastructure: 1) attributing actions, properties, and other information to specific agents, their users, or other actors; 2) shaping agents' interactions; and 3) detecting and remedying harmful actions from agents. We provide an incomplete catalog of research directions for such functions. For each direction, we include analysis of use cases, infrastructure adoption, relationships to existing (internet) infrastructure, limitations, and open questions. Making progress on agent infrastructure can prepare society for the adoption of more advanced agents.
Authors: Zeyu Gan, Yun Liao, Yong Liu
Abstract: Test-time scaling, which is also often referred to as slow-thinking, has been demonstrated to enhance multi-step reasoning in large language models (LLMs). However, despite its widespread utilization, the mechanisms underlying slow-thinking methods remain poorly understood. This paper explores the mechanisms of external slow-thinking from a theoretical standpoint. We begin by examining the snowball error effect within the LLM reasoning process and connect it to the likelihood of correct reasoning using information theory. Building on this, we show that external slow-thinking methods can be interpreted as strategies to mitigate the error probability. We further provide a comparative analysis of popular external slow-thinking approaches, ranging from simple to complex, highlighting their differences and interrelationships. Our findings suggest that the efficacy of these methods is not primarily determined by the specific framework employed, and that expanding the search scope or the model's internal reasoning capacity may yield more sustained improvements in the long term. We open-source our code at https://github.com/ZyGan1999/Snowball-Errors-and-Probability.
URLs: https://github.com/ZyGan1999/Snowball-Errors-and-Probability.
Authors: Yingce Xia, Peiran Jin, Shufang Xie, Liang He, Chuan Cao, Renqian Luo, Guoqing Liu, Yue Wang, Zequn Liu, Yuan-Jyue Chen, Zekun Guo, Yeqi Bai, Pan Deng, Yaosen Min, Ziheng Lu, Hongxia Hao, Han Yang, Jielan Li, Chang Liu, Jia Zhang, Jianwei Zhu, Ran Bi, Kehan Wu, Wei Zhang, Kaiyuan Gao, Qizhi Pei, Qian Wang, Xixian Liu, Yanting Li, Houtian Zhu, Yeqing Lu, Mingqian Ma, Zun Wang, Tian Xie, Krzysztof Maziarz, Marwin Segler, Zhao Yang, Zilong Chen, Yu Shi, Shuxin Zheng, Lijun Wu, Chen Hu, Peggy Dai, Tie-Yan Liu, Haiguang Liu, Tao Qin
Abstract: Foundation models have revolutionized natural language processing and artificial intelligence, significantly enhancing how machines comprehend and generate human languages. Inspired by the success of these foundation models, researchers have developed foundation models for individual scientific domains, including small molecules, materials, proteins, DNA, RNA and even cells. However, these models are typically trained in isolation, lacking the ability to integrate across different scientific domains. Recognizing that entities within these domains can all be represented as sequences, which together form the "language of nature", we introduce Nature Language Model (NatureLM), a sequence-based science foundation model designed for scientific discovery. Pre-trained with data from multiple scientific domains, NatureLM offers a unified, versatile model that enables various applications including: (i) generating and optimizing small molecules, proteins, RNA, and materials using text instructions; (ii) cross-domain generation/design, such as protein-to-molecule and protein-to-RNA generation; and (iii) top performance across different domains, matching or surpassing state-of-the-art specialist models. NatureLM offers a promising generalist approach for various scientific tasks, including drug discovery (hit generation/optimization, ADMET optimization, synthesis), novel material design, and the development of therapeutic proteins or nucleotides. We have developed NatureLM models in different sizes (1 billion, 8 billion, and 46.7 billion parameters) and observed a clear improvement in performance as the model size increases.
Authors: Hui Wang, Xufeng Zhang, Chaoxu Mu
Abstract: Heuristics have achieved great success in solving combinatorial optimization problems~(COPs). However, heuristics designed by humans require too much domain knowledge and testing time. Since Large Language Models~(LLMs) possess strong capabilities to understand and generate content with a knowledge base that covers various domains, they offer potential ways to automatically optimize heuristics. To this end, we propose Planning of Heuristics~(PoH), an optimization method that integrates LLM self-reflection with Monte Carlo Tree Search, a well-known planning algorithm. PoH iteratively refines generated heuristics by evaluating their performance and providing improvement suggestions. Our method enables to iteratively evaluate the generated heuristics~(states) and improve them based on the improvement suggestions~(actions) and evaluation results~(rewards), by effectively simulating future states to search for paths with higher rewards. In this paper, we apply PoH to solve the Traveling Salesman Problem and the Flow Shop Scheduling Problem. The experimental results show that PoH outperforms hand-crafted heuristics and other Automatic Heuristic Design methods based on LLMs, and achieves the state-of-the-art performance in automating heuristic optimization with LLMs to solve tested COPs, especially with large sizes.
Authors: Ilana Shapiro, Ruanqianqian Huang, Zachary Novack, Cheng-i Wang, Hao-Wen Dong, Taylor Berg-Kirkpatrick, Shlomo Dubnov, Sorin Lerner
Abstract: Western music is an innately hierarchical system of interacting levels of structure, from fine-grained melody to high-level form. In order to analyze music compositions holistically and at multiple granularities, we propose a unified, hierarchical meta-representation of musical structure called the structural temporal graph (STG). For a single piece, the STG is a data structure that defines a hierarchy of progressively finer structural musical features and the temporal relationships between them. We use the STG to enable a novel approach for deriving a representative structural summary of a music corpus, which we formalize as a nested NP-hard combinatorial optimization problem extending the Generalized Median Graph problem. Our approach first applies simulated annealing to develop a measure of structural distance between two music pieces rooted in graph isomorphism. Our approach then combines the formal guarantees of SMT solvers with nested simulated annealing over structural distances to produce a structurally sound, representative centroid STG for an entire corpus of STGs from individual pieces. To evaluate our approach, we conduct experiments verifying that structural distance accurately differentiates between music pieces, and that derived centroids accurately structurally characterize their corpora.
Authors: Alberto Coffrini, Paolo Barsocchi, Francesco Furfari, Antonino Crivello, Alessio Ferrari
Abstract: Indoor navigation presents unique challenges due to complex layouts and the unavailability of GNSS signals. Existing solutions often struggle with contextual adaptation, and typically require dedicated hardware. In this work, we explore the potential of a Large Language Model (LLM), i.e., ChatGPT, to generate natural, context-aware navigation instructions from indoor map images. We design and evaluate test cases across different real-world environments, analyzing the effectiveness of LLMs in interpreting spatial layouts, handling user constraints, and planning efficient routes. Our findings demonstrate the potential of LLMs for supporting personalized indoor navigation, with an average of 86.59% correct indications and a maximum of 97.14%. The proposed system achieves high accuracy and reasoning performance. These results have key implications for AI-driven navigation and assistive technologies.
Authors: Kexian Tang, Junyao Gao, Yanhong Zeng, Haodong Duan, Yanan Sun, Zhening Xing, Wenran Liu, Kaifeng Lyu, Kai Chen
Abstract: Multi-step spatial reasoning entails understanding and reasoning about spatial relationships across multiple sequential steps, which is crucial for tackling complex real-world applications, such as robotic manipulation, autonomous navigation, and automated assembly. To assess how well current Multimodal Large Language Models (MLLMs) have acquired this fundamental capability, we introduce LEGO-Puzzles, a scalable benchmark designed to evaluate both spatial understanding and sequential reasoning in MLLMs through LEGO-based tasks. LEGO-Puzzles consists of 1,100 carefully curated visual question-answering (VQA) samples spanning 11 distinct tasks, ranging from basic spatial understanding to complex multi-step reasoning. Based on LEGO-Puzzles, we conduct a comprehensive evaluation of 20 state-of-the-art MLLMs and uncover significant limitations in their spatial reasoning capabilities: even the most powerful MLLMs can answer only about half of the test cases, whereas human participants achieve over 90% accuracy. Furthermore, based on LEGO-Puzzles, we design generation tasks to investigate whether MLLMs can transfer their spatial understanding and reasoning abilities to image generation. Our experiments show that only GPT-4o and Gemini-2.0-Flash exhibit a limited ability to follow these instructions, while other MLLMs either replicate the input image or generate completely irrelevant outputs. Overall, LEGO-Puzzles exposes critical deficiencies in existing MLLMs' spatial understanding and sequential reasoning capabilities, and underscores the need for further advancements in multimodal spatial reasoning.
Authors: MingShan Liu, Jialing Fang
Abstract: Large language models (LLMs) have demonstrated strong mathematical reasoning capabilities but remain susceptible to hallucinations producing plausible yet incorrect statements especially in theorem proving, symbolic manipulation, and numerical computation. While self-consistency (SC) has been explored as a means to improve factuality in LLMs, existing approaches primarily apply SC to final-answer selection, neglecting the logical consistency of intermediate reasoning steps. In this work, we introduce a structured self-consistency framework designed to enhance the reliability of mathematical reasoning. Our method enforces self-consistency across intermediate steps and final outputs, reducing logical inconsistencies and hallucinations. We evaluate our approach across three core mathematical tasks: theorem proving, symbolic transformation, and numerical computation. Experimental results demonstrate that SC significantly improves proof validity, symbolic reasoning accuracy, and numerical stability while maintaining computational efficiency. Further analysis reveals that structured self-consistency not only enhances problem-solving accuracy but also reduces the variance of model-generated outputs. These findings highlight self-consistency as a robust mechanism for improving mathematical reasoning in LLMs, paving the way for more reliable and interpretable AI-driven mathematics.
Authors: Xiaojun Yuan, Haoming Ma, Yinuo Huang, Zhoufan Hua, Yong Zuo, Zhi Ding
Abstract: Semantic communications leverage artificial intelligence (AI) technologies to extract semantic information for efficient data delivery, thereby significantly reducing communication cost. With the evolution towards artificial general intelligence (AGI), the increasing demands for AGI services pose new challenges to semantic communications. In this context, an AGI application is typically defined on a general-sense task, covering a broad, even unforeseen, set of objectives, as well as driven by the need for a human-friendly interface in forms (e.g., videos, images, or text) easily understood by human users.In response, we introduce an AGI-driven communication paradigm for supporting AGI applications, called generative semantic communication (GSC). We first describe the basic concept of GSC and its difference from existing semantic communications, and then introduce a general framework of GSC based on advanced AI technologies including foundation models and generative models. Two case studies are presented to verify the advantages of GSC. Finally, open challenges and new research directions are discussed to stimulate this line of research and pave the way for practical applications.
Authors: Ning Wang, Zihan Yan, Weiyang Li, Chuan Ma, He Chen, Tao Xiang
Abstract: Embodied agents exhibit immense potential across a multitude of domains, making the assurance of their behavioral safety a fundamental prerequisite for their widespread deployment. However, existing research predominantly concentrates on the security of general large language models, lacking specialized methodologies for establishing safety benchmarks and input moderation tailored to embodied agents. To bridge this gap, this paper introduces a novel input moderation framework, meticulously designed to safeguard embodied agents. This framework encompasses the entire pipeline, including taxonomy definition, dataset curation, moderator architecture, model training, and rigorous evaluation. Notably, we introduce EAsafetyBench, a meticulously crafted safety benchmark engineered to facilitate both the training and stringent assessment of moderators specifically designed for embodied agents. Furthermore, we propose Pinpoint, an innovative prompt-decoupled input moderation scheme that harnesses a masked attention mechanism to effectively isolate and mitigate the influence of functional prompts on moderation tasks. Extensive experiments conducted on diverse benchmark datasets and models validate the feasibility and efficacy of the proposed approach. The results demonstrate that our methodologies achieve an impressive average detection accuracy of 94.58%, surpassing the performance of existing state-of-the-art techniques, alongside an exceptional moderation processing time of merely 0.002 seconds per instance.
Authors: Anita Srbinovska, Angela Srbinovska, Vivek Senthil, Adrian Martin, John McCluskey, Jonathan Bateman, Ernest Fokou\'e
Abstract: This paper proposes a novel interdisciplinary framework for analyzing police body-worn camera (BWC) footage from the Rochester Police Department (RPD) using advanced artificial intelligence (AI) and statistical machine learning (ML) techniques. Our goal is to detect, classify, and analyze patterns of interaction between police officers and civilians to identify key behavioral dynamics, such as respect, disrespect, escalation, and de-escalation. We apply multimodal data analysis by integrating image, audio, and natural language processing (NLP) techniques to extract meaningful insights from BWC footage. The framework incorporates speaker separation, transcription, and large language models (LLMs) to produce structured, interpretable summaries of police-civilian encounters. We also employ a custom evaluation pipeline to assess transcription quality and behavior detection accuracy in high-stakes, real-world policing scenarios. Our methodology, computational techniques, and findings outline a practical approach for law enforcement review, training, and accountability processes while advancing the frontiers of knowledge discovery from complex police BWC data.
Authors: Maris F. L. Galesloot, Roman Andriushchenko, Milan \v{C}e\v{s}ka, Sebastian Junges, Nils Jansen
Abstract: Partially observable Markov decision processes (POMDPs) model specific environments in sequential decision-making under uncertainty. Critically, optimal policies for POMDPs may not be robust against perturbations in the environment. Hidden-model POMDPs (HM-POMDPs) capture sets of different environment models, that is, POMDPs with a shared action and observation space. The intuition is that the true model is hidden among a set of potential models, and it is unknown which model will be the environment at execution time. A policy is robust for a given HM-POMDP if it achieves sufficient performance for each of its POMDPs.We compute such robust policies by combining two orthogonal techniques: (1) a deductive formal verification technique that supports tractable robust policy evaluation by computing a worst-case POMDP within the HM-POMDP, and (2) subgradient ascent to optimize the candidate policy for a worst-case POMDP. The empirical evaluation shows that, compared to various baselines, our approach (1) produces policies that are more robust and generalize better to unseen POMDPs, and (2) scales to HM-POMDPs that consist of over a hundred thousand environments.
Authors: Qi Li, Kun Li, Haozhi Han, Honghui Shang, Xinfu He, Yunquan Zhang, Hong An, Ting Cao, Mao Yang
Abstract: Can a scientific simulation system be physically consistent, interpretable by design, and scalable across regimes--all at once? Despite decades of progress, this trifecta remains elusive. Classical methods like Kinetic Monte Carlo ensure thermodynamic accuracy but scale poorly; learning-based methods offer efficiency but often sacrifice physical consistency and interpretability. We present SwarmThinkers, a reinforcement learning framework that recasts atomic-scale simulation as a physically grounded swarm intelligence system. Each diffusing particle is modeled as a local decision-making agent that selects transitions via a shared policy network trained under thermodynamic constraints. A reweighting mechanism fuses learned preferences with transition rates, preserving statistical fidelity while enabling interpretable, step-wise decision making. Training follows a centralized-training, decentralized-execution paradigm, allowing the policy to generalize across system sizes, concentrations, and temperatures without retraining. On a benchmark simulating radiation-induced Fe-Cu alloy precipitation, SwarmThinkers is the first system to achieve full-scale, physically consistent simulation on a single A100 GPU, previously attainable only via OpenKMC on a supercomputer. It delivers up to 4963x (3185x on average) faster computation with 485x lower memory usage. By treating particles as decision-makers, not passive samplers, SwarmThinkers marks a paradigm shift in scientific simulation--one that unifies physical consistency, interpretability, and scalability through agent-driven intelligence.
Authors: Jiahao Qiu, Fulian Xiao, Yimin Wang, Yuchen Mao, Yijia Chen, Xinzhe Juan, Shu Zhang, Siran Wang, Xuan Qi, Tongcheng Zhang, Zixin Yao, Jiacheng Guo, Yifu Lu, Charles Argon, Jundi Cui, Daixin Chen, Junran Zhou, Shuyao Zhou, Zhanpeng Zhou, Ling Yang, Shilong Liu, Hongru Wang, Kaixuan Huang, Xun Jiang, Yuming Cao, Yue Chen, Yunfei Chen, Zhengyi Chen, Ruowei Dai, Mengqiu Deng, Jiye Fu, Yunting Gu, Zijie Guan, Zirui Huang, Xiaoyan Ji, Yumeng Jiang, Delong Kong, Haolong Li, Jiaqi Li, Ruipeng Li, Tianze Li, Zhuoran Li, Haixia Lian, Mengyue Lin, Xudong Liu, Jiayi Lu, Jinghan Lu, Wanyu Luo, Ziyue Luo, Zihao Pu, Zhi Qiao, Ruihuan Ren, Liang Wan, Ruixiang Wang, Tianhui Wang, Yang Wang, Zeyu Wang, Zihua Wang, Yujia Wu, Zhaoyi Wu, Hao Xin, Weiao Xing, Ruojun Xiong, Weijie Xu, Yao Shu, Yao Xiao, Xiaorui Yang, Yuchen Yang, Nan Yi, Jiadong Yu, Yangyuxuan Yu, Huiting Zeng, Danni Zhang, Yunjie Zhang, Zhaoyu Zhang, Zhiheng Zhang, Xiaofeng Zheng, Peirong Zhou, Linyan Zhong, Xiaoyin Zong, Ying Zhao, Zhenxin Chen, Lin Ding, Xiaoyu Gao, Bingbing Gong, Yichao Li, Yang Liao, Guang Ma, Tianyuan Ma, Xinrui Sun, Tianyi Wang, Han Xia, Ruobing Xian, Gen Ye, Tengfei Yu, Wentao Zhang, Yuxi Wang, Xi Gao, Mengdi Wang
Abstract: Recent advances in large language models (LLMs) have led to remarkable progress across domains, yet their capabilities in the humanities, particularly history, remain underexplored. Historical reasoning poses unique challenges for AI, involving multimodal source interpretation, temporal inference, and cross-linguistic analysis. While general-purpose agents perform well on many existing benchmarks, they lack the domain-specific expertise required to engage with historical materials and questions. To address this gap, we introduce HistBench, a new benchmark of 414 high-quality questions designed to evaluate AI's capacity for historical reasoning and authored by more than 40 expert contributors. The tasks span a wide range of historical problems-from factual retrieval based on primary sources to interpretive analysis of manuscripts and images, to interdisciplinary challenges involving archaeology, linguistics, or cultural history. Furthermore, the benchmark dataset spans 29 ancient and modern languages and covers a wide range of historical periods and world regions. Finding the poor performance of LLMs and other agents on HistBench, we further present HistAgent, a history-specific agent equipped with carefully designed tools for OCR, translation, archival search, and image understanding in History. On HistBench, HistAgent based on GPT-4o achieves an accuracy of 27.54% pass@1 and 36.47% pass@2, significantly outperforming LLMs with online search and generalist agents, including GPT-4o (18.60%), DeepSeek-R1(14.49%) and Open Deep Research-smolagents(20.29% pass@1 and 25.12% pass@2). These results highlight the limitations of existing LLMs and generalist agents and demonstrate the advantages of HistAgent for historical reasoning.
Authors: Yongjin Yang, Euiin Yi, Jongwoo Ko, Kimin Lee, Zhijing Jin, Se-Young Yun
Abstract: The remarkable growth in large language model (LLM) capabilities has spurred exploration into multi-agent systems, with debate frameworks emerging as a promising avenue for enhanced problem-solving. These multi-agent debate (MAD) approaches, where agents collaboratively present, critique, and refine arguments, potentially offer improved reasoning, robustness, and diverse perspectives over monolithic models. Despite prior studies leveraging MAD, a systematic understanding of its effectiveness compared to self-agent methods, particularly under varying conditions, remains elusive. This paper seeks to fill this gap by conceptualizing MAD as a test-time computational scaling technique, distinguished by collaborative refinement and diverse exploration capabilities. We conduct a comprehensive empirical investigation comparing MAD with strong self-agent test-time scaling baselines on mathematical reasoning and safety-related tasks. Our study systematically examines the influence of task difficulty, model scale, and agent diversity on MAD's performance. Key findings reveal that, for mathematical reasoning, MAD offers limited advantages over self-agent scaling but becomes more effective with increased problem difficulty and decreased model capability, while agent diversity shows little benefit. Conversely, for safety tasks, MAD's collaborative refinement can increase vulnerability, but incorporating diverse agent configurations facilitates a gradual reduction in attack success through the collaborative refinement process. We believe our findings provide critical guidance for the future development of more effective and strategically deployed MAD systems.
