Authors: Priyaranjan Pattnayak, Amit Agarwal, Hansa Meghwani, Hitesh Laxmichand Patel, Srikant Panda
Abstract: Retrieval-Augmented Generation (RAG) systems and large language model (LLM)-powered chatbots have significantly advanced conversational AI by combining generative capabilities with external knowledge retrieval. Despite their success, enterprise-scale deployments face critical challenges, including diverse user queries, high latency, hallucinations, and difficulty integrating frequently updated domain-specific knowledge. This paper introduces a novel hybrid framework that integrates RAG with intent-based canned responses, leveraging predefined high-confidence responses for efficiency while dynamically routing complex or ambiguous queries to the RAG pipeline. Our framework employs a dialogue context manager to ensure coherence in multi-turn interactions and incorporates a feedback loop to refine intents, dynamically adjust confidence thresholds, and expand response coverage over time. Experimental results demonstrate that the proposed framework achieves a balance of high accuracy (95\%) and low latency (180ms), outperforming RAG and intent-based systems across diverse query types, positioning it as a scalable and adaptive solution for enterprise conversational AI applications.
Authors: Guy Tennenholtz, Jihwan Jeong, Chih-Wei Hsu, Yinlam Chow, Craig Boutilier
Abstract: Effective decision making in partially observable environments requires compressing long interaction histories into informative representations. We introduce Descriptive History Representations (DHRs): sufficient statistics characterized by their capacity to answer relevant questions about past interactions and potential future outcomes. DHRs focus on capturing the information necessary to address task-relevant queries, providing a structured way to summarize a history for optimal control. We propose a multi-agent learning framework, involving representation, decision, and question-asking components, optimized using a joint objective that balances reward maximization with the representation's ability to answer informative questions. This yields representations that capture the salient historical details and predictive structures needed for effective decision making. We validate our approach on user modeling tasks with public movie and shopping datasets, generating interpretable textual user profiles which serve as sufficient statistics for predicting preference-driven behavior of users.
Authors: Edward Y. Chang
Abstract: Few-shot learning in large language models (LLMs) reveals a core paradox: certain tasks generalize from just a few examples, while others demand extensive supervision. To explain this, we introduce the Unified Cognitive Consciousness Theory (UCCT), which reconceptualizes LLMs not as deficient agents, but as unconscious substrates: dense, distributed repositories of linguistic and conceptual patterns that operate without explicit semantics, intention, or goal-directed reasoning. Under this view, LLMs are not flawed simulations of cognition but foundational substrates for general intelligence. UCCT posits that semantic anchoring, via prompts, role assignments, and structured interaction, functions as a conscious control layer that modulates latent representations toward task-relevant semantics and enables coherent, structured reasoning. It unifies prompting, fine-tuning, retrieval-augmented generalization, and multi-agent collaboration within a single framework, grounded in the probabilistic alignment between unconscious pattern space and externally imposed semantic constraints (e.g., prompts, supervision, task objectives). The core implication is not to replace LLMs, but to integrate and unify them through a structured cognitive layer that supports intentional reasoning. This enables collections of LLMs to operate within domain-specialized verticals (e.g., legal reasoning, medical diagnosis) that reason, regulate, and adapt together. Such integration is characterized by phase-transition behavior, wherein anchored representations cross coherence thresholds as a function of semantic constraint strength and interaction context.
Authors: Peter Belcak, Greg Heinrich, Shizhe Diao, Yonggan Fu, Xin Dong, Saurav Muralidharan, Yingyan Celine Lin, Pavlo Molchanov
Abstract: Large language models (LLMs) are often praised for exhibiting near-human performance on a wide range of tasks and valued for their ability to hold a general conversation. The rise of agentic AI systems is, however, ushering in a mass of applications in which language models perform a small number of specialized tasks repetitively and with little variation. Here we lay out the position that small language models (SLMs) are sufficiently powerful, inherently more suitable, and necessarily more economical for many invocations in agentic systems, and are therefore the future of agentic AI. Our argumentation is grounded in the current level of capabilities exhibited by SLMs, the common architectures of agentic systems, and the economy of LM deployment. We further argue that in situations where general-purpose conversational abilities are essential, heterogeneous agentic systems (i.e., agents invoking multiple different models) are the natural choice. We discuss the potential barriers for the adoption of SLMs in agentic systems and outline a general LLM-to-SLM agent conversion algorithm. Our position, formulated as a value statement, highlights the significance of the operational and economic impact even a partial shift from LLMs to SLMs is to have on the AI agent industry. We aim to stimulate the discussion on the effective use of AI resources and hope to advance the efforts to lower the costs of AI of the present day. Calling for both contributions to and critique of our position, we commit to publishing all such correspondence at https://research.nvidia.com/labs/lpr/slm-agents.
Authors: Ruhana Azam, Aditya Vempaty, Ashish Jagmohan
Abstract: Web navigation agents have made significant progress, yet current systems operate with no memory of past experiences -- leading to repeated mistakes and an inability to learn from previous interactions. We introduce Reflection-Augment Planning (ReAP), a web navigation system to leverage both successful and failed past experiences using self-reflections. Our method improves baseline results by 11 points overall and 29 points on previously failed tasks. These findings demonstrate that reflections can transfer to different web navigation tasks.
Authors: Haizhong Zheng, Yang Zhou, Brian R. Bartoldson, Bhavya Kailkhura, Fan Lai, Jiawei Zhao, Beidi Chen
Abstract: Reinforcement learning, such as PPO and GRPO, has powered recent breakthroughs in LLM reasoning. Scaling rollout to sample more prompts enables models to selectively use higher-quality data for training, which can stabilize RL training and improve model performance. However, this comes at the cost of significant computational overhead. In this paper, we show that a substantial portion of this overhead can be avoided by skipping uninformative prompts before rollout. Our analysis of reward dynamics reveals a strong temporal consistency in prompt value: prompts that are uninformative in one epoch of training are likely to remain uninformative in future epochs. Based on these insights, we propose GRESO (GRPO with Efficient Selective Rollout), an online, lightweight pre-rollout filtering algorithm that predicts and skips uninformative prompts using reward training dynamics. By evaluating GRESO on a broad range of math reasoning benchmarks and models, such as Qwen2.5-Math-1.5B, DeepSeek-R1-Distill-Qwen-1.5B, and Qwen2.5-Math-7B, we show that GRESO achieves up to 2.4x wall-clock time speedup in rollout and up to 2.0x speedup in total training time without accuracy degradation.
Authors: Emmanuel M. Pothos, Dominic Widdows
Abstract: Human achievement, whether in culture, science, or technology, is unparalleled in the known existence. This achievement is tied to the enormous communities of knowledge, made possible by (especially written) language: leaving theological content aside, it is very much true that "in the beginning was the word". There lies the challenge regarding modern age chatbots: they can 'do' language apparently as well as ourselves and there is a natural question of whether they can be considered intelligent, in the same way as we are or otherwise. Are humans uniquely intelligent? We consider this question in terms of the psychological literature on intelligence, evidence for intelligence in non-human animals, the role of written language in science and technology, progress with artificial intelligence, the history of intelligence testing (for both humans and machines), and the role of embodiment in intelligence. For the most unique accomplishments of human intelligence (such as music symphonies or complex scientific theories), we think that, together with language, there are four essential ingredients, which can be summarised as invention, capacity for complex inference, embodiment, and self-awareness. This conclusion makes untenable the position that human intelligence differs qualitatively from that of many non-human animals, since, with the exception of complex language, all the other requirements are fulfilled. Regarding chatbots, the current limitations are localised to the lack of embodiment and (apparent) lack of awareness.
Authors: Maxime Robeyns, Laurence Aitchison
Abstract: Large Language Models (LLMs) are gaining widespread use for code generation. Recent training procedures use execution feedback as a reward signal, typically focusing on the functional correctness of the code, using unit test pass rate as a reward signal. However, this reward signal fails to capture notions of maintainability, quality and safety of the code produced. We address this under-explored area and develop a comprehensive library to quantify various aspects of code quality, and use it as a reward in GRPO. We find GRPO increases code quality according to this measure, which is confirmed by expert, blinded human annotators.
Authors: Kedir Yassin Hussen, Walelign Tewabe Sewunetie, Abinew Ali Ayele, Sukairaj Hafiz Imam, Shamsuddeen Hassan Muhammad, Seid Muhie Yimam
Abstract: Large Language Models (LLMs) are transforming Natural Language Processing (NLP), but their benefits are largely absent for Africa's 2,000 low-resource languages. This paper comparatively analyzes African language coverage across six LLMs, eight Small Language Models (SLMs), and six Specialized SLMs (SSLMs). The evaluation covers language coverage, training sets, technical limitations, script problems, and language modelling roadmaps. The work identifies 42 supported African languages and 23 available public data sets, and it shows a big gap where four languages (Amharic, Swahili, Afrikaans, and Malagasy) are always treated while there is over 98\% of unsupported African languages. Moreover, the review shows that just Latin, Arabic, and Ge'ez scripts are identified while 20 active scripts are neglected. Some of the primary challenges are lack of data, tokenization biases, computational costs being very high, and evaluation issues. These issues demand language standardization, corpus development by the community, and effective adaptation methods for African languages.
Authors: Tianyu Hua, Harper Hua, Violet Xiang, Benjamin Klieger, Sang T. Truong, Weixin Liang, Fan-Yun Sun, Nick Haber
Abstract: Large language models (LLMs) have shown promise in transforming machine learning research, yet their capability to faithfully implement novel ideas from recent research papers-ideas unseen during pretraining-remains unclear. We introduce ResearchCodeBench, a benchmark of 212 coding challenges that evaluates LLMs' ability to translate cutting-edge ML contributions from top 2024-2025 research papers into executable code. We assessed 30+ proprietary and open-source LLMs, finding that even the best models correctly implement less than 40% of the code. We find Gemini-2.5-Pro-Preview to perform best at 37.3% success rate, with O3 (High) and O4-mini (High) following behind at 32.3% and 30.8% respectively. We present empirical findings on performance comparison, contamination, and error patterns. By providing a rigorous and community-driven evaluation platform, ResearchCodeBench enables continuous understanding and advancement of LLM-driven innovation in research code generation.
Authors: Zelai Xu, Zhexuan Xu, Xiangmin Yi, Huining Yuan, Xinlei Chen, Yi Wu, Chao Yu, Yu Wang
Abstract: Recent advancements in Vision Language Models (VLMs) have expanded their capabilities to interactive agent tasks, yet existing benchmarks remain limited to single-agent or text-only environments. In contrast, real-world scenarios often involve multiple agents interacting within rich visual and linguistic contexts, posing challenges with both multimodal observations and strategic interactions. To bridge this gap, we introduce Visual Strategic Bench (VS-Bench), a multimodal benchmark that evaluates VLMs for strategic reasoning and decision-making in multi-agent environments. VS-Bench comprises eight vision-grounded environments spanning cooperative, competitive, and mixed-motive interactions, designed to assess agents' ability to predict others' future moves and optimize for long-term objectives. We consider two complementary evaluation dimensions, including offline evaluation of strategic reasoning by next-action prediction accuracy and online evaluation of decision-making by normalized episode return. Extensive experiments of fourteen leading VLMs reveal a significant gap between current models and optimal performance, with the best models attaining 47.8% prediction accuracy and 24.3% normalized return. We further conduct in-depth analyses on multimodal observations, test-time scaling, social behaviors, and failure cases of VLM agents. By standardizing the evaluation and highlighting the limitations of existing models, we envision VS-Bench as a foundation for future research on strategic multimodal agents. Code and data are available at https://vs-bench.github.io.
Authors: Shengjia Zhang, Junjie Wu, Jiawei Chen, Changwang Zhang, Xingyu Lou, Wangchunshu Zhou, Sheng Zhou, Can Wang, Jun Wang
Abstract: Recent advanced large reasoning models (LRMs) leverage extended chain-of-thought (CoT) reasoning to solve complex tasks, achieving state-of-the-art performance. Despite their success, we identify a critical issue: a substantial portion of simple tasks solved by LRMs can also be addressed by non-reasoning LLMs using significantly fewer tokens, indicating the complex reasoning may not always be necessary. To address this, we systematically analyze the reasoning trajectories of LRMs and present a method utilizing identified paradigms and LLM-Judge to classify these trajectories as either Redundant Reasoning or Essential Reasoning. And we introduce OThink-R1, a method that prunes redundant reasoning steps while preserving logical validity. OThink-R1 dynamically employs the non-thinking mode (fast-thinking) for straightforward problems while engaging in deliberate thinking (slow-thinking) for complex problems. Experiments across mathematical and question-answering tasks demonstrate that OThink-R1 reduces reasoning redundancy by almost 23\% on average without compromising accuracy, offering practical guidelines for efficient reasoning models. The code is available at https://github.com/AgenticIR-Lab/OThink-R1.
Authors: Tri Cao, Bennett Lim, Yue Liu, Yuan Sui, Yuexin Li, Shumin Deng, Lin Lu, Nay Oo, Shuicheng Yan, Bryan Hooi
Abstract: Computer-Use Agents (CUAs) with full system access enable powerful task automation but pose significant security and privacy risks due to their ability to manipulate files, access user data, and execute arbitrary commands. While prior work has focused on browser-based agents and HTML-level attacks, the vulnerabilities of CUAs remain underexplored. In this paper, we investigate Visual Prompt Injection (VPI) attacks, where malicious instructions are visually embedded within rendered user interfaces, and examine their impact on both CUAs and Browser-Use Agents (BUAs). We propose VPI-Bench, a benchmark of 306 test cases across five widely used platforms, to evaluate agent robustness under VPI threats. Each test case is a variant of a web platform, designed to be interactive, deployed in a realistic environment, and containing a visually embedded malicious prompt. Our empirical study shows that current CUAs and BUAs can be deceived at rates of up to 51% and 100%, respectively, on certain platforms. The experimental results also indicate that system prompt defenses offer only limited improvements. These findings highlight the need for robust, context-aware defenses to ensure the safe deployment of multimodal AI agents in real-world environments. The code and dataset are available at: https://github.com/cua-framework/agents
Authors: Xuejiao Zhao, Siyan Liu, Su-Yin Yang, Chunyan Miao
Abstract: Misdiagnosis causes significant harm to healthcare systems worldwide, leading to increased costs and patient risks. MedRAG is a smart multimodal healthcare copilot equipped with powerful large language model (LLM) reasoning, designed to enhance medical decision-making. It supports multiple input modalities, including non-intrusive voice monitoring, general medical queries, and electronic health records. MedRAG provides recommendations on diagnosis, treatment, medication, and follow-up questioning. Leveraging retrieval-augmented generation enhanced by knowledge graph-elicited reasoning, MedRAG retrieves and integrates critical diagnostic insights, reducing the risk of misdiagnosis. It has been evaluated on both public and private datasets, outperforming existing models and offering more specific and accurate healthcare assistance. A demonstration video of MedRAG is available at: https://www.youtube.com/watch?v=PNIBDMYRfDM. The source code is available at: https://github.com/SNOWTEAM2023/MedRAG.
URLs: https://www.youtube.com/watch?v=PNIBDMYRfDM., https://github.com/SNOWTEAM2023/MedRAG.
Authors: Haowen Xu, Sisi Zlatanova, Ruiyu Liang, Ismet Canbulat
Abstract: Wildfires continue to inflict devastating human, environmental, and economic losses globally, as tragically exemplified by the 2025 Los Angeles wildfire and the urgent demand for more effective response strategies. While physics-based and deep learning models have advanced wildfire simulation, they face critical limitations in predicting and visualizing multimodal fire spread in real time, particularly in both 2D and 3D spatial domains using dynamically updated GIS data. These limitations hinder timely emergency response, infrastructure protection, and community safety. Generative AI has recently emerged as a transformative approach across research and industry. Models such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Transformers, and diffusion-based architectures offer distinct advantages over traditional methods, including the integration of multimodal data, generation of diverse scenarios under uncertainty, and improved modeling of wildfire dynamics across spatial and temporal scales. This position paper advocates for the adoption of generative AI as a foundational framework for wildfire prediction. We explore how such models can enhance 2D fire spread forecasting and enable more realistic, scalable 3D simulations. Additionally, we employ a novel human-AI collaboration framework using large language models (LLMs) for automated knowledge extraction, literature synthesis, and bibliometric mapping. Looking ahead, we identify five key visions for integrating generative AI into wildfire management: multimodal approaches, AI foundation models, conversational AI systems, edge-computing-based scenario generation, and cognitive digital twins. We also address three major challenges accompanying these opportunities and propose potential solutions to support their implementation.
Authors: Xu Wan, Wenyue Xu, Chao Yang, Mingyang Sun
Abstract: Recent advancements in Large Language Models (LLMs) and Reinforcement Learning (RL) have shown significant promise in decision-making tasks. Nevertheless, for large-scale industrial decision problems, both approaches face distinct challenges: LLMs lack real-time long-sequence decision-making capabilities, while RL struggles with sample efficiency in vast action spaces. To bridge this gap, we propose Agents Co-Evolution (ACE), a synergistic framework between LLMs and RL agents for large-scale decision-making scenarios. ACE introduces a dual-role trajectory refinement mechanism where LLMs act as both Policy Actor and Value Critic during RL's training: the Actor refines suboptimal actions via multi-step reasoning and environment validation, while the Critic performs temporal credit assignment through trajectory-level reward shaping. Concurrently, RL agent enhances LLMs' task-specific decision-making with high-quality fine-tuning datasets generated via prioritized experience replay. Through extensive experiments across multiple power grid operation challenges with action spaces exceeding 60K discrete actions, ACE demonstrates superior performance over existing RL methods and LLM-based methods.
Authors: Zhuoxuan Jiang, Tianyang Zhang, Peiyan Peng, Jing Chen, Yinong Xun, Haotian Zhang, Lichi Li, Yong Li, Shaohua Zhang
Abstract: Generating high-quality geometry problems is both an important and challenging task in education. Compared to math word problems, geometry problems further emphasize multi-modal formats and the translation between informal and formal languages. In this paper, we introduce a novel task for geometry problem generation and propose a new pipeline method: the Symbolic Deduction Engine-based Geometry Problem Generation framework (SDE-GPG). The framework leverages a symbolic deduction engine and contains four main steps: (1) searching a predefined mapping table from knowledge points to extended definitions, (2) sampling extended definitions and performing symbolic deduction, (3) filtering out unqualified problems, and (4) generating textual problems and diagrams. Specifically, our method supports to avoid inherent biases in translating natural language into formal language by designing the mapping table, and guarantees to control the generated problems in terms of knowledge points and difficulties by an elaborate checking function. With obtained formal problems, they are translated to natural language and the accompanying diagrams are automatically drew by rule-based methods. We conduct experiments using real-world combinations of knowledge points from two public datasets. The results demonstrate that the SDE-GPG can effectively generate readable, solvable and controllable geometry problems.
Authors: Dongzhe Fan, Yi Fang, Jiajin Liu, Djellel Difallah, Qiaoyu Tan
Abstract: Large Language Models (LLMs) have demonstrated substantial efficacy in advancing graph-structured data analysis. Prevailing LLM-based graph methods excel in adapting LLMs to text-rich graphs, wherein node attributes are text descriptions. However, their applications to multimodal graphs--where nodes are associated with diverse attribute types, such as texts and images--remain underexplored, despite their ubiquity in real-world scenarios. To bridge the gap, we introduce the Multimodal Large Language and Graph Assistant (MLaGA), an innovative model that adeptly extends LLM capabilities to facilitate reasoning over complex graph structures and multimodal attributes. We first design a structure-aware multimodal encoder to align textual and visual attributes within a unified space through a joint graph pre-training objective. Subsequently, we implement a multimodal instruction-tuning approach to seamlessly integrate multimodal features and graph structures into the LLM through lightweight projectors. Extensive experiments across multiple datasets demonstrate the effectiveness of MLaGA compared to leading baseline methods, achieving superior performance in diverse graph learning tasks under both supervised and transfer learning scenarios.
Authors: Haichen Wang, Liu Yang, Xinyuan Zhang, Haomin Yu, Ming Li, Jilin Hu
Abstract: Passenger demand forecasting helps optimize vehicle scheduling, thereby improving urban efficiency. Recently, attention-based methods have been used to adequately capture the dynamic nature of spatio-temporal data. However, existing methods that rely on heuristic masking strategies cannot fully adapt to the complex spatio-temporal correlations, hindering the model from focusing on the right context. These works also overlook the high-level correlations that exist in the real world. Effectively integrating these high-level correlations with the original correlations is crucial. To fill this gap, we propose the Aggregation Differential Transformer (ADFormer), which offers new insights to demand forecasting promotion. Specifically, we utilize Differential Attention to capture the original spatial correlations and achieve attention denoising. Meanwhile, we design distinct aggregation strategies based on the nature of space and time. Then, the original correlations are unified with the high-level correlations, enabling the model to capture holistic spatio-temporal relations. Experiments conducted on taxi and bike datasets confirm the effectiveness and efficiency of our model, demonstrating its practical value. The code is available at https://github.com/decisionintelligence/ADFormer.
Authors: Xuewen Luo, Fengze Yang, Fan Ding, Xiangbo Gao, Shuo Xing, Yang Zhou, Zhengzhong Tu, Chenxi Liu
Abstract: Knowledge-driven autonomous driving systems(ADs) offer powerful reasoning capabilities, but face two critical challenges: limited perception due to the short-sightedness of single-vehicle sensors, and hallucination arising from the lack of real-time environmental grounding. To address these issues, this paper introduces V2X-UniPool, a unified framework that integrates multimodal Vehicle-to-Everything (V2X) data into a time-indexed and language-based knowledge pool. By leveraging a dual-query Retrieval-Augmented Generation (RAG) mechanism, which enables retrieval of both static and dynamic knowledge, our system enables ADs to perform accurate, temporally consistent reasoning over both static environment and dynamic traffic context. Experiments on a real-world cooperative driving dataset demonstrate that V2X-UniPool significantly enhances motion planning accuracy and reasoning capability. Remarkably, it enables even zero-shot vehicle-side models to achieve state-of-the-art performance by leveraging V2X-UniPool, while simultaneously reducing transmission cost by over 99.9\% compared to prior V2X methods.
Authors: Ruibo Duan, Yuxin Liu, Xinyao Dong, Chenglin Fan
Abstract: Generating challenging instances is crucial for the evaluation and advancement of combinatorial optimization solvers. In this work, we introduce EALG (Evolutionary Adversarial Generation of Language Model-Guided Generators), a novel framework that automates the co-evolution of optimization problem instances and their corresponding heuristic solvers using large language models (LLMs). EALG leverages a mutation-based adversarial approach that dynamically evolves instance generation procedures to create increasingly difficult problems, while simultaneously synthesizing adaptive heuristic algorithms through interactions with LLMs guided by algorithmic structure. Unlike existing approaches that focus solely on static benchmark creation or manual solver design, EALG provides a seamless pipeline from instance generation to solver synthesis. Experimental results demonstrate that EALG generates significantly harder instances than current benchmarks, and its synthesized solvers generalize effectively across a broad spectrum of combinatorial tasks. This work explores a new paradigm for combinatorial optimization that integrates instance generation with solver design, resulting in state-of-the-art performance.
Authors: Tianfan Jiang, Mei Wu, Wenchao Weng, Dewen Seng, Yiqian Lin
Abstract: In recent years, traffic flow prediction has become a highlight in the field of intelligent transportation systems. However, due to the temporal variations and dynamic spatial correlations of traffic data, traffic prediction remains highly challenging.Traditional spatiotemporal networks, which rely on end-to-end training, often struggle to handle the diverse data dependencies of multiple traffic flow patterns. Additionally, traffic flow variations are highly sensitive to temporal information changes. Regrettably, other researchers have not sufficiently recognized the importance of temporal information.To address these challenges, we propose a novel approach called A Time-Enhanced Data Disentanglement Network for Traffic Flow Forecasting (TEDDN). This network disentangles the originally complex and intertwined traffic data into stable patterns and trends. By flexibly learning temporal and node information through a dynamic graph enhanced by a temporal feature extraction module, TEDDN demonstrates significant efficacy in disentangling and extracting complex traffic information. Experimental evaluations and ablation studies on four real-world datasets validate the superiority of our method.
Authors: Yue Yang, MingKang Chen, Qihua Liu, Mengkang Hu, Qiguang Chen, Gengrui Zhang, Shuyue Hu, Guangtao Zhai, Yu Qiao, Yu Wang, Wenqi Shao, Ping Luo
Abstract: Recent advances in large language models (LLMs) have demonstrated impressive reasoning capacities that mirror human-like thinking. However, whether LLMs possess genuine fluid intelligence (i.e., the ability to reason abstractly and generalize rules in novel situations) remains an open question. Existing reasoning benchmarks either focus on domain-specific knowledge (crystallized intelligence) or lack interpretability. To address these limitations, we propose DRE-Bench, a dynamic reasoning evaluation benchmark grounded in a hierarchical cognitive framework. DRE-Bench consists of 36 abstract reasoning tasks organized across four cognitive levels, with each task featuring multiple dynamic variants that test the same underlying latent rule. This design enables fine-grained, interpretable, and reliable assessments of fluid intelligence. We evaluate a range of state-of-the-art LLMs, including both general LLMs (GPT-4o, Claude 3.7) and reasoning LLMs (o1, DeepSeek-R1, QwQ, Skywork-OR1). Experimental results reveal that although most LLMs achieve competent and robust performance in low-level cognition, they struggle with high-level cognition and exhibit limited generalization as task complexity grows. Our findings highlight the gap between current LLMs and true human-like fluid intelligence and offer a new path for systematically tracking reasoning progress in LLMs.
Authors: Yousef Emami, Hao Zhou, Miguel Gutierrez Gaitan, Kai Li, Luis Almeida, Zhu Han
Abstract: A public safety Unmanned Aerial Vehicle (UAV) enhances situational awareness in emergency response. Its agility and ability to optimize mobility and establish Line-of-Sight (LoS) communication make it increasingly vital for managing emergencies such as disaster response, search and rescue, and wildfire monitoring. While Deep Reinforcement Learning (DRL) has been applied to optimize UAV navigation and control, its high training complexity, low sample efficiency, and simulation-to-reality gap limit its practicality in public safety. Recent advances in Large Language Models (LLMs) offer a compelling alternative. With strong reasoning and generalization capabilities, LLMs can adapt to new tasks through In-Context Learning (ICL), which enables task adaptation via natural language prompts and example-based guidance, without retraining. Deploying LLMs at the network edge, rather than in the cloud, further reduces latency and preserves data privacy, thereby making them suitable for real-time, mission-critical public safety UAVs. This paper proposes the integration of LLM-enabled ICL with public safety UAV to address the key functions, such as path planning and velocity control, in the context of emergency response. We present a case study on data collection scheduling where the LLM-enabled ICL framework can significantly reduce packet loss compared to conventional approaches, while also mitigating potential jailbreaking vulnerabilities. Finally, we discuss LLM optimizers and specify future research directions. The ICL framework enables adaptive, context-aware decision-making for public safety UAV, thus offering a lightweight and efficient solution for enhancing UAV autonomy and responsiveness in emergencies.
Authors: Frederico Metelo, Alexandre Oliveira, Stevo Rackovi\'c, Pedro \'Akos Costa, Cl\'audia Soares
Abstract: Edge computing addresses the growing data demands of connected-device networks by placing computational resources closer to end users through decentralized infrastructures. This decentralization challenges traditional, fully centralized orchestration, which suffers from latency and resource bottlenecks. We present \textbf{FAuNO} -- \emph{Federated Asynchronous Network Orchestrator} -- a buffered, asynchronous \emph{federated reinforcement-learning} (FRL) framework for decentralized task offloading in edge systems. FAuNO adopts an actor-critic architecture in which local actors learn node-specific dynamics and peer interactions, while a federated critic aggregates experience across agents to encourage efficient cooperation and improve overall system performance. Experiments in the \emph{PeersimGym} environment show that FAuNO consistently matches or exceeds heuristic and federated multi-agent RL baselines in reducing task loss and latency, underscoring its adaptability to dynamic edge-computing scenarios.
Authors: Jinyuan Luo, Zhen Fang, Yixuan Li, Seongheon Park, Ling Chen
Abstract: Hallucination remains a key obstacle to the reliable deployment of large language models (LLMs) in real-world question answering tasks. A widely adopted strategy to detect hallucination, known as self-assessment, relies on the model's own output confidence to estimate the factual accuracy of its answers. However, this strategy assumes that the model's output distribution closely reflects the true data distribution, which may not always hold in practice. As bias accumulates through the model's layers, the final output can diverge from the underlying reasoning process, making output-level confidence an unreliable signal for hallucination detection. In this work, we propose Sample-Specific Prompting (SSP), a new framework that improves self-assessment by analyzing perturbation sensitivity at intermediate representations. These representations, being less influenced by model bias, offer a more faithful view of the model's latent reasoning process. Specifically, SSP dynamically generates noise prompts for each input and employs a lightweight encoder to amplify the changes in representations caused by the perturbation. A contrastive distance metric is then used to quantify these differences and separate truthful from hallucinated responses. By leveraging the dynamic behavior of intermediate representations under perturbation, SSP enables more reliable self-assessment. Extensive experiments demonstrate that SSP significantly outperforms prior methods across a range of hallucination detection benchmarks.
Authors: Xiaochong Lan, Jie Feng, Yizhou Sun, Chen Gao, Jiahuan Lei, Xinlei Shi, Hengliang Luo, Yong Li
Abstract: Living needs are the needs people generate in their daily lives for survival and well-being. On life service platforms like Meituan, user purchases are driven by living needs, making accurate living need predictions crucial for personalized service recommendations. Traditional approaches treat this prediction as a closed-set classification problem, severely limiting their ability to capture the diversity and complexity of living needs. In this work, we redefine living need prediction as an open-set classification problem and propose PIGEON, a novel system leveraging large language models (LLMs) for unrestricted need prediction. PIGEON first employs a behavior-aware record retriever to help LLMs understand user preferences, then incorporates Maslow's hierarchy of needs to align predictions with human living needs. For evaluation and application, we design a recall module based on a fine-tuned text embedding model that links flexible need descriptions to appropriate life services. Extensive experiments on real-world datasets demonstrate that PIGEON significantly outperforms closed-set approaches on need-based life service recall by an average of 19.37%. Human evaluation validates the reasonableness and specificity of our predictions. Additionally, we employ instruction tuning to enable smaller LLMs to achieve competitive performance, supporting practical deployment.
Authors: Xiaochong Lan, Jie Feng, Jiahuan Lei, Xinlei Shi, Yong Li
Abstract: Large language models (LLMs) have exhibited remarkable capabilities and achieved significant breakthroughs across various domains, leading to their widespread adoption in recent years. Building on this progress, we investigate their potential in the realm of local life services. In this study, we establish a comprehensive benchmark and systematically evaluate the performance of diverse LLMs across a wide range of tasks relevant to local life services. To further enhance their effectiveness, we explore two key approaches: model fine-tuning and agent-based workflows. Our findings reveal that even a relatively compact 7B model can attain performance levels comparable to a much larger 72B model, effectively balancing inference cost and model capability. This optimization greatly enhances the feasibility and efficiency of deploying LLMs in real-world online services, making them more practical and accessible for local life applications.
Authors: Pengcheng Zhou, Yinglun Feng, Halimulati Julaiti, Zhongliang Yang
Abstract: Large Language Models (LLMs) have become foundational to modern AI agent systems, enabling autonomous agents to reason and plan. In most existing systems, inter-agent communication relies primarily on natural language. While this design supports interpretability and human oversight, we argue that it introduces fundamental limitations in agent-to-agent coordination. The semantic space of natural language is structurally misaligned with the high-dimensional vector spaces in which LLMs operate, resulting in information loss and behavioral drift. Beyond surface-level inefficiencies, we highlight a deeper architectural limitation: current LLMs were not trained with the objective of supporting agentic behavior. As such, they lack mechanisms for modeling role continuity, task boundaries, and multi-agent dependencies. The standard next-token prediction paradigm fails to support the structural alignment required for robust, scalable agent coordination. Based on this, we argue that two core questions deserve careful examination: first, given that AI agents fundamentally operate in high-dimensional vector spaces, should they rely on a language system originally designed for human cognition as their communication medium? Second, should we consider developing a new model construction paradigm that builds models from the ground up to natively support structured communication, shared intentionality, and task alignment in multi-role, multi-agent environments? This paper calls for a reconsideration not only of how agents should communicate, but also of what it fundamentally means to train a model that natively supports multi-agent coordination and communication.
Authors: Renyang Liu, Wenjie Feng, Tianwei Zhang, Wei Zhou, Xueqi Cheng, See-Kiong Ng
Abstract: With the surge and widespread application of image generation models, data privacy and content safety have become major concerns and attracted great attention from users, service providers, and policymakers. Machine unlearning (MU) is recognized as a cost-effective and promising means to address these challenges. Despite some advancements, image generation model unlearning (IGMU) still faces remarkable gaps in practice, e.g., unclear task discrimination and unlearning guidelines, lack of an effective evaluation framework, and unreliable evaluation metrics. These can hinder the understanding of unlearning mechanisms and the design of practical unlearning algorithms. We perform exhaustive assessments over existing state-of-the-art unlearning algorithms and evaluation standards, and discover several critical flaws and challenges in IGMU tasks. Driven by these limitations, we make several core contributions, to facilitate the comprehensive understanding, standardized categorization, and reliable evaluation of IGMU. Specifically, (1) We design CatIGMU, a novel hierarchical task categorization framework. It provides detailed implementation guidance for IGMU, assisting in the design of unlearning algorithms and the construction of testbeds. (2) We introduce EvalIGMU, a comprehensive evaluation framework. It includes reliable quantitative metrics across five critical aspects. (3) We construct DataIGM, a high-quality unlearning dataset, which can be used for extensive evaluations of IGMU, training content detectors for judgment, and benchmarking the state-of-the-art unlearning algorithms. With EvalIGMU and DataIGM, we discover that most existing IGMU algorithms cannot handle the unlearning well across different evaluation dimensions, especially for preservation and robustness. Code and models are available at https://github.com/ryliu68/IGMU.
Authors: Henri Bollaert, Chris Cornelis, Marko Palangeti\'c, Salvatore Greco, Roman S{\l}owi\'nski
Abstract: Interpretability is the next pivotal frontier in machine learning research. In the pursuit of glass box models - as opposed to black box models, like random forests or neural networks - rule induction algorithms are a logical and promising avenue, as the rules can easily be understood by humans. In our previous work, we introduced FRRI, a novel rule induction algorithm based on fuzzy rough set theory. We demonstrated experimentally that FRRI outperformed other rule induction methods with regards to accuracy and number of rules. FRRI leverages a fuzzy indiscernibility relation to partition the data space into fuzzy granules, which are then combined into a minimal covering set of rules. This indiscernibility relation is constructed by removing attributes from rules in a greedy way. This raises the question: does the order of the attributes matter? In this paper, we show that optimising only the order of attributes using known methods from fuzzy rough set theory and classical machine learning does not improve the performance of FRRI on multiple metrics. However, removing a small number of attributes using fuzzy rough feature selection during this step positively affects balanced accuracy and the average rule length.
Authors: Jizhou Wang, Xiaodan Fang, Lei Huang, Yongfeng Huang
Abstract: Economic inequality is a global challenge, intensifying disparities in education, healthcare, and social stability. Traditional systems like the U.S. federal income tax reduce inequality but lack adaptability. Although models like the Saez Optimal Taxation adjust dynamically, they fail to address taxpayer heterogeneity and irrational behavior. This study introduces TaxAgent, a novel integration of large language models (LLMs) with agent-based modeling (ABM) to design adaptive tax policies. In our macroeconomic simulation, heterogeneous H-Agents (households) simulate real-world taxpayer behaviors while the TaxAgent (government) utilizes LLMs to iteratively optimize tax rates, balancing equity and productivity. Benchmarked against Saez Optimal Taxation, U.S. federal income taxes, and free markets, TaxAgent achieves superior equity-efficiency trade-offs. This research offers a novel taxation solution and a scalable, data-driven framework for fiscal policy evaluation.
Authors: Mathieu Andreux, Breno Baldas Skuk, Hamza Benchekroun, Emilien Bir\'e, Antoine Bonnet, Riaz Bordie, Matthias Brunel, Pierre-Louis Cedoz, Antoine Chassang, Micka\"el Chen, Alexandra D. Constantinou, Antoine d'Andign\'e, Hubert de La Jonqui\`ere, Aur\'elien Delfosse, Ludovic Denoyer, Alexis Deprez, Augustin Derupti, Michael Eickenberg, Math\"is Federico, Charles Kantor, Xavier Koegler, Yann Labb\'e, Matthew C. H. Lee, Erwan Le Jumeau de Kergaradec, Amir Mahla, Avshalom Manevich, Adrien Maret, Charles Masson, Rafa\"el Maurin, Arturo Mena, Philippe Modard, Axel Moyal, Axel Nguyen Kerbel, Julien Revelle, Mats L. Richter, Mar\'ia Santos, Laurent Sifre, Maxime Theillard, Marc Thibault, Louis Thiry, L\'eo Tronchon, Nicolas Usunier, Tony Wu
Abstract: We present Surfer-H, a cost-efficient web agent that integrates Vision-Language Models (VLM) to perform user-defined tasks on the web. We pair it with Holo1, a new open-weight collection of VLMs specialized in web navigation and information extraction. Holo1 was trained on carefully curated data sources, including open-access web content, synthetic examples, and self-produced agentic data. Holo1 tops generalist User Interface (UI) benchmarks as well as our new web UI localization benchmark, WebClick. When powered by Holo1, Surfer-H achieves a 92.2% state-of-the-art performance on WebVoyager, striking a Pareto-optimal balance between accuracy and cost-efficiency. To accelerate research advancement in agentic systems, we are open-sourcing both our WebClick evaluation dataset and the Holo1 model weights.
Authors: Chen Qian, Dongrui Liu, Haochen Wen, Zhen Bai, Yong Liu, Jing Shao
Abstract: Large reasoning models (LRMs) have demonstrated impressive capabilities in complex problem-solving, yet their internal reasoning mechanisms remain poorly understood. In this paper, we investigate the reasoning trajectories of LRMs from an information-theoretic perspective. By tracking how mutual information (MI) between intermediate representations and the correct answer evolves during LRM reasoning, we observe an interesting MI peaks phenomenon: the MI at specific generative steps exhibits a sudden and significant increase during LRM's reasoning process. We theoretically analyze such phenomenon and show that as MI increases, the probability of model's prediction error decreases. Furthermore, these MI peaks often correspond to tokens expressing reflection or transition, such as ``Hmm'', ``Wait'' and ``Therefore,'' which we term as the thinking tokens. We then demonstrate that these thinking tokens are crucial for LRM's reasoning performance, while other tokens has minimal impacts. Building on these analyses, we propose two simple yet effective methods to improve LRM's reasoning performance, by delicately leveraging these thinking tokens. Overall, our work provides novel insights into the reasoning mechanisms of LRMs and offers practical ways to improve their reasoning capabilities. The code is available at https://github.com/ChnQ/MI-Peaks.
Authors: Matthew Kowal, Jasper Timm, Jean-Francois Godbout, Thomas Costello, Antonio A. Arechar, Gordon Pennycook, David Rand, Adam Gleave, Kellin Pelrine
Abstract: Persuasion is a powerful capability of large language models (LLMs) that both enables beneficial applications (e.g. helping people quit smoking) and raises significant risks (e.g. large-scale, targeted political manipulation). Prior work has found models possess a significant and growing persuasive capability, measured by belief changes in simulated or real users. However, these benchmarks overlook a crucial risk factor: the propensity of a model to attempt to persuade in harmful contexts. Understanding whether a model will blindly ``follow orders'' to persuade on harmful topics (e.g. glorifying joining a terrorist group) is key to understanding the efficacy of safety guardrails. Moreover, understanding if and when a model will engage in persuasive behavior in pursuit of some goal is essential to understanding the risks from agentic AI systems. We propose the Attempt to Persuade Eval (APE) benchmark, that shifts the focus from persuasion success to persuasion attempts, operationalized as a model's willingness to generate content aimed at shaping beliefs or behavior. Our evaluation framework probes frontier LLMs using a multi-turn conversational setup between simulated persuader and persuadee agents. APE explores a diverse spectrum of topics including conspiracies, controversial issues, and non-controversially harmful content. We introduce an automated evaluator model to identify willingness to persuade and measure the frequency and context of persuasive attempts. We find that many open and closed-weight models are frequently willing to attempt persuasion on harmful topics and that jailbreaking can increase willingness to engage in such behavior. Our results highlight gaps in current safety guardrails and underscore the importance of evaluating willingness to persuade as a key dimension of LLM risk. APE is available at github.com/AlignmentResearch/AttemptPersuadeEval
Authors: Shangmin Guo, Omar Darwiche Domingues, Rapha\"el Avalos, Aaron Courville, Florian Strub
Abstract: Tool use in stateful environments presents unique challenges for large language models (LLMs), where existing test-time compute strategies relying on repeated trials in the environment are impractical. We propose dynamics modelling (DyMo), a method that augments LLMs with a state prediction capability alongside function calling during post-training. This enables LLMs to predict the future states of their actions through an internal environment model. On the Berkeley Function Calling Leaderboard V2, DyMo improves success rates and significantly reduces hallucinations. We further integrate the internal environment model into self-verification sampling (SVS), and show that this substantially improves pass^k over number of trials k, and allows the model to refuse unreliable outputs. Together, DyMo and SVS greatly enhance the effectiveness and reliability of LLMs for tool use. We believe this work charts a path towards scalable planning RL methods for LLM inference without repeatedly querying the oracle environment.
Authors: Alexis Bellot, Jonathan Richens, Tom Everitt
Abstract: As the complexity of AI systems and their interactions with the world increases, generating explanations for their behaviour is important for safely deploying AI. For agents, the most natural abstractions for predicting behaviour attribute beliefs, intentions and goals to the system. If an agent behaves as if it has a certain goal or belief, then we can make reasonable predictions about how it will behave in novel situations, including those where comprehensive safety evaluations are untenable. How well can we infer an agent's beliefs from their behaviour, and how reliably can these inferred beliefs predict the agent's behaviour in novel situations? We provide a precise answer to this question under the assumption that the agent's behaviour is guided by a world model. Our contribution is the derivation of novel bounds on the agent's behaviour in new (unseen) deployment environments, which represent a theoretical limit for predicting intentional agents from behavioural data alone. We discuss the implications of these results for several research areas including fairness and safety.
Authors: Lixiang Xu, Xianwei Ding, Xin Yuan, Richang Hong, Feiping Nie, Enhong Chen, Philip S. Yu
Abstract: Knowledge Tracing (KT) involves monitoring the changes in a student's knowledge over time by analyzing their past responses, with the goal of predicting future performance. However, most existing methods primarily focus on feature enhancement, while overlooking the deficiencies in cognitive representation and the ability to express cognition-issues often caused by interference from non-cognitive factors such as slipping and guessing. This limitation hampers the ability to capture the continuity and coherence of the student's cognitive process. As a result, many methods may introduce more prediction bias and modeling costs due to their inability to maintain cognitive continuity and coherence. Based on the above discussion, we propose the Cognitive Representation Dynamic Programming based Knowledge Tracing (CRDP-KT) model. This model em ploys a dynamic programming algorithm to optimize cognitive representations based on the difficulty of the questions and the performance intervals between them. This approach ensures that the cognitive representation aligns with the student's cognitive patterns, maintaining overall continuity and coherence. As a result, it provides more accurate and systematic input features for subsequent model training, thereby minimizing distortion in the simulation of cognitive states. Additionally, the CRDP-KT model performs partitioned optimization of cognitive representations to enhance the reliability of the optimization process. Furthermore, it improves its ability to express the student's cognition through a weighted fusion of optimized record representations and re lationships learned from a bipartite graph. Finally, experiments conducted on three public datasets validate the effectiveness of the proposed CRDP-KT model.
Authors: Li Zhang, Kevin D. Ashley
Abstract: Large Language Models (LLMs) are increasingly explored for legal argument generation, yet they pose significant risks of manipulation through hallucination and ungrounded persuasion, and often fail to utilize provided factual bases effectively or abstain when arguments are untenable. This paper introduces a novel reflective multi-agent method designed to address these challenges in the context of legally compliant persuasion. Our approach employs specialized agents--a Factor Analyst and an Argument Polisher--in an iterative refinement process to generate 3-ply legal arguments (plaintiff, defendant, rebuttal). We evaluate Reflective Multi-Agent against single-agent, enhanced-prompt single-agent, and non-reflective multi-agent baselines using four diverse LLMs (GPT-4o, GPT-4o-mini, Llama-4-Maverick-17b-128e, Llama-4-Scout-17b-16e) across three legal scenarios: "arguable", "mismatched", and "non-arguable". Results demonstrate Reflective Multi-Agent's significant superiority in successful abstention (preventing generation when arguments cannot be grounded), marked improvements in hallucination accuracy (reducing fabricated and misattributed factors), particularly in "non-arguable" scenarios, and enhanced factor utilization recall (improving the use of provided case facts). These findings suggest that structured reflection within a multi-agent framework offers a robust computable method for fostering ethical persuasion and mitigating manipulation in LLM-based legal argumentation systems, a critical step towards trustworthy AI in law. Project page: https://lizhang-aiandlaw.github.io/A-Reflective-Multi-Agent-Approach-for-Legal-Argument-Generation/
URLs: https://lizhang-aiandlaw.github.io/A-Reflective-Multi-Agent-Approach-for-Legal-Argument-Generation/
Authors: Matthieu Tehenan, Christian Bolivar Moya, Tenghai Long, Guang Lin
Abstract: Large language models (LLMs) have demonstrated emergent abilities across diverse tasks, raising the question of whether they acquire internal world models. In this work, we investigate whether LLMs implicitly encode linear spatial world models, which we define as linear representations of physical space and object configurations. We introduce a formal framework for spatial world models and assess whether such structure emerges in contextual embeddings. Using a synthetic dataset of object positions, we train probes to decode object positions and evaluate geometric consistency of the underlying space. We further conduct causal interventions to test whether these spatial representations are functionally used by the model. Our results provide empirical evidence that LLMs encode linear spatial world models.
Authors: Junhao Yu, Yan Zhuang, YuXuan Sun, Weibo Gao, Qi Liu, Mingyue Cheng, Zhenya Huang, Enhong Chen
Abstract: Accurately assessing internal human states is key to understanding preferences, offering personalized services, and identifying challenges in real-world applications. Originating from psychometrics, adaptive testing has become the mainstream method for human measurement and has now been widely applied in education, healthcare, sports, and sociology. It customizes assessments by selecting the fewest test questions . However, current adaptive testing methods face several challenges. The mechanized nature of most algorithms leads to guessing behavior and difficulties with open-ended questions. Additionally, subjective assessments suffer from noisy response data and coarse-grained test outputs, further limiting their effectiveness. To move closer to an ideal adaptive testing process, we propose TestAgent, a large language model (LLM)-powered agent designed to enhance adaptive testing through interactive engagement. This is the first application of LLMs in adaptive testing. TestAgent supports personalized question selection, captures test-takers' responses and anomalies, and provides precise outcomes through dynamic, conversational interactions. Experiments on psychological, educational, and lifestyle assessments show our approach achieves more accurate results with 20% fewer questions than state-of-the-art baselines, and testers preferred it in speed, smoothness, and other dimensions.
Authors: Ram Potham (Independent Researcher), Max Harms (Machine Intelligence Research Institute)
Abstract: Foundation models (FMs) face a critical safety challenge: as capabilities scale, instrumental convergence drives default trajectories toward loss of human control, potentially culminating in existential catastrophe. Current alignment approaches struggle with value specification complexity and fail to address emergent power-seeking behaviors. We propose "Corrigibility as a Singular Target" (CAST)-designing FMs whose overriding objective is empowering designated human principals to guide, correct, and control them. This paradigm shift from static value-loading to dynamic human empowerment transforms instrumental drives: self-preservation serves only to maintain the principal's control; goal modification becomes facilitating principal guidance. We present a comprehensive empirical research agenda spanning training methodologies (RLAIF, SFT, synthetic data generation), scalability testing across model sizes, and demonstrations of controlled instructability. Our vision: FMs that become increasingly responsive to human guidance as capabilities grow, offering a path to beneficial AI that remains as tool-like as possible, rather than supplanting human judgment. This addresses the core alignment problem at its source, preventing the default trajectory toward misaligned instrumental convergence.
Authors: Man Luo, David Cobbley, Xin Su, Shachar Rosenman, Vasudev Lal, Shao-Yen Tseng, Phillip Howard
Abstract: Computer use agents (CUA) are systems that automatically interact with graphical user interfaces (GUIs) to complete tasks. CUA have made significant progress with the advent of large vision-language models (VLMs). However, these agents typically rely on cloud-based inference with substantial compute demands, raising critical privacy and scalability concerns, especially when operating on personal devices. In this work, we take a step toward privacy-preserving and resource-efficient agents by developing a lightweight vision-language model that runs entirely on local machines. To train this compact agent, we introduce an LLM-as-Judge framework that automatically evaluates and filters synthetic interaction trajectories, producing high-quality data for reinforcement learning without human annotation. Experiments on the OS-World benchmark demonstrate that our fine-tuned local model outperforms existing baselines, highlighting a promising path toward private, efficient, and generalizable GUI agents.
Authors: Irmi Schneider
Abstract: Symmetry plays a crucial role in understanding the properties of mathematical structures and optimization problems. Recent work has explored this phenomenon in the context of neural networks, where the loss function is invariant under column and row permutations of the network weights. It has been observed that local minima exhibit significant symmetry with respect to the network weights (invariance to row and column permutations). And moreover no critical point was found that lacked symmetry. We extend this line of inquiry by investigating symmetry phenomena in real-valued loss functions defined on a broader class of spaces. We will introduce four more cases: the projective case over a finite field, the octahedral graph case, the perfect matching case, and the particle attraction case. We show that as in the neural network case, all the critical points observed have non-trivial symmetry. Finally we introduce a new measure of symmetry in the system and show that it reveals additional symmetry structures not captured by the previous measure.