Authors: Jingyi Yang, Shuai Shao, Dongrui Liu, Jing Shao
Abstract: With the rapid development of multimodal large language models (MLLMs), they are increasingly deployed as autonomous computer-use agents capable of accomplishing complex computer tasks. However, a pressing issue arises: Can the safety risk principles designed and aligned for general MLLMs in dialogue scenarios be effectively transferred to real-world computer-use scenarios? Existing research on evaluating the safety risks of MLLM-based computer-use agents suffers from several limitations: it either lacks realistic interactive environments, or narrowly focuses on one or a few specific risk types. These limitations ignore the complexity, variability, and diversity of real-world environments, thereby restricting comprehensive risk evaluation for computer-use agents. To this end, we introduce \textbf{RiOSWorld}, a benchmark designed to evaluate the potential risks of MLLM-based agents during real-world computer manipulations. Our benchmark includes 492 risky tasks spanning various computer applications, involving web, social media, multimedia, os, email, and office software. We categorize these risks into two major classes based on their risk source: (i) User-originated risks and (ii) Environmental risks. For the evaluation, we evaluate safety risks from two perspectives: (i) Risk goal intention and (ii) Risk goal completion. Extensive experiments with multimodal agents on \textbf{RiOSWorld} demonstrate that current computer-use agents confront significant safety risks in real-world scenarios. Our findings highlight the necessity and urgency of safety alignment for computer-use agents in real-world computer manipulation, providing valuable insights for developing trustworthy computer-use agents. Our benchmark is publicly available at https://yjyddq.github.io/RiOSWorld.github.io/.
Authors: Yongjin Yang, Sinjae Kang, Juyong Lee, Dongjun Lee, Se-Young Yun, Kimin Lee
Abstract: Training large language model (LLM) agents to acquire necessary skills and perform diverse tasks within an environment is gaining interest as a means to enable open-endedness. However, creating the training dataset for their skill acquisition faces several challenges. Manual trajectory collection requires significant human effort. Another approach, where LLMs directly propose tasks to learn, is often invalid, as the LLMs lack knowledge of which tasks are actually feasible. Moreover, the generated data may not provide a meaningful learning signal, as agents often already perform well on the proposed tasks. To address this, we propose a novel automatic skill discovery framework EXIF for LLM-powered agents, designed to improve the feasibility of generated target behaviors while accounting for the agents' capabilities. Our method adopts an exploration-first strategy by employing an exploration agent (Alice) to train the target agent (Bob) to learn essential skills in the environment. Specifically, Alice first interacts with the environment to retrospectively generate a feasible, environment-grounded skill dataset, which is then used to train Bob. Crucially, we incorporate an iterative feedback loop, where Alice evaluates Bob's performance to identify areas for improvement. This feedback then guides Alice's next round of exploration, forming a closed-loop data generation process. Experiments on Webshop and Crafter demonstrate EXIF's ability to effectively discover meaningful skills and iteratively expand the capabilities of the trained agent without any human intervention, achieving substantial performance improvements. Interestingly, we observe that setting Alice to the same model as Bob also notably improves performance, demonstrating EXIF's potential for building a self-evolving system.
Authors: Qiyao Wei, Samuel Holt, Jing Yang, Markus Wulfmeier, Mihaela van der Schaar
Abstract: Peer review, the bedrock of scientific advancement in machine learning (ML), is strained by a crisis of scale. Exponential growth in manuscript submissions to premier ML venues such as NeurIPS, ICML, and ICLR is outpacing the finite capacity of qualified reviewers, leading to concerns about review quality, consistency, and reviewer fatigue. This position paper argues that AI-assisted peer review must become an urgent research and infrastructure priority. We advocate for a comprehensive AI-augmented ecosystem, leveraging Large Language Models (LLMs) not as replacements for human judgment, but as sophisticated collaborators for authors, reviewers, and Area Chairs (ACs). We propose specific roles for AI in enhancing factual verification, guiding reviewer performance, assisting authors in quality improvement, and supporting ACs in decision-making. Crucially, we contend that the development of such systems hinges on access to more granular, structured, and ethically-sourced peer review process data. We outline a research agenda, including illustrative experiments, to develop and validate these AI assistants, and discuss significant technical and ethical challenges. We call upon the ML community to proactively build this AI-assisted future, ensuring the continued integrity and scalability of scientific validation, while maintaining high standards of peer review.
Authors: Mingfeng Fan, Jianan Zhou, Yifeng Zhang, Yaoxin Wu, Jinbiao Chen, Guillaume Adrien Sartoretti
Abstract: Recent deep reinforcement learning methods have achieved remarkable success in solving multi-objective combinatorial optimization problems (MOCOPs) by decomposing them into multiple subproblems, each associated with a specific weight vector. However, these methods typically treat all subproblems equally and solve them using a single model, hindering the effective exploration of the solution space and thus leading to suboptimal performance. To overcome the limitation, we propose POCCO, a novel plug-and-play framework that enables adaptive selection of model structures for subproblems, which are subsequently optimized based on preference signals rather than explicit reward values. Specifically, we design a conditional computation block that routes subproblems to specialized neural architectures. Moreover, we propose a preference-driven optimization algorithm that learns pairwise preferences between winning and losing solutions. We evaluate the efficacy and versatility of POCCO by applying it to two state-of-the-art neural methods for MOCOPs. Experimental results across four classic MOCOP benchmarks demonstrate its significant superiority and strong generalization.
Authors: Xialie Zhuang, Peixian Ma, Zhikai Jia, Shiwei Liu, Zheng Cao
Abstract: The ongoing evolution of language models has led to the development of large-scale architectures that demonstrate exceptional performance across a wide range of tasks. However, these models come with significant computational and energy demands, as well as potential privacy implications. In this context, Small Reasoning Language Models (SRLMs) with approximately 0.5 billion parameters present a compelling alternative due to their remarkable computational efficiency and cost effectiveness, particularly in resource-constrained environments. Despite these advantages, the limited capacity of 0.5 billion parameter models poses challenges in handling complex tasks such as mathematical reasoning and code generation. This research investigates various training strategies, including supervised fine-tuning (SFT), knowledge distillation (KD), and reinforcement learning (RL), as well as their hybrid implementations, to enhance the performance of 0.5B SRLMs. We analyze effective methodologies to bridge the performance gap between SRLMS and larger models and present insights into optimal training pipelines tailored for these smaller architectures. Through extensive experimental validation and analysis, our work aims to provide actionable recommendations for maximizing the reasoning capabilities of 0.5B models.
Authors: Yining Hong, Rui Sun, Bingxuan Li, Xingcheng Yao, Maxine Wu, Alexander Chien, Da Yin, Ying Nian Wu, Zhecan James Wang, Kai-Wei Chang
Abstract: AI agents today are mostly siloed - they either retrieve and reason over vast amount of digital information and knowledge obtained online; or interact with the physical world through embodied perception, planning and action - but rarely both. This separation limits their ability to solve tasks that require integrated physical and digital intelligence, such as cooking from online recipes, navigating with dynamic map data, or interpreting real-world landmarks using web knowledge. We introduce Embodied Web Agents, a novel paradigm for AI agents that fluidly bridge embodiment and web-scale reasoning. To operationalize this concept, we first develop the Embodied Web Agents task environments, a unified simulation platform that tightly integrates realistic 3D indoor and outdoor environments with functional web interfaces. Building upon this platform, we construct and release the Embodied Web Agents Benchmark, which encompasses a diverse suite of tasks including cooking, navigation, shopping, tourism, and geolocation - all requiring coordinated reasoning across physical and digital realms for systematic assessment of cross-domain intelligence. Experimental results reveal significant performance gaps between state-of-the-art AI systems and human capabilities, establishing both challenges and opportunities at the intersection of embodied cognition and web-scale knowledge access. All datasets, codes and websites are publicly available at our project page https://embodied-web-agent.github.io/.
Authors: Yiwen Ding, Jiarui Liu, Zhiheng Lyu, Kun Zhang, Bernhard Schoelkopf, Zhijing Jin, Rada Mihalcea
Abstract: While several previous studies have analyzed gender bias in research, we are still missing a comprehensive analysis of gender differences in the AI community, covering diverse topics and different development trends. Using the AI Scholar dataset of 78K researchers in the field of AI, we identify several gender differences: (1) Although female researchers tend to have fewer overall citations than males, this citation difference does not hold for all academic-age groups; (2) There exist large gender homophily in co-authorship on AI papers; (3) Female first-authored papers show distinct linguistic styles, such as longer text, more positive emotion words, and more catchy titles than male first-authored papers. Our analysis provides a window into the current demographic trends in our AI community, and encourages more gender equality and diversity in the future. Our code and data are at https://github.com/causalNLP/ai-scholar-gender.
Authors: Leander Weber, Jim Berend, Moritz Weckbecker, Alexander Binder, Thomas Wiegand, Wojciech Samek, Sebastian Lapuschkin
Abstract: Gradient-based optimization has been a cornerstone of machine learning that enabled the vast advances of Artificial Intelligence (AI) development over the past decades. However, this type of optimization requires differentiation, and with recent evidence of the benefits of non-differentiable (e.g. neuromorphic) architectures over classical models w.r.t. efficiency, such constraints can become limiting in the future. We present Layer-wise Feedback Propagation (LFP), a novel training principle for neural network-like predictors that utilizes methods from the domain of explainability to decompose a reward to individual neurons based on their respective contributions. Leveraging these neuron-wise rewards, our method then implements a greedy approach reinforcing helpful parts of the network and weakening harmful ones. While having comparable computational complexity to gradient descent, LFP does not require gradient computation and generates sparse and thereby memory- and energy-efficient parameter updates and models. We establish the convergence of LFP theoretically and empirically, demonstrating its effectiveness on various models and datasets. Via two applications - neural network pruning and the approximation-free training of Spiking Neural Networks (SNNs) - we demonstrate that LFP combines increased efficiency in terms of computation and representation with flexibility w.r.t. choice of model architecture and objective function. Our code is available at https://github.com/leanderweber/layerwise-feedback-propagation.
URLs: https://github.com/leanderweber/layerwise-feedback-propagation.
Authors: Qizhou Wang, Guansong Pang, Mahsa Salehi, Xiaokun Xia, Christopher Leckie
Abstract: This paper considers an important Graph Anomaly Detection (GAD) task, namely open-set GAD, which aims to train a detection model using a small number of normal and anomaly nodes (referred to as seen anomalies) to detect both seen anomalies and unseen anomalies (i.e., anomalies that cannot be illustrated the training anomalies). Those labelled training data provide crucial prior knowledge about abnormalities for GAD models, enabling substantially reduced detection errors. However, current supervised GAD methods tend to over-emphasise fitting the seen anomalies, leading to many errors of detecting the unseen anomalies as normal nodes. Further, existing open-set AD models were introduced to handle Euclidean data, failing to effectively capture discriminative features from graph structure and node attributes for GAD. In this work, we propose a novel open-set GAD approach, namely normal structure regularisation (NSReg), to achieve generalised detection ability to unseen anomalies, while maintaining its effectiveness on detecting seen anomalies. The key idea in NSReg is to introduce a regularisation term that enforces the learning of compact, semantically-rich representations of normal nodes based on their structural relations to other nodes. When being optimised with supervised anomaly detection losses, the regularisation term helps incorporate strong normality into the modelling, and thus, it effectively avoids over-fitting the seen anomalies and learns a better normality decision boundary, largely reducing the false negatives of detecting unseen anomalies as normal. Extensive empirical results on seven real-world datasets show that NSReg significantly outperforms state-of-the-art competing methods by at least 14% AUC-ROC on the unseen anomaly classes and by 10% AUC-ROC on all anomaly classes.
Authors: Zifan Xu, Haozhu Wang, Dmitriy Bespalov, Xian Wu, Peter Stone, Yanjun Qi
Abstract: Chain-of-thought (CoT) prompting is a popular in-context learning (ICL) approach for large language models (LLMs), especially when tackling complex reasoning tasks. Traditional ICL approaches construct prompts using examples that contain questions similar to the input question. However, CoT prompting, which includes crucial intermediate reasoning steps (rationales) within its examples, necessitates selecting examples based on these rationales rather than the questions themselves. Existing methods require human experts or pre-trained LLMs to describe the skill, a high-level abstraction of rationales, to guide the selection. These methods, however, are often costly and difficult to scale. Instead, this paper introduces a new approach named Latent Reasoning Skills (LaRS) that employs unsupervised learning to create a latent space representation of rationales, with a latent variable called a reasoning skill. Concurrently, LaRS learns a reasoning policy to determine the required reasoning skill for a given question. Then the ICL examples are selected by aligning the reasoning skills between past examples and the question. This approach is theoretically grounded and compute-efficient, eliminating the need for auxiliary LLM inference or manual prompt design. Empirical results demonstrate that LaRS consistently outperforms SOTA skill-based selection methods, processing example banks four times faster, reducing LLM inferences during the selection stage by half, and showing greater robustness to sub-optimal example banks.
Authors: Masahiro Kato, Kyohei Okumura, Takuya Ishihara, Toru Kitagawa
Abstract: This study investigates the contextual best arm identification (BAI) problem, aiming to design an adaptive experiment to identify the best treatment arm conditioned on contextual information (covariates). We consider a decision-maker who assigns treatment arms to experimental units during an experiment and recommends the estimated best treatment arm based on the contexts at the end of the experiment. The decision-maker uses a policy for recommendations, which is a function that provides the estimated best treatment arm given the contexts. In our evaluation, we focus on the worst-case expected regret, a relative measure between the expected outcomes of an optimal policy and our proposed policy. We derive a lower bound for the expected simple regret and then propose a strategy called Adaptive Sampling-Policy Learning (PLAS). We prove that this strategy is minimax rate-optimal in the sense that its leading factor in the regret upper bound matches the lower bound as the number of experimental units increases.
Authors: Leuson Da Silva, Jordan Samhi, Foutse Khomh
Abstract: Since its release in November 2022, ChatGPT has shaken up Stack Overflow, the premier platform for developers queries on programming and software development. Demonstrating an ability to generate instant, human-like responses to technical questions, ChatGPT has ignited debates within the developer community about the evolving role of human-driven platforms in the age of generative AI. Two months after ChatGPT release, Meta released its answer with its own Large Language Model (LLM) called LLaMA: the race was on. We conducted an empirical study analyzing questions from Stack Overflow and using these LLMs to address them. This way, we aim to (i) quantify the reliability of LLMs answers and their potential to replace Stack Overflow in the long term; (ii) identify and understand why LLMs fail; (iii) measure users activity evolution with Stack Overflow over time; and (iv) compare LLMs together. Our empirical results are unequivocal: ChatGPT and LLaMA challenge human expertise, yet do not outperform it for some domains, while a significant decline in user posting activity has been observed. Furthermore, we also discuss the impact of our findings regarding the usage and development of new LLMs and provide guidelines for future challenges faced by users and researchers.
Authors: Siwei Yang, Bingchen Zhao, Cihang Xie
Abstract: This paper introduces AQA-Bench, a novel benchmark to assess the sequential reasoning capabilities of large language models (LLMs) in algorithmic contexts, such as depth-first search (DFS). The key feature of our evaluation benchmark lies in its interactive evaluation protocol - for example, in DFS, the availability of each node's connected edge is contingent upon the model's traversal to that node, thereby necessitating the LLM's ability to effectively remember visited nodes and strategize subsequent moves considering the possible environmental feedback in the future steps. We comprehensively build AQA-Bench with three different algorithms, namely binary search, depth-first search, and breadth-first search, and to evaluate the sequential reasoning ability of 14 different LLMs. Our investigations reveal several interesting findings: (1) Closed-source models like GPT-4 and Gemini generally show much stronger sequential reasoning ability, significantly outperforming open-source LLMs. (2) Naively providing in-context examples may inadvertently hurt few-shot performance in an interactive environment due to over-fitting to examples. (3) Instead of using optimal steps from another test case as the in-context example, a very limited number of predecessor steps in the current test case following the optimal policy can substantially boost small models' performance. (4) The performance gap between weak models and strong models is greatly due to the incapability of weak models to start well. (5) The scaling correlation between performance and model size is not always significant, sometimes even showcasing an inverse trend. We hope our study can catalyze future work on advancing the understanding and enhancement of LLMs' capabilities in sequential reasoning. The code is available at https://github.com/UCSC-VLAA/AQA-Bench.
Authors: Soikat Hasan Ahmed, Jan Finkbeiner, Emre Neftci
Abstract: Event cameras offer high temporal resolution and dynamic range with minimal motion blur, making them promising for robust object detection. While Spiking Neural Networks (SNNs) on neuromorphic hardware are often considered for energy-efficient and low latency event-based data processing, they often fall short of Artificial Neural Networks (ANNs) in accuracy and flexibility. Here, we introduce Attention-based Hybrid SNN-ANN backbones for event-based object detection to leverage the strengths of both SNN and ANN architectures. A novel Attention-based SNN-ANN bridge module captures sparse spatial and temporal relations from the SNN layer and converts them into dense feature maps for the ANN part of the backbone. Additionally, we present a variant that integrates DWConvL-STMs to the ANN blocks to capture slower dynamics. This multi-timescale network combines fast SNN processing for short timesteps with long-term dense RNN processing, effectively capturing both fast and slow dynamics. Experimental results demonstrate that our proposed method surpasses SNN-based approaches by significant margins, with results comparable to existing ANN and RNN-based methods. Unlike ANN-only networks, the hybrid setup allows us to implement the SNN blocks on digital neuromorphic hardware to investigate the feasibility of our approach. Extensive ablation studies and implementation on neuromorphic hardware confirm the effectiveness of our proposed modules and architectural choices. Our hybrid SNN-ANN architectures pave the way for ANN-like performance at a drastically reduced parameter, latency, and power budget.
Authors: Kyla H. Levin, Nicolas van Kempen, Emery D. Berger, Stephen N. Freund
Abstract: Debugging is a critical but challenging task for programmers. This paper proposes ChatDBG, an AI-powered debugging assistant. ChatDBG integrates large language models (LLMs) to significantly enhance the capabilities and user-friendliness of conventional debuggers. ChatDBG lets programmers engage in a collaborative dialogue with the debugger, allowing them to pose complex questions about program state, perform root cause analysis for crashes or assertion failures, and explore open-ended queries like "why is x null?". To handle these queries, ChatDBG grants the LLM autonomy to "take the wheel": it can act as an independent agent capable of querying and controlling the debugger to navigate through stacks and inspect program state. It then reports its findings and yields back control to the programmer. By leveraging the real-world knowledge embedded in LLMs, ChatDBG can diagnose issues identifiable only through the use of domain-specific reasoning. Our ChatDBG prototype integrates with standard debuggers including LLDB and GDB for native code and Pdb for Python. Our evaluation across a diverse set of code, including C/C++ code with known bugs and a suite of Python code including standalone scripts and Jupyter notebooks, demonstrates that ChatDBG can successfully analyze root causes, explain bugs, and generate accurate fixes for a wide range of real-world errors. For the Python programs, a single query led to an actionable bug fix 67% of the time; one additional follow-up query increased the success rate to 85%. ChatDBG has seen rapid uptake; it has already been downloaded more than 75,000 times.
Authors: Siya Qi, Lin Gui, Yulan He, Zheng Yuan
Abstract: The proliferation of Large Language Models (LLMs) has introduced a critical challenge: accurate hallucination evaluation that ensures model reliability. While Automatic Hallucination Evaluation (AHE) has emerged as essential, the field suffers from methodological fragmentation, hindering both theoretical understanding and practical advancement. This survey addresses this critical gap through a comprehensive analysis of 74 evaluation methods, revealing that 74% specifically target LLMs, a paradigm shift that demands new evaluation frameworks. We formulate a unified evaluation pipeline encompassing datasets and benchmarks, evidence collection strategies, and comparison mechanisms, systematically documenting the evolution from pre-LLM to post-LLM methodologies. Beyond taxonomical organization, we identify fundamental limitations in current approaches and their implications for real-world deployment. To guide future research, we delineate key challenges and propose strategic directions, including enhanced interpretability mechanisms and integration of application-specific evaluation criteria, ultimately providing a roadmap for developing more robust and practical hallucination evaluation systems.
Authors: Jun Huang, Yan Liu
Abstract: This paper proposes a new gradient-based XAI method called Guided AbsoluteGrad for saliency map explanations. We utilize both positive and negative gradient magnitudes and employ gradient variance to distinguish the important areas for noise deduction. We also introduce a novel evaluation metric named ReCover And Predict (RCAP), which considers the Localization and Visual Noise Level objectives of the explanations. We propose two propositions for these two objectives and prove the necessity of evaluating them. We evaluate Guided AbsoluteGrad with seven gradient-based XAI methods using the RCAP metric and other SOTA metrics in three case studies: (1) ImageNet dataset with ResNet50 model; (2) International Skin Imaging Collaboration (ISIC) dataset with EfficientNet model; (3) the Places365 dataset with DenseNet161 model. Our method surpasses other gradient-based approaches, showcasing the quality of enhanced saliency map explanations through gradient magnitude.
Authors: Shaina Raza, Mizanur Rahman, Michael R. Zhang
Abstract: Recent advancements in large language models (LLMs) have significantly improved natural language processing (NLP) applications. However, these models often inherit biases from their training data. While several datasets exist for bias detection, most are limited to one or two NLP tasks, typically classification or evaluation, and lack comprehensive coverage across a broader range of tasks. To address this gap, we introduce the Bias Evaluations Across Domains (BEADs) dataset, designed to support a wide range of NLP tasks, including text classification, token classification, bias quantification, and benign language generation. A key contribution of this work is the gold-standard annotation provided by GPT-4 for scalability, with expert verification to ensure high reliability. BEADs can be used for both fine-tuning models (for classification and generation tasks) and evaluating LLM behavior. Our findings show that BEADs effectively surfaces various biases during model fine-tuning and helps reduce biases in language generation tasks while maintaining output quality. The dataset also highlights prevalent demographic biases in LLMs during evaluation. We release BEADs as a practical resource for detecting and mitigating bias across domains, supporting the development of responsible AI systems. Project: https://vectorinstitute.github.io/BEAD/ Data: https://huggingface.co/datasets/shainar/BEAD
URLs: https://vectorinstitute.github.io/BEAD/, https://huggingface.co/datasets/shainar/BEAD
Authors: Kailas Dayanandan, Nikhil Kumar, Anand Sinha, Brejesh Lall
Abstract: The dual thinking framework considers fast, intuitive, and slower logical processing. The perception of dual thinking in vision requires images where inferences from intuitive and logical processing differ, and the latter is under-explored in current studies. We introduce a novel adversarial dataset to provide evidence for the dual thinking framework in human vision, which also facilitates the study of the qualitative behavior of deep learning models. Our psychophysical studies show the presence of multiple inferences in rapid succession, and analysis of errors shows that the early stopping of visual processing can result in missing relevant information. MLLMs (Multi-modal Large Language Models) and VLMs (Vision Language Models) have made significant progress in correcting errors in intuitive processing in human vision and showed enhanced performance on images requiring logical processing. However, their improvements in logical processing have not kept pace with their advancements in intuitive processing. In contrast, segmentation models exhibit errors similar to those seen in intuitive human processing and lack understanding of sub-structures, as indicated by errors related to sub-components in identified instances. As AI (Artificial Intelligence)-based systems find increasing applications in safety-critical domains like autonomous driving, the integration of logical processing capabilities becomes essential. This not only enhances performance but also addresses the limitations of scaling-based approaches while ensuring robustness and reliability in real-world environments.
Authors: Tuan-Luc Huynh, Thuy-Trang Vu, Weiqing Wang, Yinwei Wei, Trung Le, Dragan Gasevic, Yuan-Fang Li, Thanh-Toan Do
Abstract: Differentiable Search Index (DSI) utilizes pre-trained language models to perform indexing and document retrieval via end-to-end learning without relying on external indexes. However, DSI requires full re-training to index new documents, causing significant computational inefficiencies. Continual learning (CL) offers a solution by enabling the model to incrementally update without full re-training. Existing CL solutions in document retrieval rely on memory buffers or generative models for rehearsal, which is infeasible when accessing previous training data is restricted due to privacy concerns. To this end, we introduce PromptDSI, a prompt-based, rehearsal-free continual learning approach for document retrieval. PromptDSI follows the Prompt-based Continual Learning (PCL) framework, using learnable prompts to efficiently index new documents without accessing previous documents or queries. To improve retrieval latency, we remove the initial forward pass of PCL, which otherwise greatly increases training and inference time, with a negligible trade-off in performance. Additionally, we introduce a novel topic-aware prompt pool that employs neural topic embeddings as fixed keys, eliminating the instability of prompt key optimization while maintaining competitive performance with existing PCL prompt pools. In a challenging rehearsal-free continual learning setup, we demonstrate that PromptDSI variants outperform rehearsal-based baselines, match the strong cache-based baseline in mitigating forgetting, and significantly improving retrieval performance on new corpora.
Authors: Yunpeng Jiang, Yutong Ban, Paul Weng
Abstract: Data augmentation is widely applied and has shown its benefits in different machine learning tasks. However, as recently observed, it may have an unfair effect in multi-class classification. While data augmentation generally improves the overall performance (and therefore is beneficial for many classes), it can actually be detrimental for other classes, which can be problematic in some application domains. In this paper, to counteract this phenomenon, we propose CLAM, a CLAss-dependent Multiplicative-weights method. To derive it, we first formulate the training of a classifier as a non-linear optimization problem that aims at simultaneously maximizing the individual class performances and balancing them. By rewriting this optimization problem as an adversarial two-player game, we propose a novel multiplicative weight algorithm, for which we prove the convergence. Interestingly, our formulation also reveals that the class-dependent effects of data augmentation is not due to data augmentation only, but is in fact a general phenomenon. Our empirical results over six datasets demonstrate that the performance of learned classifiers is indeed more fairly distributed over classes, with only limited impact on the average accuracy.
Authors: Le Xue, Manli Shu, Anas Awadalla, Jun Wang, An Yan, Senthil Purushwalkam, Honglu Zhou, Viraj Prabhu, Yutong Dai, Michael S Ryoo, Shrikant Kendre, Jieyu Zhang, Shaoyen Tseng, Gustavo A Lujan-Moreno, Matthew L Olson, Musashi Hinck, David Cobbley, Vasudev Lal, Can Qin, Shu Zhang, Chia-Chih Chen, Ning Yu, Juntao Tan, Tulika Manoj Awalgaonkar, Shelby Heinecke, Huan Wang, Yejin Choi, Ludwig Schmidt, Zeyuan Chen, Silvio Savarese, Juan Carlos Niebles, Caiming Xiong, Ran Xu
Abstract: This paper introduces BLIP-3, an open framework for developing Large Multimodal Models (LMMs). The framework comprises meticulously curated datasets, a training recipe, model architectures, and a resulting suite of LMMs. We release 4B and 14B models, including both the pre-trained base model and the instruction fine-tuned ones. Our models undergo rigorous evaluation across a range of tasks, including both single and multi-image benchmarks. Our models demonstrate competitive performance among open-source LMMs with similar model sizes. Our resulting LMMs demonstrate competitive performance among open-source LMMs with similar model sizes, with the ability to comprehend interleaved image-text inputs. Our training code, models, and all datasets used in this work, including the three largescale datasets we create and the preprocessed ones, will be open-sourced to better support the research community.
Authors: Junwei You, Haotian Shi, Zhuoyu Jiang, Zilin Huang, Rui Gan, Keshu Wu, Xi Cheng, Xiaopeng Li, Bin Ran
Abstract: Vehicle-to-everything (V2X) cooperation has emerged as a promising paradigm to overcome the perception limitations of classical autonomous driving by leveraging information from both ego-vehicle and infrastructure sensors. However, effectively fusing heterogeneous visual and semantic information while ensuring robust trajectory planning remains a significant challenge. This paper introduces V2X-VLM, a novel end-to-end (E2E) cooperative autonomous driving framework based on vision-language models (VLMs). V2X-VLM integrates multiperspective camera views from vehicles and infrastructure with text-based scene descriptions to enable a more comprehensive understanding of driving environments. Specifically, we propose a contrastive learning-based mechanism to reinforce the alignment of heterogeneous visual and textual characteristics, which enhances the semantic understanding of complex driving scenarios, and employ a knowledge distillation strategy to stabilize training. Experiments on a large real-world dataset demonstrate that V2X-VLM achieves state-of-the-art trajectory planning accuracy, significantly reducing L2 error and collision rate compared to existing cooperative autonomous driving baselines. Ablation studies validate the contributions of each component. Moreover, the evaluation of robustness and efficiency highlights the practicality of V2X-VLM for real-world deployment to enhance overall autonomous driving safety and decision-making.
Authors: Nicolas Yax, Pierre-Yves Oudeyer, Stefano Palminteri
Abstract: In machine learning, contamination refers to situations where testing data leak into the training set. The issue is particularly relevant for the evaluation of the performance of Large Language Models (LLMs), which are generally trained on gargantuan, and generally opaque, corpora of text scraped from the world wide web. Developing tools to detect contamination is therefore crucial to be able to fairly and properly track the evolution of the performance of LLMs. To date, only a few recent studies have attempted to address the issue of quantifying and detecting contamination in short text sequences, such as those commonly found in benchmarks. However, these methods have limitations that can sometimes render them impractical. In the present paper, we introduce LogProber, a novel, efficient algorithm that we show to be able to detect contamination in a black box setting that tries to tackle some of these drawbacks by focusing on the familiarity with the question rather than the answer. Here, we explore the properties of the proposed method in comparison with concurrent approaches, identify its advantages and limitations, and illustrate how different forms of contamination can go undetected depending on the design of the detection algorithm.
Authors: Ziyan Cui, Ning Li, Huaikang Zhou
Abstract: Artificial Intelligence (AI) is increasingly being integrated into scientific research, particularly in the social sciences, where understanding human behavior is critical. Large Language Models (LLMs) have shown promise in replicating human-like responses in various psychological experiments. We conducted a large-scale study replicating 156 psychological experiments from top social science journals using three state-of-the-art LLMs (GPT-4, Claude 3.5 Sonnet, and DeepSeek v3). Our results reveal that while LLMs demonstrate high replication rates for main effects (73-81%) and moderate to strong success with interaction effects (46-63%), They consistently produce larger effect sizes than human studies, with Fisher Z values approximately 2-3 times higher than human studies. Notably, LLMs show significantly lower replication rates for studies involving socially sensitive topics such as race, gender and ethics. When original studies reported null findings, LLMs produced significant results at remarkably high rates (68-83%) - while this could reflect cleaner data with less noise, as evidenced by narrower confidence intervals, it also suggests potential risks of effect size overestimation. Our results demonstrate both the promise and challenges of LLMs in psychological research, offering efficient tools for pilot testing and rapid hypothesis validation while enriching rather than replacing traditional human subject studies, yet requiring more nuanced interpretation and human validation for complex social phenomena and culturally sensitive research questions.
Authors: Ting Liu, Zunnan Xu, Yue Hu, Liangtao Shi, Zhiqiang Wang, Quanjun Yin
Abstract: Referring Expression Comprehension (REC), which aims to ground a local visual region via natural language, is a task that heavily relies on multimodal alignment. Most existing methods utilize powerful pre-trained models to transfer visual/linguistic knowledge by full fine-tuning. However, full fine-tuning the entire backbone not only breaks the rich prior knowledge embedded in the pre-training, but also incurs significant computational costs. Motivated by the recent emergence of Parameter-Efficient Transfer Learning (PETL) methods, we aim to solve the REC task in an effective and efficient manner. Directly applying these PETL methods to the REC task is inappropriate, as they lack the specific-domain abilities for precise local visual perception and visual-language alignment. Therefore, we propose a novel framework of Multimodal Prior-guided Parameter Efficient Tuning, namely MaPPER. Specifically, MaPPER comprises Dynamic Prior Adapters guided by an aligned prior, and Local Convolution Adapters to extract precise local semantics for better visual perception. Moreover, the Prior-Guided Text module is proposed to further utilize the prior for facilitating the cross-modal alignment. Experimental results on three widely-used benchmarks demonstrate that MaPPER achieves the best accuracy compared to the full fine-tuning and other PETL methods with only 1.41% tunable backbone parameters. Our code is available at https://github.com/liuting20/MaPPER.
Authors: Nan Sun, Bo Mao, Yongchang Li, Di Guo, Huaping Liu
Abstract: Current service robots suffer from limited natural language communication abilities, heavy reliance on predefined commands, ongoing human intervention, and, most notably, a lack of proactive collaboration awareness in human-populated environments. This results in narrow applicability and low utility. In this paper, we introduce AssistantX, an LLM-powered proactive assistant designed for autonomous operation in realworld scenarios with high accuracy. AssistantX employs a multi-agent framework consisting of 4 specialized LLM agents, each dedicated to perception, planning, decision-making, and reflective review, facilitating advanced inference capabilities and comprehensive collaboration awareness, much like a human assistant by your side. We built a dataset of 210 real-world tasks to validate AssistantX, which includes instruction content and status information on whether relevant personnel are available. Extensive experiments were conducted in both text-based simulations and a real office environment over the course of a month and a half. Our experiments demonstrate the effectiveness of the proposed framework, showing that AssistantX can reactively respond to user instructions, actively adjust strategies to adapt to contingencies, and proactively seek assistance from humans to ensure successful task completion. More details and videos can be found at https://assistantx-agent.github.io/AssistantX/.
Authors: Yijiang Li, Qingying Gao, Tianwei Zhao, Bingyang Wang, Haoran Sun, Haiyun Lyu, Robert D. Hawkins, Nuno Vasconcelos, Tal Golan, Dezhi Luo, Hokin Deng
Abstract: While Multi-modal Large Language Models (MLLMs) demonstrate impressive abilities over high-level perception and reasoning, their robustness in the wild remains limited, often falling short on tasks that are intuitive and effortless for humans. We examine the hypothesis that these deficiencies stem from the absence of core knowledge--rudimentary cognitive abilities innate to humans from early childhood. To explore the core knowledge representation in MLLMs, we introduce CoreCognition, a large-scale benchmark encompassing 12 core knowledge concepts grounded in developmental cognitive science. We evaluate 230 models with 11 different prompts, leading to a total of 2,530 data points for analysis. Our experiments uncover four key findings, collectively demonstrating core knowledge deficits in MLLMs: they consistently underperform and show reduced, or even absent, scalability on low-level abilities relative to high-level ones. Finally, we propose Concept Hacking, a novel controlled evaluation method that reveals MLLMs fail to progress toward genuine core knowledge understanding, but instead rely on shortcut learning as they scale.
Authors: Yu-Neng Chuang, Prathusha Kameswara Sarma, Parikshit Gopalan, John Boccio, Sara Bolouki, Xia Hu, Helen Zhou
Abstract: Large language models (LLMs) have demonstrated impressive performance on several tasks and are increasingly deployed in real-world applications. However, especially in high-stakes settings, it becomes vital to know when the output of an LLM may be unreliable. Depending on whether an answer is trustworthy, a system can then choose to route the question to another expert, or otherwise fall back on a safe default behavior. In this work, we study the extent to which LLMs can reliably indicate confidence in their answers, and how this notion of confidence can translate into downstream accuracy gains. We propose Self-Reflection with Error-based Feedback (Self-REF), a lightweight training strategy to teach LLMs to express confidence in whether their answers are correct in a reliable manner. Self-REF introduces confidence tokens into the LLM, from which a confidence score can be extracted. Compared to conventional approaches such as verbalizing confidence and examining token probabilities, we demonstrate empirically that confidence tokens show significant improvements in downstream routing and rejection learning tasks.
Authors: Peter Shaw, James Cohan, Jacob Eisenstein, Kenton Lee, Jonathan Berant, Kristina Toutanova
Abstract: We propose a new programming language called ALTA and a compiler that can map ALTA programs to Transformer weights. ALTA is inspired by RASP, a language proposed by Weiss et al. (2021), and Tracr (Lindner et al., 2023), a compiler from RASP programs to Transformer weights. ALTA complements and extends this prior work, offering the ability to express loops and to compile programs to Universal Transformers, among other advantages. ALTA allows us to constructively show how Transformers can represent length-invariant algorithms for computing parity and addition, as well as a solution to the SCAN benchmark of compositional generalization tasks, without requiring intermediate scratchpad decoding steps. We also propose tools to analyze cases where the expressibility of an algorithm is established, but end-to-end training on a given training set fails to induce behavior consistent with the desired algorithm. To this end, we explore training from ALTA execution traces as a more fine-grained supervision signal. This enables additional experiments and theoretical analyses relating the learnability of various algorithms to data availability and modeling decisions, such as positional encodings. We make the ALTA framework -- language specification, symbolic interpreter, and weight compiler -- available to the community to enable further applications and insights.
Authors: Jiyao Wang, Xiao Yang, Zhenyu Wang, Ximeng Wei, Ange Wang, Dengbo He, Kaishun Wu
Abstract: Road safety remains a critical challenge worldwide, with approximately 1.35 million fatalities annually attributed to traffic accidents, often due to human errors. As we advance towards higher levels of vehicle automation, challenges still exist, as driving with automation can cognitively over-demand drivers if they engage in non-driving-related tasks (NDRTs), or lead to drowsiness if driving was the sole task. This calls for the urgent need for an effective Driver Monitoring System (DMS) that can evaluate cognitive load and drowsiness in SAE Level-2/3 autonomous driving contexts. In this study, we propose a novel multi-task DMS, termed VDMoE, which leverages RGB video input to monitor driver states non-invasively. By utilizing key facial features to minimize computational load and integrating remote Photoplethysmography (rPPG) for physiological insights, our approach enhances detection accuracy while maintaining efficiency. Additionally, we optimize the Mixture-of-Experts (MoE) framework to accommodate multi-modal inputs and improve performance across different tasks. A novel prior-inclusive regularization method is introduced to align model outputs with statistical priors, thus accelerating convergence and mitigating overfitting risks. We validate our method with the creation of a new dataset (MCDD), which comprises RGB video and physiological indicators from 42 participants, and two public datasets. Our findings demonstrate the effectiveness of VDMoE in monitoring driver states, contributing to safer autonomous driving systems. The code and data will be released.
Authors: Zeyuan Li, Yangfan He, Lewei He, Jianhui Wang, Tianyu Shi, Bin Lei, Yuchen Li, Qiuwu Chen
Abstract: Recently, large language models (LLMs) have achieved significant progress in automated code generation. Despite their strong instruction-following capabilities, these models frequently struggled to align with user intent in coding scenarios. In particular, they were hampered by datasets that lacked diversity and failed to address specialized tasks or edge cases. Furthermore, challenges in supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) led to failures in generating precise, human-intent-aligned code. To tackle these challenges and improve the code generation performance for automated programming systems, we propose Feedback-driven Adaptive Long/short-term memory reinforced Coding Optimization (i.e., FALCON). FALCON is structured into two hierarchical levels. From the global level, long-term memory improves code quality by retaining and applying learned knowledge. At the local level, short-term memory allows for the incorporation of immediate feedback from compilers and AI systems. Additionally, we introduce meta-reinforcement learning with feedback rewards to solve the global-local bi-level optimization problem and enhance the model's adaptability across diverse code generation tasks. Extensive experiments demonstrate that our technique achieves state-of-the-art performance, leading other reinforcement learning methods by more than 4.5 percentage points on the MBPP benchmark and 6.1 percentage points on the Humaneval benchmark. The open-sourced code is publicly available at https://github.com/titurte/FALCON.
Authors: Bohan Lyu, Yadi Cao, Duncan Watson-Parris, Leon Bergen, Taylor Berg-Kirkpatrick, Rose Yu
Abstract: Large Language Models (LLMs) demonstrate promising capabilities in solving scientific problems but often suffer from the issue of hallucination. While integrating LLMs with tools can mitigate this issue, models fine-tuned on tool usage become overreliant on them and incur unnecessary costs. Inspired by how human experts assess problem complexity before selecting solutions, we propose a novel two-component fine-tuning method, Adapting While Learning (AWL). In the first component, World Knowledge Learning (WKL), LLMs internalize scientific knowledge by learning from tool-generated solutions. In the second component, Tool Usage Adaptation (TUA), we categorize problems as easy or hard based on the model's accuracy, and train it to maintain direct reasoning for easy problems while switching to tools for hard ones. We validate our method on six scientific benchmark datasets across climate science, epidemiology, physics, and other domains. Compared to the original instruct model (8B), models post-trained with AWL achieve 29.11% higher answer accuracy and 12.72% better tool usage accuracy, even surpassing state-of-the-art models including GPT-4o and Claude-3.5 on four custom-created datasets. Our code is open-source at https://github.com/Rose-STL-Lab/Adapting-While-Learning.
URLs: https://github.com/Rose-STL-Lab/Adapting-While-Learning.
Authors: Guan Zhe Hong, Nishanth Dikkala, Enming Luo, Cyrus Rashtchian, Xin Wang, Rina Panigrahy
Abstract: Due to the size and complexity of modern large language models (LLMs), it has proven challenging to uncover the underlying mechanisms that models use to solve reasoning problems. For instance, is their reasoning for a specific problem localized to certain parts of the network? Do they break down the reasoning problem into modular components that are then executed as sequential steps as we go deeper in the model? To better understand the reasoning capability of LLMs, we study a minimal propositional logic problem that requires combining multiple facts to arrive at a solution. By studying this problem on Mistral and Gemma models, up to 27B parameters, we illuminate the core components the models use to solve such logic problems. From a mechanistic interpretability point of view, we use causal mediation analysis to uncover the pathways and components of the LLMs' reasoning processes. Then, we offer fine-grained insights into the functions of attention heads in different layers. We not only find a sparse circuit that computes the answer, but we decompose it into sub-circuits that have four distinct and modular uses. Finally, we reveal that three distinct models -- Mistral-7B, Gemma-2-9B and Gemma-2-27B -- contain analogous but not identical mechanisms.