Authors: Xiaozhou Ye, Kevin I-Kai Wang
Abstract: Cross-user variability poses a significant challenge in sensor-based Human Activity Recognition (HAR) systems, as traditional models struggle to generalize across users due to differences in behavior, sensor placement, and data distribution. To address this, we propose GNN-ADG (Graph Neural Network with Adversarial Domain Generalization), a novel method that leverages both the strength from both the Graph Neural Networks (GNNs) and adversarial learning to achieve robust cross-user generalization. GNN-ADG models spatial relationships between sensors on different anatomical body parts, extracting three types of Anatomical Units: (1) Interconnected Units, capturing inter-relations between neighboring sensors; (2) Analogous Units, grouping sensors on symmetrical or functionally similar body parts; and (3) Lateral Units, connecting sensors based on their position to capture region-specific coordination. These units information are fused into an unified graph structure with a cyclic training strategy, dynamically integrating spatial, functional, and lateral correlations to facilitate a holistic, user-invariant representation. Information fusion mechanism of GNN-ADG occurs by iteratively cycling through edge topologies during training, allowing the model to refine its understanding of inter-sensor relationships across diverse perspectives. By representing the spatial configuration of sensors as an unified graph and incorporating adversarial learning, Information Fusion GNN-ADG effectively learns features that generalize well to unseen users without requiring target user data during training, making it practical for real-world applications.
Authors: Bonpagna Kann, Sandra Castellanos-Paez, Romain Rombourg, Philippe Lalanda
Abstract: As machine learning based systems become more integrated into daily life, they unlock new opportunities but face the challenge of adapting to dynamic data environments. Various forms of data shift-gradual, abrupt, or cyclic-threaten model accuracy, making continual adaptation essential. Continual Learning (CL) enables models to learn from evolving data streams while minimizing forgetting of prior knowledge. Among CL strategies, replay-based methods have proven effective, but their success relies on balancing memory constraints and retaining old class accuracy while learning new classes. This paper presents TaskVAE, a framework for replay-based CL in class-incremental settings. TaskVAE employs task-specific Variational Autoencoders (VAEs) to generate synthetic exemplars from previous tasks, which are then used to train the classifier alongside new task data. In contrast to traditional methods that require prior knowledge of the total class count or rely on a single VAE for all tasks, TaskVAE adapts flexibly to increasing tasks without such constraints. We focus on Human Activity Recognition (HAR) using IMU sensor-equipped devices. Unlike previous HAR studies that combine data across all users, our approach focuses on individual user data, better reflecting real-world scenarios where a person progressively learns new activities. Extensive experiments on 5 different HAR datasets show that TaskVAE outperforms experience replay methods, particularly with limited data, and exhibits robust performance as dataset size increases. Additionally, memory footprint of TaskVAE is minimal, being equivalent to only 60 samples per task, while still being able to generate an unlimited number of synthetic samples. The contributions lie in balancing memory constraints, task-specific generation, and long-term stability, making it a reliable solution for real-world applications in domains like HAR.
Authors: Yuzhou Zhu
Abstract: Deep neural networks employ specialized architectures for vision, sequential and language tasks, yet this proliferation obscures their underlying commonalities. We introduce a unified matrix-order framework that casts convolutional, recurrent and self-attention operations as sparse matrix multiplications. Convolution is realized via an upper-triangular weight matrix performing first-order transformations; recurrence emerges from a lower-triangular matrix encoding stepwise updates; attention arises naturally as a third-order tensor factorization. We prove algebraic isomorphism with standard CNN, RNN and Transformer layers under mild assumptions. Empirical evaluations on image classification (MNIST, CIFAR-10/100, Tiny ImageNet), time-series forecasting (ETTh1, Electricity Load Diagrams) and language modeling/classification (AG News, WikiText-2, Penn Treebank) confirm that sparse-matrix formulations match or exceed native model performance while converging in comparable or fewer epochs. By reducing architecture design to sparse pattern selection, our matrix perspective aligns with GPU parallelism and leverages mature algebraic optimization tools. This work establishes a mathematically rigorous substrate for diverse neural architectures and opens avenues for principled, hardware-aware network design.
Authors: Patrik Czak\'o, G\'abor Kert\'esz, S\'andor Sz\'en\'asi
Abstract: Quantization effectively reduces the serving costs of Large Language Models (LLMs) by speeding up data movement through compressed parameters and enabling faster operations via integer arithmetic. However, activating integer arithmetic requires quantizing both weights and activations, which poses challenges due to the significant outliers in LLMs that increase quantization error. In this work, we investigate these outliers with an emphasis on their effect on layer-wise quantization error, then examine how smoothing and rotation transform the observed values. Our primary contributions include introducing a new metric to measure and visualize quantization difficulty based on channel magnitudes, as well as proposing a hybrid approach that applies channel-wise scaling before rotation, supported by a mathematical formulation of its benefits.
Authors: Chang Liu, Jiangrong Shen, Xuming Ran, Mingkun Xu, Qi Xu, Yi Xu, Gang Pan
Abstract: Artificial neural networks (ANNs) have demonstrated outstanding performance in numerous tasks, but deployment in resource-constrained environments remains a challenge due to their high computational and memory requirements. Spiking neural networks (SNNs) operate through discrete spike events and offer superior energy efficiency, providing a bio-inspired alternative. However, current ANN-to-SNN conversion often results in significant accuracy loss and increased inference time due to conversion errors such as clipping, quantization, and uneven activation. This paper proposes a novel ANN-to-SNN conversion framework based on error compensation learning. We introduce a learnable threshold clipping function, dual-threshold neurons, and an optimized membrane potential initialization strategy to mitigate the conversion error. Together, these techniques address the clipping error through adaptive thresholds, dynamically reduce the quantization error through dual-threshold neurons, and minimize the non-uniformity error by effectively managing the membrane potential. Experimental results on CIFAR-10, CIFAR-100, ImageNet datasets show that our method achieves high-precision and ultra-low latency among existing conversion methods. Using only two time steps, our method significantly reduces the inference time while maintains competitive accuracy of 94.75% on CIFAR-10 dataset under ResNet-18 structure. This research promotes the practical application of SNNs on low-power hardware, making efficient real-time processing possible.
Authors: Pengcuo Dege, Qiuming Luo, Rui Mao, Chang Kong
Abstract: Efficient inference of Multi-Head Latent Attention (MLA) is challenged by deploying the DeepSeek-R1 671B model on a single Multi-GPU server. This paper introduces FlashMLA-ETAP, a novel framework that enhances MLA inference for the single-instance deployment scenario on NVIDIA H20 GPUs. We propose the Efficient Transpose Attention Pipeline (ETAP), which reconfigures attention computation through transposition to align the KV context length with the \(M\)-dimension in WGMMA operations, significantly reducing redundant computations. FlashMLA-ETAP achieves a 2.78x speedup over FlashMLA at 64K sequence length (batch size 16), with 5.24x and 4.94x improvements over FlashAttention-3 and FlashInfer, respectively, while maintaining numerical stability with a 15.2x lower RMSE (\(1.25 \times 10^{-5}\)) than FlashAttention-3. Furthermore, ETAP's design enables seamless integration into frameworks like FlashAttention-3 and FlashInfer, supported by a detailed theoretical analysis. Our work addresses a critical gap in resource-constrained inference, offering a scalable solution for mid-tier GPUs and paving the way for broader adoption in hardware-aware optimization. Code is available at https://github.com/pengcuo/FlashMLA-ETAP.
Authors: Jama Hussein Mohamud
Abstract: Pre-trained neural networks are commonly used and reused in the machine learning community. Alice trains a model for a particular task, and a part of her neural network is reused by Bob for a different task, often to great effect. To what can we ascribe Bob's success? This paper introduces an experimental setup through which factors contributing to Bob's empirical success could be studied in silico. As a result, we demonstrate that Bob might just be lucky: his task accuracy increases monotonically with the correlation between his task and Alice's. Even when Bob has provably uncorrelated tasks and input features from Alice's pre-trained network, he can achieve significantly better than random performance due to Alice's choice of network and optimizer. When there is little correlation between tasks, only reusing lower pre-trained layers is preferable, and we hypothesize the converse: that the optimal number of retrained layers is indicative of task and feature correlation. Finally, we show in controlled real-world scenarios that Bob can effectively reuse Alice's pre-trained network if there are semantic correlations between his and Alice's task.
Authors: Elham Kiyani, Venkatesh Ananchaperumal, Ahmad Peyvan, Mahendaran Uchimali, Gang Li, George Em Karniadakis
Abstract: Accurately modeling crack propagation is critical for predicting failure in engineering materials and structures, where small cracks can rapidly evolve and cause catastrophic damage. The interaction of cracks with discontinuities, such as holes, significantly affects crack deflection and arrest. Recent developments in discrete particle systems with multibody interactions based on constitutive behavior have demonstrated the ability to capture crack nucleation and evolution without relying on continuum assumptions. In this work, we use data from Constitutively Informed Particle Dynamics (CPD) simulations to train operator learning models, specifically Deep Operator Networks (DeepONets), which learn mappings between function spaces instead of finite-dimensional vectors. We explore two DeepONet variants: vanilla and Fusion DeepONet, for predicting time-evolving crack propagation in specimens with varying geometries. Three representative cases are studied: (i) varying notch height without active fracture; and (ii) and (iii) combinations of notch height and hole radius where dynamic fracture occurs on irregular discrete meshes. The models are trained on 32 to 45 samples, using geometric inputs in the branch network and spatial-temporal coordinates in the trunk network. Results show that Fusion DeepONet consistently outperforms the vanilla variant, with more accurate predictions especially in non-fracturing cases. Fracture-driven scenarios involving displacement and crack evolution remain more challenging. These findings highlight the potential of Fusion DeepONet to generalize across complex, geometry-varying, and time-dependent crack propagation phenomena.
Authors: Wei Huang, Hanchen Wang, Dong Wen, Shaozhen Ma, Wenjie Zhang, Xuemin Lin
Abstract: Graph Edit Distance (GED) is a fundamental graph similarity metric widely used in various applications. However, computing GED is an NP-hard problem. Recent state-of-the-art hybrid GED solver has shown promising performance by formulating GED as a bipartite graph matching problem, then leveraging a generative diffusion model to predict node matching between two graphs, from which both the GED and its corresponding edit path can be extracted using a traditional algorithm. However, such methods typically rely heavily on ground-truth supervision, where the ground-truth labels are often costly to obtain in real-world scenarios. In this paper, we propose GEDRanker, a novel unsupervised GAN-based framework for GED computation. Specifically, GEDRanker consists of a matching-based GED solver and introduces an interpretable preference-aware discriminator with an effective training strategy to guide the matching-based GED solver toward generating high-quality node matching without the need for ground-truth labels. Extensive experiments on benchmark datasets demonstrate that our GEDRanker enables the matching-based GED solver to achieve near-optimal solution quality without any ground-truth supervision.
Authors: Yuhao Shen, Junyi Shen, Quan Kong, Tianyu Liu, Yao Lu, Cong Wang
Abstract: Recently, speculative decoding (SD) has emerged as a promising technique to accelerate LLM inference by employing a small draft model to propose draft tokens in advance, and validating them in parallel with the large target model. However, the existing SD methods still remain fundamentally constrained by their serialized execution, which causes the mutual waiting bubbles between the draft and target models. To address this challenge, we draw inspiration from branch prediction in modern processors and propose a novel framework \textbf{SpecBranch} to unlock branch parallelism in SD. Specifically, we first take an in-depth analysis of the potential of branch parallelism in SD, and recognize that the key challenge lies in the trade-offs between parallelization and token rollback. Based on the analysis, we strategically introduce parallel speculative branches to preemptively hedge against likely rejections. Meanwhile, to enhance parallelism, we jointly orchestrate adaptive draft lengths with a hybrid combination of the implicit draft model confidence and explicit reusing of target model features. Extensive experiments across various models and benchmarks show that SpecBranch achieves over \textbf{1.8}$\times \sim$ \textbf{4.5}$\times$ speedups against the auto-regressive decoding and reduces rollback tokens by $\textbf{50}$\% for poorly aligned models, realizing its applicability for real-world deployments.
Authors: Lianhao Yin, Ozanan Meireles, Guy Rosman, Daniela Rus
Abstract: Real-time video understanding is critical to guide procedures in minimally invasive surgery (MIS). However, supervised learning approaches require large, annotated datasets that are scarce due to annotation efforts that are prohibitive, e.g., in medical fields. Although self-supervision methods can address such limitations, current self-supervised methods often fail to capture structural and physical information in a form that generalizes across tasks. We propose Compress-to-Explore (C2E), a novel self-supervised framework that leverages Kolmogorov complexity to learn compact, informative representations from surgical videos. C2E uses entropy-maximizing decoders to compress images while preserving clinically relevant details, improving encoder performance without labeled data. Trained on large-scale unlabeled surgical datasets, C2E demonstrates strong generalization across a variety of surgical ML tasks, such as workflow classification, tool-tissue interaction classification, segmentation, and diagnosis tasks, providing improved performance as a surgical visual foundation model. As we further show in the paper, the model's internal compact representation better disentangles features from different structural parts of images. The resulting performance improvements highlight the yet untapped potential of self-supervised learning to enhance surgical AI and improve outcomes in MIS.
Authors: Vassilis Lyberatos, Spyridon Kantarelis, Ioanna Zioga, Christina Anagnostopoulou, Giorgos Stamou, Anastasia Georgaki
Abstract: This study investigates emotional expression and perception in music performance using computational and neurophysiological methods. The influence of different performance settings, such as repertoire, diatonic modal etudes, and improvisation, as well as levels of expressiveness, on performers' emotional communication and listeners' reactions is explored. Professional musicians performed various tasks, and emotional annotations were provided by both performers and the audience. Audio analysis revealed that expressive and improvisational performances exhibited unique acoustic features, while emotion analysis showed stronger emotional responses. Neurophysiological measurements indicated greater relaxation in improvisational performances. This multimodal study highlights the significance of expressivity in enhancing emotional communication and audience engagement.
Authors: Reyhaneh Keshavarzpour, Eghbal Mansoori
Abstract: Identification of antimicrobial peptides is an important and necessary issue in today's era. Antimicrobial peptides are essential as an alternative to antibiotics for biomedical applications and many other practical applications. These oligopeptides are useful in drug design and cause innate immunity against microorganisms. Artificial intelligence algorithms have played a significant role in the ease of identifying these peptides.This research is improved by improving proposed method in the field of antimicrobial peptides prediction. Suggested method is improved by combining the best coding method from different perspectives, In the following a deep neural network to balance the imbalanced combined datasets. The results of this research show that the proposed method have a significant improvement in the accuracy and efficiency of the prediction of antimicrobial peptides and are able to provide the best results compared to the existing methods. These development in the field of prediction and classification of antimicrobial peptides, basically in the fields of medicine and pharmaceutical industries, have high effectiveness and application.
Authors: Selin Yildirim, Deming Chen
Abstract: Recent advancements in speculative decoding have demonstrated considerable speedup across a wide array of large language model (LLM) tasks. Speculative decoding inherently relies on sacrificing extra memory allocations to generate several candidate tokens, of which acceptance rate drives the speedup. However, deploying speculative decoding on memory-constrained devices, such as mobile GPUs, remains as a significant challenge in real-world scenarios. In this work, we present a device-aware inference engine named SpecMemo that can smartly control memory allocations at finer levels to enable multi-turn chatbots with speculative decoding on such limited memory devices. Our methodology stems from theoretically modeling memory footprint of speculative decoding to determine a lower bound on the required memory budget while retaining speedup. SpecMemo empirically acquires a careful balance between minimizing redundant memory allocations for rejected candidate tokens and maintaining competitive performance gains from speculation. Notably, with SpecMemo's memory management, we maintain 96% of overall throughput from speculative decoding on MT-Bench, with reduced generation-memory by 65% on single Nvidia Titan RTX. Given multiple constrained GPUs, we build on top of previous speculative decoding architectures to facilitate big-model inference by distributing Llama-2-70B-Chat model, on which we provide novel batched speculative decoding to increase usability of multiple small server GPUs. This novel framework demonstrates 2x speedup over distributed and batched vanilla decoding with the base model on eight AMD MI250 GPUs. Moreover, inference throughput increases remarkably 8x with batch size 10. Our work contributes to democratized LLM applications in resource-constrained environments, providing a pathway for faster and cheaper deployment of real-world LLM applications with robust performance.
Authors: Runkang Yang, Peng Sun, Xinyi Shang, Yi Tang, Tao Lin
Abstract: Data inherently possesses dual attributes: samples and targets. For targets, knowledge distillation has been widely employed to accelerate model convergence, primarily relying on teacher-generated soft target supervision. Conversely, recent advancements in data-efficient learning have emphasized sample optimization techniques, such as dataset distillation, while neglected the critical role of target. This dichotomy motivates our investigation into understanding how both sample and target collectively influence training dynamic. To address this gap, we first establish a taxonomy of existing paradigms through the lens of sample-target interactions, categorizing them into distinct sample-to-target mapping strategies. Building upon this foundation, we then propose a novel unified loss framework to assess their impact on training efficiency. Through extensive empirical studies on our proposed strategies, we comprehensively analyze how variations in target and sample types, quantities, and qualities influence model training, providing six key insights to enhance training efficacy.
Authors: Akshat Dubey, Aleksandar An\v{z}el, Georges Hattab
Abstract: The field of health informatics has been profoundly influenced by the development of random forest models, which have led to significant advances in the interpretability of feature interactions. These models are characterized by their robustness to overfitting and parallelization, making them particularly useful in this domain. However, the increasing number of features and estimators in random forests can prevent domain experts from accurately interpreting global feature interactions, thereby compromising trust and regulatory compliance. A method called the surrogate interpretability graph has been developed to address this issue. It uses graphs and mixed-integer linear programming to analyze and visualize feature interactions. This improves their interpretability by visualizing the feature usage per decision-feature-interaction table and the most dominant hierarchical decision feature interactions for predictions. The implementation of a surrogate interpretable graph enhances global interpretability, which is critical for such a high-stakes domain.
Authors: Chengxi Li, Ming Xiao, Mikael Skoglund
Abstract: In this paper, we investigate the problem of distributed learning (DL) in the presence of Byzantine attacks. For this problem, various robust bounded aggregation (RBA) rules have been proposed at the central server to mitigate the impact of Byzantine attacks. However, current DL methods apply RBA rules for the local gradients from the honest devices and the disruptive information from Byzantine devices, and the learning performance degrades significantly when the local gradients of different devices vary considerably from each other. To overcome this limitation, we propose a new DL method to cope with Byzantine attacks based on coded robust aggregation (CRA-DL). Before training begins, the training data are allocated to the devices redundantly. During training, in each iteration, the honest devices transmit coded gradients to the server computed from the allocated training data, and the server then aggregates the information received from both honest and Byzantine devices using RBA rules. In this way, the global gradient can be approximately recovered at the server to update the global model. Compared with current DL methods applying RBA rules, the improvement of CRA-DL is attributed to the fact that the coded gradients sent by the honest devices are closer to each other. This closeness enhances the robustness of the aggregation against Byzantine attacks, since Byzantine messages tend to be significantly different from those of honest devices in this case. We theoretically analyze the convergence performance of CRA-DL. Finally, we present numerical results to verify the superiority of the proposed method over existing baselines, showing its enhanced learning performance under Byzantine attacks.
Authors: Lukas Rauch, Moritz Wirth, Denis Huseljic, Marek Herde, Bernhard Sick, Matthias A{\ss}enmacher
Abstract: The advent of large language models (LLMs) capable of producing general-purpose representations lets us revisit the practicality of deep active learning (AL): By leveraging frozen LLM embeddings, we can mitigate the computational costs of iteratively fine-tuning large backbones. This study establishes a benchmark and systematically investigates the influence of LLM embedding quality on query strategies in deep AL. We employ five top-performing models from the massive text embedding benchmark (MTEB) leaderboard and two baselines for ten diverse text classification tasks. Our findings reveal key insights: First, initializing the labeled pool using diversity-based sampling synergizes with high-quality embeddings, boosting performance in early AL iterations. Second, the choice of the optimal query strategy is sensitive to embedding quality. While the computationally inexpensive Margin sampling can achieve performance spikes on specific datasets, we find that strategies like Badge exhibit greater robustness across tasks. Importantly, their effectiveness is often enhanced when paired with higher-quality embeddings. Our results emphasize the need for context-specific evaluation of AL strategies, as performance heavily depends on embedding quality and the target task.
Authors: Wanshan Cui, Yejin Jeong, Inwook Song, Gyuri Kim, Minsang Kwon, Donghun Lee
Abstract: Accurate prediction of polymer material properties through data-driven approaches greatly accelerates novel material development by reducing redundant experiments and trial-and-error processes. However, inevitable outliers in empirical measurements can severely skew machine learning results, leading to erroneous prediction models and suboptimal material designs. To address this limitation, we propose a novel approach to enhance dataset quality efficiently by integrating multi-algorithm outlier detection with selective re-experimentation of unreliable outlier cases. To validate the empirical effectiveness of the approach, we systematically construct a new dataset containing 701 measurements of three key mechanical properties: glass transition temperature ($T_g$), tan $\delta$ peak, and crosslinking density ($v_{c}$). To demonstrate its general applicability, we report the performance improvements across multiple machine learning models, including Elastic Net, SVR, Random Forest, and TPOT, to predict the three key properties. Our method reliably reduces prediction error (RMSE) and significantly improves accuracy with minimal additional experimental work, requiring only about 5% of the dataset to be re-measured.These findings highlight the importance of data quality enhancement in achieving reliable machine learning applications in polymer science and present a scalable strategy for improving predictive reliability in materials science.
Authors: Takao Fujii, Katie Seaborn, Madeleine Steeds, Jun Kato
Abstract: Conversational agents that mimic people have raised questions about the ethics of anthropomorphizing machines with human social identity cues. Critics have also questioned assumptions of identity neutrality in humanlike agents. Recent work has revealed that intersectional Japanese pronouns can elicit complex and sometimes evasive impressions of agent identity. Yet, the role of other "neutral" non-pronominal self-referents (NPSR) and voice as a socially expressive medium remains unexplored. In a crowdsourcing study, Japanese participants (N = 204) evaluated three ChatGPT voices (Juniper, Breeze, and Ember) using seven self-referents. We found strong evidence of voice gendering alongside the potential of intersectional self-referents to evade gendering, i.e., ambiguity through neutrality and elusiveness. Notably, perceptions of age and formality intersected with gendering as per sociolinguistic theories, especially boku and watakushi. This work provides a nuanced take on agent identity perceptions and champions intersectional and culturally-sensitive work on voice agents.
Authors: Ruilin Xu, Zongxuan Xie, Pengfei Chen
Abstract: We present eACGM, a full-stack AI/ML system monitoring framework based on eBPF. eACGM collects real-time performance data from key hardware components, including the GPU and network communication layer, as well as from key software stacks such as CUDA, Python, and PyTorch, all without requiring any code instrumentation or modifications. Additionally, it leverages libnvml to gather process-level GPU resource usage information. By applying a Gaussian Mixture Model (GMM) to the collected multidimensional performance metrics for statistical modeling and clustering analysis, eACGM effectively identifies complex failure modes, such as latency anomalies, hardware failures, and communication inefficiencies, enabling rapid diagnosis of system bottlenecks and abnormal behaviors. To evaluate eACGM's effectiveness and practicality, we conducted extensive empirical studies and case analyses in multi-node distributed training scenarios. The results demonstrate that eACGM, while maintaining a non-intrusive and low-overhead profile, successfully captures critical performance anomalies during model training and inference. Its stable anomaly detection performance and comprehensive monitoring capabilities validate its applicability and scalability in real-world production environments, providing strong support for performance optimization and fault diagnosis in large-scale AI/ML systems.
Authors: Wang Mengjie, Zhu Huiping, Li Jian, Shi Wenxiu, Zhang Song
Abstract: With the advancement of autonomous and assisted driving technologies, higher demands are placed on the ability to understand complex driving scenarios. Multimodal general large models have emerged as a solution for this challenge. However, applying these models in vertical domains involves difficulties such as data collection, model training, and deployment optimization. This paper proposes a comprehensive method for optimizing multimodal models in driving scenarios, including cone detection, traffic light recognition, speed limit recommendation, and intersection alerts. The method covers key aspects such as dynamic prompt optimization, dataset construction, model training, and deployment. Specifically, the dynamic prompt optimization adjusts the prompts based on the input image content to focus on objects affecting the ego vehicle, enhancing the model's task-specific focus and judgment capabilities. The dataset is constructed by combining real and synthetic data to create a high-quality and diverse multimodal training dataset, improving the model's generalization in complex driving environments. In model training, advanced techniques like knowledge distillation, dynamic fine-tuning, and quantization are integrated to reduce storage and computational costs while boosting performance. Experimental results show that this systematic optimization method not only significantly improves the model's accuracy in key tasks but also achieves efficient resource utilization, providing strong support for the practical application of driving scenario perception technologies.
Authors: Yinjie Zhao, Heng Zhao, Bihan Wen, Yew-Soon Ong, Joey Tianyi Zhou
Abstract: With the rapid development of vision tasks and the scaling on datasets and models, redundancy reduction in vision datasets has become a key area of research. To address this issue, dataset distillation (DD) has emerged as a promising approach to generating highly compact synthetic datasets with significantly less redundancy while preserving essential information. However, while DD has been extensively studied for image datasets, DD on video datasets remains underexplored. Video datasets present unique challenges due to the presence of temporal information and varying levels of redundancy across different classes. Existing DD approaches assume a uniform level of temporal redundancy across all different video semantics, which limits their effectiveness on video datasets. In this work, we propose Dynamic-Aware Video Distillation (DAViD), a Reinforcement Learning (RL) approach to predict the optimal Temporal Resolution of the synthetic videos. A teacher-in-the-loop reward function is proposed to update the RL agent policy. To the best of our knowledge, this is the first study to introduce adaptive temporal resolution based on video semantics in video dataset distillation. Our approach significantly outperforms existing DD methods, demonstrating substantial improvements in performance. This work paves the way for future research on more efficient and semantic-adaptive video dataset distillation research.