Authors: Yingying Fang, Zihao Jin, Shaojie Guo, Jinda Liu, Zhiling Yue, Yijian Gao, Junzhi Ning, Zhi Li, Simon Walsh, Guang Yang
Abstract: Despite significant advancements in automated report generation, the opaqueness of text interpretability continues to cast doubt on the reliability of the content produced. This paper introduces a novel approach to identify specific image features in X-ray images that influence the outputs of report generation models. Specifically, we propose Cyclic Vision-Language Manipulator CVLM, a module to generate a manipulated X-ray from an original X-ray and its report from a designated report generator. The essence of CVLM is that cycling manipulated X-rays to the report generator produces altered reports aligned with the alterations pre-injected into the reports for X-ray generation, achieving the term "cyclic manipulation". This process allows direct comparison between original and manipulated X-rays, clarifying the critical image features driving changes in reports and enabling model users to assess the reliability of the generated texts. Empirical evaluations demonstrate that CVLM can identify more precise and reliable features compared to existing explanation methods, significantly enhancing the transparency and applicability of AI-generated reports.
Authors: Ashwin Nair, Zhen Rong Goh, Tianrui Liu, Abigail Yongping Huang
Abstract: Current metadata creation for web archives is time consuming and costly due to reliance on human effort. This paper explores the use of gpt-4o for metadata generation within the Web Archive Singapore, focusing on scalability, efficiency, and cost effectiveness. We processed 112 Web ARChive (WARC) files using data reduction techniques, achieving a notable 99.9% reduction in metadata generation costs. By prompt engineering, we generated titles and abstracts, which were evaluated both intrinsically using Levenshtein Distance and BERTScore, and extrinsically with human cataloguers using McNemar's test. Results indicate that while our method offers significant cost savings and efficiency gains, human curated metadata maintains an edge in quality. The study identifies key challenges including content inaccuracies, hallucinations, and translation issues, suggesting that Large Language Models (LLMs) should serve as complements rather than replacements for human cataloguers. Future work will focus on refining prompts, improving content filtering, and addressing privacy concerns through experimentation with smaller models. This research advances the integration of LLMs in web archiving, offering valuable insights into their current capabilities and outlining directions for future enhancements. The code is available at https://github.com/masamune-prog/warc2summary for further development and use by institutions facing similar challenges.
Authors: Alkis Kalavasis, Anay Mehrotra, Grigoris Velegkas
Abstract: Specifying all desirable properties of a language model is challenging, but certain requirements seem essential. Given samples from an unknown language, the trained model should produce valid strings not seen in training and be expressive enough to capture the language's full richness. Otherwise, outputting invalid strings constitutes "hallucination," and failing to capture the full range leads to "mode collapse." We ask if a language model can meet both requirements. We investigate this within a statistical language generation setting building on Gold and Angluin. Here, the model receives random samples from a distribution over an unknown language K, which belongs to a possibly infinite collection of languages. The goal is to generate unseen strings from K. We say the model generates from K with consistency and breadth if, as training size increases, its output converges to all unseen strings in K. Kleinberg and Mullainathan [KM24] asked if consistency and breadth in language generation are possible. We answer this negatively: for a large class of language models, including next-token prediction models, this is impossible for most collections of candidate languages. This contrasts with [KM24]'s result, showing consistent generation without breadth is possible for any countable collection of languages. Our finding highlights that generation with breadth fundamentally differs from generation without breadth. As a byproduct, we establish near-tight bounds on the number of samples needed for generation with or without breadth. Finally, our results offer hope: consistent generation with breadth is achievable for any countable collection of languages when negative examples (strings outside K) are available alongside positive ones. This suggests that post-training feedback, which encodes negative examples, can be crucial in reducing hallucinations while limiting mode collapse.
Authors: Xu Chen, Zida Cheng, Yuangang Pan, Shuai Xiao, Xiaoming Liu, Jinsong Lan, Xiaoyong Zhu, Bo Zheng, Ivor W. Tsang
Abstract: Existing click-through rate (CTR) prediction works have studied the role of feature interaction through a variety of techniques. Each interaction technique exhibits its own strength, and solely using one type usually constrains the model's capability to capture the complex feature relationships, especially for industrial data with enormous input feature fields. Recent research shows that effective CTR models often combine an MLP network with a dedicated feature interaction network in a two-parallel structure. However, the interplay and cooperative dynamics between different streams or branches remain under-researched. In this work, we introduce a novel Multi-Branch Cooperation Network (MBCnet) which enables multiple branch networks to collaborate with each other for better complex feature interaction modeling. Specifically, MBCnet consists of three branches: the Extensible Feature Grouping and Crossing (EFGC) branch that promotes the model's memorization ability of specific feature fields, the low rank Cross Net branch and Deep branch to enhance explicit and implicit feature crossing for improved generalization. Among these branches, a novel cooperation scheme is proposed based on two principles: Branch co-teaching and moderate differentiation. Branch co-teaching encourages well-learned branches to support poorly-learned ones on specific training samples. Moderate differentiation advocates branches to maintain a reasonable level of difference in their feature representations on the same inputs. This cooperation strategy improves learning through mutual knowledge sharing and boosts the discovery of diverse feature interactions across branches. Experiments on large-scale industrial datasets and online A/B test at Taobao app demonstrate MBCnet's superior performance, delivering a 0.09 point increase in CTR, 1.49% growth in deals, and 1.62% rise in GMV. Core codes are available online.
Authors: Yunkee Chae, Eunsik Shin, Suntae Hwang, Seungryeol Paik, Kyogu Lee
Abstract: Lyrics generation presents unique challenges, particularly in achieving precise syllable control while adhering to song form structures such as verses and choruses. Conventional line-by-line approaches often lead to unnatural phrasing, underscoring the need for more granular syllable management. We propose a framework for lyrics generation that enables multi-level syllable control at the word, phrase, line, and paragraph levels, aware of song form. Our approach generates complete lyrics conditioned on input text and song form, ensuring alignment with specified syllable constraints. Generated lyrics samples are available at: https://tinyurl.com/lyrics9999
Authors: Svetlana Churina, Kokil Jaidka
Abstract: The incivility prevalent on platforms like Twitter (now X) and Reddit poses a challenge for developing AI systems that can support productive and rhetorically sound political argumentation. In this study, we report experiments with GPT-3.5 Turbo, fine-tuned on two contrasting datasets of political discussions: high-variance, high-incivility Twitter replies to U.S. Congress, and low-variance, low-incivility posts from Reddit's r/ChangeMyView. We systematically evaluate how these data sources and prompting strategies shape the rhetorical framing and deliberative quality of model-generated arguments. Our results show that Reddit-finetuned models produce safer but rhetorically rigid arguments, while cross-platform fine-tuning amplifies toxicity. Prompting reduces specific toxic behaviors, such as personal attacks, but fails to fully mitigate the influence of high-incivility training data. We introduce and validate a rhetorical evaluation rubric and provide practical guidelines for deploying LLMs in content authoring, moderation, and deliberation support.
Authors: Taneesh Gupta, Rahul Madhavan, Xuchao Zhang, Nagarajan Natarajan, Chetan Bansal, Saravan Rajmohan
Abstract: Direct Preference Optimization (DPO) has become a popular approach for aligning language models using pairwise preferences. However, in practical post-training pipelines, on-policy generation typically yields multiple candidate responses per prompt, which are scored by a reward model to guide learning. In this setting, we propose $\textbf{Multi-Preference Optimization (MPO)}$, a generalization of DPO that optimizes over entire sets of responses by extending the Bradley-Terry model to groupwise comparisons between chosen and rejected sets. To further enhance learning, MPO employs deviation-based weighting, which emphasizes outlier responses that deviate most from the mean reward, effectively inducing a self-paced curriculum. We theoretically prove that MPO reduces alignment bias at a rate of $\mathcal{O}\left(\frac{1}{\sqrt{n}}\right)$ with respect to the number of responses per query. Empirically, MPO achieves state-of-the-art performance on the UltraFeedback benchmark and yields up to $\sim 17.5\%$ improvement over the state-of-the-art baseline in length-controlled win rate on AlpacaEval2, establishing a new baseline for preference-based alignment
Authors: Tongxu Zhang, Bei Wang
Abstract: In recent years, point cloud upsampling has been widely applied in tasks such as 3D reconstruction and object recognition. This study proposed a novel framework, ReLPU, which enhances upsampling performance by explicitly learning from both global and local structural features of point clouds. Specifically, we extracted global features from uniformly segmented inputs (Average Segments) and local features from patch-based inputs of the same point cloud. These two types of features were processed through parallel autoencoders, fused, and then fed into a shared decoder for upsampling. This dual-input design improved feature completeness and cross-scale consistency, especially in sparse and noisy regions. Our framework was applied to several state-of-the-art autoencoder-based networks and validated on standard datasets. Experimental results demonstrated consistent improvements in geometric fidelity and robustness. In addition, saliency maps confirmed that parallel global-local learning significantly enhanced the interpretability and performance of point cloud upsampling.
Authors: Jan Skvrna, Lukas Neumann
Abstract: Inferring object 3D position and orientation from a single RGB camera is a foundational task in computer vision with many important applications. Traditionally, 3D object detection methods are trained in a fully-supervised setup, requiring LiDAR and vast amounts of human annotations, which are laborious, costly, and do not scale well with the ever-increasing amounts of data being captured. We present a novel method to train a 3D object detector from a single RGB camera without domain-specific human annotations, making orders of magnitude more data available for training. The method uses newly proposed Local Object Motion Model to disentangle object movement source between subsequent frames, is approximately 700 times faster than previous work and compensates camera focal length differences to aggregate multiple datasets. The method is evaluated on three public datasets, where despite using no human labels, it outperforms prior work by a significant margin. It also shows its versatility as a pre-training tool for fully-supervised training and shows that combining pseudo-labels from multiple datasets can achieve comparable accuracy to using human labels from a single dataset. The source code and model are available at https://github.com/jskvrna/MonoSOWA.
Authors: Ziyu Zhao, Yixiao Zhou, Zhi Zhang, Didi Zhu, Tao Shen, Zexi Li, Jinluan Yang, Xuwu Wang, Jing Su, Kun Kuang, Zhongyu Wei, Fei Wu, Yu Cheng
Abstract: Low-Rank Adaptation (LoRA) is widely used for adapting large language models (LLMs) to specific domains due to its efficiency and modularity. Meanwhile, vanilla LoRA struggles with task conflicts in multi-task scenarios. Recent works adopt Mixture of Experts (MoE) by treating each LoRA module as an expert, thereby mitigating task interference through multiple specialized LoRA modules. While effective, these methods often isolate knowledge within individual tasks, failing to fully exploit the shared knowledge across related tasks. In this paper, we establish a connection between single LoRA and multi-LoRA MoE, integrating them into a unified framework. We demonstrate that the dynamic routing of multiple LoRAs is functionally equivalent to rank partitioning and block-level activation within a single LoRA. We further empirically demonstrate that finer-grained LoRA partitioning, within the same total and activated parameter constraints, leads to better performance gains across heterogeneous tasks. Building on these findings, we propose Single-ranked Mixture of Experts LoRA (\textbf{SMoRA}), which embeds MoE into LoRA by \textit{treating each rank as an independent expert}. With a \textit{dynamic rank-wise activation} mechanism, SMoRA promotes finer-grained knowledge sharing while mitigating task conflicts. Experiments demonstrate that SMoRA activates fewer parameters yet achieves better performance in multi-task scenarios.
Authors: Zhiyuan Fu, Junfan Chen, Lan Zhang, Ting Yang, Jun Niu, Hongyu Sun, Ruidong Li, Peng Liu, Jice Wang, Fannv He, Yuqing Zhang
Abstract: Large Language Models (LLMs) are rapidly transforming the landscape of digital content creation. However, the prevalent black-box Application Programming Interface (API) access to many LLMs introduces significant challenges in accountability, governance, and security. LLM fingerprinting, which aims to identify the source model by analyzing statistical and stylistic features of generated text, offers a potential solution. Current progress in this area is hindered by a lack of dedicated datasets and the need for efficient, practical methods that are robust against adversarial manipulations. To address these challenges, we introduce FD-Dataset, a comprehensive bilingual fingerprinting benchmark comprising 90,000 text samples from 20 famous proprietary and open-source LLMs. Furthermore, we present FDLLM, a novel fingerprinting method that leverages parameter-efficient Low-Rank Adaptation (LoRA) to fine-tune a foundation model. This approach enables LoRA to extract deep, persistent features that characterize each source LLM. Through our analysis, we find that LoRA adaptation promotes the aggregation of outputs from the same LLM in representation space while enhancing the separation between different LLMs. This mechanism explains why LoRA proves particularly effective for LLM fingerprinting. Extensive empirical evaluations on FD-Dataset demonstrate FDLLM's superiority, achieving a Macro F1 score 22.1% higher than the strongest baseline. FDLLM also exhibits strong generalization to newly released models, achieving an average accuracy of 95% on unseen models. Notably, FDLLM remains consistently robust under various adversarial attacks, including polishing, translation, and synonym substitution. Experimental results show that FDLLM reduces the average attack success rate from 49.2% (LM-D) to 23.9%.
Authors: Ander Gray, Vignesh Gopakumar, Sylvain Rousseau, S\'ebastien Destercke
Abstract: We propose a method for obtaining statistically guaranteed prediction sets for functional machine learning methods: surrogate models which map between function spaces, motivated by the need to build reliable PDE emulators. The method constructs nested prediction sets on a low-dimensional representation (an SVD) of the surrogate model's error, and then maps these sets to the prediction space using set-propagation techniques. This results in prediction sets for functional surrogate models with conformal prediction coverage guarantees. We use zonotopes as basis of the set construction, which allow an exact linear propagation and are closed under Cartesian products, making them well-suited to this high-dimensional problem. The method is model agnostic and can thus be applied to complex Sci-ML models, including Neural Operators, but also in simpler settings. We also introduce a technique to capture the truncation error of the SVD, preserving the guarantees of the method.
Authors: Feng Chen, Allan Raventos, Nan Cheng, Surya Ganguli, Shaul Druckmann
Abstract: Recent progress in large language models (LLMs) highlights the power of scaling test-time compute to achieve strong performance on complex tasks, such as mathematical reasoning and code generation. This raises a critical question: how should model training be modified to optimize performance under a subsequent test-time compute strategy and budget? To explore this, we focus on pass@N, a simple test-time strategy that searches for a correct answer in $N$ independent samples. We show, surprisingly, that training with cross-entropy (CE) loss can be ${\it misaligned}$ with pass@N in that pass@N accuracy ${\it decreases}$ with longer training. We explain the origins of this misalignment in terms of model overconfidence induced by CE, and experimentally verify our prediction of overconfidence as an impediment to scaling test-time compute via pass@N. Furthermore we suggest a principled, modified training loss that is better aligned to pass@N by limiting model confidence and rescuing pass@N test performance. Our algorithm demonstrates improved mathematical reasoning on MATH and MiniF2F benchmarks under several scenarios: (1) providing answers to math questions; and (2) proving theorems by searching over proof trees of varying shapes. Overall our work underscores the importance of co-designing two traditionally separate phases of LLM development: training-time protocols and test-time search and reasoning strategies.
Authors: R. Kenny Jones, Paul Guerrero, Niloy J. Mitra, Daniel Ritchie
Abstract: We present ShapeLib, the first method that leverages the priors of LLMs to design libraries of programmatic 3D shape abstractions. Our system accepts two forms of design intent: text descriptions of functions to include in the library and a seed set of exemplar shapes. We discover abstractions that match this design intent with a guided LLM workflow that first proposes, and then validates, different ways of applying and implementing functions. We learn recognition networks that map shapes to programs with these newly discovered abstractions by training on data produced by LLM authored synthetic data generation procedures. Across modeling domains (split by shape category), we find that LLMs, when thoughtfully combined with geometric reasoning, can be guided to author a library of abstraction functions that generalize to shapes outside of the seed set. This framework addresses a long-standing shape analysis problem of how to discover reusable abstraction functions while exposing interpretable, semantically aligned interfaces. We find that ShapeLib provides distinct advantages over prior alternative abstraction discovery works in terms of generalization, usability, and maintaining plausibility under manipulation. Finally, we demonstrate that ShapeLib's abstraction functions unlock a number of downstream applications, combining LLM reasoning over shape programs with geometry processing to support shape editing and generation.
Authors: Sunay Joshi, Shayan Kiyani, George Pappas, Edgar Dobriban, Hamed Hassani
Abstract: We consider the problem of conformal prediction under covariate shift. Given labeled data from a source domain and unlabeled data from a covariate shifted target domain, we seek to construct prediction sets with valid marginal coverage in the target domain. Most existing methods require estimating the unknown likelihood ratio function, which can be prohibitive for high-dimensional data such as images. To address this challenge, we introduce the likelihood ratio regularized quantile regression (LR-QR) algorithm, which combines the pinball loss with a novel choice of regularization in order to construct a threshold function without directly estimating the unknown likelihood ratio. We show that the LR-QR method has coverage at the desired level in the target domain, up to a small error term that we can control. Our proofs draw on a novel analysis of coverage via stability bounds from learning theory. Our experiments demonstrate that the LR-QR algorithm outperforms existing methods on high-dimensional prediction tasks, including a regression task for the Communities and Crime dataset, an image classification task from the WILDS repository, and an LLM question-answering task on the MMLU benchmark.
Authors: Bernal Jim\'enez Guti\'errez, Yiheng Shu, Weijian Qi, Sizhe Zhou, Yu Su
Abstract: Our ability to continuously acquire, organize, and leverage knowledge is a key feature of human intelligence that AI systems must approximate to unlock their full potential. Given the challenges in continual learning with large language models (LLMs), retrieval-augmented generation (RAG) has become the dominant way to introduce new information. However, its reliance on vector retrieval hinders its ability to mimic the dynamic and interconnected nature of human long-term memory. Recent RAG approaches augment vector embeddings with various structures like knowledge graphs to address some of these gaps, namely sense-making and associativity. However, their performance on more basic factual memory tasks drops considerably below standard RAG. We address this unintended deterioration and propose HippoRAG 2, a framework that outperforms standard RAG comprehensively on factual, sense-making, and associative memory tasks. HippoRAG 2 builds upon the Personalized PageRank algorithm used in HippoRAG and enhances it with deeper passage integration and more effective online use of an LLM. This combination pushes this RAG system closer to the effectiveness of human long-term memory, achieving a 7% improvement in associative memory tasks over the state-of-the-art embedding model while also exhibiting superior factual knowledge and sense-making memory capabilities. This work paves the way for non-parametric continual learning for LLMs. Code and data are available at https://github.com/OSU-NLP-Group/HippoRAG.
Authors: Jann Railey Montalan, Jimson Paulo Layacan, David Demitri Africa, Richell Isaiah Flores, Michael T. Lopez II, Theresa Denise Magsajo, Anjanette Cayabyab, William Chandra Tjhi
Abstract: Recent advances in large language models (LLMs) have demonstrated remarkable capabilities on widely benchmarked high-resource languages. However, linguistic nuances of under-resourced languages remain unexplored. We introduce Batayan, a holistic Filipino benchmark that systematically evaluates LLMs across three key natural language processing (NLP) competencies: understanding, reasoning, and generation. Batayan consolidates eight tasks, three of which have not existed prior for Filipino corpora, covering both Tagalog and code-switched Taglish utterances. Our rigorous, native-speaker-driven adaptation and validation processes ensures fluency and authenticity to the complex morphological and syntactic structures of Filipino, alleviating the pervasive translationese bias in existing Filipino corpora. We report empirical results on a variety of open-source and commercial LLMs, highlighting significant performance gaps that signal the under-representation of Filipino in pre-training corpora, the unique hurdles in modeling Filipino's rich morphology and construction, and the importance of explicit Filipino language support. Moreover, we discuss the practical challenges encountered in dataset construction and propose principled solutions for building culturally and linguistically-faithful resources in under-represented languages. We also provide a public evaluation suite as a clear foundation for iterative, community-driven progress in Filipino NLP.
Authors: Mollie Shichman, Claire Bonial, Austin Blodgett, Taylor Hudson, Francis Ferraro, Rachel Rudinger
Abstract: During Human Robot Interactions in disaster relief scenarios, Large Language Models (LLMs) have the potential for substantial physical reasoning to assist in mission objectives. However, these capabilities are often found only in larger models, which are frequently not reasonable to deploy on robotic systems. To meet our problem space requirements, we introduce a dataset and pipeline to create Field Reasoning and Instruction Decoding Agent (FRIDA) models. In our pipeline, domain experts and linguists combine their knowledge to make high-quality few-shot prompts used to generate synthetic data for fine-tuning. We hand-curate datasets for this few-shot prompting and for evaluation to improve LLM reasoning on both general and disaster-specific objects. We concurrently run an ablation study to understand which kinds of synthetic data most affect performance. We fine-tune several small instruction-tuned models and find that ablated FRIDA models only trained on objects' physical state and function data outperformed both the FRIDA models trained on all synthetic data and the base models in our customized evaluation. We demonstrate that the FRIDA pipeline is capable of instilling physical common sense with minimal data.
Authors: Daniela B\"ohm, Georg Gottlob, Matthias Lanzinger, Davide Longo, Cem Okulmus, Reinhard Pichler, Alexander Selzer
Abstract: Query optimization has played a central role in database research for decades. However, more often than not, the proposed optimization techniques lead to a performance improvement in some, but not in all, situations. Therefore, we urgently need a methodology for designing a decision procedure that decides for a given query whether the optimization technique should be applied or not. In this work, we propose such a methodology with a focus on Yannakakis-style query evaluation as our optimization technique of interest. More specifically, we formulate this decision problem as an algorithm selection problem and we present a Machine Learning based approach for its solution. Empirical results with several benchmarks on a variety of database systems show that our approach indeed leads to a statistically significant performance improvement.
Authors: Yoonyoung Cho, Junhyek Han, Jisu Han, Beomjoon Kim
Abstract: For robots to operate in general environments like households, they must be able to perform non-prehensile manipulation actions such as toppling and rolling to manipulate ungraspable objects. However, prior works on non-prehensile manipulation cannot yet generalize across environments with diverse geometries. The main challenge lies in adapting to varying environmental constraints: within a cabinet, the robot must avoid walls and ceilings; to lift objects to the top of a step, the robot must account for the step's pose and extent. While deep reinforcement learning (RL) has demonstrated impressive success in non-prehensile manipulation, accounting for such variability presents a challenge for the generalist policy, as it must learn diverse strategies for each new combination of constraints. To address this, we propose a modular and reconfigurable architecture that adaptively reconfigures network modules based on task requirements. To capture the geometric variability in environments, we extend the contact-based object representation (CORN) to environment geometries, and propose a procedural algorithm for generating diverse environments to train our agent. Taken together, the resulting policy can zero-shot transfer to novel real-world environments and objects despite training entirely within a simulator. We additionally release a simulation-based benchmark featuring nine digital twins of real-world scenes with 353 objects to facilitate non-prehensile manipulation research in realistic domains.
Authors: Th\'eo Vincent, Tim Faust, Yogesh Tripathi, Jan Peters, Carlo D'Eramo
Abstract: Recent works have successfully demonstrated that sparse deep reinforcement learning agents can be competitive against their dense counterparts. This opens up opportunities for reinforcement learning applications in fields where inference time and memory requirements are cost-sensitive or limited by hardware. Until now, dense-to-sparse methods have relied on hand-designed sparsity schedules that are not synchronized with the agent's learning pace. Crucially, the final sparsity level is chosen as a hyperparameter, which requires careful tuning as setting it too high might lead to poor performances. In this work, we address these shortcomings by crafting a dense-to-sparse algorithm that we name Eau De $Q$-Network (EauDeQN). To increase sparsity at the agent's learning pace, we consider multiple online networks with different sparsity levels, where each online network is trained from a shared target network. At each target update, the online network with the smallest loss is chosen as the next target network, while the other networks are replaced by a pruned version of the chosen network. We evaluate the proposed approach on the Atari $2600$ benchmark and the MuJoCo physics simulator, showing that EauDeQN reaches high sparsity levels while keeping performances high.
Authors: Songming Zhang, Xue Zhang, Tong Zhang, Bojie Hu, Yufeng Chen, Jinan Xu
Abstract: In modern large language models (LLMs), LLM alignment is of crucial importance and is typically achieved through methods such as reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO). However, in most existing methods for LLM alignment, all tokens in the response are optimized using a sparse, response-level reward or preference annotation. The ignorance of token-level rewards may erroneously punish high-quality tokens or encourage low-quality tokens, resulting in suboptimal performance and slow convergence speed. To address this issue, we propose AlignDistil, an RLHF-equivalent distillation method for token-level reward optimization. Specifically, we introduce the reward learned by DPO into the RLHF objective and theoretically prove the equivalence between this objective and a token-level distillation process, where the teacher distribution linearly combines the logits from the DPO model and a reference model. On this basis, we further bridge the accuracy gap between the reward from the DPO model and the pure reward model, by building a contrastive DPO reward with a normal and a reverse DPO model. Moreover, to avoid under- and over-optimization on different tokens, we design a token adaptive logit extrapolation mechanism to construct an appropriate teacher distribution for each token. Experimental results demonstrate the superiority of our AlignDistil over existing methods and showcase fast convergence due to its token-level distributional reward optimization.
Authors: Anar Yeginbergen, Maite Oronoz, Rodrigo Agerri
Abstract: This paper investigates the role of dynamic external knowledge integration in improving counter-argument generation using Large Language Models (LLMs). While LLMs have shown promise in argumentative tasks, their tendency to generate lengthy, potentially unfactual responses highlights the need for more controlled and evidence-based approaches. We introduce a new manually curated dataset of argument and counter-argument pairs specifically designed to balance argumentative complexity with evaluative feasibility. We also propose a new LLM-as-a-Judge evaluation methodology that shows a stronger correlation with human judgments compared to traditional reference-based metrics. Our experimental results demonstrate that integrating dynamic external knowledge from the web significantly improves the quality of generated counter-arguments, particularly in terms of relatedness, persuasiveness, and factuality. The findings suggest that combining LLMs with real-time external knowledge retrieval offers a promising direction for developing more effective and reliable counter-argumentation systems.
Authors: Bang Nguyen, Tingting Du, Mengxia Yu, Lawrence Angrave, Meng Jiang
Abstract: While the Question Generation (QG) task has been increasingly adopted in educational assessments, its evaluation remains limited by approaches that lack a clear connection to the educational values of test items. In this work, we introduce test item analysis, a method frequently used by educators to assess test question quality, into QG evaluation. Specifically, we construct pairs of candidate questions that differ in quality across dimensions such as topic coverage, item difficulty, item discrimination, and distractor efficiency. We then examine whether existing QG evaluation approaches can effectively distinguish these differences. Our findings reveal significant shortcomings in these approaches with respect to accurately assessing test item quality in relation to student performance. To address this gap, we propose a novel QG evaluation framework, QG-SMS, which leverages Large Language Model for Student Modeling and Simulation to perform test item analysis. As demonstrated in our extensive experiments and human evaluation study, the additional perspectives introduced by the simulated student profiles lead to a more effective and robust assessment of test items.
Authors: Chen Xu, Tony Khuong Nguyen, Emma Dixon, Christopher Rodriguez, Patrick Miller, Robert Lee, Paarth Shah, Rares Ambrus, Haruki Nishimura, Masha Itkina
Abstract: Recent years have witnessed impressive robotic manipulation systems driven by advances in imitation learning and generative modeling, such as diffusion- and flow-based approaches. As robot policy performance increases, so does the complexity and time horizon of achievable tasks, inducing unexpected and diverse failure modes that are difficult to predict a priori. To enable trustworthy policy deployment in safety-critical human environments, reliable runtime failure detection becomes important during policy inference. However, most existing failure detection approaches rely on prior knowledge of failure modes and require failure data during training, which imposes a significant challenge in practicality and scalability. In response to these limitations, we present FAIL-Detect, a modular two-stage approach for failure detection in imitation learning-based robotic manipulation. To accurately identify failures from successful training data alone, we frame the problem as sequential out-of-distribution (OOD) detection. We first distill policy inputs and outputs into scalar signals that correlate with policy failures and capture epistemic uncertainty. FAIL-Detect then employs conformal prediction (CP) as a versatile framework for uncertainty quantification with statistical guarantees. Empirically, we thoroughly investigate both learned and post-hoc scalar signal candidates on diverse robotic manipulation tasks. Our experiments show learned signals to be mostly consistently effective, particularly when using our novel flow-based density estimator. Furthermore, our method detects failures more accurately and faster than state-of-the-art (SOTA) failure detection baselines. These results highlight the potential of FAIL-Detect to enhance the safety and reliability of imitation learning-based robotic systems as they progress toward real-world deployment.
Authors: Gaurav Kumar Gupta, Pranal Pande, Nirajan Acharya, Aniket Kumar Singh, Suman Niroula
Abstract: Large Language Models (LLMs) are revolutionizing medical diagnostics by enhancing both disease classification and clinical decision-making. In this study, we evaluate the performance of two LLM- based diagnostic tools, DeepSeek R1 and O3 Mini, using a structured dataset of symptoms and diagnoses. We assessed their predictive accuracy at both the disease and category levels, as well as the reliability of their confidence scores. DeepSeek R1 achieved a disease-level accuracy of 76% and an overall accuracy of 82%, outperforming O3 Mini, which attained 72% and 75% respectively. Notably, DeepSeek R1 demonstrated exceptional performance in Mental Health, Neurological Disorders, and Oncology, where it reached 100% accuracy, while O3 Mini excelled in Autoimmune Disease classification with 100% accuracy. Both models, however, struggled with Respiratory Disease classification, recording accuracies of only 40% for DeepSeek R1 and 20% for O3 Mini. Additionally, the analysis of confidence scores revealed that DeepSeek R1 provided high-confidence predictions in 92% of cases, compared to 68% for O3 Mini. Ethical considerations regarding bias, model interpretability, and data privacy are also discussed to ensure the responsible integration of LLMs into clinical practice. Overall, our findings offer valuable insights into the strengths and limitations of LLM-based diagnostic systems and provide a roadmap for future enhancements in AI-driven healthcare.
Authors: Murong Yue, Ziyu Yao
Abstract: Batch prompting, which combines a batch of multiple queries sharing the same context in one inference, has emerged as a promising solution to reduce inference costs. However, our study reveals a significant security vulnerability in batch prompting: malicious users can inject attack instructions into a batch, leading to unwanted interference across all queries, which can result in the inclusion of harmful content, such as phishing links, or the disruption of logical reasoning. In this paper, we construct BATCHSAFEBENCH, a comprehensive benchmark comprising 150 attack instructions of two types and 8k batch instances, to study the batch prompting vulnerability systematically. Our evaluation of both closed-source and open-weight LLMs demonstrates that all LLMs are susceptible to batch-prompting attacks. We then explore multiple defending approaches. While the prompting-based defense shows limited effectiveness for smaller LLMs, the probing-based approach achieves about 95% accuracy in detecting attacks. Additionally, we perform a mechanistic analysis to understand the attack and identify attention heads that are responsible for it.
Authors: Enrico Marchesini, Benjamin Donnot, Constance Crozier, Ian Dytham, Christian Merz, Lars Schewe, Nico Westerbeck, Cathy Wu, Antoine Marot, Priya L. Donti
Abstract: Reinforcement learning (RL) can provide adaptive and scalable controllers essential for power grid decarbonization. However, RL methods struggle with power grids' complex dynamics, long-horizon goals, and hard physical constraints. For these reasons, we present RL2Grid, a benchmark designed in collaboration with power system operators to accelerate progress in grid control and foster RL maturity. Built on RTE France's power simulation framework, RL2Grid standardizes tasks, state and action spaces, and reward structures for a systematic evaluation and comparison of RL algorithms. Moreover, we integrate operational heuristics and design safety constraints based on human expertise to ensure alignment with physical requirements. By establishing reference performance metrics for classic RL baselines on RL2Grid's tasks, we highlight the need for novel methods capable of handling real systems and discuss future directions for RL-based grid control.
Authors: Kentaro Nomura, Tatsuya Aoki, Tadahiro Taniguchi, Takato Horii
Abstract: We propose a fully decentralized multi-agent world model that enables both symbol emergence for communication and coordinated behavior through temporal extension of collective predictive coding. Unlike previous research that focuses on either communication or coordination separately, our approach achieves both simultaneously. Our method integrates world models with communication channels, enabling agents to predict environmental dynamics, estimate states from partial observations, and share critical information through bidirectional message exchange with contrastive learning for message alignment. Using a two-agent trajectory drawing task, we demonstrate that our communication-based approach outperforms non-communicative models when agents have divergent perceptual capabilities, achieving the second-best coordination after centralized models. Importantly, our decentralized approach with constraints preventing direct access to other agents' internal states facilitates the emergence of more meaningful symbol systems that accurately reflect environmental states. These findings demonstrate the effectiveness of decentralized communication for supporting coordination while developing shared representations of the environment.
Authors: Sher Badshah, Ali Emami, Hassan Sajjad
Abstract: As Large Language Models (LLMs) become increasingly integrated into real-world, autonomous applications, relying on static, pre-annotated references for evaluation poses significant challenges in cost, scalability, and completeness. We propose Tool-Augmented LLM Evaluation (TALE), a framework to assess LLM outputs without predetermined ground-truth answers. Unlike conventional metrics that compare to fixed references or depend solely on LLM-as-a-judge knowledge, TALE employs an agent with tool-access capabilities that actively retrieves and synthesizes external evidence. It iteratively generates web queries, collects information, summarizes findings, and refines subsequent searches through reflection. By shifting away from static references, TALE aligns with free-form question-answering tasks common in real-world scenarios. Experimental results on multiple free-form QA benchmarks show that TALE not only outperforms standard reference-based metrics for measuring response accuracy but also achieves substantial to near-perfect agreement with human evaluations. TALE enhances the reliability of LLM evaluations in real-world, dynamic scenarios without relying on static references.
Authors: Yang Jiao, Xiaodong Wang, Kai Yang
Abstract: Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of applications, e.g., medical question-answering, mathematical sciences, and code generation. However, they also exhibit inherent limitations, such as outdated knowledge and susceptibility to hallucinations. Retrieval-Augmented Generation (RAG) has emerged as a promising paradigm to address these issues, but it also introduces new vulnerabilities. Recent efforts have focused on the security of RAG-based LLMs, yet existing attack methods face three critical challenges: (1) their effectiveness declines sharply when only a limited number of poisoned texts can be injected into the knowledge database, (2) they lack sufficient stealth, as the attacks are often detectable by anomaly detection systems, which compromises their effectiveness, and (3) they rely on heuristic approaches to generate poisoned texts, lacking formal optimization frameworks and theoretic guarantees, which limits their effectiveness and applicability. To address these issues, we propose coordinated Prompt-RAG attack (PR-attack), a novel optimization-driven attack that introduces a small number of poisoned texts into the knowledge database while embedding a backdoor trigger within the prompt. When activated, the trigger causes the LLM to generate pre-designed responses to targeted queries, while maintaining normal behavior in other contexts. This ensures both high effectiveness and stealth. We formulate the attack generation process as a bilevel optimization problem leveraging a principled optimization framework to develop optimal poisoned texts and triggers. Extensive experiments across diverse LLMs and datasets demonstrate the effectiveness of PR-Attack, achieving a high attack success rate even with a limited number of poisoned texts and significantly improved stealth compared to existing methods.
Authors: Junli Liu, Qizhi Chen, Zhigang Wang, Yiwen Tang, Yiting Zhang, Chi Yan, Dong Wang, Bin Zhao, Xuelong Li
Abstract: Visual grounding (VG) aims to localize target objects in an image based on natural language descriptions. In this paper, we propose AerialVG, a new task focusing on visual grounding from aerial views. Compared to traditional VG, AerialVG poses new challenges, \emph{e.g.}, appearance-based grounding is insufficient to distinguish among multiple visually similar objects, and positional relations should be emphasized. Besides, existing VG models struggle when applied to aerial imagery, where high-resolution images cause significant difficulties. To address these challenges, we introduce the first AerialVG dataset, consisting of 5K real-world aerial images, 50K manually annotated descriptions, and 103K objects. Particularly, each annotation in AerialVG dataset contains multiple target objects annotated with relative spatial relations, requiring models to perform comprehensive spatial reasoning. Furthermore, we propose an innovative model especially for the AerialVG task, where a Hierarchical Cross-Attention is devised to focus on target regions, and a Relation-Aware Grounding module is designed to infer positional relations. Experimental results validate the effectiveness of our dataset and method, highlighting the importance of spatial reasoning in aerial visual grounding. The code and dataset will be released.
Authors: Mahshad Lotfinia, Arash Tayebiarasteh, Samaneh Samiei, Mehdi Joodaki, Soroosh Tayebi Arasteh
Abstract: Reliable artificial intelligence (AI) models for medical image analysis often depend on large and diverse labeled datasets. Federated learning (FL) offers a decentralized and privacy-preserving approach to training but struggles in highly non-independent and identically distributed (non-IID) settings, where institutions with more representative data may experience degraded performance. Moreover, existing large-scale FL studies have been limited to adult datasets, neglecting the unique challenges posed by pediatric data, which introduces additional non-IID variability. To address these limitations, we analyzed n=398,523 adult chest radiographs from diverse institutions across multiple countries and n=9,125 pediatric images, leveraging transfer learning from general-purpose self-supervised image representations to classify pneumonia and cases with no abnormality. Using state-of-the-art vision transformers, we found that FL improved performance only for smaller adult datasets (P<0.001) but degraded performance for larger datasets (P<0.064) and pediatric cases (P=0.242). However, equipping FL with self-supervised weights significantly enhanced outcomes across pediatric cases (P=0.031) and most adult datasets (P<0.008), except the largest dataset (P=0.052). These findings underscore the potential of easily deployable general-purpose self-supervised image representations to address non-IID challenges in clinical FL applications and highlight their promise for enhancing patient outcomes and advancing pediatric healthcare, where data scarcity and variability remain persistent obstacles.
Authors: Ruochi Zhang, Qian Yang, Xiaoyang Wang, Tian Wang, Qiong Zhou, Ziqi Deng, Kewei Li, Yueying Wang, Yusi Fan, Jiale Zhang, Lan Huang, Chang Liu, Fengfeng Zhou
Abstract: The rapid accumulation of Electronic Health Records (EHRs) has transformed healthcare by providing valuable data that enhance clinical predictions and diagnoses. While conventional machine learning models have proven effective, they often lack robust representation learning and depend heavily on expert-crafted features. Although deep learning offers powerful solutions, it is often criticized for its lack of interpretability. To address these challenges, we propose DeepSelective, a novel end to end deep learning framework for predicting patient prognosis using EHR data, with a strong emphasis on enhancing model interpretability. DeepSelective combines data compression techniques with an innovative feature selection approach, integrating custom-designed modules that work together to improve both accuracy and interpretability. Our experiments demonstrate that DeepSelective not only enhances predictive accuracy but also significantly improves interpretability, making it a valuable tool for clinical decision-making. The source code is freely available at http://www.healthinformaticslab.org/supp/resources.php .
URLs: http://www.healthinformaticslab.org/supp/resources.php
Authors: Jang Hyun Cho, Andrea Madotto, Effrosyni Mavroudi, Triantafyllos Afouras, Tushar Nagarajan, Muhammad Maaz, Yale Song, Tengyu Ma, Shuming Hu, Suyog Jain, Miguel Martin, Huiyu Wang, Hanoona Rasheed, Peize Sun, Po-Yao Huang, Daniel Bolya, Nikhila Ravi, Shashank Jain, Tammy Stark, Shane Moon, Babak Damavandi, Vivian Lee, Andrew Westbury, Salman Khan, Philipp Kr\"ahenb\"uhl, Piotr Doll\'ar, Lorenzo Torresani, Kristen Grauman, Christoph Feichtenhofer
Abstract: Vision-language models are integral to computer vision research, yet many high-performing models remain closed-source, obscuring their data, design and training recipe. The research community has responded by using distillation from black-box models to label training data, achieving strong benchmark results, at the cost of measurable scientific progress. However, without knowing the details of the teacher model and its data sources, scientific progress remains difficult to measure. In this paper, we study building a Perception Language Model (PLM) in a fully open and reproducible framework for transparent research in image and video understanding. We analyze standard training pipelines without distillation from proprietary models and explore large-scale synthetic data to identify critical data gaps, particularly in detailed video understanding. To bridge these gaps, we release 2.8M human-labeled instances of fine-grained video question-answer pairs and spatio-temporally grounded video captions. Additionally, we introduce PLM-VideoBench, a suite for evaluating challenging video understanding tasks focusing on the ability to reason about "what", "where", "when", and "how" of a video. We make our work fully reproducible by providing data, training recipes, code & models. https://github.com/facebookresearch/perception_models
Authors: Wenqing Peng, Zhi-Song Liu, Michael Boy
Abstract: Estimating rate coefficients from complex chemical reactions is essential for advancing detailed chemistry. However, the stiffness inherent in real-world atmospheric chemistry systems poses severe challenges, leading to training instability and poor convergence that hinder effective rate coefficient estimation using learning-based approaches. To address this, we propose a Stiff Physics-Informed Neural ODE framework (SPIN-ODE) for chemical reaction modelling. Our method introduces a three-stage optimisation process: first, a latent neural ODE learns the continuous and differentiable trajectory between chemical concentrations and their time derivatives; second, an explicit Chemical Reaction Neural Network (CRNN) extracts the underlying rate coefficients based on the learned dynamics; and third, fine-tune CRNN using a neural ODE solver to further improve rate coefficient estimation. Extensive experiments on both synthetic and newly proposed real-world datasets validate the effectiveness and robustness of our approach. As the first work on stiff Neural ODEs for chemical rate coefficient discovery, our study opens promising directions for integrating neural networks with detailed chemistry.