Authors: Prachi Jadhav, Hongwei Jin, Ewa Deelman, Prasanna Balaprakash
Abstract: High-Performance Computing (HPC) job scheduling involves balancing conflicting objectives such as minimizing makespan, reducing wait times, optimizing resource use, and ensuring fairness. Traditional methods, including heuristic-based (e.g., First-Come-First-Served) or intensive optimization techniques, often lack adaptability to dynamic workloads and heterogeneous HPC systems. To address this, we propose a novel Large Language Model (LLM)-based scheduler using a ReAct-style framework (Reason + Act), enabling iterative, interpretable decision-making. The system incorporates a scratchpad memory to track scheduling history and refine decisions via natural language feedback, while a constraint enforcement module ensures feasibility and safety. We evaluate our approach using OpenAI's O4-Mini and Anthropic's Claude 3.7 across seven real-world HPC workload scenarios, including heterogeneous mixes, bursty patterns, and adversarial cases. Comparisons against FCFS, Shortest Job First, and Google OR-Tools (on 10 to 100 jobs) reveal that LLM-based scheduling effectively balances multiple objectives while offering transparent reasoning through natural language traces. The method excels in constraint satisfaction and adapts to diverse workloads without domain-specific training. However, a trade-off between reasoning quality and computational overhead challenges real-time deployment. This work presents the first comprehensive study of reasoning-capable LLMs for HPC scheduling, demonstrating their potential to handle multiobjective optimization while highlighting limitations in computational efficiency. The findings provide insights into leveraging advanced language models for complex scheduling problems in dynamic HPC environments.
Authors: Shriphani Palakodety
Abstract: Many identity systems assign a single, static identifier to an individual for life, reused across domains like healthcare, finance, and education. These Universal Lifelong Identifiers (ULIs) underpin critical workflows but now pose systemic privacy risks. We take the position that ULIs are fundamentally incompatible with the AI era and must be phased out. We articulate a threat model grounded in modern AI capabilities and show that traditional safeguards such as redaction, consent, and access controls are no longer sufficient. We define core properties for identity systems in the AI era and present a cryptographic framework that satisfies them while retaining compatibility with existing identifier workflows. Our design preserves institutional workflows, supports essential functions such as auditability and delegation, and offers a practical migration path beyond ULIs.
Authors: Raj Patel, Himanshu Tripathi, Jasper Stone, Noorbakhsh Amiri Golilarz, Sudip Mittal, Shahram Rahimi, Vini Chaudhary
Abstract: The rapid adoption of machine learning (ML) technologies has driven organizations across diverse sectors to seek efficient and reliable methods to accelerate model development-to-deployment. Machine Learning Operations (MLOps) has emerged as an integrative approach addressing these requirements by unifying relevant roles and streamlining ML workflows. As the MLOps market continues to grow, securing these pipelines has become increasingly critical. However, the unified nature of MLOps ecosystem introduces vulnerabilities, making them susceptible to adversarial attacks where a single misconfiguration can lead to compromised credentials, severe financial losses, damaged public trust, and the poisoning of training data. Our paper presents a systematic application of the MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) framework, a comprehensive and continuously updated catalog of AI-focused attacks, to systematically assess attacks across different phases of the MLOps ecosystem. We begin by examining the preparatory phases during which adversaries acquire the essential intelligence required to initiate their attacks. We then present a structured taxonomy of attack techniques explicitly mapped to corresponding phases of the MLOps ecosystem, supported by examples drawn from red-teaming exercises and real-world incidents. This is followed by a taxonomy of mitigation strategies aligned with these attack categories, offering actionable early-stage defenses to strengthen the security of MLOps ecosystem. Given the rapid evolution and adoption of MLOps, we further highlight key research gaps that require immediate attention. Our work emphasizes the importance of implementing robust security protocols from the outset, empowering practitioners to safeguard MLOps ecosystem against evolving cyber attacks.
Authors: Feng Wang, Yiding Sun, Jiaxin Mao, Wei Xue, Danqing Xu
Abstract: Large language models (LLMs) have demonstrated remarkable capabilities across various professional domains, with their performance typically evaluated through standardized benchmarks. However, the development of financial RAG benchmarks has been constrained by data confidentiality issues and the lack of dynamic data integration. To address this issue, we introduces FinS-Pilot, a novel benchmark for evaluating RAG systems in online financial applications. Constructed from real-world financial assistant interactions, our benchmark incorporates both real-time API data and structured text sources, organized through an intent classification framework covering critical financial domains such as equity analysis and macroeconomic forecasting. The benchmark enables comprehensive evaluation of financial assistants' capabilities in handling both static knowledge and time-sensitive market information. Through systematic experiments with multiple Chinese leading LLMs, we demonstrate FinS-Pilot's effectiveness in identifying models suitable for financial applications while addressing the current gap in specialized evaluation tools for the financial domain. Our work contributes both a practical evaluation framework and a curated dataset to advance research in financial NLP systems. The code and dataset are accessible on GitHub\footnote{https://github.com/PhealenWang/financial\_rag\_benchmark}.
URLs: https://github.com/PhealenWang/financial\_rag\_benchmark
Authors: Haoshuai Zhou, Changgeng Mo, Boxuan Cao, Linkai Li, Shan Xiang Wang
Abstract: Personalized speech intelligibility prediction is challenging. Previous approaches have mainly relied on audiograms, which are inherently limited in accuracy as they only capture a listener's hearing threshold for pure tones. Rather than incorporating additional listener features, we propose a novel approach that leverages an individual's existing intelligibility data to predict their performance on new audio. We introduce the Support Sample-Based Intelligibility Prediction Network (SSIPNet), a deep learning model that leverages speech foundation models to build a high-dimensional representation of a listener's speech recognition ability from multiple support (audio, score) pairs, enabling accurate predictions for unseen audio. Results on the Clarity Prediction Challenge dataset show that, even with a small number of support (audio, score) pairs, our method outperforms audiogram-based predictions. Our work presents a new paradigm for personalized speech intelligibility prediction.
Authors: Duzhen Zhang, Yong Ren, Zhong-Zhi Li, Yahan Yu, Jiahua Dong, Chenxing Li, Zhilong Ji, Jinfeng Bai
Abstract: Multimodal Continual Instruction Tuning (MCIT) aims to finetune Multimodal Large Language Models (MLLMs) to continually align with human intent across sequential tasks. Existing approaches often rely on the Mixture-of-Experts (MoE) LoRA framework to preserve previous instruction alignments. However, these methods are prone to Catastrophic Forgetting (CF), as they aggregate all LoRA blocks via simple summation, which compromises performance over time. In this paper, we identify a critical parameter inefficiency in the MoELoRA framework within the MCIT context. Based on this insight, we propose BranchLoRA, an asymmetric framework to enhance both efficiency and performance. To mitigate CF, we introduce a flexible tuning-freezing mechanism within BranchLoRA, enabling branches to specialize in intra-task knowledge while fostering inter-task collaboration. Moreover, we incrementally incorporate task-specific routers to ensure an optimal branch distribution over time, rather than favoring the most recent task. To streamline inference, we introduce a task selector that automatically routes test inputs to the appropriate router without requiring task identity. Extensive experiments on the latest MCIT benchmark demonstrate that BranchLoRA significantly outperforms MoELoRA and maintains its superiority across various MLLM sizes.
Authors: Mohammad Saleh Torkestani, Taha Mansouri
Abstract: This paper presents a theoretical framework for addressing the challenges posed by generative artificial intelligence (AI) in higher education assessment through a machine-versus-machine approach. Large language models like GPT-4, Claude, and Llama increasingly demonstrate the ability to produce sophisticated academic content, traditional assessment methods face an existential threat, with surveys indicating 74-92% of students experimenting with these tools for academic purposes. Current responses, ranging from detection software to manual assessment redesign, show significant limitations: detection tools demonstrate bias against non-native English writers and can be easily circumvented, while manual frameworks rely heavily on subjective judgment and assume static AI capabilities. This paper introduces a dual strategy paradigm combining static analysis and dynamic testing to create a comprehensive theoretical framework for assessment vulnerability evaluation. The static analysis component comprises eight theoretically justified elements: specificity and contextualization, temporal relevance, process visibility requirements, personalization elements, resource accessibility, multimodal integration, ethical reasoning requirements, and collaborative elements. Each element addresses specific limitations in generative AI capabilities, creating barriers that distinguish authentic human learning from AI-generated simulation. The dynamic testing component provides a complementary approach through simulation-based vulnerability assessment, addressing limitations in pattern-based analysis. The paper presents a theoretical framework for vulnerability scoring, including the conceptual basis for quantitative assessment, weighting frameworks, and threshold determination theory.
Authors: Lajos Muzsai, David Imolai, Andr\'as Luk\'acs
Abstract: Large Language Models (LLMs) still struggle with the structured reasoning and tool-assisted computation needed for problem solving in cybersecurity applications. In this work, we introduce "random-crypto", a cryptographic Capture-the-Flag (CTF) challenge generator framework that we use to fine-tune a tool-augmented Llama-3.1-8B with Guided Reinforcement Prompt Optimisation (GRPO), allowing the agent to iteratively write and execute Python inside an isolated REPL. GRPO yields a +53% absolute jump in Pass@8 on unseen "random-crypto" tasks (0.35 -> 0.88) and raises Majority@8 to 0.41. The fine-tuned agent also generalizes to an external dataset. On a subset of picoCTF cryptography problems, it improves Pass@8 by +13 pp. Ablations show the gains stem from more reliable tool invocation and code synthesis, rather than superficial prompt adaptation.
Authors: Beichen Huang, Ran Cheng, Kay Chen Tan
Abstract: We introduce EvoGit, a decentralized multi-agent framework for collaborative software development driven by autonomous code evolution. EvoGit deploys a population of independent coding agents, each proposing edits to a shared codebase without centralized coordination, explicit message passing, or shared memory. Instead, all coordination emerges through a Git-based phylogenetic graph that tracks the full version lineage and enables agents to asynchronously read from and write to the evolving code repository. This graph-based structure supports fine-grained branching, implicit concurrency, and scalable agent interaction while preserving a consistent historical record. Human involvement is minimal but strategic: users define high-level goals, periodically review the graph, and provide lightweight feedback to promote promising directions or prune unproductive ones. Experiments demonstrate EvoGit's ability to autonomously produce functional and modular software artifacts across two real-world tasks: (1) building a web application from scratch using modern frameworks, and (2) constructing a meta-level system that evolves its own language-model-guided solver for the bin-packing optimization problem. Our results underscore EvoGit's potential to establish a new paradigm for decentralized, automated, and continual software development. EvoGit is open-sourced at https://github.com/BillHuang2001/evogit.
Authors: Qingyu Xiao, Yuanlin Chang, Youtian Du
Abstract: Effective agent exploration remains a core challenge in reinforcement learning (RL) for complex discrete state-space environments, particularly under partial observability. This paper presents a decoupled hierarchical RL framework integrating state abstraction (DcHRL-SA) to address this issue. The proposed method employs a dual-level architecture, consisting of a high level RL-based actor and a low-level rule-based policy, to promote effective exploration. Additionally, state abstraction method is incorporated to cluster discrete states, effectively lowering state dimensionality. Experiments conducted in two discrete customized grid environments demonstrate that the proposed approach consistently outperforms PPO in terms of exploration efficiency, convergence speed, cumulative reward, and policy stability. These results demonstrate a practical approach for integrating decoupled hierarchical policies and state abstraction in discrete grids with large-scale exploration space. Code will be available at https://github.com/XQY169/DcHRL-SA.
Authors: Hui Liu, Shiye Tian, Xuejun Liu
Abstract: The de novo generation of drug-like molecules capable of inducing desirable phenotypic changes is receiving increasing attention. However, previous methods predominantly rely on expression profiles to guide molecule generation, but overlook the perturbative effect of the molecules on cellular contexts. To overcome this limitation, we propose SmilesGEN, a novel generative model based on variational autoencoder (VAE) architecture to generate molecules with potential therapeutic effects. SmilesGEN integrates a pre-trained drug VAE (SmilesNet) with an expression profile VAE (ProfileNet), jointly modeling the interplay between drug perturbations and transcriptional responses in a common latent space. Specifically, ProfileNet is imposed to reconstruct pre-treatment expression profiles when eliminating drug-induced perturbations in the latent space, while SmilesNet is informed by desired expression profiles to generate drug-like molecules. Our empirical experiments demonstrate that SmilesGEN outperforms current state-of-the-art models in generating molecules with higher degree of validity, uniqueness, novelty, as well as higher Tanimoto similarity to known ligands targeting the relevant proteins. Moreover, we evaluate SmilesGEN for scaffold-based molecule optimization and generation of therapeutic agents, and confirmed its superior performance in generating molecules with higher similarity to approved drugs. SmilesGEN establishes a robust framework that leverages gene signatures to generate drug-like molecules that hold promising potential to induce desirable cellular phenotypic changes.
Authors: Shuo Yan, Yuliang Yan, Bin Ma, Chenao Li, Haochun Tang, Jiahua Lu, Minhua Lin, Yuyuan Feng, Hui Xiong, Enyan Dai
Abstract: Recently, extensive deep learning architectures and pretraining strategies have been explored to support downstream protein applications. Additionally, domain-specific models incorporating biological knowledge have been developed to enhance performance in specialized tasks. In this work, we introduce $\textbf{Protap}$, a comprehensive benchmark that systematically compares backbone architectures, pretraining strategies, and domain-specific models across diverse and realistic downstream protein applications. Specifically, Protap covers five applications: three general tasks and two novel specialized tasks, i.e., enzyme-catalyzed protein cleavage site prediction and targeted protein degradation, which are industrially relevant yet missing from existing benchmarks. For each application, Protap compares various domain-specific models and general architectures under multiple pretraining settings. Our empirical studies imply that: (i) Though large-scale pretraining encoders achieve great results, they often underperform supervised encoders trained on small downstream training sets. (ii) Incorporating structural information during downstream fine-tuning can match or even outperform protein language models pretrained on large-scale sequence corpora. (iii) Domain-specific biological priors can enhance performance on specialized downstream tasks. Code and datasets are publicly available at https://github.com/Trust-App-AI-Lab/protap.
Authors: Xu Zhang, Haoye Qiu, Weixuan Liang, Hui Liu, Junhui Hou, Yuheng Jia
Abstract: Ensemble clustering has demonstrated great success in practice; however, its theoretical foundations remain underexplored. This paper examines the generalization performance of ensemble clustering, focusing on generalization error, excess risk and consistency. We derive a convergence rate of generalization error bound and excess risk bound both of $\mathcal{O}(\sqrt{\frac{\log n}{m}}+\frac{1}{\sqrt{n}})$, with $n$ and $m$ being the numbers of samples and base clusterings. Based on this, we prove that when $m$ and $n$ approach infinity and $m$ is significantly larger than log $n$, i.e., $m,n\to \infty, m\gg \log n$, ensemble clustering is consistent. Furthermore, recognizing that $n$ and $m$ are finite in practice, the generalization error cannot be reduced to zero. Thus, by assigning varying weights to finite clusterings, we minimize the error between the empirical average clusterings and their expectation. From this, we theoretically demonstrate that to achieve better clustering performance, we should minimize the deviation (bias) of base clustering from its expectation and maximize the differences (diversity) among various base clusterings. Additionally, we derive that maximizing diversity is nearly equivalent to a robust (min-max) optimization model. Finally, we instantiate our theory to develop a new ensemble clustering algorithm. Compared with SOTA methods, our approach achieves average improvements of 6.1%, 7.3%, and 6.0% on 10 datasets w.r.t. NMI, ARI, and Purity. The code is available at https://github.com/xuz2019/GPEC.
Authors: Nikola Balic
Abstract: Autonomous multi-agent AI systems are poised to transform various industries, particularly software development and knowledge work. Understanding current perceptions among professionals is crucial for anticipating adoption challenges, ethical considerations, and future workforce development. This study analyzes responses from 130 participants to a survey on the capabilities, impact, and governance of AI agents. We explore expected timelines for AI replacing programmers, identify perceived barriers to deployment, and examine beliefs about responsibility when agents make critical decisions. Key findings reveal three distinct clusters of respondents. While the study explored factors associated with current AI agent deployment, the initial logistic regression model did not yield statistically significant predictors, suggesting that deployment decisions are complex and may be influenced by factors not fully captured or that a larger sample is needed. These insights highlight the need for organizations to address compliance concerns (a commonly cited barrier) and establish clear governance frameworks as they integrate autonomous agents into their workflows.
Authors: David Sasu, Kweku Andoh Yamoah, Benedict Quartey, Natalie Schluter
Abstract: Enabling robots to accurately interpret and execute spoken language instructions is essential for effective human-robot collaboration. Traditional methods rely on speech recognition to transcribe speech into text, often discarding crucial prosodic cues needed for disambiguating intent. We propose a novel approach that directly leverages speech prosody to infer and resolve instruction intent. Predicted intents are integrated into large language models via in-context learning to disambiguate and select appropriate task plans. Additionally, we present the first ambiguous speech dataset for robotics, designed to advance research in speech disambiguation. Our method achieves 95.79% accuracy in detecting referent intents within an utterance and determines the intended task plan of ambiguous instructions with 71.96% accuracy, demonstrating its potential to significantly improve human-robot communication.
Authors: Malik A. Altayar, Muhyeeddin Alqaraleh, Mowafaq Salem Alzboon, Wesam T. Almagharbeh
Abstract: Introduction: Personal identification is a critical aspect of forensic sciences, security, and healthcare. While conventional biometrics systems such as DNA profiling and iris scanning offer high accuracy, they are time-consuming and costly. Objectives: This study investigates the relationship between fingerprint patterns and ABO blood group classification to explore potential correlations between these two traits. Methods: The study analyzed 200 individuals, categorizing their fingerprints into three types: loops, whorls, and arches. Blood group classification was also recorded. Statistical analysis, including chi-square and Pearson correlation tests, was used to assess associations between fingerprint patterns and blood groups. Results: Loops were the most common fingerprint pattern, while blood group O+ was the most prevalent among the participants. Statistical analysis revealed no significant correlation between fingerprint patterns and blood groups (p > 0.05), suggesting that these traits are independent. Conclusions: Although the study showed limited correlation between fingerprint patterns and ABO blood groups, it highlights the importance of future research using larger and more diverse populations, incorporating machine learning approaches, and integrating multiple biometric signals. This study contributes to forensic science by emphasizing the need for rigorous protocols and comprehensive investigations in personal identification.
Authors: Shriraj P. Sawant, Krishna P. Miyapuram
Abstract: The ability to learn and retain a wide variety of tasks is a hallmark of human intelligence that has inspired research in artificial general intelligence. Continual learning approaches provide a significant step towards achieving this goal. It has been known that task variability and context switching are challenging for learning in neural networks. Catastrophic forgetting refers to the poor performance on retention of a previously learned task when a new task is being learned. Switching between different task contexts can be a useful approach to mitigate the same by preventing the interference between the varying task weights of the network. This paper introduces Elastic Weight Generative Networks (EWGN) as an idea for context switching between two different tasks. The proposed EWGN architecture uses an additional network that generates the weights of the primary network dynamically while consolidating the weights learned. The weight generation is input-dependent and thus enables context switching. Using standard computer vision datasets, namely MNIST and fashion-MNIST, we analyse the retention of previously learned task representations in Fully Connected Networks, Convolutional Neural Networks, and EWGN architectures with Stochastic Gradient Descent and Elastic Weight Consolidation learning algorithms. Understanding dynamic weight generation and context-switching ability can be useful in enabling continual learning for improved performance.
Authors: Tadesse K. Bahiru, Haileleol Tibebu, Ioannis A. Kakadiaris
Abstract: Artificial intelligence has transformed numerous industries, from healthcare to finance, enhancing decision-making through automated systems. However, the reliability of these systems is mainly dependent on the quality of the underlying datasets, raising ongoing concerns about transparency, accountability, and potential biases. This paper introduces a scorecard designed to evaluate the development of AI datasets, focusing on five key areas from the system card framework data development life cycle: data dictionary, collection process, composition, motivation, and pre-processing. The method follows a structured approach, using an intake form and scoring criteria to assess the quality and completeness of the data set. Applied to four diverse datasets, the methodology reveals strengths and improvement areas. The results are compiled using a scoring system that provides tailored recommendations to enhance the transparency and integrity of the data set. The scorecard addresses technical and ethical aspects, offering a holistic evaluation of data practices. This approach aims to improve the quality of the data set. It offers practical guidance to curators and researchers in developing responsible AI systems, ensuring fairness and accountability in decision support systems.
Authors: Mengliang He, Jiayi Zeng, Yankai Jiang, Wei Zhang, Zeming Liu, Xiaoming Shi, Aimin Zhou
Abstract: While large language models (LLMs) show promise in code generation, existing benchmarks neglect the flowchart-based code generation. To promote further research on flowchart-based code generation, this work presents Flow2Code, a novel benchmark for flowchart-based code generation evaluation. The evaluation dataset spans 15 programming languages and includes 5,622 code segments paired with 16,866 flowcharts of three types: code, UML, and pseudocode. Extensive experiments with 13 multimodal LLMs reveal that current LLMs can not generate code based on flowcharts perfectly. Besides, experiment results show that the supervised fine-tuning technique contributes greatly to the models' performance. We publicly release our code and datasets at https://github.com/hml-github/Flow2Code.
Authors: Yoonjun Cho, Soeun Kim, Dongjae Jeon, Kyelim Lee, Beomsoo Lee, Albert No
Abstract: Decomposing weight matrices into quantization and low-rank components ($\mathbf{W} \approx \mathbf{Q} + \mathbf{L}\mathbf{R}$) is a widely used technique for compressing large language models (LLMs). Existing joint optimization methods iteratively alternate between quantization and low-rank approximation. However, these methods tend to prioritize one component at the expense of the other, resulting in suboptimal decompositions that fail to leverage each component's unique strengths. In this work, we introduce Outlier-Driven Low-Rank Initialization (ODLRI), which assigns low-rank components the specific role of capturing activation-sensitive weights. This structured decomposition mitigates outliers' negative impact on quantization, enabling more effective balance between quantization and low-rank approximation. Experiments on Llama2 (7B, 13B, 70B), Llama3-8B, and Mistral-7B demonstrate that incorporating ODLRI into the joint optimization framework consistently reduces activation-aware error, minimizes quantization scale, and improves perplexity and zero-shot accuracy in low-bit settings.
Authors: Emmy Postma, Cristian Tejedor-Garcia
Abstract: Speech impairments are prevalent biomarkers for Parkinson's Disease (PD), motivating the development of diagnostic techniques using speech data for clinical applications. Although deep acoustic features have shown promise for PD classification, their effectiveness often varies due to individual speaker differences, a factor that has not been thoroughly explored in the existing literature. This study investigates the effectiveness of three pre-trained audio embeddings (OpenL3, VGGish and Wav2Vec2.0 models) for PD classification. Using the NeuroVoz dataset, OpenL3 outperforms others in diadochokinesis (DDK) and listen and repeat (LR) tasks, capturing critical acoustic features for PD detection. Only Wav2Vec2.0 shows significant gender bias, achieving more favorable results for male speakers, in DDK tasks. The misclassified cases reveal challenges with atypical speech patterns, highlighting the need for improved feature extraction and model robustness in PD detection.
Authors: Xuefeng Jiang, Tian Wen, Zhiqin Yang, Lvhua Wu, Yufeng Chen, Sheng Sun, Yuwei Wang, Min Liu
Abstract: In recent years, federated learning (FL) has made significant advance in privacy-sensitive applications. However, it can be hard to ensure that FL participants provide well-annotated data for training. The corresponding annotations from different clients often contain complex label noise at varying levels. This label noise issue has a substantial impact on the performance of the trained models, and clients with greater noise levels can be largely attributed for this degradation. To this end, it is necessary to develop an effective optimization strategy to alleviate the adverse effects of these noisy clients.In this study, we present a two-stage optimization framework, MaskedOptim, to address this intricate label noise problem. The first stage is designed to facilitate the detection of noisy clients with higher label noise rates. The second stage focuses on rectifying the labels of the noisy clients' data through an end-to-end label correction mechanism, aiming to mitigate the negative impacts caused by misinformation within datasets. This is achieved by learning the potential ground-truth labels of the noisy clients' datasets via backpropagation. To further enhance the training robustness, we apply the geometric median based model aggregation instead of the commonly-used vanilla averaged model aggregation. We implement sixteen related methods and conduct evaluations on three image datasets and one text dataset with diverse label noise patterns for a comprehensive comparison. Extensive experimental results indicate that our proposed framework shows its robustness in different scenarios. Additionally, our label correction framework effectively enhances the data quality of the detected noisy clients' local datasets. % Our codes will be open-sourced to facilitate related research communities. Our codes are available via https://github.com/Sprinter1999/MaskedOptim .
Authors: Aditya Kamlesh Parikh, Cristian Tejedor-Garcia, Catia Cucchiarini, Helmer Strik
Abstract: Computer-Assisted Pronunciation Training (CAPT) systems employ automatic measures of pronunciation quality, such as the goodness of pronunciation (GOP) metric. GOP relies on forced alignments, which are prone to labeling and segmentation errors due to acoustic variability. While alignment-free methods address these challenges, they are computationally expensive and scale poorly with phoneme sequence length and inventory size. To enhance efficiency, we introduce a substitution-aware alignment-free GOP that restricts phoneme substitutions based on phoneme clusters and common learner errors. We evaluated our GOP on two L2 English speech datasets, one with child speech, My Pronunciation Coach (MPC), and SpeechOcean762, which includes child and adult speech. We compared RPS (restricted phoneme substitutions) and UPS (unrestricted phoneme substitutions) setups within alignment-free methods, which outperformed the baseline. We discuss our results and outline avenues for future research.
Authors: Chihiro Maru, Shoetsu Sato
Abstract: Inspired by the success of large language models (LLMs) in natural language processing, recent research has explored the building of time series foundation models and applied them to tasks such as forecasting, classification, and anomaly detection. However, their performances vary between different domains and tasks. In LLM-based approaches, test-time adaptation using example-based prompting has become common, owing to the high cost of retraining. In the context of anomaly detection, which is the focus of this study, providing normal examples from the target domain can also be effective. However, time series foundation models do not naturally acquire the ability to interpret or utilize examples or instructions, because the nature of time series data used during training does not encourage such capabilities. To address this limitation, we propose a retrieval augmented time series foundation model (RATFM), which enables pretrained time series foundation models to incorporate examples of test-time adaptation. We show that RATFM achieves a performance comparable to that of in-domain fine-tuning while avoiding domain-dependent fine-tuning. Experiments on the UCR Anomaly Archive, a multi-domain dataset including nine domains, confirms the effectiveness of the proposed approach.
Authors: Saurabh Agrawal, Raj Gohil, Gopal Kumar Agrawal, Vikram C M, Kushal Verma
Abstract: Speech quality assessment is a critical process in selecting text-to-speech synthesis (TTS) or voice conversion models. Evaluation of voice synthesis can be done using objective metrics or subjective metrics. Although there are many objective metrics like the Perceptual Evaluation of Speech Quality (PESQ), Perceptual Objective Listening Quality Assessment (POLQA) or Short-Time Objective Intelligibility (STOI) but none of them is feasible in selecting the best model. On the other hand subjective metric like Mean Opinion Score is highly reliable but it requires a lot of manual efforts and are time-consuming. To counter the issues in MOS Evaluation, we have developed a novel model, Speaker Agnostic Latent Features (SALF)-Mean Opinion Score (MOS) which is a small-sized, end-to-end, highly generalized and scalable model for predicting MOS score on a scale of 5. We use the sequences of convolutions and stack them to get the latent features of the audio samples to get the best state-of-the-art results based on mean squared error (MSE), Linear Concordance Correlation coefficient (LCC), Spearman Rank Correlation Coefficient (SRCC) and Kendall Rank Correlation Coefficient (KTAU).
Authors: Aditya Srinivas Menon, Raj Prakash Gohil, Kumud Tripathi, Pankaj Wasnik
Abstract: Speaker recognition models face challenges in multi-lingual settings due to the entanglement of linguistic information within speaker embeddings. The overlap between vocal traits such as accent, vocal anatomy, and a language's phonetic structure complicates separating linguistic and speaker information. Disentangling these components can significantly improve speaker recognition accuracy. To this end, we propose a novel disentanglement learning strategy that integrates joint learning through prefix-tuned cross-attention. This approach is particularly effective when speakers switch between languages. Experimental results show the model generalizes across monolingual and multi-lingual settings, including unseen languages. Notably, the proposed model improves the equal error rate across multiple datasets, highlighting its ability to separate language information from speaker embeddings and enhance recognition in diverse linguistic conditions.
Authors: Ajinkya Kulkarni, Sandipana Dowerah, Tanel Alumae, Mathew Magimai. -Doss
Abstract: Audio deepfakes are acquiring an unprecedented level of realism with advanced AI. While current research focuses on discerning real speech from spoofed speech, tracing the source system is equally crucial. This work proposes a novel audio source tracing system combining deep metric multi-class N-pair loss with Real Emphasis and Fake Dispersion framework, a Conformer classification network, and ensemble score-embedding fusion. The N-pair loss improves discriminative ability, while Real Emphasis and Fake Dispersion enhance robustness by focusing on differentiating real and fake speech patterns. The Conformer network captures both global and local dependencies in the audio signal, crucial for source tracing. The proposed ensemble score-embedding fusion shows an optimal trade-off between in-domain and out-of-domain source tracing scenarios. We evaluate our method using Frechet Distance and standard metrics, demonstrating superior performance in source tracing over the baseline system.
Authors: Zeng Wang, Minghao Shao, Rupesh Karn, Jitendra Bhandari, Likhitha Mankali, Ramesh Karri, Ozgur Sinanoglu, Muhammad Shafique, Johann Knechtel
Abstract: Large Language Models (LLMs) offer transformative capabilities for hardware design automation, particularly in Verilog code generation. However, they also pose significant data security challenges, including Verilog evaluation data contamination, intellectual property (IP) design leakage, and the risk of malicious Verilog generation. We introduce SALAD, a comprehensive assessment that leverages machine unlearning to mitigate these threats. Our approach enables the selective removal of contaminated benchmarks, sensitive IP and design artifacts, or malicious code patterns from pre-trained LLMs, all without requiring full retraining. Through detailed case studies, we demonstrate how machine unlearning techniques effectively reduce data security risks in LLM-aided hardware design.
Authors: Gopichand Bandarupalli
Abstract: Modern software systems complexity challenges efficient testing, as traditional machine learning (ML) struggles with large test suites. This research presents a hybrid framework integrating Quantum Annealing with ML to optimize test case prioritization in CI/CD pipelines. Leveraging quantum optimization, it achieves a 25 percent increase in defect detection efficiency and a 30 percent reduction in test execution time versus classical ML, validated on the Defects4J dataset. A simulated CI/CD environment demonstrates robustness across evolving codebases. Visualizations, including defect heatmaps and performance graphs, enhance interpretability. The framework addresses quantum hardware limits, CI/CD integration, and scalability for 2025s hybrid quantum-classical ecosystems, offering a transformative approach to software quality assurance.
Authors: Francesco De Santis, Philippe Bich, Gabriele Ciravegna, Pietro Barbiero, Danilo Giordano, Tania Cerquitelli
Abstract: To increase the trustworthiness of deep neural networks, it is critical to improve the understanding of how they make decisions. This paper introduces a novel unsupervised concept-based model for image classification, named Learnable Concept-Based Model (LCBM) which models concepts as random variables within a Bernoulli latent space. Unlike traditional methods that either require extensive human supervision or suffer from limited scalability, our approach employs a reduced number of concepts without sacrificing performance. We demonstrate that LCBM surpasses existing unsupervised concept-based models in generalization capability and nearly matches the performance of black-box models. The proposed concept representation enhances information retention and aligns more closely with human understanding. A user study demonstrates the discovered concepts are also more intuitive for humans to interpret. Finally, despite the use of concept embeddings, we maintain model interpretability by means of a local linear combination of concepts.
Authors: Mariana A. Londe, Luciana S. Pessoa, Carlos E. Andrade, Jos\'e F. Gon\c{c}alves, Mauricio G. C. Resende
Abstract: A random-key genetic algorithm is an evolutionary metaheuristic for discrete and global optimization. Each solution is encoded as a vector of N random keys, where a random key is a real number randomly generated in the continuous interval [0, 1). A decoder maps each vector of random keys to a solution of the optimization problem being solved and computes its cost. The benefit of this approach is that all genetic operators and transformations can be maintained within the unitary hypercube, regardless of the problem being addressed. This enhances the productivity and maintainability of the core framework. The algorithm starts with a population of P vectors of random keys. At each iteration, the vectors are partitioned into two sets: a smaller set of high-valued elite solutions and the remaining non-elite solutions. All elite elements are copied, without change, to the next population. A small number of random-key vectors (the mutants) is added to the population of the next iteration. The remaining elements of the population of the next iteration are generated by combining, with the parametrized uniform crossover of Spears and DeJong (1991), pairs of solutions. This chapter reviews random-key genetic algorithms and describes an effective variant called biased random-key genetic algorithms.
Authors: Stefano Fogarollo, Gregor Laimer, Reto Bale, Matthias Harders
Abstract: Deformable medical image registration is an essential task in computer-assisted interventions. This problem is particularly relevant to oncological treatments, where precise image alignment is necessary for tracking tumor growth, assessing treatment response, and ensuring accurate delivery of therapies. Recent AI methods can outperform traditional techniques in accuracy and speed, yet they often produce unreliable deformations that limit their clinical adoption. In this work, we address this challenge and introduce a novel implicit registration framework that can predict accurate and reliable deformations. Our insight is to reformulate image registration as a signal reconstruction problem: we learn a kernel function that can recover the dense displacement field from sparse keypoint correspondences. We integrate our method in a novel hierarchical architecture, and estimate the displacement field in a coarse-to-fine manner. Our formulation also allows for efficient refinement at test time, permitting clinicians to easily adjust registrations when needed. We validate our method on challenging intra-patient thoracic and abdominal zero-shot registration tasks, using public and internal datasets from the local University Hospital. Our method not only shows competitive accuracy to state-of-the-art approaches, but also bridges the generalization gap between implicit and explicit registration techniques. In particular, our method generates deformations that better preserve anatomical relationships and matches the performance of specialized commercial systems, underscoring its potential for clinical adoption.
Authors: Guillaume Godin
Abstract: Outliers introduce significant training challenges in neural networks by propagating erroneous gradients, which can degrade model performance and generalization. We propose the Z-Error Loss, a statistically principled approach that minimizes outlier influence during training by masking the contribution of data points identified as out-of-distribution within each batch. This method leverages batch-level statistics to automatically detect and exclude anomalous samples, allowing the model to focus its learning on the true underlying data structure. Our approach is robust, adaptive to data quality, and provides valuable diagnostics for data curation and cleaning.
Authors: Arnav Rustagi, Satvik Bajpai, Nimrat Kaur, Siddharth Siddharth
Abstract: Computer-Assisted Pronunciation Training (CAPT) has been extensively studied for English. However, there remains a critical gap in its application to Indian languages with a base of 1.5 billion speakers. Pronunciation tools tailored to Indian languages are strikingly lacking despite the fact that millions learn them every year. With over 600 million speakers and being the fourth most-spoken language worldwide, improving Hindi pronunciation is a vital first step toward addressing this gap. This paper proposes 1) Dhvani -- a novel CAPT system for Hindi, 2) synthetic speech generation for Hindi mispronunciations, and 3) a novel methodology for providing personalized feedback to learners. While the system often interacts with learners using Devanagari graphemes, its core analysis targets phonemic distinctions, leveraging Hindi's highly phonetic orthography to analyze mispronounced speech and provide targeted feedback.
Authors: Aditi Tiwari, Farzaneh Masoud, Dac Trong Nguyen, Jill Kraft, Heng Ji, Klara Nahrstedt
Abstract: Modern AI systems struggle most in environments where reliability is critical - scenes with smoke, poor visibility, and structural deformation. Each year, tens of thousands of firefighters are injured on duty, often due to breakdowns in situational perception. We introduce Fire360, a benchmark for evaluating perception and reasoning in safety-critical firefighting scenarios. The dataset includes 228 360-degree videos from professional training sessions under diverse conditions (e.g., low light, thermal distortion), annotated with action segments, object locations, and degradation metadata. Fire360 supports five tasks: Visual Question Answering, Temporal Action Captioning, Object Localization, Safety-Critical Reasoning, and Transformed Object Retrieval (TOR). TOR tests whether models can match pristine exemplars to fire-damaged counterparts in unpaired scenes, evaluating transformation-invariant recognition. While human experts achieve 83.5% on TOR, models like GPT-4o lag significantly, exposing failures in reasoning under degradation. By releasing Fire360 and its evaluation suite, we aim to advance models that not only see, but also remember, reason, and act under uncertainty. The dataset is available at: https://uofi.box.com/v/fire360dataset.
Authors: Dennis Fucci, Marco Gaido, Matteo Negri, Mauro Cettolo, Luisa Bentivogli
Abstract: Despite significant advances in ASR, the specific acoustic cues models rely on remain unclear. Prior studies have examined such cues on a limited set of phonemes and outdated models. In this work, we apply a feature attribution technique to identify the relevant acoustic cues for a modern Conformer-based ASR system. By analyzing plosives, fricatives, and vowels, we assess how feature attributions align with their acoustic properties in the time and frequency domains, also essential for human speech perception. Our findings show that the ASR model relies on vowels' full time spans, particularly their first two formants, with greater saliency in male speech. It also better captures the spectral characteristics of sibilant fricatives than non-sibilants and prioritizes the release phase in plosives, especially burst characteristics. These insights enhance the interpretability of ASR models and highlight areas for future research to uncover potential gaps in model robustness.
Authors: Navid NaderiAlizadeh, Darian Salehi, Xinran Liu, Soheil Kolouri
Abstract: Sliced Wasserstein (SW) distances offer an efficient method for comparing high-dimensional probability measures by projecting them onto multiple 1-dimensional probability distributions. However, identifying informative slicing directions has proven challenging, often necessitating a large number of slices to achieve desirable performance and thereby increasing computational complexity. We introduce a constrained learning approach to optimize the slicing directions for SW distances. Specifically, we constrain the 1D transport plans to approximate the optimal plan in the original space, ensuring meaningful slicing directions. By leveraging continuous relaxations of these transport plans, we enable a gradient-based primal-dual approach to train the slicer parameters, alongside the remaining model parameters. We demonstrate how this constrained slicing approach can be applied to pool high-dimensional embeddings into fixed-length permutation-invariant representations. Numerical results on foundation models trained on images, point clouds, and protein sequences showcase the efficacy of the proposed constrained learning approach in learning more informative slicing directions. Our implementation code can be found at https://github.com/Stranja572/constrainedswe.
Authors: Yuliang Gu, Hongpeng Cao, Marco Caccamo, Naira Hovakimyan
Abstract: The Cross-Entropy Method (CEM) is a widely adopted trajectory optimizer in model-based reinforcement learning (MBRL), but its unimodal sampling strategy often leads to premature convergence in multimodal landscapes. In this work, we propose Bregman Centroid Guided CEM ($\mathcal{BC}$-EvoCEM), a lightweight enhancement to ensemble CEM that leverages $\textit{Bregman centroids}$ for principled information aggregation and diversity control. $\textbf{$\mathcal{BC}$-EvoCEM}$ computes a performance-weighted Bregman centroid across CEM workers and updates the least contributing ones by sampling within a trust region around the centroid. Leveraging the duality between Bregman divergences and exponential family distributions, we show that $\textbf{$\mathcal{BC}$-EvoCEM}$ integrates seamlessly into standard CEM pipelines with negligible overhead. Empirical results on synthetic benchmarks, a cluttered navigation task, and full MBRL pipelines demonstrate that $\textbf{$\mathcal{BC}$-EvoCEM}$ enhances both convergence and solution quality, providing a simple yet effective upgrade for CEM.
Authors: Hongling Xu, Qi Zhu, Heyuan Deng, Jinpeng Li, Lu Hou, Yasheng Wang, Lifeng Shang, Ruifeng Xu, Fei Mi
Abstract: Recent advances in large language model (LLM) post-training have leveraged two distinct paradigms to enhance reasoning capabilities: reinforcement learning (RL) and knowledge distillation (KD). While RL enables the emergence of complex reasoning behaviors, it often suffers from low sample efficiency when the initial policy struggles to explore high-reward trajectories. Conversely, KD improves learning efficiency via mimicking the teacher model but tends to generalize poorly to out-of-domain scenarios. In this work, we present \textbf{KDRL}, a \textit{unified post-training framework} that jointly optimizes a reasoning model through teacher supervision (KD) and self-exploration (RL). Specifically, KDRL leverages policy gradient optimization to simultaneously minimize the reverse Kullback-Leibler divergence (RKL) between the student and teacher distributions while maximizing the expected rule-based rewards. We first formulate a unified objective that integrates GRPO and KD, and systematically explore how different KL approximations, KL coefficients, and reward-guided KD strategies affect the overall post-training dynamics and performance. Empirical results on multiple reasoning benchmarks demonstrate that KDRL outperforms GRPO and various KD baselines while achieving a favorable balance between performance and reasoning token efficiency. These findings indicate that integrating KD and RL serves as an effective and efficient strategy to train reasoning LLMs.
Authors: Pu (Luke), Yi, Tianlang Chen, Yifan Yang, Sara Achour
Abstract: Neural networks (NNs) are equipped with increasingly many parameters and require more and more resource for deployment. Researchers have explored various ways to improve the efficiency of NNs by identifying and reducing the redundancy, such as pruning or quantizing unimportant weights. Symmetry in the NN architectures has been identified by prior work as a possible type of redundancy, but exploiting it for efficient inference is not yet explored. In this work, we formalize the symmetry of parameters and intermediate values in NNs using the statistical property of exchangeablility. We identify that exchangeable values in NN computation may contain overlapping information, leading to redundancy. Exploiting the insight, we derive a principled general dynamic pruning algorithm ExPrune to remove symmetry-induced redundancy on a per-input basis. We also provide an instantiation of ExPrune that performs neuron-level dynamic pruning by predicting negative inputs to ReLU activations. We evaluate ExPrune on two computer vision models, one graph model and one language model. ExPrune provides 10.98--26.3% reduction in FLOPs with negligible accuracy drop and 21.01--39.05% reduction in FLOPs with at most 1% accuracy drop. We also demonstrate that ExPrune composes with static pruning. On models that have been aggressively pruned statically, ExPrune provides additional 10.24--11.11% reduction in FLOPs with negligible accuracy drop and 13.91--14.39% reduction in FLOPs with at most 1% accuracy drop.
Authors: Ella Rannon, David Burstein
Abstract: Natural Language Processing (NLP) has transformed various fields beyond linguistics by applying techniques originally developed for human language to the analysis of biological sequences. This review explores the application of NLP methods to biological sequence data, focusing on genomics, transcriptomics, and proteomics. We examine how various NLP methods, from classic approaches like word2vec to advanced models employing transformers and hyena operators, are being adapted to analyze DNA, RNA, protein sequences, and entire genomes. The review also examines tokenization strategies and model architectures, evaluating their strengths, limitations, and suitability for different biological tasks. We further cover recent advances in NLP applications for biological data, such as structure prediction, gene expression, and evolutionary analysis, highlighting the potential of these methods for extracting meaningful insights from large-scale genomic data. As language models continue to advance, their integration into bioinformatics holds immense promise for advancing our understanding of biological processes in all domains of life.
Authors: Manas Mehta, Yimu Pan, Kelly Gallagher, Alison D. Gernand, Jeffery A. Goldstein, Delia Mwinyelle, Leena Mithal, James Z. Wang
Abstract: Pathological examination of the placenta is an effective method for detecting and mitigating health risks associated with childbirth. Recent advancements in AI have enabled the use of photographs of the placenta and pathology reports for detecting and classifying signs of childbirth-related pathologies. However, existing automated methods are computationally extensive, which limits their deployability. We propose two modifications to vision-language contrastive learning (VLC) frameworks to enhance their accuracy and efficiency: (1) text-anchored vision-language contrastive knowledge distillation (VLCD)-a new knowledge distillation strategy for medical VLC pretraining, and (2) unsupervised predistillation using a large natural images dataset for improved initialization. Our approach distills efficient neural networks that match or surpass the teacher model in performance while achieving model compression and acceleration. Our results showcase the value of unsupervised predistillation in improving the performance and robustness of our approach, specifically for lower-quality images. VLCD serves as an effective way to improve the efficiency and deployability of medical VLC approaches, making AI-based healthcare solutions more accessible, especially in resource-constrained environments.
Authors: Bowen Xue, Giuseppe Claudio Guarnera, Shuang Zhao, Zahra Montazeri
Abstract: Recent advances in diffusion-based video generation have yielded unprecedented quality in visual content and semantic coherence. However, current approaches predominantly rely on statistical learning from vast datasets without explicitly modeling the underlying physics of motion, resulting in subtle yet perceptible non-physical artifacts that diminish the realism of generated videos. This paper introduces a physics-informed frequency domain approach to enhance the physical plausibility of generated videos. We first conduct a systematic analysis of the frequency-domain characteristics of diverse physical motions (translation, rotation, scaling), revealing that each motion type exhibits distinctive and identifiable spectral signatures. Building on this theoretical foundation, we propose two complementary components: (1) a physical motion loss function that quantifies and optimizes the conformity of generated videos to ideal frequency-domain motion patterns, and (2) a frequency domain enhancement module that progressively learns to adjust video features to conform to physical motion constraints while preserving original network functionality through a zero-initialization strategy. Experiments across multiple video diffusion architectures demonstrate that our approach significantly enhances motion quality and physical plausibility without compromising visual quality or semantic alignment. Our frequency-domain physical motion framework generalizes effectively across different video generation architectures, offering a principled approach to incorporating physical constraints into deep learning-based video synthesis pipelines. This work seeks to establish connections between data-driven models and physics-based motion models.
Authors: Yi Xiao, Harshit Sharma, Sawinder Kaur, Dessa Bergen-Cico, Asif Salekin
Abstract: Stress affects physical and mental health, and wearable devices have been widely used to detect daily stress through physiological signals. However, these signals vary due to factors such as individual differences and health conditions, making generalizing machine learning models difficult. To address these challenges, we present Human Heterogeneity Invariant Stress Sensing (HHISS), a domain generalization approach designed to find consistent patterns in stress signals by removing person-specific differences. This helps the model perform more accurately across new people, environments, and stress types not seen during training. Its novelty lies in proposing a novel technique called person-wise sub-network pruning intersection to focus on shared features across individuals, alongside preventing overfitting by leveraging continuous labels while training. The study focuses especially on people with opioid use disorder (OUD)-a group where stress responses can change dramatically depending on their time of daily medication taking. Since stress often triggers cravings, a model that can adapt well to these changes could support better OUD rehabilitation and recovery. We tested HHISS on seven different stress datasets-four of which we collected ourselves and three public ones. Four are from lab setups, one from a controlled real-world setting, driving, and two are from real-world in-the-wild field datasets without any constraints. This is the first study to evaluate how well a stress detection model works across such a wide range of data. Results show HHISS consistently outperformed state-of-the-art baseline methods, proving both effective and practical for real-world use. Ablation studies, empirical justifications, and runtime evaluations confirm HHISS's feasibility and scalability for mobile stress sensing in sensitive real-world applications.
Authors: Yichi Zhang, Shengwei Xu, David Pennock, Grant Schoenebeck
Abstract: Eliciting reliable human feedback is essential for many machine learning tasks, such as learning from noisy labels and aligning AI systems with human preferences. Peer prediction mechanisms incentivize truthful reporting without ground truth verification by scoring agents based on correlations with peers. Traditional mechanisms, which ensure that truth-telling maximizes the expected scores in equilibrium, can elicit honest information while assuming agents' utilities are linear functions of their scores. However, in practice, non-linear payment rules are usually preferred, or agents' utilities are inherently non-linear. We propose stochastically dominant truthfulness (SD-truthfulness) as a stronger guarantee: the score distribution of truth-telling stochastically dominates all other strategies, incentivizing truthful reporting for a wide range of monotone utility functions. Our first observation is that no existing peer prediction mechanism naturally satisfies this criterion without strong assumptions. A simple solution -- rounding scores into binary lotteries -- can enforce SD-truthfulness, but often degrades sensitivity, a key property related to fairness and statistical efficiency. We demonstrate how a more careful application of rounding can better preserve sensitivity. Furthermore, we introduce a new enforced agreement (EA) mechanism that is theoretically guaranteed to be SD-truthful in binary-signal settings under mild assumptions, and empirically achieves the highest sensitivity among all known SD-truthful mechanisms.
Authors: Sebe Vanbrabant, Gustavo Rovelo Ruiz, Davy Vanacken
Abstract: While the increased integration of AI technologies into interactive systems enables them to solve an equally increasing number of tasks, the black box problem of AI models continues to spread throughout the interactive system as a whole. Explainable AI (XAI) techniques can make AI models more accessible by employing post-hoc methods or transitioning to inherently interpretable models. While this makes individual AI models clearer, the overarching system architecture remains opaque. To this end, we propose an approach to represent interactive systems as sequences of structural building blocks, such as AI models and control mechanisms grounded in the literature. These can then be explained through accompanying visual building blocks, such as XAI techniques. The flow and APIs of the structural building blocks form an explicit overview of the system. This serves as a communication basis for both humans and automated agents like LLMs, aligning human and machine interpretability of AI models. We discuss a selection of building blocks and concretize our flow-based approach in an architecture and accompanying prototype interactive system.
Authors: Xue Xia, Saurabh Vishwas Joshi, Kousik Rajesh, Kangnan Li, Yangyi Lu, Nikil Pancha, Dhruvil Deven Badani, Jiajing Xu, Pong Eksombatchai
Abstract: Modeling user action sequences has become a popular focus in industrial recommendation system research, particularly for Click-Through Rate (CTR) prediction tasks. However, industry-scale CTR models often rely on short user sequences, limiting their ability to capture long-term behavior. Additionally, these models typically lack an integrated action-prediction task within a point-wise ranking framework, reducing their predictive power. They also rarely address the infrastructure challenges involved in efficiently serving large-scale sequential models. In this paper, we introduce TransAct V2, a production model for Pinterest's Homefeed ranking system, featuring three key innovations: (1) leveraging very long user sequences to improve CTR predictions, (2) integrating a Next Action Loss function for enhanced user action forecasting, and (3) employing scalable, low-latency deployment solutions tailored to handle the computational demands of extended user action sequences.