Authors: Hiari Pizzini Cavagna, Daniele Cesarini, Andrea Bartolini
Abstract: The increasing demand for generative AI as Large Language Models (LLMs) services has driven the need for specialized hardware architectures that optimize computational efficiency and energy consumption. This paper evaluates the performance of the Tenstorrent Grayskull e75 RISC-V accelerator for basic linear algebra kernels at reduced numerical precision, a fundamental operation in LLM computations. We present a detailed characterization of Grayskull's execution model, gridsize, matrix dimensions, data formats, and numerical precision impact computational efficiency. Furthermore, we compare Grayskull's performance against state-of-the-art architectures with tensor acceleration, including Intel Sapphire Rapids processors and two NVIDIA GPUs (V100 and A100). Whilst NVIDIA GPUs dominate raw performance, Grayskull demonstrates a competitive trade-off between power consumption and computational throughput, reaching a peak of 1.55 TFLOPs/Watt with BF16.
Authors: Feilong Jiang, Xiaonan Hou, Jianqiao Ye, Min Xia
Abstract: Physics-Informed Neural Networks (PINNs) have emerged as a powerful framework for solving partial differential equations (PDEs) by embedding physical laws directly into the loss function. However, effective training of PINNs remains challenging due to internal covariate shift, which destabilizes feature distributions and impairs model expressiveness. While normalization techniques like Batch Normalization and Layer Normalization are standard remedies in deep learning, they disrupt the pointwise input-output mappings critical to preserving the physical consistency in PINNs. In this work, we introduce Mask-PINNs, a novel architecture that regulates internal feature distributions through a smooth, learnable mask function applied pointwise across hidden layers. Unlike conventional normalization methods, the proposed mask function preserves the deterministic nature of input-output relationships while suppressing activation drift and saturation. Theoretically, we demonstrate that Mask-PINNs control feature spread near initialization by attenuating gradient variance growth through a tailored modulation mechanism. Empirically, we validate the method on multiple PDE benchmarks across diverse activation functions. Our results show consistent improvements in prediction accuracy, convergence stability, and robustness, with relative L2 errors reduced by up to two orders of magnitude over baseline models. Furthermore, we demonstrate that Mask-PINNs enable the effective use of wider networks, overcoming a key limitation in existing PINN frameworks.
Authors: Xingjin Wang, Howe Tissue, Lu Wang, Linjing Li, Daniel Dajun Zeng
Abstract: Continual Pre-Training (CPT) has become a popular and effective method to apply strong foundation models to specific downstream tasks. In this work, we explore the learning dynamics throughout the CPT process for large language models. We specifically focus on how general and downstream domain performance evolves at each training step, with domain performance measured via validation losses. We have observed that the CPT loss curve fundamentally characterizes the transition from one curve to another hidden curve, and could be described by decoupling the effects of distribution shift and learning rate annealing. We derive a CPT scaling law that combines the two factors, enabling the prediction of loss at any (continual) training steps and across learning rate schedules (LRS) in CPT. Our formulation presents a comprehensive understanding of several critical factors in CPT, including loss potential, peak learning rate, training steps, replay ratio, etc. Moreover, our approach can be adapted to customize training hyper-parameters to different CPT goals such as balancing general and domain-specific performance. Extensive experiments demonstrate that our scaling law holds across various CPT datasets and training hyper-parameters.
Authors: Jiafan Li, Jiaqi Zhu, Liang Chang, Yilin Li, Miaomiao Li, Yang Wang, Hongan Wang
Abstract: Nowadays, numerous online platforms can be described as multi-modal heterogeneous networks (MMHNs), such as Douban's movie networks and Amazon's product review networks. Accurately categorizing nodes within these networks is crucial for analyzing the corresponding entities, which requires effective representation learning on nodes. However, existing multi-modal fusion methods often adopt either early fusion strategies which may lose the unique characteristics of individual modalities, or late fusion approaches overlooking the cross-modal guidance in GNN-based information propagation. In this paper, we propose a novel model for node classification in MMHNs, named Heterogeneous Graph Neural Network with Inter-Modal Attention (HGNN-IMA). It learns node representations by capturing the mutual influence of multiple modalities during the information propagation process, within the framework of heterogeneous graph transformer. Specifically, a nested inter-modal attention mechanism is integrated into the inter-node attention to achieve adaptive multi-modal fusion, and modality alignment is also taken into account to encourage the propagation among nodes with consistent similarities across all modalities. Moreover, an attention loss is augmented to mitigate the impact of missing modalities. Extensive experiments validate the superiority of the model in the node classification task, providing an innovative view to handle multi-modal data, especially when accompanied with network structures.
Authors: Xiaohui Wang, Peng Ye, Chenyu Huang, Shenghe Zheng, Bo Zhang, Lei Bai, Wanli Ouyang, Tao Chen
Abstract: With the rise of the fine-tuned--pretrained paradigm, storing numerous fine-tuned models for multi-tasking creates significant storage overhead. Delta compression alleviates this by storing only the pretrained model and the highly compressed delta weights (the differences between fine-tuned and pretrained model weights). However, existing methods fail to maintain both high compression and performance, and often rely on data. To address these challenges, we propose UltraDelta, the first data-free delta compression pipeline that achieves both ultra-high compression and strong performance. UltraDelta is designed to minimize redundancy, maximize information, and stabilize performance across inter-layer, intra-layer, and global dimensions, using three key components: (1) Variance-Based Mixed Sparsity Allocation assigns sparsity based on variance, giving lower sparsity to high-variance layers to preserve inter-layer information. (2) Distribution-Aware Compression applies uniform quantization and then groups parameters by value, followed by group-wise pruning, to better preserve intra-layer distribution. (3) Trace-Norm-Guided Rescaling uses the trace norm of delta weights to estimate a global rescaling factor, improving model stability under higher compression. Extensive experiments across (a) large language models (fine-tuned on LLaMA-2 7B and 13B) with up to 133x, (b) general NLP models (RoBERTa-base, T5-base) with up to 800x, (c) vision models (ViT-B/32, ViT-L/14) with up to 400x, and (d) multi-modal models (BEiT-3) with 40x compression ratio, demonstrate that UltraDelta consistently outperforms existing methods, especially under ultra-high compression.
Authors: Kaiyuan Chen, Shuangyu Xie, Zehan Ma, Pannag R Sanketi, Ken Goldberg
Abstract: Vision-Language Models (VLMs) acquire real-world knowledge and general reasoning ability through Internet-scale image-text corpora. They can augment robotic systems with scene understanding and task planning, and assist visuomotor policies that are trained on robot trajectory data. We explore the reverse paradigm - using rich, real, multi-modal robot trajectory data to enhance and evaluate VLMs. In this paper, we present Robo2VLM, a Visual Question Answering (VQA) dataset generation framework for VLMs. Given a human tele-operated robot trajectory, Robo2VLM derives ground-truth from non-visual and non-descriptive sensory modalities, such as end-effector pose, gripper aperture, and force sensing. Based on these modalities, it segments the robot trajectory into a sequence of manipulation phases. At each phase, Robo2VLM uses scene and interaction understanding to identify 3D properties of the robot, task goal, and the target object. The properties are used to generate representative VQA queries - images with textural multiple-choice questions - based on spatial, goal-conditioned, and interaction reasoning question templates. We curate Robo2VLM-1, a large-scale in-the-wild dataset with 684,710 questions covering 463 distinct scenes and 3,396 robotic manipulation tasks from 176k real robot trajectories. Results suggest that Robo2VLM-1 can benchmark and improve VLM capabilities in spatial and interaction reasoning.
Authors: Wenjie Yang, Mao Zheng, Mingyang Song, Zheng Li, Sitong Wang
Abstract: Large language models (LLMs) have recently demonstrated remarkable capabilities in machine translation (MT). However, most advanced MT-specific LLMs heavily rely on external supervision signals during training, such as human-annotated reference data or trained reward models (RMs), which are often expensive to obtain and challenging to scale. To overcome this limitation, we propose a Simple Self-Rewarding (SSR) Reinforcement Learning (RL) framework for MT that is reference-free, fully online, and relies solely on self-judging rewards. Training with SSR using 13K monolingual examples and Qwen-2.5-7B as the backbone, our model SSR-Zero-7B outperforms existing MT-specific LLMs, e.g., TowerInstruct-13B and GemmaX-28-9B, as well as larger general LLMs like Qwen2.5-32B-Instruct in English $\leftrightarrow$ Chinese translation tasks from WMT23, WMT24, and Flores200 benchmarks. Furthermore, by augmenting SSR with external supervision from COMET, our strongest model, SSR-X-Zero-7B, achieves state-of-the-art performance in English $\leftrightarrow$ Chinese translation, surpassing all existing open-source models under 72B parameters and even outperforming closed-source models, e.g., GPT-4o and Gemini 1.5 Pro. Our analysis highlights the effectiveness of the self-rewarding mechanism compared to the external LLM-as-a-judge approach in MT and demonstrates its complementary benefits when combined with trained RMs. Our findings provide valuable insight into the potential of self-improving RL methods. We have publicly released our code, data and models.
Authors: Boyi Wei, Benedikt Stroebl, Jiacen Xu, Joie Zhang, Zhou Li, Peter Henderson
Abstract: Foundation models are increasingly becoming better autonomous programmers, raising the prospect that they could also automate dangerous offensive cyber-operations. Current frontier model audits probe the cybersecurity risks of such agents, but most fail to account for the degrees of freedom available to adversaries in the real world. In particular, with strong verifiers and financial incentives, agents for offensive cybersecurity are amenable to iterative improvement by would-be adversaries. We argue that assessments should take into account an expanded threat model in the context of cybersecurity, emphasizing the varying degrees of freedom that an adversary may possess in stateful and non-stateful environments within a fixed compute budget. We show that even with a relatively small compute budget (8 H100 GPU Hours in our study), adversaries can improve an agent's cybersecurity capability on InterCode CTF by more than 40\% relative to the baseline -- without any external assistance. These results highlight the need to evaluate agents' cybersecurity risk in a dynamic manner, painting a more representative picture of risk.
Authors: Liqin Ye, Agam Shah, Chao Zhang, Sudheer Chava
Abstract: The traditional process of creating labeled datasets is labor-intensive and expensive. Recent breakthroughs in open-source large language models (LLMs) have opened up a new avenue in generating labeled datasets automatically for various natural language processing (NLP) tasks, providing an alternative to such an expensive annotation process. However, the reliability of such auto-generated labels remains a significant concern due to inherent inaccuracies. When learning from noisy labels, the model's generalization is likely to be harmed as it is prone to overfit to those label noises. While previous studies in learning from noisy labels mainly focus on synthetic noise and real-world noise, LLM-generated label noise receives less attention. In this paper, we propose SiDyP: Simplex Label Diffusion with Dynamic Prior to calibrate the classifier's prediction, thus enhancing its robustness towards LLM-generated noisy labels. SiDyP retrieves potential true label candidates by neighborhood label distribution in text embedding space and iteratively refines noisy candidates using a simplex diffusion model. Our framework can increase the performance of the BERT classifier fine-tuned on both zero-shot and few-shot LLM-generated noisy label datasets by an average of 7.21% and 7.30% respectively. We demonstrate the effectiveness of SiDyP by conducting extensive benchmarking for different LLMs over a variety of NLP tasks. Our code is available on Github.
Authors: Chengzhi Liu, Zhongxing Xu, Qingyue Wei, Juncheng Wu, James Zou, Xin Eric Wang, Yuyin Zhou, Sheng Liu
Abstract: Test-time compute has empowered multimodal large language models to generate extended reasoning chains, yielding strong performance on tasks such as multimodal math reasoning. However, this improved reasoning ability often comes with increased hallucination: as generations become longer, models tend to drift away from image-grounded content and rely more heavily on language priors. Attention analysis shows that longer reasoning chains lead to reduced focus on visual inputs, which contributes to hallucination. To systematically study this phenomenon, we introduce RH-AUC, a metric that quantifies how a model's perception accuracy changes with reasoning length, allowing us to evaluate whether the model preserves visual grounding during reasoning. We also release RH-Bench, a diagnostic benchmark that spans a variety of multimodal tasks, designed to assess the trade-off between reasoning ability and hallucination. Our analysis reveals that (i) larger models typically achieve a better balance between reasoning and perception, and (ii) this balance is influenced more by the types and domains of training data than by its overall volume. These findings underscore the importance of evaluation frameworks that jointly consider both reasoning quality and perceptual fidelity.
Authors: Runmin Jiang, Genpei Zhang, Yuntian Yang, Siqi Wu, Yuheng Zhang, Wanyue Feng, Yizhou Zhao, Xi Xiao, Xiao Wang, Tianyang Wang, Xingjian Li, Min Xu
Abstract: Cryo-electron microscopy (cryo-EM) offers near-atomic resolution imaging of macromolecules, but developing robust models for downstream analysis is hindered by the scarcity of high-quality annotated data. While synthetic data generation has emerged as a potential solution, existing methods often fail to capture both the structural diversity of biological specimens and the complex, spatially varying noise inherent in cryo-EM imaging. To overcome these limitations, we propose CryoCCD, a synthesis framework that integrates biophysical modeling with generative techniques. Specifically, CryoCCD produces multi-scale cryo-EM micrographs that reflect realistic biophysical variability through compositional heterogeneity, cellular context, and physics-informed imaging. To generate realistic noise, we employ a conditional diffusion model, enhanced by cycle consistency to preserve structural fidelity and mask-aware contrastive learning to capture spatially adaptive noise patterns. Extensive experiments show that CryoCCD generates structurally accurate micrographs and enhances performance in downstream tasks, outperforming state-of-the-art baselines in both particle picking and reconstruction.
Authors: Junaid Muzaffar, Khubaib Ahmed, Ingo Frommholz, Zeeshan Pervez, Ahsan ul Haq
Abstract: Training reinforcement learning (RL) agents often requires significant computational resources and prolonged training durations. To address this challenge, we build upon prior work that introduced a neural architecture with permutation-invariant sensory processing. We propose a modified attention mechanism that applies a non-linear transformation to the key vectors (K), producing enriched representations (K') through a custom mapping function. This Nonlinear Attention (NLA) mechanism enhances the representational capacity of the attention layer, enabling the agent to learn more expressive feature interactions. As a result, our model achieves significantly faster convergence and improved training efficiency, while maintaining performance on par with the baseline. These results highlight the potential of nonlinear attention mechanisms to accelerate reinforcement learning without sacrificing effectiveness.
Authors: Wenhao Song, Xuan Wu, Bo Yang, You Zhou, Yubin Xiao, Yanchun Liang, Hongwei Ge, Heow Pueh Lee, Chunguo Wu
Abstract: To address the weight coupling problem, certain studies introduced few-shot Neural Architecture Search (NAS) methods, which partition the supernet into multiple sub-supernets. However, these methods often suffer from computational inefficiency and tend to provide suboptimal partitioning schemes. To address this problem more effectively, we analyze the weight coupling problem from a novel perspective, which primarily stems from distinct modules in succeeding layers imposing conflicting gradient directions on the preceding layer modules. Based on this perspective, we propose the Gradient Contribution (GC) method that efficiently computes the cosine similarity of gradient directions among modules by decomposing the Vector-Jacobian Product during supernet backpropagation. Subsequently, the modules with conflicting gradient directions are allocated to distinct sub-supernets while similar ones are grouped together. To assess the advantages of GC and address the limitations of existing Graph Neural Architecture Search methods, which are limited to searching a single type of Graph Neural Networks (Message Passing Neural Networks (MPNNs) or Graph Transformers (GTs)), we propose the Unified Graph Neural Architecture Search (UGAS) framework, which explores optimal combinations of MPNNs and GTs. The experimental results demonstrate that GC achieves state-of-the-art (SOTA) performance in supernet partitioning quality and time efficiency. In addition, the architectures searched by UGAS+GC outperform both the manually designed GNNs and those obtained by existing NAS methods. Finally, ablation studies further demonstrate the effectiveness of all proposed methods.
Authors: Yilin Xiao, Junnan Dong, Chuang Zhou, Su Dong, Qian-wen Zhang, Di Yin, Xing Sun, Xiao Huang
Abstract: Graph Retrieval Augmented Generation (GraphRAG) has garnered increasing recognition for its potential to enhance large language models (LLMs) by structurally organizing domain-specific corpora and facilitating complex reasoning. However, current evaluations of GraphRAG models predominantly rely on traditional question-answering datasets. Their limited scope in questions and evaluation metrics fails to comprehensively assess the reasoning capacity improvements enabled by GraphRAG models. To address this gap, we introduce GraphRAG-Bench, a large-scale, domain-specific benchmark designed to rigorously evaluate GraphRAG models. Our benchmark offers three key superiorities: \((i)\) Challenging question design. Featuring college-level, domain-specific questions that demand multi-hop reasoning, the benchmark ensures that simple content retrieval is insufficient for problem-solving. For example, some questions require mathematical reasoning or programming. \((ii)\) Diverse task coverage. The dataset includes a broad spectrum of reasoning tasks, multiple-choice, true/false, multi-select, open-ended, and fill-in-the-blank. It spans 16 disciplines in twenty core textbooks. \((iii)\) Holistic evaluation framework. GraphRAG-Bench provides comprehensive assessment across the entire GraphRAG pipeline, including graph construction, knowledge retrieval, and answer generation. Beyond final-answer correctness, it evaluates the logical coherence of the reasoning process. By applying nine contemporary GraphRAG methods to GraphRAG-Bench, we demonstrate its utility in quantifying how graph-based structuring improves model reasoning capabilities. Our analysis reveals critical insights about graph architectures, retrieval efficacy, and reasoning capabilities, offering actionable guidance for the research community.
Authors: Jiahao Wang, Jinbo Han, Xingda Wei, Sijie Shen, Dingyan Zhang, Chenguang Fang, Rong Chen, Wenyuan Yu, Haibo Chen
Abstract: Serving large language models (LLMs) is important for cloud providers, and caching intermediate results (KV\$) after processing each request substantially improves serving throughput and latency. However, there is limited understanding of how LLM serving benefits from KV\$ caching, where system design decisions like cache eviction policies are highly workload-dependent. In this paper, we present the first systematic characterization of the KV\$ workload patterns from one of the leading LLM service providers. We draw observations that were not covered by previous studies focusing on synthetic workloads, including: KV\$ reuses are skewed across requests, where reuses between single-turn requests are equally important as multi-turn requests; the reuse time and probability are diverse considering all requests, but for a specific request category, the pattern tends to be predictable; and the overall cache size required for an ideal cache hit ratio is moderate. Based on the characterization, we further propose a workload-aware cache eviction policy that improves the serving performance under real-world traces, especially with limited cache capacity.
Authors: Bin Lin, Zongjian Li, Xinhua Cheng, Yuwei Niu, Yang Ye, Xianyi He, Shenghai Yuan, Wangbo Yu, Shaodong Wang, Yunyang Ge, Yatian Pang, Li Yuan
Abstract: Although existing unified models achieve strong performance in vision-language understanding and text-to-image generation, they remain limited in addressing image perception and manipulation -- capabilities increasingly demanded in practical applications. Recently, OpenAI introduced the powerful GPT-4o-Image model, which showcases advanced capabilities in comprehensive image perception and manipulation, sparking widespread interest. Through carefully designed experiments, we observe that GPT-4o-Image likely relies on semantic encoders rather than VAEs for feature extraction, despite VAEs being commonly regarded as crucial for image manipulation tasks. Inspired by this insight, we propose UniWorld-V1, a unified generative framework built upon semantic features extracted from powerful multimodal large language models and contrastive semantic encoders. Using only 2.7M training data, UniWorld-V1 achieves impressive performance across diverse tasks, including image understanding, generation, manipulation, and perception. We fully open-source the UniWorld-V1 framework, including model weights, training and evaluation scripts, and datasets to promote reproducibility and further research.
Authors: Xinghang Li, Jingzhe Ding, Chao Peng, Bing Zhao, Xiang Gao, Hongwan Gao, Xinchen Gu
Abstract: The code generation capabilities of large language models(LLMs) have emerged as a critical dimension in evaluating their overall performance. However, prior research has largely overlooked the security risks inherent in the generated code. In this work, we introduce SafeGenBench, a benchmark specifically designed to assess the security of LLM-generated code. The dataset encompasses a wide range of common software development scenarios and vulnerability types. Building upon this benchmark, we develop an automatic evaluation framework that leverages both static application security testing(SAST) and LLM-based judging to assess the presence of security vulnerabilities in model-generated code. Through the empirical evaluation of state-of-the-art LLMs on SafeGenBench, we reveal notable deficiencies in their ability to produce vulnerability-free code. Our findings highlight pressing challenges and offer actionable insights for future advancements in the secure code generation performance of LLMs. The data and code will be released soon.
Authors: Wenxuan Xie, Yaxun Dai, Wenhao Jiang
Abstract: Recent advancements in large language models (LLMs) have significantly improved performance on the Text-to-SQL task. However, prior approaches typically rely on static, pre-processed database information provided at inference time, which limits the model's ability to fully understand the database contents. Without dynamic interaction, LLMs are constrained to fixed, human-provided context and cannot autonomously explore the underlying data. To address this limitation, we propose SDE-SQL, a framework that enables large language models to perform self-driven exploration of databases during inference. This is accomplished by generating and executing SQL probes, which allow the model to actively retrieve information from the database and iteratively update its understanding of the data. Unlike prior methods, SDE-SQL operates in a zero-shot setting, without relying on any question-SQL pairs as in-context demonstrations. When evaluated on the BIRD benchmark with Qwen2.5-72B-Instruct, SDE-SQL achieves an 8.02% relative improvement in execution accuracy over the vanilla Qwen2.5-72B-Instruct baseline, establishing a new state-of-the-art among methods based on open-source models without supervised fine-tuning (SFT) or model ensembling. Moreover, with SFT, the performance of SDE-SQL can be further enhanced, yielding an additional 0.52% improvement.
Authors: Ziheng Qin, Hailun Xu, Wei Chee Yew, Qi Jia, Yang Luo, Kanchan Sarkar, Danhui Guan, Kai Wang, Yang You
Abstract: Machine learning relies heavily on data, yet the continuous growth of real-world data poses challenges for efficient dataset construction and training. A fundamental yet unsolved question is: given our current model and data, does a new data (sample/batch) need annotation/learning? Conventional approaches retain all available data, leading to non-optimal data and training efficiency. Active learning aims to reduce data redundancy by selecting a subset of samples to annotate, while it increases pipeline complexity and introduces bias. In this work, we propose Info-Coevolution, a novel framework that efficiently enables models and data to coevolve through online selective annotation with no bias. Leveraging task-specific models (and open-source models), it selectively annotates and integrates online and web data to improve datasets efficiently. For real-world datasets like ImageNet-1K, Info-Coevolution reduces annotation and training costs by 32\% without performance loss. It is able to automatically give the saving ratio without tuning the ratio. It can further reduce the annotation ratio to 50\% with semi-supervised learning. We also explore retrieval-based dataset enhancement using unlabeled open-source data. Code is available at https://github.com/NUS-HPC-AI-Lab/Info-Coevolution/.
Authors: Hiba Khey, Amine Lakhder, Salma Rouichi, Imane El Ghabi, Kamal Hejjaoui, Younes En-nahli, Fahd Kalloubi, Moez Amri
Abstract: The rapid advancement of transformer-based language models has catalyzed breakthroughs in biomedical and clinical natural language processing; however, plant science remains markedly underserved by such domain-adapted tools. In this work, we present PlantBert, a high-performance, open-source language model specifically tailored for extracting structured knowledge from plant stress-response literature. Built upon the DeBERTa architecture-known for its disentangled attention and robust contextual encoding-PlantBert is fine-tuned on a meticulously curated corpus of expert-annotated abstracts, with a primary focus on lentil (Lens culinaris) responses to diverse abiotic and biotic stressors. Our methodology combines transformer-based modeling with rule-enhanced linguistic post-processing and ontology-grounded entity normalization, enabling PlantBert to capture biologically meaningful relationships with precision and semantic fidelity. The underlying corpus is annotated using a hierarchical schema aligned with the Crop Ontology, encompassing molecular, physiological, biochemical, and agronomic dimensions of plant adaptation. PlantBert exhibits strong generalization capabilities across entity types and demonstrates the feasibility of robust domain adaptation in low-resource scientific fields. By providing a scalable and reproducible framework for high-resolution entity recognition, PlantBert bridges a critical gap in agricultural NLP and paves the way for intelligent, data-driven systems in plant genomics, phenomics, and agronomic knowledge discovery. Our model is publicly released to promote transparency and accelerate cross-disciplinary innovation in computational plant science.
Authors: Junfei Wu, Jian Guan, Kaituo Feng, Qiang Liu, Shu Wu, Liang Wang, Wei Wu, Tieniu Tan
Abstract: As textual reasoning with large language models (LLMs) has advanced significantly, there has been growing interest in enhancing the multimodal reasoning capabilities of large vision-language models (LVLMs). However, existing methods primarily approach multimodal reasoning in a straightforward, text-centric manner, where both reasoning and answer derivation are conducted purely through text, with the only difference being the presence of multimodal input. As a result, these methods often encounter fundamental limitations in spatial reasoning tasks that demand precise geometric understanding and continuous spatial tracking-capabilities that humans achieve through mental visualization and manipulation. To address the limitations, we propose drawing to reason in space, a novel paradigm that enables LVLMs to reason through elementary drawing operations in the visual space. By equipping models with basic drawing operations, including annotating bounding boxes and drawing auxiliary lines, we empower them to express and analyze spatial relationships through direct visual manipulation, meanwhile avoiding the performance ceiling imposed by specialized perception tools in previous tool-integrated reasoning approaches. To cultivate this capability, we develop a three-stage training framework: cold-start training with synthetic data to establish basic drawing abilities, reflective rejection sampling to enhance self-reflection behaviors, and reinforcement learning to directly optimize for target rewards. Extensive experiments demonstrate that our model, named VILASR, consistently outperforms existing methods across diverse spatial reasoning benchmarks, involving maze navigation, static spatial reasoning, video-based reasoning, and multi-view-based reasoning tasks, with an average improvement of 18.4%.
Authors: Akshar Tumu, Henrik I. Christensen, Marcell Vazquez-Chanlatte, Chikao Tsuchiya, Dhaval Bhanderi
Abstract: Lane-topology prediction is a critical component of safe and reliable autonomous navigation. An accurate understanding of the road environment aids this task. We observe that this information often follows conventions encoded in natural language, through design codes that reflect the road structure and road names that capture the road functionality. We augment this information in a lightweight manner to SMERF, a map-prior-based online lane-topology prediction model, by combining structured road metadata from OSM maps and lane-width priors from Road design manuals with the road centerline encodings. We evaluate our method on two geo-diverse complex intersection scenarios. Our method shows improvement in both lane and traffic element detection and their association. We report results using four topology-aware metrics to comprehensively assess the model performance. These results demonstrate the ability of our approach to generalize and scale to diverse topologies and conditions.
Authors: Lianghong Guo, Yanlin Wang, Caihua Li, Pengyu Yang, Jiachi Chen, Wei Tao, Yingtian Zou, Duyu Tang, Zibin Zheng
Abstract: Constructing large-scale datasets for the GitHub issue resolution task is crucial for both training and evaluating the software engineering capabilities of Large Language Models (LLMs). However, the traditional process for creating such benchmarks is notoriously challenging and labor-intensive, particularly in the stages of setting up evaluation environments, grading test outcomes, and validating task instances. In this paper, we propose SWE-Factory, an automated pipeline designed to address these challenges. To tackle these issues, our pipeline integrates three core automated components. First, we introduce SWE-Builder, a multi-agent system that automates evaluation environment construction, which employs four specialized agents that work in a collaborative, iterative loop and leverages an environment memory pool to enhance efficiency. Second, we introduce a standardized, exit-code-based grading method that eliminates the need for manually writing custom parsers. Finally, we automate the fail2pass validation process using these reliable exit code signals. Experiments on 671 issues across four programming languages show that our pipeline can effectively construct valid task instances; for example, with GPT-4.1-mini, our SWE-Builder constructs 269 valid instances at $0.045 per instance, while with Gemini-2.5-flash, it achieves comparable performance at the lowest cost of $0.024 per instance. We also demonstrate that our exit-code-based grading achieves 100% accuracy compared to manual inspection, and our automated fail2pass validation reaches a precision of 0.92 and a recall of 1.00. We hope our automated pipeline will accelerate the collection of large-scale, high-quality GitHub issue resolution datasets for both training and evaluation. Our code and datasets are released at https://github.com/DeepSoftwareAnalytics/swe-factory.
Authors: Barbara Oakley, Michael Johnston, Ken-Zen Chen, Eulho Jung, Terrence J. Sejnowski
Abstract: In the age of generative AI and ubiquitous digital tools, human cognition faces a structural paradox: as external aids become more capable, internal memory systems risk atrophy. Drawing on neuroscience and cognitive psychology, this paper examines how heavy reliance on AI systems and discovery-based pedagogies may impair the consolidation of declarative and procedural memory -- systems essential for expertise, critical thinking, and long-term retention. We review how tools like ChatGPT and calculators can short-circuit the retrieval, error correction, and schema-building processes necessary for robust neural encoding. Notably, we highlight striking parallels between deep learning phenomena such as "grokking" and the neuroscience of overlearning and intuition. Empirical studies are discussed showing how premature reliance on AI during learning inhibits proceduralization and intuitive mastery. We argue that effective human-AI interaction depends on strong internal models -- biological "schemata" and neural manifolds -- that enable users to evaluate, refine, and guide AI output. The paper concludes with policy implications for education and workforce training in the age of large language models.
Authors: Jin Kim, Muhammad Wahi-Anwa, Sangyun Park, Shawn Shin, John M. Hoffman, Matthew S. Brown
Abstract: Agentic Artificial Intelligence (AI) systems leveraging Large Language Models (LLMs) exhibit significant potential for complex reasoning, planning, and tool utilization. We demonstrate that a specialized computer vision system can be built autonomously from a natural language prompt using Agentic AI methods. This involved extending SimpleMind (SM), an open-source Cognitive AI environment with configurable tools for medical image analysis, with an LLM-based agent, implemented using OpenManus, to automate the planning (tool configuration) for a particular computer vision task. We provide a proof-of-concept demonstration that an agentic system can interpret a computer vision task prompt, plan a corresponding SimpleMind workflow by decomposing the task and configuring appropriate tools. From the user input prompt, "provide sm (SimpleMind) config for lungs, heart, and ribs segmentation for cxr (chest x-ray)"), the agent LLM was able to generate the plan (tool configuration file in YAML format), and execute SM-Learn (training) and SM-Think (inference) scripts autonomously. The computer vision agent automatically configured, trained, and tested itself on 50 chest x-ray images, achieving mean dice scores of 0.96, 0.82, 0.83, for lungs, heart, and ribs, respectively. This work shows the potential for autonomous planning and tool configuration that has traditionally been performed by a data scientist in the development of computer vision applications.
Authors: H\'ector Carri\'on, Yutong Bai, V\'ictor A. Hern\'andez Castro, Kishan Panaganti, Ayush Zenith, Matthew Trang, Tony Zhang, Pietro Perona, Jitendra Malik
Abstract: World models aim to simulate environments and enable effective agent behavior. However, modeling real-world environments presents unique challenges as they dynamically change across both space and, crucially, time. To capture these composed dynamics, we introduce a Spatio-Temporal Road Image Dataset for Exploration (STRIDE) permuting 360-degree panoramic imagery into rich interconnected observation, state and action nodes. Leveraging this structure, we can simultaneously model the relationship between egocentric views, positional coordinates, and movement commands across both space and time. We benchmark this dataset via TARDIS, a transformer-based generative world model that integrates spatial and temporal dynamics through a unified autoregressive framework trained on STRIDE. We demonstrate robust performance across a range of agentic tasks such as controllable photorealistic image synthesis, instruction following, autonomous self-control, and state-of-the-art georeferencing. These results suggest a promising direction towards sophisticated generalist agents--capable of understanding and manipulating the spatial and temporal aspects of their material environments--with enhanced embodied reasoning capabilities. Training code, datasets, and model checkpoints are made available at https://huggingface.co/datasets/Tera-AI/STRIDE.
Authors: Shikun Li, Shipeng Li, Zhiqin Yang, Xinghua Zhang, Gaode Chen, Xiaobo Xia, Hengyu Liu, Zhe Peng
Abstract: Reinforcement learning (RL) has become a key technique for enhancing LLMs' reasoning abilities, yet its data inefficiency remains a major bottleneck. To address this critical yet challenging issue, we present a novel gradient-alignment-based method, named LearnAlign, which intelligently selects the learnable and representative training reasoning data for RL post-training. To overcome the issue of response-length bias in gradient norms, we introduce the data learnability based on the success rate, which can indicate the learning potential of each data point. Experiments across three mathematical reasoning benchmarks demonstrate that our method significantly reduces training data requirements while achieving minor performance degradation or even improving performance compared to full-data training. For example, it reduces data requirements by up to 1,000 data points with better performance (77.53%) than that on the full dataset on GSM8K benchmark (77.04%). Furthermore, we show its effectiveness in the staged RL setting. This work provides valuable insights into data-efficient RL post-training and establishes a foundation for future research in optimizing reasoning data selection. To facilitate future work, we will release code.
Authors: Anna Soligo, Edward Turner, Senthooran Rajamanoharan, Neel Nanda
Abstract: Fine-tuning large language models on narrow datasets can cause them to develop broadly misaligned behaviours: a phenomena known as emergent misalignment. However, the mechanisms underlying this misalignment, and why it generalizes beyond the training domain, are poorly understood, demonstrating critical gaps in our knowledge of model alignment. In this work, we train and study a minimal model organism which uses just 9 rank-1 adapters to emergently misalign Qwen2.5-14B-Instruct. Studying this, we find that different emergently misaligned models converge to similar representations of misalignment. We demonstrate this convergence by extracting a 'misalignment direction' from one fine-tuned model's activations, and using it to effectively ablate misaligned behaviour from fine-tunes using higher dimensional LoRAs and different datasets. Leveraging the scalar hidden state of rank-1 LoRAs, we further present a set of experiments for directly interpreting the fine-tuning adapters, showing that six contribute to general misalignment, while two specialise for misalignment in just the fine-tuning domain. Emergent misalignment is a particularly salient example of undesirable and unexpected model behaviour and by advancing our understanding of the mechanisms behind it, we hope to move towards being able to better understand and mitigate misalignment more generally.
Authors: Dylan Kline
Abstract: This study bridges cognitive science and neural network design by examining whether artificial models exhibit human-like forgetting curves. Drawing upon Ebbinghaus' seminal work on memory decay and principles of spaced repetition, we propose a quantitative framework to measure information retention in neural networks. Our approach computes the recall probability by evaluating the similarity between a network's current hidden state and previously stored prototype representations. This retention metric facilitates the scheduling of review sessions, thereby mitigating catastrophic forgetting during deployment and enhancing training efficiency by prompting targeted reviews. Our experiments with Multi-Layer Perceptrons reveal human-like forgetting curves, with knowledge becoming increasingly robust through scheduled reviews. This alignment between neural network forgetting curves and established human memory models identifies neural networks as an architecture that naturally emulates human memory decay and can inform state-of-the-art continual learning algorithms.
Authors: Yanting Miao, William Loh, Suraj Kothawade, Pacal Poupart
Abstract: Recent work uses reinforcement learning (RL) to fine-tune text-to-image diffusion models, improving text-image alignment and sample quality. However, existing approaches introduce unnecessary complexity: they cache the full sampling trajectory, depend on differentiable reward models or large preference datasets, or require specialized guidance techniques. Motivated by the "golden noise" hypothesis -- that certain initial noise samples can consistently yield superior alignment -- we introduce Noise PPO, a minimalist RL algorithm that leaves the pre-trained diffusion model entirely frozen and learns a prompt-conditioned initial noise generator. Our approach requires no trajectory storage, reward backpropagation, or complex guidance tricks. Extensive experiments show that optimizing the initial noise distribution consistently improves alignment and sample quality over the original model, with the most significant gains at low inference steps. As the number of inference steps increases, the benefit of noise optimization diminishes but remains present. These findings clarify the scope and limitations of the golden noise hypothesis and reinforce the practical value of minimalist RL fine-tuning for diffusion models.
Authors: Benjamin Marais, Tony Quertier, Gr\'egoire Barrue
Abstract: In a context of malware analysis, numerous approaches rely on Artificial Intelligence to handle a large volume of data. However, these techniques focus on data view (images, sequences) and not on an expert's view. Noticing this issue, we propose a preprocessing that focuses on expert knowledge to improve malware semantic analysis and result interpretability. We propose a new preprocessing method which creates JSON reports for Portable Executable files. These reports gather features from both static and behavioral analysis, and incorporate packer signature detection, MITRE ATT\&CK and Malware Behavior Catalog (MBC) knowledge. The purpose of this preprocessing is to gather a semantic representation of binary files, understandable by malware analysts, and that can enhance AI models' explainability for malicious files analysis. Using this preprocessing to train a Large Language Model for Malware classification, we achieve a weighted-average F1-score of 0.94 on a complex dataset, representative of market reality.
Authors: Naomi Fridman, Bubby Solway, Tomer Fridman, Itamar Barnea, Anat Goldstein
Abstract: Breast cancer remains a leading cause of cancer-related mortality worldwide, making early detection and accurate treatment response monitoring critical priorities. We present BreastDCEDL, a curated, deep learning-ready dataset comprising pre-treatment 3D Dynamic Contrast-Enhanced MRI (DCE-MRI) scans from 2,070 breast cancer patients drawn from the I-SPY1, I-SPY2, and Duke cohorts, all sourced from The Cancer Imaging Archive. The raw DICOM imaging data were rigorously converted into standardized 3D NIfTI volumes with preserved signal integrity, accompanied by unified tumor annotations and harmonized clinical metadata including pathologic complete response (pCR), hormone receptor (HR), and HER2 status. Although DCE-MRI provides essential diagnostic information and deep learning offers tremendous potential for analyzing such complex data, progress has been limited by lack of accessible, public, multicenter datasets. BreastDCEDL addresses this gap by enabling development of advanced models, including state-of-the-art transformer architectures that require substantial training data. To demonstrate its capacity for robust modeling, we developed the first transformer-based model for breast DCE-MRI, leveraging Vision Transformer (ViT) architecture trained on RGB-fused images from three contrast phases (pre-contrast, early post-contrast, and late post-contrast). Our ViT model achieved state-of-the-art pCR prediction performance in HR+/HER2- patients (AUC 0.94, accuracy 0.93). BreastDCEDL includes predefined benchmark splits, offering a framework for reproducible research and enabling clinically meaningful modeling in breast cancer imaging.
Authors: Hantao Yu, Josh Alman
Abstract: The quadratic complexity of self-attention prevents transformers from scaling effectively to long input sequences. On the other hand, modern GPUs and other specialized hardware accelerators are well-optimized for processing small input sequences in transformers during both training and inference. A natural question arises: can we take advantage of the efficiency of small transformers to deal with long input sequences? In this paper, we show that transformers with long input sequences (large transformers) can be efficiently simulated by transformers that can only take short input sequences (small transformers). Specifically, we prove that any transformer with input length $N$ can be efficiently simulated by only $O((N/M)^2)$ transformers with input length $M \ll N$, and that this cannot be improved in the worst case. However, we then prove that in various natural scenarios including average-case inputs, sliding window masking and attention sinks, the optimal number $O(N/M)$ of small transformers suffice.
Authors: Xiaotian Zhang, Yuan Wang, Zhaopeng Feng, Ruizhe Chen, Zhijie Zhou, Yan Zhang, Hongxia Xu, Jian Wu, Zuozhu Liu
Abstract: Medical Question-Answering (QA) encompasses a broad spectrum of tasks, including multiple choice questions (MCQ), open-ended text generation, and complex computational reasoning. Despite this variety, a unified framework for delivering high-quality medical QA has yet to emerge. Although recent progress in reasoning-augmented large language models (LLMs) has shown promise, their ability to achieve comprehensive medical understanding is still largely unexplored. In this paper, we present Med-U1, a unified framework for robust reasoning across medical QA tasks with diverse output formats, ranging from MCQs to complex generation and computation tasks. Med-U1 employs pure large-scale reinforcement learning with mixed rule-based binary reward functions, incorporating a length penalty to manage output verbosity. With multi-objective reward optimization, Med-U1 directs LLMs to produce concise and verifiable reasoning chains. Empirical results reveal that Med-U1 significantly improves performance across multiple challenging Med-QA benchmarks, surpassing even larger specialized and proprietary models. Furthermore, Med-U1 demonstrates robust generalization to out-of-distribution (OOD) tasks. Extensive analysis presents insights into training strategies, reasoning chain length control, and reward design for medical LLMs. Our code is available here.