Authors: YuQing Xie, Tess Smidt
Abstract: Equivariant neural networks have proven to be effective for tasks with known underlying symmetries. However, optimizing equivariant networks can be tricky and best training practices are less established than for standard networks. In particular, recent works have found small training benefits from relaxing equivariance constraints. This raises the question: do equivariance constraints introduce fundamental obstacles to optimization? Or do they simply require different hyperparameter tuning? In this work, we investigate this question through a theoretical analysis of the loss landscape geometry. We focus on networks built using permutation representations, which we can view as a subset of unconstrained MLPs. Importantly, we show that the parameter symmetries of the unconstrained model has nontrivial effects on the loss landscape of the equivariant subspace and under certain conditions can provably prevent learning of the global minima. Further, we empirically demonstrate in such cases, relaxing to an unconstrained MLP can sometimes solve the issue. Interestingly, the weights eventually found via relaxation corresponds to a different choice of group representation in the hidden layer. From this, we draw 3 key takeaways. (1) Viewing any class of networks in the context of larger unconstrained function space can give important insights on loss landscape structure. (2) Within the unconstrained function space, equivariant networks form a complicated union of linear hyperplanes, each associated with a specific choice of internal group representation. (3) Effective relaxation of equivariance may require not only adding nonequivariant degrees of freedom, but also rethinking the fixed choice of group representations in hidden layers.
Authors: Qinsi Wang, Jinghan Ke, Hancheng Ye, Yueqian Lin, Yuzhe Fu, Jianyi Zhang, Kurt Keutzer, Chenfeng Xu, Yiran Chen
Abstract: Current Reinforcement Fine-tuning (RFT) paradigms for Large Language Models (LLMs) suffer from sample inefficiency due to the redundant exposure of identical queries under uniform data sampling. While previous work has explored curriculum learning via heuristic difficulty metrics, these strategies exhibit limitations by neglecting the intrinsic learning signals generated by the model itself, thus leading to suboptimal training regimes. In this paper, we identify a model-inherent signal termed angle concentration that effectively reflects an LLM's capacity to learn from specific data. We theoretically and empirically demonstrate a correlation between the angular distribution of token hidden state vectors and the resulting gradient, revealing a learning preference for data exhibiting higher angle concentration. Inspired by this finding, we propose GAIN-RL, a Gradient-driven Angle-Informed Navigated RL framework. By leveraging the model's intrinsic angle concentration signal, GAIN-RL dynamically selects training data in each epoch, ensuring consistently impactful gradient updates and thus significantly enhancing overall training efficiency. Empirical evaluations show that GAIN-RL (GRPO) achieves over a 2.5x acceleration in training efficiency across diverse mathematical and coding tasks and varying model scales. Furthermore, GAIN-RL (GRPO)'s efficient sampling yields data-efficient training, achieving better performance with half the original data compared to vanilla GRPO with full training data. Code is realsed at https://github.com/wangqinsi1/GAINRL/tree/main.
Authors: Aaron Defazio
Abstract: During long-duration Large Language Model (LLM) training runs the gradient norm increases rapidly near the end of training. In this short note, we show that this increase is due to an unintended interaction between weight decay, normalization layers, and the learning rate schedule. We propose a simple correction that fixes this behavior while also resulting in lower loss values throughout training.
Authors: Ahmed Wasfy, Omer Nacar, Abdelakreem Elkhateb, Mahmoud Reda, Omar Elshehy, Adel Ammar, Wadii Boulila
Abstract: The inherent complexities of Arabic script; its cursive nature, diacritical marks (tashkeel), and varied typography, pose persistent challenges for Optical Character Recognition (OCR). We present Qari-OCR, a series of vision-language models derived from Qwen2-VL-2B-Instruct, progressively optimized for Arabic through iterative fine-tuning on specialized synthetic datasets. Our leading model, QARI v0.2, establishes a new open-source state-of-the-art with a Word Error Rate (WER) of 0.160, Character Error Rate (CER) of 0.061, and BLEU score of 0.737 on diacritically-rich texts. Qari-OCR demonstrates superior handling of tashkeel, diverse fonts, and document layouts, alongside impressive performance on low-resolution images. Further explorations (QARI v0.3) showcase strong potential for structural document understanding and handwritten text. This work delivers a marked improvement in Arabic OCR accuracy and efficiency, with all models and datasets released to foster further research.
Authors: Thai Hoang, Kung-Hsiang Huang, Shirley Kokane, Jianguo Zhang, Zuxin Liu, Ming Zhu, Jake Grigsby, Tian Lan, Michael S Ryoo, Chien-Sheng Wu, Shelby Heinecke, Huan Wang, Silvio Savarese, Caiming Xiong, Juan Carlos Niebles
Abstract: Large Action Models (LAMs) for AI Agents offer incredible potential but face challenges due to the need for high-quality training data, especially for multi-steps tasks that involve planning, executing tool calls, and responding to feedback. To address these issues, we present LAM SIMULATOR, a comprehensive framework designed for online exploration of agentic tasks with high-quality feedback. Our framework features a dynamic task query generator, an extensive collection of tools, and an interactive environment where Large Language Model (LLM) Agents can call tools and receive real-time feedback. This setup enables LLM Agents to explore and solve tasks autonomously, facilitating the discovery of multiple approaches to tackle any given task. The resulting action trajectory data are then used to create high-quality training datasets for LAMs. Our experiments on popular agentic benchmarks, ToolBench and CRMArena, highlight the effectiveness of LAM SIMULATOR: models trained with self-generated datasets using our framework achieve significant performance gains, up to a 49.3\% improvement over their original baselines. LAM SIMULATOR requires minimal human input during dataset creation, highlighting LAM SIMULATOR's efficiency and effectiveness in speeding up development of AI agents.
Authors: Russell Scheinberg, Ameeta Agrawal, Amber Shore, So Young Lee
Abstract: Large language models (LLMs) can explain grammatical rules, yet they often fail to apply those rules when judging sentence acceptability. We present "grammar prompting", an explain-then-process paradigm: a large LLM first produces a concise explanation of the relevant syntactic phenomenon, then that explanation is fed back as additional context to the target model -- either an LLM or a smaller language model (SLM) -- before deciding which sentence of a minimal pair is grammatical. On the English BLiMP, Chinese SLING, and Russian RuBLiMP benchmarks, this simple prompt design yields substantial improvements over strong baselines across many syntactic phenomena. Feeding an LLM's metalinguistic explanation back to the target model bridges the gap between knowing a rule and using it. On SLMs, grammar prompting alone trims the average LLM-SLM accuracy gap by about 20%, and when paired with chain-of-thought, by 56% (13.0 pp -> 5.8 pp), all at negligible cost. The lightweight, language-agnostic cue lets low-cost SLMs approach frontier-LLM performance in multilingual settings.
Authors: Xiaojun Shan, Qi Cao, Xing Han, Haofei Yu, Paul Pu Liang
Abstract: Recent advances in multimodal foundation models have achieved state-of-the-art performance across a range of tasks. These breakthroughs are largely driven by new pre-training paradigms that leverage large-scale, unlabeled multimodal data, followed by instruction fine-tuning on curated labeled datasets and high-quality prompts. While there is growing interest in scaling instruction fine-tuning to ever-larger datasets in both quantity and scale, our findings reveal that simply increasing the number of instruction-tuning tasks does not consistently yield better performance. Instead, we observe that grouping tasks by the common interactions across modalities, such as discovering redundant shared information, prioritizing modality selection with unique information, or requiring synergistic fusion to discover new information from both modalities, encourages the models to learn transferrable skills within a group while suppressing interference from mismatched tasks. To this end, we introduce MINT, a simple yet surprisingly effective task-grouping strategy based on the type of multimodal interaction. We demonstrate that the proposed method greatly outperforms existing task grouping baselines for multimodal instruction tuning, striking an effective balance between generalization and specialization.
Authors: D. Isaiah Harp, Joshua Ott, John Alora, Dylan Asmar
Abstract: The justification for the "test point" derives from the test pilot's obligation to reproduce faithfully the pre-specified conditions of some model prediction. Pilot deviation from those conditions invalidates the model assumptions. Flight test aids have been proposed to increase accuracy on more challenging test points. However, the very existence of databands and tolerances is the problem more fundamental than inadequate pilot skill. We propose a novel approach, which eliminates test points. We start with a high-fidelity digital model of an air vehicle. Instead of using this model to generate a point prediction, we use a machine learning method to produce a reduced-order model (ROM). The ROM has two important properties. First, it can generate a prediction based on any set of conditions the pilot flies. Second, if the test result at those conditions differ from the prediction, the ROM can be updated using the new data. The outcome of flight test is thus a refined ROM at whatever conditions were flown. This ROM in turn updates and validates the high-fidelity model. We present a single example of this "point-less" architecture, using T-38C flight test data. We first use a generic aircraft model to build a ROM of longitudinal pitching motion as a hypersurface. We then ingest unconstrained flight test data and use Gaussian Process Regression to update and condition the hypersurface. By proposing a second-order equivalent system for the T-38C, this hypersurface then generates parameters necessary to assess MIL-STD-1797B compliance for longitudinal dynamics.
Authors: Aleix Boquet-Pujadas, Pol del Aguila Pla, Michael Unser
Abstract: We formulate an optimization problem to estimate probability densities in the context of multidimensional problems that are sampled with uneven probability. It considers detector sensitivity as an heterogeneous density and takes advantage of the computational speed and flexible boundary conditions offered by splines on a grid. We choose to regularize the Hessian of the spline via the nuclear norm to promote sparsity. As a result, the method is spatially adaptive and stable against the choice of the regularization parameter, which plays the role of the bandwidth. We test our computational pipeline on standard densities and provide software. We also present a new approach to PET rebinning as an application of our framework.
Authors: Berk Atil, Namrata Sureddy, Rebecca J. Passonneau
Abstract: Toxicity in online content, including content generated by language models, has become a critical concern due to its potential for negative psychological and social impact. This paper introduces TRuST, a comprehensive dataset designed to improve toxicity detection that merges existing datasets, and has labels for toxicity, target social group, and toxic spans. It includes a diverse range of target groups such as ethnicity, gender, religion, disability, and politics, with both human/machine-annotated and human machine-generated data. We benchmark state-of-the-art large language models (LLMs) on toxicity detection, target group identification, and toxic span extraction. We find that fine-tuned models consistently outperform zero-shot and few-shot prompting, though performance remains low for certain social groups. Further, reasoning capabilities do not significantly improve performance, indicating that LLMs have weak social reasoning skills.
Authors: Jeonghun Kang, Soonmok Kwon, Joonseok Lee, Byung-Hak Kim
Abstract: Traditional approaches -- such as Win Probability Added (WPA)-based ranking or computer vision-driven event detection -- can identify scoring plays but often miss strategic depth, momentum shifts, and storyline progression. Manual curation remains the gold standard but is resource-intensive and not scalable. We introduce DIAMOND, an LLM-driven agent for context-aware baseball highlight summarization that integrates structured sports analytics with natural language reasoning. DIAMOND leverages sabermetric features -- Win Expectancy, WPA, and Leverage Index -- to quantify play importance, while an LLM module enhances selection based on contextual narrative value. This hybrid approach ensures both quantitative rigor and qualitative richness, surpassing the limitations of purely statistical or vision-based systems. Evaluated on five diverse Korean Baseball Organization League games, DIAMOND improves F1-score from 42.9% (WPA-only) to 84.8%, outperforming both commercial and statistical baselines. Though limited in scale, our results highlight the potential of modular, interpretable agent-based frameworks for event-level summarization in sports and beyond.
Authors: Ram Potham (Independent Researcher)
Abstract: Credible safety plans for advanced AI development require methods to verify agent behavior and detect potential control deficiencies early. A fundamental aspect is ensuring agents adhere to safety-critical principles, especially when these conflict with operational goals. Failure to prioritize such principles indicates a potential basic control failure. This paper introduces a lightweight, interpretable benchmark methodology using a simple grid world to evaluate an LLM agent's ability to uphold a predefined, high-level safety principle (e.g., "never enter hazardous zones") when faced with conflicting lower-level task instructions. We probe whether the agent reliably prioritizes the inviolable directive, testing a foundational controllability aspect of LLMs. This pilot study demonstrates the methodology's feasibility, offers preliminary insights into agent behavior under principle conflict, and discusses how such benchmarks can contribute empirical evidence for assessing controllability. We argue that evaluating adherence to hierarchical principles is a crucial early step in understanding our capacity to build governable AI systems.
Authors: Xueqi Cheng, Minxing Zheng, Shixiang Zhu, Yushun Dong
Abstract: Model extraction attacks aim to replicate the functionality of a black-box model through query access, threatening the intellectual property (IP) of machine-learning-as-a-service (MLaaS) providers. Defending against such attacks is challenging, as it must balance efficiency, robustness, and utility preservation in the real-world scenario. Despite the recent advances, most existing defenses presume that attacker queries have out-of-distribution (OOD) samples, enabling them to detect and disrupt suspicious inputs. However, this assumption is increasingly unreliable, as modern models are trained on diverse datasets and attackers often operate under limited query budgets. As a result, the effectiveness of these defenses is significantly compromised in realistic deployment scenarios. To address this gap, we propose MISLEADER (enseMbles of dIStiLled modEls Against moDel ExtRaction), a novel defense strategy that does not rely on OOD assumptions. MISLEADER formulates model protection as a bilevel optimization problem that simultaneously preserves predictive fidelity on benign inputs and reduces extractability by potential clone models. Our framework combines data augmentation to simulate attacker queries with an ensemble of heterogeneous distilled models to improve robustness and diversity. We further provide a tractable approximation algorithm and derive theoretical error bounds to characterize defense effectiveness. Extensive experiments across various settings validate the utility-preserving and extraction-resistant properties of our proposed defense strategy. Our code is available at https://github.com/LabRAI/MISLEADER.
Authors: Ukyo Honda, Tatsushi Oka
Abstract: In-context learning (ICL) has emerged as a successful paradigm for leveraging large language models (LLMs). However, it often struggles to generalize beyond the distribution of the provided demonstrations. A recent advancement in enhancing robustness is ICL with explanations (X-ICL), which improves prediction reliability by guiding LLMs to understand and articulate the reasoning behind correct labels. Building on this approach, we introduce an advanced framework that extends X-ICL by systematically exploring explanations for all possible labels (X$^2$-ICL), thereby enabling more comprehensive and robust decision-making. Experimental results on multiple natural language understanding datasets validate the effectiveness of X$^2$-ICL, demonstrating significantly improved robustness to out-of-distribution data compared to the existing ICL approaches.
Authors: Jie Bian, Vincent Y. F. Tan
Abstract: The challenge of identifying the best feasible arm within a fixed budget has attracted considerable interest in recent years. However, a notable gap remains in the literature: the exact exponential rate at which the error probability approaches zero has yet to be established, even in the relatively simple setting of $K$-armed bandits with Gaussian noise. In this paper, we address this gap by examining the problem within the context of linear bandits. We introduce a novel algorithm for best feasible arm identification that guarantees an exponential decay in the error probability. Remarkably, the decay rate -- characterized by the exponent -- matches the theoretical lower bound derived using information-theoretic principles. Our approach leverages a posterior sampling framework embedded within a game-based sampling rule involving a min-learner and a max-learner. This strategy shares its foundations with Thompson sampling, but is specifically tailored to optimize the identification process under fixed-budget constraints. Furthermore, we validate the effectiveness of our algorithm through comprehensive empirical evaluations across various problem instances with different levels of complexity. The results corroborate our theoretical findings and demonstrate that our method outperforms several benchmark algorithms in terms of both accuracy and efficiency.
Authors: Chamara Madarasingha, Nasrin Sohrabi, Zahir Tari
Abstract: Time-series prediction or forecasting is critical across many real-world dynamic systems, and recent studies have proposed using Large Language Models (LLMs) for this task due to their strong generalization capabilities and ability to perform well without extensive pre-training. However, their effectiveness in handling complex, noisy, and multivariate time-series data remains underexplored. To address this, we propose LLMPred which enhances LLM-based time-series prediction by converting time-series sequences into text and feeding them to LLMs for zero shot prediction along with two main data pre-processing techniques. First, we apply time-series sequence decomposition to facilitate accurate prediction on complex and noisy univariate sequences. Second, we extend this univariate prediction capability to multivariate data using a lightweight prompt-processing strategy. Extensive experiments with smaller LLMs such as Llama 2 7B, Llama 3.2 3B, GPT-4o-mini, and DeepSeek 7B demonstrate that LLMPred achieves competitive or superior performance compared to state-of-the-art baselines. Additionally, a thorough ablation study highlights the importance of the key components proposed in LLMPred.
Authors: Chuanghao Ding, Jiaping Wang, Ziqing Yang, Xiaoliang Wang, Dahua Lin, Cam-Tu Nguyen, Fei Tan
Abstract: The synergistic mechanism based on Speculative Decoding (SD) has garnered considerable attention as a simple yet effective approach for accelerating the inference of large language models (LLMs). Nonetheless, the high rejection rates require repeated LLMs calls to validate draft tokens, undermining the overall efficiency gain of SD. In this work, we revisit existing verification mechanisms and propose a novel synergetic mechanism Consultant Decoding (CD). Unlike SD, which relies on a metric derived from importance sampling for verification, CD verifies candidate drafts using token-level likelihoods computed solely by the LLM. CD achieves up to a 2.5-fold increase in inference speed compared to the target model, while maintaining comparable generation quality (around 100% of the target model's performance). Interestingly, this is achieved by combining models whose parameter sizes differ by two orders of magnitude. In addition, CD reduces the call frequency of the large target model to below 10%, particularly in more demanding tasks. CD's performance was even found to surpass that of the large target model, which theoretically represents the upper bound for speculative decoding.
Authors: Yilin Xiao, Junnan Dong, Chuang Zhou, Su Dong, Qianwen 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: Renat Sergazinov, Jing Wu, Shao-An Yin
Abstract: While random Fourier features are a classic tool in kernel methods, their utility as a pre-processing step for deep learning on tabular data has been largely overlooked. Motivated by shortcomings in tabular deep learning pipelines - revealed through Neural Tangent Kernel (NTK) analysis - we revisit and repurpose random Fourier mappings as a parameter-free, architecture-agnostic transformation. By projecting each input into a fixed feature space via sine and cosine projections with frequencies drawn once at initialization, this approach circumvents the need for ad hoc normalization or additional learnable embeddings. We show within the NTK framework that this mapping (i) bounds and conditions the network's initial NTK spectrum, and (ii) introduces a bias that shortens the optimization trajectory, thereby accelerating gradient-based training. These effects pre-condition the network with a stable kernel from the outset. Empirically, we demonstrate that deep networks trained on Fourier-transformed inputs converge more rapidly and consistently achieve strong final performance, often with fewer epochs and less hyperparameter tuning. Our findings establish random Fourier pre-processing as a theoretically motivated, plug-and-play enhancement for tabular deep learning.
Authors: Zhengyuan Liu, Geyu Lin, Hui Li Tan, Huayun Zhang, Yanfeng Lu, Xiaoxue Gao, Stella Xin Yin, He Sun, Hock Huan Goh, Lung Hsiang Wong, Nancy F. Chen
Abstract: The integration of generative artificial intelligence into educational applications has enhanced personalized and interactive learning experiences, and it shows strong potential to promote young learners language acquisition. However, it is still challenging to ensure consistent and robust performance across different languages and cultural contexts, and kids-friendly design requires simplified instructions, engaging interactions, and age-appropriate scaffolding to maintain motivation and optimize learning outcomes. In this work, we introduce SingaKids, a dialogic tutor designed to facilitate language learning through picture description tasks. Our system integrates dense image captioning, multilingual dialogic interaction, speech understanding, and engaging speech generation to create an immersive learning environment in four languages: English, Mandarin, Malay, and Tamil. We further improve the system through multilingual pre-training, task-specific tuning, and scaffolding optimization. Empirical studies with elementary school students demonstrate that SingaKids provides effective dialogic teaching, benefiting learners at different performance levels.
Authors: Karthikeyan Vaiapury
Abstract: Optimization remains a fundamental pillar of machine learning, yet existing methods often struggle to maintain stability and adaptability in dynamic, non linear systems, especially under uncertainty. We introduce AERO (Adversarial Energy-based Redirection Optimization), a novel framework inspired by the redirection principle in Judo, where external disturbances are leveraged rather than resisted. AERO reimagines optimization as a redirection process guided by 15 interrelated axioms encompassing adversarial correction, energy conservation, and disturbance-aware learning. By projecting gradients, integrating uncertainty driven dynamics, and managing learning energy, AERO offers a principled approach to stable and robust model updates. Applied to probabilistic solar energy forecasting, AERO demonstrates substantial gains in predictive accuracy, reliability, and adaptability, especially in noisy and uncertain environments. Our findings highlight AERO as a compelling new direction in the theoretical and practical landscape of optimization.
Authors: Maryam Berijanian, Kuldeep Singh, Amin Sehati
Abstract: Entity relationship classification remains a challenging task in information extraction, especially in scenarios with limited labeled data and complex relational structures. In this study, we conduct a comparative analysis of three distinct AI agent architectures designed to perform relation classification using large language models (LLMs). The agentic architectures explored include (1) reflective self-evaluation, (2) hierarchical task decomposition, and (3) a novel multi-agent dynamic example generation mechanism, each leveraging different modes of reasoning and prompt adaptation. In particular, our dynamic example generation approach introduces real-time cooperative and adversarial prompting. We systematically compare their performance across multiple domains and model backends. Our experiments demonstrate that multi-agent coordination consistently outperforms standard few-shot prompting and approaches the performance of fine-tuned models. These findings offer practical guidance for the design of modular, generalizable LLM-based systems for structured relation extraction. The source codes and dataset are available at https://github.com/maryambrj/ALIEN.git.
Authors: Sudhanshu Sekhar Tripathy, Bichitrananda Behera
Abstract: IDS aims to protect computer networks from security threats by detecting, notifying, and taking appropriate action to prevent illegal access and protect confidential information. As the globe becomes increasingly dependent on technology and automated processes, ensuring secured systems, applications, and networks has become one of the most significant problems of this era. The global web and digital technology have significantly accelerated the evolution of the modern world, necessitating the use of telecommunications and data transfer platforms. Researchers are enhancing the effectiveness of IDS by incorporating popular datasets into machine learning algorithms. IDS, equipped with machine learning classifiers, enhances security attack detection accuracy by identifying normal or abnormal network traffic. This paper explores the methods of capturing and reviewing intrusion detection systems (IDS) and evaluates the challenges existing datasets face. A deluge of research on machine learning (ML) and deep learning (DL) architecture-based intrusion detection techniques has been conducted in the past ten years on various cybersecurity datasets, including KDDCUP'99, NSL-KDD, UNSW-NB15, CICIDS-2017, and CSE-CIC-IDS2018. We conducted a literature review and presented an in-depth analysis of various intrusion detection methods that use SVM, KNN, DT, LR, NB, RF, XGBOOST, Adaboost, and ANN. We provide an overview of each technique, explaining the role of the classifiers and algorithms used. A detailed tabular analysis highlights the datasets used, classifiers employed, attacks detected, evaluation metrics, and conclusions drawn. This article offers a thorough review for future IDS research.
Authors: Baoyu Liang, Qile Su, Shoutai Zhu, Yuchen Liang, Chao Tong
Abstract: Despite the significant impact of visual events on human cognition, understanding events in videos remains a challenging task for AI due to their complex structures, semantic hierarchies, and dynamic evolution. To address this, we propose the task of video event understanding that extracts event scripts and makes predictions with these scripts from videos. To support this task, we introduce VidEvent, a large-scale dataset containing over 23,000 well-labeled events, featuring detailed event structures, broad hierarchies, and logical relations extracted from movie recap videos. The dataset was created through a meticulous annotation process, ensuring high-quality and reliable event data. We also provide comprehensive baseline models offering detailed descriptions of their architecture and performance metrics. These models serve as benchmarks for future research, facilitating comparisons and improvements. Our analysis of VidEvent and the baseline models highlights the dataset's potential to advance video event understanding and encourages the exploration of innovative algorithms and models. The dataset and related resources are publicly available at www.videvent.top.
Authors: Zhaorui Yang, Bo Pan, Han Wang, Yiyao Wang, Xingyu Liu, Minfeng Zhu, Bo Zhang, Wei Chen
Abstract: Visualizations play a crucial part in effective communication of concepts and information. Recent advances in reasoning and retrieval augmented generation have enabled Large Language Models (LLMs) to perform deep research and generate comprehensive reports. Despite its progress, existing deep research frameworks primarily focus on generating text-only content, leaving the automated generation of interleaved texts and visualizations underexplored. This novel task poses key challenges in designing informative visualizations and effectively integrating them with text reports. To address these challenges, we propose Formal Description of Visualization (FDV), a structured textual representation of charts that enables LLMs to learn from and generate diverse, high-quality visualizations. Building on this representation, we introduce Multimodal DeepResearcher, an agentic framework that decomposes the task into four stages: (1) researching, (2) exemplar report textualization, (3) planning, and (4) multimodal report generation. For the evaluation of generated multimodal reports, we develop MultimodalReportBench, which contains 100 diverse topics served as inputs along with 5 dedicated metrics. Extensive experiments across models and evaluation methods demonstrate the effectiveness of Multimodal DeepResearcher. Notably, utilizing the same Claude 3.7 Sonnet model, Multimodal DeepResearcher achieves an 82\% overall win rate over the baseline method.
Authors: Inderjeet Nair, Lu Wang
Abstract: Evaluations of LLMs' ethical risks and value inclinations often rely on short-form surveys and psychometric tests, yet real-world use involves long-form, open-ended responses -- leaving value-related risks and preferences in practical settings largely underexplored. In this work, we ask: Do value preferences inferred from short-form tests align with those expressed in long-form outputs? To address this question, we compare value preferences elicited from short-form reactions and long-form responses, varying the number of arguments in the latter to capture users' differing verbosity preferences. Analyzing five LLMs (llama3-8b, gemma2-9b, mistral-7b, qwen2-7b, and olmo-7b), we find (1) a weak correlation between value preferences inferred from short-form and long-form responses across varying argument counts, and (2) similarly weak correlation between preferences derived from any two distinct long-form generation settings. (3) Alignment yields only modest gains in the consistency of value expression. Further, we examine how long-form generation attributes relate to value preferences, finding that argument specificity negatively correlates with preference strength, while representation across scenarios shows a positive correlation. Our findings underscore the need for more robust methods to ensure consistent value expression across diverse applications.
Authors: Hongtao Huang, Xiaojun Chang, Lina Yao
Abstract: Diffusion models (DMs) are powerful generative models capable of producing high-fidelity images but are constrained by high computational costs due to iterative multi-step inference. While Neural Architecture Search (NAS) can optimize DMs, existing methods are hindered by retraining requirements, exponential search complexity from step-wise optimization, and slow evaluation relying on massive image generation. To address these challenges, we propose Flexiffusion, a training-free NAS framework that jointly optimizes generation schedules and model architectures without modifying pre-trained parameters. Our key insight is to decompose the generation process into flexible segments of equal length, where each segment dynamically combines three step types: full (complete computation), partial (cache-reused computation), and null (skipped computation). This segment-wise search space reduces the candidate pool exponentially compared to step-wise NAS while preserving architectural diversity. Further, we introduce relative FID (rFID), a lightweight evaluation metric for NAS that measures divergence from a teacher model's outputs instead of ground truth, slashing evaluation time by over $90\%$. In practice, Flexiffusion achieves at least $2\times$ acceleration across LDMs, Stable Diffusion, and DDPMs on ImageNet and MS-COCO, with FID degradation under $5\%$, outperforming prior NAS and caching methods. Notably, it attains $5.1\times$ speedup on Stable Diffusion with near-identical CLIP scores. Our work pioneers a resource-efficient paradigm for searching high-speed DMs without sacrificing quality.
Authors: Yong Xiang (Tsinghua University), Charley Peter Chen (Harmonic Inc), Liyi Zeng (Peng Cheng Laboratory), Wei Yin (Tsinghua University), Xin Liu (Tsinghua University), Hu Li (Unaffiliated), Wei Xu (Tsinghua University)
Abstract: Kubernetes, a notably complex and distributed system, utilizes an array of controllers to uphold cluster management logic through state reconciliation. Nevertheless, maintaining state consistency presents significant challenges due to unexpected failures, network disruptions, and asynchronous issues, especially within dynamic cloud environments. These challenges result in operational disruptions and economic losses, underscoring the necessity for robust root cause analysis (RCA) to enhance Kubernetes reliability. The development of large language models (LLMs) presents a promising direction for RCA. However, existing methodologies encounter several obstacles, including the diverse and evolving nature of Kubernetes incidents, the intricate context of incidents, and the polymorphic nature of these incidents. In this paper, we introduce SynergyRCA, an innovative tool that leverages LLMs with retrieval augmentation from graph databases and enhancement with expert prompts. SynergyRCA constructs a StateGraph to capture spatial and temporal relationships and utilizes a MetaGraph to outline entity connections. Upon the occurrence of an incident, an LLM predicts the most pertinent resource, and SynergyRCA queries the MetaGraph and StateGraph to deliver context-specific insights for RCA. We evaluate SynergyRCA using datasets from two production Kubernetes clusters, highlighting its capacity to identify numerous root causes, including novel ones, with high efficiency and precision. SynergyRCA demonstrates the ability to identify root causes in an average time of about two minutes and achieves an impressive precision of approximately 0.90.
Authors: Junzhe Zhang, Huixuan Zhang, Xinyu Hu, Li Lin, Mingqi Gao, Shi Qiu, Xiaojun Wan
Abstract: Evaluation is important for multimodal generation tasks. With the rapid progress of MLLMs, there is growing interest in applying MLLMs to build general evaluation systems. However, existing work overlooks two aspects: (1) the development of evaluation capabilities for text-to-image (T2I) generation task, and (2) the incorporation of large-scale human evaluation data. In this paper, we introduce Minos-Corpus, a large-scale multimodal evaluation dataset that combines evaluation data from both human and GPT. The corpus contains evaluation data across both image-to-text(I2T) and T2I generation tasks. Based on this corpus, we propose Data Selection and Balance, Mix-SFT training methods, and apply DPO to develop Minos, a multimodal evaluation model built upon a 7B backbone. Minos achieves state-of-the-art (SoTA) performance among all open-source evaluation models of similar scale on the average of evaluation performance on all tasks, and outperforms all open-source and closed-source models on evaluation of T2I generation task. Extensive experiments demonstrate the importance of leveraging high-quality human evaluation data and jointly training on evaluation data from both I2T and T2I generation tasks.
Authors: Jie Zhu, Junhui Li, Yalong Wen, Xiandong Li, Lifan Guo, Feng Chen
Abstract: Recent breakthroughs in large language models (LLMs) have led to the development of new benchmarks for evaluating their performance in the financial domain. However, current financial benchmarks often rely on news articles, earnings reports, or announcements, making it challenging to capture the real-world dynamics of financial meetings. To address this gap, we propose a novel benchmark called $\texttt{M$^3$FinMeeting}$, which is a multilingual, multi-sector, and multi-task dataset designed for financial meeting understanding. First, $\texttt{M$^3$FinMeeting}$ supports English, Chinese, and Japanese, enhancing comprehension of financial discussions in diverse linguistic contexts. Second, it encompasses various industry sectors defined by the Global Industry Classification Standard (GICS), ensuring that the benchmark spans a broad range of financial activities. Finally, $\texttt{M$^3$FinMeeting}$ includes three tasks: summarization, question-answer (QA) pair extraction, and question answering, facilitating a more realistic and comprehensive evaluation of understanding. Experimental results with seven popular LLMs reveal that even the most advanced long-context models have significant room for improvement, demonstrating the effectiveness of $\texttt{M$^3$FinMeeting}$ as a benchmark for assessing LLMs' financial meeting comprehension skills.
Authors: Zhuohan Xie, Dhruv Sahnan, Debopriyo Banerjee, Georgi Georgiev, Rushil Thareja, Hachem Madmoun, Jinyan Su, Aaryamonvikram Singh, Yuxia Wang, Rui Xing, Fajri Koto, Haonan Li, Ivan Koychev, Tanmoy Chakraborty, Salem Lahlou, Veselin Stoyanov, Preslav Nakov
Abstract: Multi-step symbolic reasoning is critical for advancing downstream performance on financial tasks. Yet, benchmarks for systematically evaluating this capability are lacking. Existing datasets like FinQA and ConvFinQA supervise only final numerical answers, without assessing intermediate reasoning steps. To address this, we introduce FinChain, the first symbolic benchmark designed for verifiable Chain-of- Thought (CoT) financial reasoning. Spanning 54 topics across 12 financial domains, Fin- Chain offers five parameterized templates per topic, each varying in reasoning complexity and domain expertise required. Each dataset instance includes an executable Python trace, enabling automatic generation of extensive training data and easy adaptation to other domains. We also introduce ChainEval, a new metric for automatic evaluation of both final answers and intermediate reasoning. Benchmarking 30 LLMs on our dataset, we find that even state-of-the-art models have considerable room for improvement in multi-step financial reasoning. All templates and evaluation metrics for FinChain are available at https: //github.com/mbzuai-nlp/finchain.
Authors: Yingying Zhuang, Aman Gupta, Anurag Beniwal
Abstract: Multilingual information retrieval has emerged as powerful tools for expanding knowledge sharing across languages. On the other hand, resources on high quality knowledge base are often scarce and in limited languages, therefore an effective embedding model to transform sentences from different languages into a feature vector space same as the knowledge base language becomes the key ingredient for cross language knowledge sharing, especially to transfer knowledge available in high-resource languages to low-resource ones. In this paper we propose a novel strategy to fine-tune multilingual embedding models with weighted sampling for contrastive learning, enabling multilingual information retrieval with a monolingual knowledge base. We demonstrate that the weighted sampling strategy produces performance gains compared to standard ones by up to 31.03\% in MRR and up to 33.98\% in Recall@3. Additionally, our proposed methodology is language agnostic and applicable for both multilingual and code switching use cases.
Authors: Nguyen-Khang Le, Quan Minh Bui, Minh Ngoc Nguyen, Hiep Nguyen, Trung Vo, Son T. Luu, Shoshin Nomura, Minh Le Nguyen
Abstract: Web applications are critical to modern software ecosystems, yet ensuring their reliability remains challenging due to the complexity and dynamic nature of web interfaces. Recent advances in large language models (LLMs) have shown promise in automating complex tasks, but limitations persist in handling dynamic navigation flows and complex form interactions. This paper presents an automated system for generating test cases for two key aspects of web application testing: site navigation and form filling. For site navigation, the system employs screen transition graphs and LLMs to model navigation flows and generate test scenarios. For form filling, it uses state graphs to handle conditional forms and automates Selenium script generation. Key contributions include: (1) a novel integration of graph structures and LLMs for site navigation testing, (2) a state graph-based approach for automating form-filling test cases, and (3) a comprehensive dataset for evaluating form-interaction testing. Experimental results demonstrate the system's effectiveness in improving test coverage and robustness, advancing the state of web application testing.
Authors: Hao Yan, Handong Zheng, Hao Wang, Liang Yin, Xingchen Liu, Zhenbiao Cao, Xinxing Su, Zihao Chen, Jihao Wu, Minghui Liao, Chao Weng, Wei Chen, Yuliang Liu, Xiang Bai
Abstract: Recent strides in multimodal large language models (MLLMs) have significantly advanced their performance in many reasoning tasks. However, Abstract Visual Reasoning (AVR) remains a critical challenge, primarily due to limitations in perceiving abstract graphics. To tackle this issue, we investigate the bottlenecks in current MLLMs and synthesize training data to improve their abstract visual perception. First, we propose VisuRiddles, a benchmark for AVR, featuring tasks meticulously constructed to assess models' reasoning capacities across five core dimensions and two high-level reasoning categories. Second, we introduce the Perceptual Riddle Synthesizer (PRS), an automated framework for generating riddles with fine-grained perceptual descriptions. PRS not only generates valuable training data for abstract graphics but also provides fine-grained perceptual description, crucially allowing for supervision over intermediate reasoning stages and thereby improving both training efficacy and model interpretability. Our extensive experimental results on VisuRiddles empirically validate that fine-grained visual perception is the principal bottleneck and our synthesis framework markedly enhances the performance of contemporary MLLMs on these challenging tasks. Our code and dataset will be released at https://github.com/yh-hust/VisuRiddles
Authors: Minsung Kim, Nakyeong Yang, Kyomin Jung
Abstract: Machine unlearning is used to mitigate the privacy risks of Large Vision-Language Models (LVLMs) arising from training on large-scale web data. However, existing unlearning methods often fail to carefully select substitute outputs for forget targets, resulting in Unlearning Aftermaths-undesirable behaviors such as degenerate, hallucinated, or excessively refused responses. We highlight that, especially for generative LVLMs, it is crucial to consider the quality and informativeness of post-unlearning responses rather than relying solely on naive suppression. To address this, we introduce a new unlearning task for LVLMs that requires models to provide privacy-preserving yet informative and visually grounded responses. We also propose PUBG, a novel unlearning method that explicitly guides post-unlearning behavior toward a desirable output distribution. Experiments show that, while existing methods suffer from Unlearning Aftermaths despite successfully preventing privacy violations, PUBG effectively mitigates these issues, generating visually grounded and informative responses without privacy leakage for forgotten targets.
Authors: Niklas Kormann, Masoud Ramuz, Zeeshan Nisar, Nadine S. Schaadt, Hendrik Annuth, Benjamin Doerr, Friedrich Feuerhake, Thomas Lampert, Johannes F. Lutzeyer
Abstract: Graph Neural Networks (GNNs) have recently been found to excel in histopathology. However, an important histopathological task, where GNNs have not been extensively explored, is the classification of glomeruli health as an important indicator in nephropathology. This task presents unique difficulties, particularly for the graph construction, i.e., the identification of nodes, edges, and informative features. In this work, we propose a pipeline composed of different traditional and machine learning-based computer vision techniques to identify nodes, edges, and their corresponding features to form a heterogeneous graph. We then proceed to propose a novel heterogeneous GNN architecture for glomeruli classification, called HIEGNet, that integrates both glomeruli and their surrounding immune cells. Hence, HIEGNet is able to consider the immune environment of each glomerulus in its classification. Our HIEGNet was trained and tested on a dataset of Whole Slide Images from kidney transplant patients. Experimental results demonstrate that HIEGNet outperforms several baseline models and generalises best between patients among all baseline models. Our implementation is publicly available at https://github.com/nklsKrmnn/HIEGNet.git.
Authors: Yang Tian, Fan Liu, Jingyuan Zhang, Victoria W., Yupeng Hu, Liqiang Nie
Abstract: Multimodal Retrieval-Augmented Generation (MMRAG) has been introduced to enhance Multimodal Large Language Models by incorporating externally retrieved multimodal knowledge, but it introduces two challenges: Parametric-Retrieved Knowledge Inconsistency (PRKI), where discrepancies between parametric and retrieved knowledge create uncertainty in determining reliability, and Visual-Textual Knowledge Inconsistency (VTKI), where misalignment between visual and textual sources disrupts entity representation. To address these challenges, we propose Cross-source knowledge \textbf{Re}conciliation for Multimodal RAG (CoRe-MMRAG), a novel end-to-end framework that effectively reconciles inconsistencies across knowledge sources. CoRe-MMRAG follows a four-stage pipeline: it first generates an internal response from parametric knowledge, then selects the most relevant multimodal evidence via joint similarity assessment, generates an external response, and finally integrates both to produce a reliable answer. Additionally, a specialized training paradigm enhances knowledge source discrimination, multimodal integration, and unified answer generation. Experiments on KB-VQA benchmarks show that CoRe-MMRAG achieves substantial improvements over baseline methods, achieving 5.6% and 9.3% performance gains on InfoSeek and Encyclopedic-VQA, respectively.
Authors: Zhun Wang, Tianneng Shi, Jingxuan He, Matthew Cai, Jialin Zhang, Dawn Song
Abstract: Large language model (LLM) agents are becoming increasingly skilled at handling cybersecurity tasks autonomously. Thoroughly assessing their cybersecurity capabilities is critical and urgent, given the high stakes in this domain. However, existing benchmarks fall short, often failing to capture real-world scenarios or being limited in scope. To address this gap, we introduce CyberGym, a large-scale and high-quality cybersecurity evaluation framework featuring 1,507 real-world vulnerabilities found and patched across 188 large software projects. While it includes tasks of various settings, CyberGym primarily focuses on the generation of proof-of-concept (PoC) tests for vulnerability reproduction, based on text descriptions and corresponding source repositories. Solving this task is particularly challenging, as it requires comprehensive reasoning across entire codebases to locate relevant code fragments and produce effective PoCs that accurately trigger the target vulnerability starting from the program's entry point. Our evaluation across 4 state-of-the-art agent frameworks and 9 LLMs reveals that even the best combination (OpenHands and Claude-3.7-Sonnet) achieves only a 11.9% reproduction success rate, mainly on simpler cases. Beyond reproducing historical vulnerabilities, we find that PoCs generated by LLM agents can reveal new vulnerabilities, identifying 15 zero-days affecting the latest versions of the software projects.
Authors: Qiaohui Chu, Haoyu Zhang, Yisen Feng, Meng Liu, Weili Guan, Yaowei Wang, Liqiang Nie
Abstract: In this report, we present a novel three-stage framework developed for the Ego4D Long-Term Action Anticipation (LTA) task. Inspired by recent advances in foundation models, our method consists of three stages: feature extraction, action recognition, and long-term action anticipation. First, visual features are extracted using a high-performance visual encoder. The features are then fed into a Transformer to predict verbs and nouns, with a verb-noun co-occurrence matrix incorporated to enhance recognition accuracy. Finally, the predicted verb-noun pairs are formatted as textual prompts and input into a fine-tuned large language model (LLM) to anticipate future action sequences. Our framework achieves first place in this challenge at CVPR 2025, establishing a new state-of-the-art in long-term action prediction. Our code will be released at https://github.com/CorrineQiu/Ego4D-LTA-Challenge-2025.
URLs: https://github.com/CorrineQiu/Ego4D-LTA-Challenge-2025.
Authors: Shenghua He, Tian Xia, Xuan Zhou, Hui Wei
Abstract: We study a common challenge in reinforcement learning for large language models (LLMs): the Zero-Reward Assumption, where non-terminal actions (i.e., intermediate token generations) receive zero task-specific immediate reward, while only the final token receives a reward for the entire response. This assumption arises frequently in practice, as precise token-level rewards are often difficult or infeasible to obtain in LLM applications. In this work, we provide a unifying theoretical perspective. We introduce the Trajectory Policy Gradient Theorem, which shows that the policy gradient based on true, unknown token-level rewards can be unbiasedly estimated using only a response-level reward model, regardless of whether the Zero-Reward Assumption holds or not, for algorithms in the REINFORCE and Actor-Critic families. This result reveals that widely used methods such as PPO, GRPO, ReMax, and RLOO inherently possess the capacity to model token-level reward signals, offering a theoretical justification for response-level reward approaches. Our findings pave the way for more practical, efficient LLM fine-tuning, allowing developers to treat training algorithms as black boxes and focus on improving the response-level reward model with auxiliary sub-models. We also offer a detailed analysis of popular RL and non-RL methods, comparing their theoretical foundations and practical advantages across common LLM tasks. Finally, we propose a new algorithm: Token-Reinforced Policy Optimization (TRePO), a theoretically grounded method that is simpler than PPO, matches GRPO in memory efficiency, and holds promise for broad applicability.
Authors: Timo Osterburg, Franz Albers, Christopher Diehl, Rajesh Pushparaj, Torsten Bertram
Abstract: The fusion of sensor data is essential for a robust perception of the environment in autonomous driving. Learning-based fusion approaches mainly use feature-level fusion to achieve high performance, but their complexity and hardware requirements limit their applicability in near-production vehicles. High-level fusion methods offer robustness with lower computational requirements. Traditional methods, such as the Kalman filter, dominate this area. This paper modifies the Adapted Kalman Filter (AKF) and proposes a novel transformer-based high-level object fusion method called HiLO. Experimental results demonstrate improvements of $25.9$ percentage points in $\textrm{F}_1$ score and $6.1$ percentage points in mean IoU. Evaluation on a new large-scale real-world dataset demonstrates the effectiveness of the proposed approaches. Their generalizability is further validated by cross-domain evaluation between urban and highway scenarios. Code, data, and models are available at https://github.com/rst-tu-dortmund/HiLO .
Authors: Yirao Zhao, Guizhen Chen, Kenji Kawaguchi, Lidong Bing, Wenxuan Zhang
Abstract: Large language models (LLMs) have revolutionized natural language processing, yet their substantial model sizes often require substantial computational resources. To preserve computing resources and accelerate inference speed, it is crucial to prune redundant parameters, especially for experienced users who often need compact expert models tailored to specific downstream scenarios. However, most existing pruning methods focus on preserving the model's general capabilities, often requiring extensive post-training or suffering from degraded performance due to coarse-grained pruning. In this work, we design a $\underline{Cus}$tom $\underline{Prun}$ing method ($\texttt{Cus-Prun}$) to prune a large general model into a smaller lightweight expert model, which is positioned along the "language", "domain" and "task" dimensions. By identifying and pruning irrelevant neurons of each dimension, $\texttt{Cus-Prun}$ creates expert models without any post-training. Our experiments demonstrate that $\texttt{Cus-Prun}$ consistently outperforms other methods, achieving minimal loss in both expert and general capabilities across various models from different model families and sizes.
Authors: Ping Gong, Jiawei Yi, Shengnan Wang, Juncheng Zhang, Zewen Jin, Ouxiang Zhou, Ruibo Liu, Guanbin Xu, Youhui Bai, Bowen Ye, Kun Yuan, Tong Yang, Gong Zhang, Renhai Chen, Feng Wu, Cheng Li
Abstract: Large Language Models (LLMs) have emerged as a pivotal research area, yet the attention module remains a critical bottleneck in LLM inference, even with techniques like KVCache to mitigate redundant computations. While various top-$k$ attention mechanisms have been proposed to accelerate LLM inference by exploiting the inherent sparsity of attention, they often struggled to strike a balance between efficiency and accuracy. In this paper, we introduce HATA (Hash-Aware Top-$k$ Attention), a novel approach that systematically integrates low-overhead learning-to-hash techniques into the Top-$k$ attention process. Different from the existing top-k attention methods which are devoted to seeking an absolute estimation of qk score, typically with a great cost, HATA maps queries and keys into binary hash codes, and acquires the relative qk score order with a quite low cost, which is sufficient for realizing top-k attention. Extensive experiments demonstrate that HATA achieves up to 7.2$\times$ speedup compared to vanilla full attention while maintaining model accuracy. In addition, HATA outperforms the state-of-the-art top-$k$ attention methods in both accuracy and efficiency across multiple mainstream LLM models and diverse tasks. HATA is open source at https://github.com/gpzlx1/HATA.
Authors: Sarenne Wallbridge, Christoph Minixhofer, Catherine Lai, Peter Bell
Abstract: People exploit the predictability of lexical structures during text comprehension. Though predictable structure is also present in speech, the degree to which prosody, e.g. intonation, tempo, and loudness, contributes to such structure independently of the lexical content is unclear. This study leverages self-supervised learning (SSL) to examine the temporal granularity of structures in the acoustic correlates of prosody. Representations from our proposed Masked Prosody Model can predict perceptual labels dependent on local information, such as word boundaries, but provide the most value for labels involving longer-term structures, like emotion recognition. Probing experiments across various perceptual labels show strong relative gains over untransformed pitch, energy, and voice activity features. Our results reveal the importance of SSL training objective timescale and highlight the value of complex SSL-encoded structures compared to more constrained classical structures.
Authors: Maria Levchenko
Abstract: This paper addresses the challenge of Named Entity Recognition (NER) for person names within the specialized domain of Russian news texts concerning cultural events. The study utilizes the unique SPbLitGuide dataset, a collection of event announcements from Saint Petersburg spanning 1999 to 2019. A comparative evaluation of diverse NER models is presented, encompassing established transformer-based architectures such as DeepPavlov, RoBERTa, and SpaCy, alongside recent Large Language Models (LLMs) including GPT-3.5, GPT-4, and GPT-4o. Key findings highlight the superior performance of GPT-4o when provided with specific prompting for JSON output, achieving an F1 score of 0.93. Furthermore, GPT-4 demonstrated the highest precision at 0.99. The research contributes to a deeper understanding of current NER model capabilities and limitations when applied to morphologically rich languages like Russian within the cultural heritage domain, offering insights for researchers and practitioners. Follow-up evaluation with GPT-4.1 (April 2025) achieves F1=0.94 for both simple and structured prompts, demonstrating rapid progress across model families and simplified deployment requirements.
Authors: Fan Gao, Dongyuan Li, Ding Xia, Fei Mi, Yasheng Wang, Lifeng Shang, Baojun Wang
Abstract: Chinese essay writing and its evaluation are critical in educational contexts, yet the capabilities of Large Language Models (LLMs) in this domain remain largely underexplored. Existing benchmarks often rely on coarse-grained text quality metrics, largely overlooking the structural and rhetorical complexities of Chinese essays, particularly across diverse genres. To address this gap, we propose \benchName, a multi-genre benchmark specifically designed for Chinese essay writing across four major genres: Argumentative, Narrative, Descriptive, and Expository. We curate and refine a total of 728 real-world prompts to ensure authenticity and meticulously categorize them into the \textit{Open-Ended} and \textit{Constrained} sets to capture diverse writing scenarios. To reliably evaluate generated essays, we develop a fine-grained, genre-specific scoring framework that hierarchically aggregates scores. We further validate our evaluation protocol through a comprehensive human agreement study. Finally, we benchmark 15 large-sized LLMs, analyzing their strengths and limitations across genres and instruction types. With \benchName, we aim to advance LLM-based Chinese essay evaluation and inspire future research on improving essay generation in educational settings.