Authors: Johnny Peng, Thanh Tung Khuat, Katarzyna Musial, Bogdan Gabrys
Abstract: Data is crucial for machine learning (ML) applications, yet acquiring large datasets can be costly and time-consuming, especially in complex, resource-intensive fields like biopharmaceuticals. A key process in this industry is upstream bioprocessing, where living cells are cultivated and optimised to produce therapeutic proteins and biologics. The intricate nature of these processes, combined with high resource demands, often limits data collection, resulting in smaller datasets. This comprehensive review explores ML methods designed to address the challenges posed by small data and classifies them into a taxonomy to guide practical applications. Furthermore, each method in the taxonomy was thoroughly analysed, with a detailed discussion of its core concepts and an evaluation of its effectiveness in tackling small data challenges, as demonstrated by application results in the upstream bioprocessing and other related domains. By analysing how these methods tackle small data challenges from different perspectives, this review provides actionable insights, identifies current research gaps, and offers guidance for leveraging ML in data-constrained environments.
Authors: Pengfei Zuo, Huimin Lin, Junbo Deng, Nan Zou, Xingkun Yang, Yingyu Diao, Weifeng Gao, Ke Xu, Zhangyu Chen, Shirui Lu, Zhao Qiu, Peiyang Li, Xianyu Chang, Zhengzhong Yu, Fangzheng Miao, Jia Zheng, Ying Li, Yuan Feng, Bei Wang, Zaijian Zong, Mosong Zhou, Wenli Zhou, Houjiang Chen, Xingyu Liao, Yipeng Li, Wenxiao Zhang, Ping Zhu, Yinggang Wang, Chuanjie Xiao, Depeng Liang, Dong Cao, Juncheng Liu, Yongqiang Yang, Xiaolong Bai, Yi Li, Huaguo Xie, Huatao Wu, Zhibin Yu, Lv Chen, Hu Liu, Yujun Ding, Haipei Zhu, Jing Xia, Yi Xiong, Zhou Yu, Heng Liao
Abstract: The rapid evolution of large language models (LLMs), driven by growing parameter scales, adoption of mixture-of-experts (MoE) architectures, and expanding context lengths, imposes unprecedented demands on AI infrastructure. Traditional AI clusters face limitations in compute intensity, memory bandwidth, inter-chip communication, and latency, compounded by variable workloads and strict service-level objectives. Addressing these issues requires fundamentally redesigned hardware-software integration. This paper introduces Huawei CloudMatrix, a next-generation AI datacenter architecture, realized in the production-grade CloudMatrix384 supernode. It integrates 384 Ascend 910 NPUs and 192 Kunpeng CPUs interconnected via an ultra-high-bandwidth Unified Bus (UB) network, enabling direct all-to-all communication and dynamic pooling of resources. These features optimize performance for communication-intensive operations, such as large-scale MoE expert parallelism and distributed key-value cache access. To fully leverage CloudMatrix384, we propose CloudMatrix-Infer, an advanced LLM serving solution incorporating three core innovations: a peer-to-peer serving architecture that independently scales prefill, decode, and caching; a large-scale expert parallelism strategy supporting EP320 via efficient UB-based token dispatch; and hardware-aware optimizations including specialized operators, microbatch-based pipelining, and INT8 quantization. Evaluation with the DeepSeek-R1 model shows CloudMatrix-Infer achieves state-of-the-art efficiency: prefill throughput of 6,688 tokens/s per NPU and decode throughput of 1,943 tokens/s per NPU (<50 ms TPOT). It effectively balances throughput and latency, sustaining 538 tokens/s per NPU even under stringent 15 ms latency constraints, while INT8 quantization maintains model accuracy across benchmarks.
Authors: Ye Li, Yuan Meng, Zewen Sun, Kangye Ji, Chen Tang, Jiajun Fan, Xinzhu Ma, Shutao Xia, Zhi Wang, Wenwu Zhu
Abstract: Vision-Language-Action (VLA) models have attracted increasing attention for their strong control capabilities. However, their high computational cost and low execution frequency hinder their suitability for real-time tasks such as robotic manipulation and autonomous navigation. Existing VLA acceleration methods primarily focus on structural optimization, overlooking the fact that these models operate in sequential decision-making environments. As a result, temporal redundancy in sequential action generation and spatial redundancy in visual input remain unaddressed. To this end, we propose SP-VLA, a unified framework that accelerates VLA models by jointly scheduling models and pruning tokens. Specifically, we design an action-aware model scheduling mechanism that reduces temporal redundancy by dynamically switching between VLA model and a lightweight generator. Inspired by the human motion pattern of focusing on key decision points while relying on intuition for other actions, we categorize VLA actions into deliberative and intuitive, assigning the former to the VLA model and the latter to the lightweight generator, enabling frequency-adaptive execution through collaborative model scheduling. To address spatial redundancy, we further develop a spatio-semantic dual-aware token pruning method. Tokens are classified into spatial and semantic types and pruned based on their dual-aware importance to accelerate VLA inference. These two mechanisms work jointly to guide the VLA in focusing on critical actions and salient visual information, achieving effective acceleration while maintaining high accuracy. Experimental results demonstrate that our method achieves up to 1.5$\times$ acceleration with less than 3% drop in accuracy, outperforming existing approaches in multiple tasks.
Authors: Xuan Wang, Siyuan Liang, Zhe Liu, Yi Yu, Yuliang Lu, Xiaochun Cao, Ee-Chien Chang
Abstract: With the growing integration of vision-language models (VLMs), mobile agents are now widely used for tasks like UI automation and camera-based user assistance. These agents are often fine-tuned on limited user-generated datasets, leaving them vulnerable to covert threats during the training process. In this work we present GHOST, the first clean-label backdoor attack specifically designed for mobile agents built upon VLMs. Our method manipulates only the visual inputs of a portion of the training samples - without altering their corresponding labels or instructions - thereby injecting malicious behaviors into the model. Once fine-tuned with this tampered data, the agent will exhibit attacker-controlled responses when a specific visual trigger is introduced at inference time. The core of our approach lies in aligning the gradients of poisoned samples with those of a chosen target instance, embedding backdoor-relevant features into the poisoned training data. To maintain stealth and enhance robustness, we develop three realistic visual triggers: static visual patches, dynamic motion cues, and subtle low-opacity overlays. We evaluate our method across six real-world Android apps and three VLM architectures adapted for mobile use. Results show that our attack achieves high attack success rates (up to 94.67 percent) while maintaining high clean-task performance (FSR up to 95.85 percent). Additionally, ablation studies shed light on how various design choices affect the efficacy and concealment of the attack. Overall, this work is the first to expose critical security flaws in VLM-based mobile agents, highlighting their susceptibility to clean-label backdoor attacks and the urgent need for effective defense mechanisms in their training pipelines.
Authors: Haoru Xue, Xiaoyu Huang, Dantong Niu, Qiayuan Liao, Thomas Kragerud, Jan Tommy Gravdahl, Xue Bin Peng, Guanya Shi, Trevor Darrell, Koushil Screenath, Shankar Sastry
Abstract: Vision-language-action (VLA) models have demonstrated strong semantic understanding and zero-shot generalization, yet most existing systems assume an accurate low-level controller with hand-crafted action "vocabulary" such as end-effector pose or root velocity. This assumption confines prior work to quasi-static tasks and precludes the agile, whole-body behaviors required by humanoid whole-body control (WBC) tasks. To capture this gap in the literature, we start by introducing the first sim-to-real-ready, vision-language, closed-loop benchmark for humanoid WBC, comprising over 150 tasks from 10 categories. We then propose LeVERB: Latent Vision-Language-Encoded Robot Behavior, a hierarchical latent instruction-following framework for humanoid vision-language WBC, the first of its kind. At the top level, a vision-language policy learns a latent action vocabulary from synthetically rendered kinematic demonstrations; at the low level, a reinforcement-learned WBC policy consumes these latent verbs to generate dynamics-level commands. In our benchmark, LeVERB can zero-shot attain a 80% success rate on simple visual navigation tasks, and 58.5% success rate overall, outperforming naive hierarchical whole-body VLA implementation by 7.8 times.
Authors: Vaskar Nath, Elaine Lau, Anisha Gunjal, Manasi Sharma, Nikhil Baharte, Sean Hendryx
Abstract: We study the process through which reasoning models trained with reinforcement learning on verifiable rewards (RLVR) can learn to solve new problems. We find that RLVR drives performance in two main ways: (1) by compressing pass@$k$ into pass@1 and (2) via "capability gain" in which models learn to solve new problems that they previously could not solve even at high $k$. We find that while capability gain exists across model scales, learning to solve new problems is primarily driven through self-distillation. We demonstrate these findings across model scales ranging from 0.5B to 72B parameters on >500,000 reasoning problems with prompts and verifiable final answers across math, science, and code domains. We further show that we can significantly improve pass@$k$ rates by leveraging natural language guidance for the model to consider within context while still requiring the model to derive a solution chain from scratch. Based of these insights, we derive $\text{Guide}$ -- a new class of online training algorithms. $\text{Guide}$ adaptively incorporates hints into the model's context on problems for which all rollouts were initially incorrect and adjusts the importance sampling ratio for the "off-policy" trajectories in order to optimize the policy for contexts in which the hints are no longer present. We describe variants of $\text{Guide}$ for GRPO and PPO and empirically show that Guide-GRPO on 7B and 32B parameter models improves generalization over its vanilla counterpart with up to 4$\%$ macro-average improvement across math benchmarks. We include careful ablations to analyze $\text{Guide}$'s components and theoretically analyze Guide's learning efficiency.
Authors: Essential AI, :, Andrew Hojel, Michael Pust, Tim Romanski, Yash Vanjani, Ritvik Kapila, Mohit Parmar, Adarsh Chaluvaraju, Alok Tripathy, Anil Thomas, Ashish Tanwer, Darsh J Shah, Ishaan Shah, Karl Stratos, Khoi Nguyen, Kurt Smith, Michael Callahan, Peter Rushton, Philip Monk, Platon Mazarakis, Saad Jamal, Saurabh Srivastava, Somanshu Singla, Ashish Vaswani
Abstract: Data plays the most prominent role in how language models acquire skills and knowledge. The lack of massive, well-organized pre-training datasets results in costly and inaccessible data pipelines. We present Essential-Web v1.0, a 24-trillion-token dataset in which every document is annotated with a twelve-category taxonomy covering topic, format, content complexity, and quality. Taxonomy labels are produced by EAI-Distill-0.5b, a fine-tuned 0.5b-parameter model that achieves an annotator agreement within 3% of Qwen2.5-32B-Instruct. With nothing more than SQL-style filters, we obtain competitive web-curated datasets in math (-8.0% relative to SOTA), web code (+14.3%), STEM (+24.5%) and medical (+8.6%). Essential-Web v1.0 is available on HuggingFace: https://huggingface.co/datasets/EssentialAI/essential-web-v1.0
URLs: https://huggingface.co/datasets/EssentialAI/essential-web-v1.0
Authors: Tian Xia, Fabio De Sousa Ribeiro, Rajat R Rasal, Avinash Kori, Raghav Mehta, Ben Glocker
Abstract: Counterfactual image generation aims to simulate realistic visual outcomes under specific causal interventions. Diffusion models have recently emerged as a powerful tool for this task, combining DDIM inversion with conditional generation via classifier-free guidance (CFG). However, standard CFG applies a single global weight across all conditioning variables, which can lead to poor identity preservation and spurious attribute changes - a phenomenon known as attribute amplification. To address this, we propose Decoupled Classifier-Free Guidance (DCFG), a flexible and model-agnostic framework that introduces group-wise conditioning control. DCFG builds on an attribute-split embedding strategy that disentangles semantic inputs, enabling selective guidance on user-defined attribute groups. For counterfactual generation, we partition attributes into intervened and invariant sets based on a causal graph and apply distinct guidance to each. Experiments on CelebA-HQ, MIMIC-CXR, and EMBED show that DCFG improves intervention fidelity, mitigates unintended changes, and enhances reversibility, enabling more faithful and interpretable counterfactual image generation.
Authors: Aditya Bhattacharjee, Ivan Meresman Higgs, Mark Sandler, Emmanouil Benetos
Abstract: Automatic sample identification (ASID), the detection and identification of portions of audio recordings that have been reused in new musical works, is an essential but challenging task in the field of audio query-based retrieval. While a related task, audio fingerprinting, has made significant progress in accurately retrieving musical content under "real world" (noisy, reverberant) conditions, ASID systems struggle to identify samples that have undergone musical modifications. Thus, a system robust to common music production transformations such as time-stretching, pitch-shifting, effects processing, and underlying or overlaying music is an important open challenge. In this work, we propose a lightweight and scalable encoding architecture employing a Graph Neural Network within a contrastive learning framework. Our model uses only 9% of the trainable parameters compared to the current state-of-the-art system while achieving comparable performance, reaching a mean average precision (mAP) of 44.2%. To enhance retrieval quality, we introduce a two-stage approach consisting of an initial coarse similarity search for candidate selection, followed by a cross-attention classifier that rejects irrelevant matches and refines the ranking of retrieved candidates - an essential capability absent in prior models. In addition, because queries in real-world applications are often short in duration, we benchmark our system for short queries using new fine-grained annotations for the Sample100 dataset, which we publish as part of this work.
Authors: Jules Leguy, Pierre-Antoine Jean, Felipe Torres Figueroa, S\'ebastien Harispe
Abstract: This article introduces WebXAII, an open-source web framework designed to facilitate research on human interaction with eXplainable Artificial Intelligence (XAI) systems. The field of XAI is rapidly expanding, driven by the growing societal implications of the widespread adoption of AI (and in particular machine learning) across diverse applications. Researchers who study the interaction between humans and XAI techniques typically develop ad hoc interfaces in order to conduct their studies. These interfaces are usually not shared alongside the results of the studies, which limits their reusability and the reproducibility of experiments. In response, we design and implement WebXAII, a web-based platform that can embody full experimental protocols, meaning that it can present all aspects of the experiment to human participants and record their responses. The experimental protocols are translated into a composite architecture of generic views and modules, which offers a lot of flexibility. The architecture is defined in a structured configuration file, so that protocols can be implemented with minimal programming skills. We demonstrate that WebXAII can effectively embody relevant protocols, by reproducing the protocol of a state-of-the-art study of the literature.
Authors: Varun Mannam, Zhenyu Shi
Abstract: Accurate video annotation plays a vital role in modern retail applications, including customer behavior analysis, product interaction detection, and in-store activity recognition. However, conventional annotation methods heavily rely on time-consuming manual labeling by human annotators, introducing non-robust frame selection and increasing operational costs. To address these challenges in the retail domain, we propose a deep learning-based approach that automates key-frame identification in retail videos and provides automatic annotations of products and customers. Our method leverages deep neural networks to learn discriminative features by embedding video frames and incorporating object detection-based techniques tailored for retail environments. Experimental results showcase the superiority of our approach over traditional methods, achieving accuracy comparable to human annotator labeling while enhancing the overall efficiency of retail video annotation. Remarkably, our approach leads to an average of 2 times cost savings in video annotation. By allowing human annotators to verify/adjust less than 5% of detected frames in the video dataset, while automating the annotation process for the remaining frames without reducing annotation quality, retailers can significantly reduce operational costs. The automation of key-frame detection enables substantial time and effort savings in retail video labeling tasks, proving highly valuable for diverse retail applications such as shopper journey analysis, product interaction detection, and in-store security monitoring.
Authors: Laura Kopf, Nils Feldhus, Kirill Bykov, Philine Lou Bommer, Anna Hedstr\"om, Marina M. -C. H\"ohne, Oliver Eberle
Abstract: Automated interpretability research aims to identify concepts encoded in neural network features to enhance human understanding of model behavior. Current feature description methods face two critical challenges: limited robustness and the flawed assumption that each neuron encodes only a single concept (monosemanticity), despite growing evidence that neurons are often polysemantic. This assumption restricts the expressiveness of feature descriptions and limits their ability to capture the full range of behaviors encoded in model internals. To address this, we introduce Polysemantic FeatuRe Identification and Scoring Method (PRISM), a novel framework that captures the inherent complexity of neural network features. Unlike prior approaches that assign a single description per feature, PRISM provides more nuanced descriptions for both polysemantic and monosemantic features. We apply PRISM to language models and, through extensive benchmarking against existing methods, demonstrate that our approach produces more accurate and faithful feature descriptions, improving both overall description quality (via a description score) and the ability to capture distinct concepts when polysemanticity is present (via a polysemanticity score).
Authors: Yujing Sun, Lingchen Sun, Shuaizheng Liu, Rongyuan Wu, Zhengqiang Zhang, Lei Zhang
Abstract: It is a challenging problem to reproduce rich spatial details while maintaining temporal consistency in real-world video super-resolution (Real-VSR), especially when we leverage pre-trained generative models such as stable diffusion (SD) for realistic details synthesis. Existing SD-based Real-VSR methods often compromise spatial details for temporal coherence, resulting in suboptimal visual quality. We argue that the key lies in how to effectively extract the degradation-robust temporal consistency priors from the low-quality (LQ) input video and enhance the video details while maintaining the extracted consistency priors. To achieve this, we propose a Dual LoRA Learning (DLoRAL) paradigm to train an effective SD-based one-step diffusion model, achieving realistic frame details and temporal consistency simultaneously. Specifically, we introduce a Cross-Frame Retrieval (CFR) module to aggregate complementary information across frames, and train a Consistency-LoRA (C-LoRA) to learn robust temporal representations from degraded inputs. After consistency learning, we fix the CFR and C-LoRA modules and train a Detail-LoRA (D-LoRA) to enhance spatial details while aligning with the temporal space defined by C-LoRA to keep temporal coherence. The two phases alternate iteratively for optimization, collaboratively delivering consistent and detail-rich outputs. During inference, the two LoRA branches are merged into the SD model, allowing efficient and high-quality video restoration in a single diffusion step. Experiments show that DLoRAL achieves strong performance in both accuracy and speed. Code and models are available at https://github.com/yjsunnn/DLoRAL.
Authors: Vincent Roca, Marc Tommasi, Paul Andrey, Aur\'elien Bellet, Markus D. Schirmer, Hilde Henon, Laurent Puy, Julien Ramon, Gr\'egory Kuchcinski, Martin Bretzner, Renaud Lopes
Abstract: $\textbf{Objective:}$ Brain-predicted age difference (BrainAGE) is a neuroimaging biomarker reflecting brain health. However, training robust BrainAGE models requires large datasets, often restricted by privacy concerns. This study evaluates the performance of federated learning (FL) for BrainAGE estimation in ischemic stroke patients treated with mechanical thrombectomy, and investigates its association with clinical phenotypes and functional outcomes. $\textbf{Methods:}$ We used FLAIR brain images from 1674 stroke patients across 16 hospital centers. We implemented standard machine learning and deep learning models for BrainAGE estimates under three data management strategies: centralized learning (pooled data), FL (local training at each site), and single-site learning. We reported prediction errors and examined associations between BrainAGE and vascular risk factors (e.g., diabetes mellitus, hypertension, smoking), as well as functional outcomes at three months post-stroke. Logistic regression evaluated BrainAGE's predictive value for these outcomes, adjusting for age, sex, vascular risk factors, stroke severity, time between MRI and arterial puncture, prior intravenous thrombolysis, and recanalisation outcome. $\textbf{Results:}$ While centralized learning yielded the most accurate predictions, FL consistently outperformed single-site models. BrainAGE was significantly higher in patients with diabetes mellitus across all models. Comparisons between patients with good and poor functional outcomes, and multivariate predictions of these outcomes showed the significance of the association between BrainAGE and post-stroke recovery. $\textbf{Conclusion:}$ FL enables accurate age predictions without data centralization. The strong association between BrainAGE, vascular risk factors, and post-stroke recovery highlights its potential for prognostic modeling in stroke care.
Authors: Zhen Li, Chuanhao Li, Xiaofeng Mao, Shaoheng Lin, Ming Li, Shitian Zhao, Zhaopan Xu, Xinyue Li, Yukang Feng, Jianwen Sun, Zizhen Li, Fanrui Zhang, Jiaxin Ai, Zhixiang Wang, Yuwei Wu, Tong He, Jiangmiao Pang, Yu Qiao, Yunde Jia, Kaipeng Zhang
Abstract: Video generation techniques have made remarkable progress, promising to be the foundation of interactive world exploration. However, existing video generation datasets are not well-suited for world exploration training as they suffer from some limitations: limited locations, short duration, static scenes, and a lack of annotations about exploration and the world. In this paper, we introduce Sekai (meaning ``world'' in Japanese), a high-quality first-person view worldwide video dataset with rich annotations for world exploration. It consists of over 5,000 hours of walking or drone view (FPV and UVA) videos from over 100 countries and regions across 750 cities. We develop an efficient and effective toolbox to collect, pre-process and annotate videos with location, scene, weather, crowd density, captions, and camera trajectories. Experiments demonstrate the quality of the dataset. And, we use a subset to train an interactive video world exploration model, named YUME (meaning ``dream'' in Japanese). We believe Sekai will benefit the area of video generation and world exploration, and motivate valuable applications. The project page is https://lixsp11.github.io/sekai-project/.