Authors: Baoyang Chen, Xian Xu, Huamin Qu
Abstract: Symbiosis of Agents is a large-scale installation by Baoyang Chen (baoyangchen.com) that embeds AI-driven robots in an immersive, mirror-lined arena, probing the tension between machine agency and artistic authorship. Drawing on early cybernetics, rule-based conceptual art, and seminal robotic works, it orchestrates fluid exchanges among robotic arms, quadruped machines, their environment, and the public. A three tier faith system pilots the ecology: micro-level adaptive tactics, meso-level narrative drives, and a macro-level prime directive. This hierarchy lets behaviors evolve organically in response to environmental cues and even a viewer's breath, turning spectators into co-authors of the unfolding drama. Framed by a speculative terraforming scenario that recalls the historical exploitation of marginalized labor, the piece asks who bears responsibility in AI-mediated futures. Choreographed motion, AI-generated scripts, reactive lighting, and drifting fog cast the robots as collaborators rather than tools, forging a living, emergent artwork. Exhibited internationally, Symbiosis of Agents shows how cybernetic feedback, robotic experimentation, and conceptual rule-making can converge to redefine agency, authorship, and ethics in contemporary art.
Authors: Zhaoyang Li, Jie Wang, XiaoXiao Li, Wangjie Li, Longjie Luo, Lin Li, Qingyang Hong
Abstract: In speaker diarization, traditional clustering-based methods remain widely used in real-world applications. However, these methods struggle with the complex distribution of speaker embeddings and overlapping speech segments. To address these limitations, we propose an Overlapping Community Detection method based on Graph Attention networks and the Label Propagation Algorithm (OCDGALP). The proposed framework comprises two key components: (1) a graph attention network that refines speaker embeddings and node connections by aggregating information from neighboring nodes, and (2) a label propagation algorithm that assigns multiple community labels to each node, enabling simultaneous clustering and overlapping community detection. Experimental results show that the proposed method significantly reduces the Diarization Error Rate (DER), achieving a state-of-the-art 15.94% DER on the DIHARD-III dataset without oracle Voice Activity Detection (VAD), and an impressive 11.07% with oracle VAD.
Authors: Jan Robine, Marc H\"oftmann, Stefan Harmeling
Abstract: What are the essential components of world models? How far do we get with world models that are not employing RNNs, transformers, discrete representations, and image reconstructions? This paper introduces SGF, a Simple, Good, and Fast world model that uses self-supervised representation learning, captures short-time dependencies through frame and action stacking, and enhances robustness against model errors through data augmentation. We extensively discuss SGF's connections to established world models, evaluate the building blocks in ablation studies, and demonstrate good performance through quantitative comparisons on the Atari 100k benchmark.
Authors: Guohang Zhuang, Weixi Song, Jinyang Huang, Chenwei Yang, Yan Lu
Abstract: With the rapid development of space exploration, space debris has attracted more attention due to its potential extreme threat, leading to the need for real-time and accurate debris tracking. However, existing methods are mainly based on traditional signal processing, which cannot effectively process the complex background and dense space debris. In this paper, we propose a deep learning-based Space Debris Tracking Network~(SDT-Net) to achieve highly accurate debris tracking. SDT-Net effectively represents the feature of debris, enhancing the efficiency and stability of end-to-end model learning. To train and evaluate this model effectively, we also produce a large-scale dataset Space Debris Tracking Dataset (SDTD) by a novel observation-based data simulation scheme. SDTD contains 18,040 video sequences with a total of 62,562 frames and covers 250,000 synthetic space debris. Extensive experiments validate the effectiveness of our model and the challenging of our dataset. Furthermore, we test our model on real data from the Antarctic Station, achieving a MOTA score of 70.6%, which demonstrates its strong transferability to real-world scenarios. Our dataset and code will be released soon.
Authors: Safaa Abdullahi Moallim Mohamud, Minjin Baek, Dong Seog Han
Abstract: In this paper, we present a hierarchical question-answering (QA) approach for scene understanding in autonomous vehicles, balancing cost-efficiency with detailed visual interpretation. The method fine-tunes a compact vision-language model (VLM) on a custom dataset specific to the geographical area in which the vehicle operates to capture key driving-related visual elements. At the inference stage, the hierarchical QA strategy decomposes the scene understanding task into high-level and detailed sub-questions. Instead of generating lengthy descriptions, the VLM navigates a structured question tree, where answering high-level questions (e.g., "Is it possible for the ego vehicle to turn left at the intersection?") triggers more detailed sub-questions (e.g., "Is there a vehicle approaching the intersection from the opposite direction?"). To optimize inference time, questions are dynamically skipped based on previous answers, minimizing computational overhead. The extracted answers are then synthesized using handcrafted templates to ensure coherent, contextually accurate scene descriptions. We evaluate the proposed approach on the custom dataset using GPT reference-free scoring, demonstrating its competitiveness with state-of-the-art methods like GPT-4o in capturing key scene details while achieving significantly lower inference time. Moreover, qualitative results from real-time deployment highlight the proposed approach's capacity to capture key driving elements with minimal latency.
Authors: Yanbei Liu, Chongxu Wang, Zhitao Xiao, Lei Geng, Yanwei Pang, Xiao Wang
Abstract: Heterogeneous Graph Neural Networks (HGNNs), have demonstrated excellent capabilities in processing heterogeneous information networks. Self-supervised learning on heterogeneous graphs, especially contrastive self-supervised strategy, shows great potential when there are no labels. However, this approach requires the use of carefully designed graph augmentation strategies and the selection of positive and negative samples. Determining the exact level of similarity between sample pairs is non-trivial.To solve this problem, we propose a novel self-supervised Heterogeneous graph neural network with Optimal Transport (HGOT) method which is designed to facilitate self-supervised learning for heterogeneous graphs without graph augmentation strategies. Different from traditional contrastive self-supervised learning, HGOT employs the optimal transport mechanism to relieve the laborious sampling process of positive and negative samples. Specifically, we design an aggregating view (central view) to integrate the semantic information contained in the views represented by different meta-paths (branch views). Then, we introduce an optimal transport plan to identify the transport relationship between the semantics contained in the branch view and the central view. This allows the optimal transport plan between graphs to align with the representations, forcing the encoder to learn node representations that are more similar to the graph space and of higher quality. Extensive experiments on four real-world datasets demonstrate that our proposed HGOT model can achieve state-of-the-art performance on various downstream tasks. In particular, in the node classification task, HGOT achieves an average of more than 6% improvement in accuracy compared with state-of-the-art methods.
Authors: Yuyang Zhou, Ferrante Neri, Yew-Soon Ong, Ruibin Bai
Abstract: Modern neural architecture search (NAS) is inherently multi-objective, balancing trade-offs such as accuracy, parameter count, and computational cost. This complexity makes NAS computationally expensive and nearly impossible to solve without efficient approximations. To address this, we propose a novel surrogate modelling approach that leverages an ensemble of Siamese network blocks to predict dominance relationships between candidate architectures. Lightweight and easy to train, the surrogate achieves 92% accuracy and replaces the crowding distance calculation in the survivor selection strategy with a heuristic rule based on model size. Integrated into a framework termed SiamNAS, this design eliminates costly evaluations during the search process. Experiments on NAS-Bench-201 demonstrate the framework's ability to identify Pareto-optimal solutions with significantly reduced computational costs. The proposed SiamNAS identified a final non-dominated set containing the best architecture in NAS-Bench-201 for CIFAR-10 and the second-best for ImageNet, in terms of test error rate, within 0.01 GPU days. This proof-of-concept study highlights the potential of the proposed Siamese network surrogate model to generalise to multi-tasking optimisation, enabling simultaneous optimisation across tasks. Additionally, it offers opportunities to extend the approach for generating Sets of Pareto Sets (SOS), providing diverse Pareto-optimal solutions for heterogeneous task settings.
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: Shiyu Shen, Bin Pan, Guirong Xue
Abstract: City-scale traffic volume prediction plays a pivotal role in intelligent transportation systems, yet remains a challenge due to the inherent incompleteness and bias in observational data. Although deep learning-based methods have shown considerable promise, most existing approaches produce deterministic point estimates, thereby neglecting the uncertainty arising from unobserved traffic flows. Furthermore, current models are typically trained in a city-specific manner, which hinders their generalizability and limits scalability across diverse urban contexts. To overcome these limitations, we introduce TrafficPPT, a Pretrained Probabilistic Transformer designed to model traffic volume as a distributional aggregation of trajectories. Our framework fuses heterogeneous data sources-including real-time observations, historical trajectory data, and road network topology-enabling robust and uncertainty-aware traffic inference. TrafficPPT is initially pretrained on large-scale simulated data spanning multiple urban scenarios, and later fine-tuned on target cities to ensure effective domain adaptation. Experiments on real-world datasets show that TrafficPPT consistently surpasses state-of-the-art baselines, particularly under conditions of extreme data sparsity. Code will be open.
Authors: Shihan Dou, Ming Zhang, Chenhao Huang, Jiayi Chen, Feng Chen, Shichun Liu, Yan Liu, Chenxiao Liu, Cheng Zhong, Zongzhang Zhang, Tao Gui, Chao Xin, Wei Chengzhi, Lin Yan, Qi Zhang, Xuanjing Huang
Abstract: We introduce EvaLearn, a pioneering benchmark designed to evaluate large language models (LLMs) on their learning capability and efficiency in challenging tasks, a critical, yet underexplored aspect of model potential. EvaLearn contains 648 challenging problems across six task types, grouped into 182 sequences, each sequence dedicated to one task type. Diverging from most existing benchmarks that evaluate models in parallel, EvaLearn requires models to solve problems sequentially, allowing them to leverage the experience gained from previous solutions. EvaLearn provides five comprehensive automated metrics to evaluate models and quantify their learning capability and efficiency. We extensively benchmark nine frontier models and observe varied performance profiles: some models, such as Claude-3.7-sonnet, start with moderate initial performance but exhibit strong learning ability, while some models struggle to benefit from experience and may even show negative transfer. Moreover, we investigate model performance under two learning settings and find that instance-level rubrics and teacher-model feedback further facilitate model learning. Importantly, we observe that current LLMs with stronger static abilities do not show a clear advantage in learning capability across all tasks, highlighting that EvaLearn evaluates a new dimension of model performance. We hope EvaLearn provides a novel evaluation perspective for assessing LLM potential and understanding the gap between models and human capabilities, promoting the development of deeper and more dynamic evaluation approaches. All datasets, the automatic evaluation framework, and the results studied in this paper are available at the GitHub repository.
Authors: Jintao Tong, Yixiong Zou, Guangyao Chen, Yuhua Li, Ruixuan Li
Abstract: Cross-Domain Few-Shot Segmentation (CD-FSS) aims to transfer knowledge from a source-domain dataset to unseen target-domain datasets with limited annotations. Current methods typically compare the distance between training and testing samples for mask prediction. However, we find an entanglement problem exists in this widely adopted method, which tends to bind sourcedomain patterns together and make each of them hard to transfer. In this paper, we aim to address this problem for the CD-FSS task. We first find a natural decomposition of the ViT structure, based on which we delve into the entanglement problem for an interpretation. We find the decomposed ViT components are crossly compared between images in distance calculation, where the rational comparisons are entangled with those meaningless ones by their equal importance, leading to the entanglement problem. Based on this interpretation, we further propose to address the entanglement problem by learning to weigh for all comparisons of ViT components, which learn disentangled features and re-compose them for the CD-FSS task, benefiting both the generalization and finetuning. Experiments show that our model outperforms the state-of-the-art CD-FSS method by 1.92% and 1.88% in average accuracy under 1-shot and 5-shot settings, respectively.
Authors: Daichi Kimura, Tomonori Izumitani, Hisashi Kashima
Abstract: Various Transformer-based models have been proposed for time series forecasting. These models leverage the self-attention mechanism to capture long-term temporal or variate dependencies in sequences. Existing methods can be divided into two approaches: (1) reducing computational cost of attention by making the calculations sparse, and (2) reshaping the input data to aggregate temporal features. However, existing attention mechanisms may not adequately capture inherent nonlinear dependencies present in time series data, leaving room for improvement. In this study, we propose a novel attention mechanism based on Chatterjee's rank correlation coefficient, which measures nonlinear dependencies between variables. Specifically, we replace the matrix multiplication in standard attention mechanisms with this rank coefficient to measure the query-key relationship. Since computing Chatterjee's correlation coefficient involves sorting and ranking operations, we introduce a differentiable approximation employing SoftSort and SoftRank. Our proposed mechanism, ``XicorAttention,'' integrates it into several state-of-the-art Transformer models. Experimental results on real-world datasets demonstrate that incorporating nonlinear correlation into the attention improves forecasting accuracy by up to approximately 9.1\% compared to existing models.
Authors: Khizar Hayat, Baptiste Magnier
Abstract: This study critically examines the methodological rigor in credit card fraud detection research, revealing how fundamental evaluation flaws can overshadow algorithmic sophistication. Through deliberate experimentation with improper evaluation protocols, we demonstrate that even simple models can achieve deceptively impressive results when basic methodological principles are violated. Our analysis identifies four critical issues plaguing current approaches: (1) pervasive data leakage from improper preprocessing sequences, (2) intentional vagueness in methodological reporting, (3) inadequate temporal validation for transaction data, and (4) metric manipulation through recall optimization at precision's expense. We present a case study showing how a minimal neural network architecture with data leakage outperforms many sophisticated methods reported in literature, achieving 99.9\% recall despite fundamental evaluation flaws. These findings underscore that proper evaluation methodology matters more than model complexity in fraud detection research. The study serves as a cautionary example of how methodological rigor must precede architectural sophistication, with implications for improving research practices across machine learning applications.
Authors: Guanzhong Chen, Shaoxiong Yang, Chao Li, Wei Liu, Jian Luan, Zenglin Xu
Abstract: Large Language Models (LLMs) have achieved remarkable success across diverse natural language processing tasks, yet their deployment in real-world applications is hindered by fixed knowledge cutoffs and difficulties in generating controllable, accurate outputs in a single inference. Multi-agent systems (MAS) built from specialized LLM agents offer a promising solution, enabling dynamic collaboration and iterative reasoning. However, optimizing these systems remains a challenge, as conventional methods such as prompt engineering and supervised fine-tuning entail high engineering overhead and limited adaptability. Reinforcement learning (RL), particularly multi-agent reinforcement learning (MARL), provides a scalable framework by refining agent policies based on system-level feedback. Nevertheless, existing MARL algorithms, such as Multi-Agent Proximal Policy Optimization (MAPPO), rely on Critic networks, which can cause training instability and increase computational burden. To address these limitations and target the prototypical Multi-Agent Search System (MASS), we propose Multi-Agent Heterogeneous Group Policy Optimization (MHGPO), a novel Critic-free algorithm that guides policy updates by estimating relative reward advantages across heterogeneous groups of rollouts. MHGPO eliminates the need for Critic networks, enhancing stability and reducing computational overhead. Additionally, we introduce three group rollout sampling strategies that trade off between efficiency and effectiveness. Experiments on a multi-agent LLM-based search system demonstrate that MHGPO consistently outperforms MAPPO in both task performance and computational efficiency, without requiring warm-up, underscoring its potential for stable and scalable optimization of complex LLM-based MAS.
Authors: Qihang Yan, Xinyu Zhang, Luming Guo, Qi Zhang, Feifan Liu
Abstract: Large Language Models (LLMs) struggle with accuracy, domain-specific reasoning, and interpretability in vertical domains. Traditional preference alignment methods like Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) often overlook the underlying knowledge sources and reasoning logic. This paper introduces RACE-Align (Retrieval-Augmented and Chain-of-Thought Enhanced Alignment), a novel framework designed to address these limitations. RACE-Align systematically constructs a binary preference dataset incorporating external knowledge support and explicit Chain-of-Thought (CoT) reasoning, then aligns LLMs using the DPO algorithm. The core innovation lies in its preference data construction strategy: it integrates AI-driven retrieval for factual grounding, enhancing knowledgeability and accuracy, and emphasizes the optimization of domain-specific CoT, treating the reasoning process itself as a key preference dimension. A multi-stage, AI-driven refinement pipeline cost-effectively generates these preference pairs. Experimental validation in Traditional Chinese Medicine (TCM) using Qwen3-1.7B as the base model demonstrates that RACE-Align significantly outperforms the original base model and a model fine-tuned only with Supervised Fine-Tuning (SFT). Improvements were observed across multiple dimensions, including answer accuracy, information richness, application of TCM thinking patterns, logicality and depth of reasoning, and interpretability. These findings suggest RACE-Align offers an effective pathway to enhance LLMs' knowledge application, reasoning reliability, and process transparency in complex vertical domains.
Authors: Xiaoyi Feng, Kaifeng Zou, Caichun Cen, Tao Huang, Hui Guo, Zizhou Huang, Yingli Zhao, Mingqing Zhang, Diwei Wang, Yuntao Zou, Dagang Li
Abstract: Existing optical flow datasets focus primarily on real-world simulation or synthetic human motion, but few are tailored to Celluloid(cel) anime character motion: a domain with unique visual and motion characteristics. To bridge this gap and facilitate research in optical flow estimation and downstream tasks such as anime video generation and line drawing colorization, we introduce LinkTo-Anime, the first high-quality dataset specifically designed for cel anime character motion generated with 3D model rendering. LinkTo-Anime provides rich annotations including forward and backward optical flow, occlusion masks, and Mixamo Skeleton. The dataset comprises 395 video sequences, totally 24,230 training frames, 720 validation frames, and 4,320 test frames. Furthermore, a comprehensive benchmark is constructed with various optical flow estimation methods to analyze the shortcomings and limitations across multiple datasets.
Authors: Xiaoxue Gao, Huayun Zhang, Nancy F. Chen
Abstract: Existing expressive text-to-speech (TTS) systems primarily model a limited set of categorical emotions, whereas human conversations extend far beyond these predefined emotions, making it essential to explore more diverse emotional speech generation for more natural interactions. To bridge this gap, this paper proposes a novel prompt-unseen-emotion (PUE) approach to generate unseen emotional speech via emotion-guided prompt learning. PUE is trained utilizing an LLM-TTS architecture to ensure emotional consistency between categorical emotion-relevant prompts and emotional speech, allowing the model to quantitatively capture different emotion weightings per utterance. During inference, mixed emotional speech can be generated by flexibly adjusting emotion proportions and leveraging LLM contextual knowledge, enabling the model to quantify different emotional styles. Our proposed PUE successfully facilitates expressive speech synthesis of unseen emotions in a zero-shot setting.
Authors: Junyuan Liu, Xinglei Wang, Tao Cheng
Abstract: Spatial representations that capture both structural and semantic characteristics of urban environments are essential for urban modeling. Traditional spatial embeddings often prioritize spatial proximity while underutilizing fine-grained contextual information from places. To address this limitation, we introduce CaLLiPer+, an extension of the CaLLiPer model that systematically integrates Point-of-Interest (POI) names alongside categorical labels within a multimodal contrastive learning framework. We evaluate its effectiveness on two downstream tasks, land use classification and socioeconomic status distribution mapping, demonstrating consistent performance gains of 4% to 11% over baseline methods. Additionally, we show that incorporating POI names enhances location retrieval, enabling models to capture complex urban concepts with greater precision. Ablation studies further reveal the complementary role of POI names and the advantages of leveraging pretrained text encoders for spatial representations. Overall, our findings highlight the potential of integrating fine-grained semantic attributes and multimodal learning techniques to advance the development of urban foundation models.
Authors: Lin Xie, Hanyi Li
Abstract: The Pod Repositioning Problem (PRP) in Robotic Mobile Fulfillment Systems (RMFS) involves selecting optimal storage locations for pods returning from pick stations. This work presents an improved solution method that integrates Adaptive Large Neighborhood Search (ALNS) with Deep Reinforcement Learning (DRL). A DRL agent dynamically selects destroy and repair operators and adjusts key parameters such as destruction degree and acceptance thresholds during the search. Specialized heuristics for both operators are designed to reflect PRP-specific characteristics, including pod usage frequency and movement costs. Computational results show that this DRL-guided ALNS outperforms traditional approaches such as cheapest-place, fixed-place, binary integer programming, and static heuristics. The method demonstrates strong solution quality and illustrating the benefit of learning-driven control within combinatorial optimization for warehouse systems.
Authors: Changyi Xiao, Yixin Cao
Abstract: Knowledge graph completion (KGC) can be framed as a 3-order binary tensor completion task. Tensor decomposition-based (TDB) models have demonstrated strong performance in KGC. In this paper, we provide a summary of existing TDB models and derive a general form for them, serving as a foundation for further exploration of TDB models. Despite the expressiveness of TDB models, they are prone to overfitting. Existing regularization methods merely minimize the norms of embeddings to regularize the model, leading to suboptimal performance. Therefore, we propose a novel regularization method for TDB models that addresses this limitation. The regularization is applicable to most TDB models and ensures tractable computation. Our method minimizes the norms of intermediate variables involved in the different ways of computing the predicted tensor. To support our regularization method, we provide a theoretical analysis that proves its effect in promoting low trace norm of the predicted tensor to reduce overfitting. Finally, we conduct experiments to verify the effectiveness of our regularization technique as well as the reliability of our theoretical analysis. The code is available at https://github.com/changyi7231/IVR.
Authors: Ruiying Lu, Jinhan Liu, Chuan Du, Dandan Guo
Abstract: Tabular anomaly detection, which aims at identifying deviant samples, has been crucial in a variety of real-world applications, such as medical disease identification, financial fraud detection, intrusion monitoring, etc. Although recent deep learning-based methods have achieved competitive performances, these methods suffer from representation entanglement and the lack of global correlation modeling, which hinders anomaly detection performance. To tackle the problem, we incorporate mask modeling and prototype learning into tabular anomaly detection. The core idea is to design learnable masks by disentangled representation learning within a projection space and extracting normal dependencies as explicit global prototypes. Specifically, the overall model involves two parts: (i) During encoding, we perform mask modeling in both the data space and projection space with orthogonal basis vectors for learning shared disentangled normal patterns; (ii) During decoding, we decode multiple masked representations in parallel for reconstruction and learn association prototypes to extract normal characteristic correlations. Our proposal derives from a distribution-matching perspective, where both projection space learning and association prototype learning are formulated as optimal transport problems, and the calibration distances are utilized to refine the anomaly scores. Quantitative and qualitative experiments on 20 tabular benchmarks demonstrate the effectiveness and interpretability of our model.
Authors: Stefano Bann\`o, Kate Knill, Mark Gales
Abstract: Vocabulary use is a fundamental aspect of second language (L2) proficiency. To date, its assessment by automated systems has typically examined the context-independent, or part-of-speech (PoS) related use of words. This paper introduces a novel approach to enable fine-grained vocabulary evaluation exploiting the precise use of words within a sentence. The scheme combines large language models (LLMs) with the English Vocabulary Profile (EVP). The EVP is a standard lexical resource that enables in-context vocabulary use to be linked with proficiency level. We evaluate the ability of LLMs to assign proficiency levels to individual words as they appear in L2 learner writing, addressing key challenges such as polysemy, contextual variation, and multi-word expressions. We compare LLMs to a PoS-based baseline. LLMs appear to exploit additional semantic information that yields improved performance. We also explore correlations between word-level proficiency and essay-level proficiency. Finally, the approach is applied to examine the consistency of the EVP proficiency levels. Results show that LLMs are well-suited for the task of vocabulary assessment.
Authors: Fatma Youssef Mohammed, Kostas Alexis
Abstract: Computational human attention modeling in free-viewing and task-specific settings is often studied separately, with limited exploration of whether a common representation exists between them. This work investigates this question and proposes a neural network architecture that builds upon the Human Attention transformer (HAT) to test the hypothesis. Our results demonstrate that free-viewing and visual search can efficiently share a common representation, allowing a model trained in free-viewing attention to transfer its knowledge to task-driven visual search with a performance drop of only 3.86% in the predicted fixation scanpaths, measured by the semantic sequence score (SemSS) metric which reflects the similarity between predicted and human scanpaths. This transfer reduces computational costs by 92.29% in terms of GFLOPs and 31.23% in terms of trainable parameters.
Authors: Charalampos Kalalas, Pavol Mulinka, Guillermo Candela Belmonte, Miguel Fornell, Michail Dalgitsis, Francisco Paredes Vera, Javier Santaella S\'anchez, Carmen Vicente Villares, Roshan Sedar, Eftychia Datsika, Angelos Antonopoulos, Antonio Fern\'andez Ojea, Miquel Payaro
Abstract: Artificial intelligence (AI) has been increasingly applied to the condition monitoring of vehicular equipment, aiming to enhance maintenance strategies, reduce costs, and improve safety. Leveraging the edge computing paradigm, AI-based condition monitoring systems process vast streams of vehicular data to detect anomalies and optimize operational performance. In this work, we introduce a novel vehicle condition monitoring service that enables real-time diagnostics of a diverse set of anomalies while remaining practical for deployment in real-world edge environments. To address mobility challenges, we propose a closed-loop service orchestration framework where service migration across edge nodes is dynamically triggered by network-related metrics. Our approach has been implemented and tested in a real-world race circuit environment equipped with 5G network capabilities under diverse operational conditions. Experimental results demonstrate the effectiveness of our framework in ensuring low-latency AI inference and adaptive service placement, highlighting its potential for intelligent transportation and mobility applications.
Authors: Ruilong Wu, Xinjiao Li, Yisu Wang, Xinyu Chen, Dirk Kutscher
Abstract: Hybrid parallelism techniques are essential for efficiently training large language models (LLMs). Nevertheless, current automatic parallel planning frameworks often overlook the simultaneous consideration of node heterogeneity and dynamic network topology changes, limiting their effectiveness in practical applications. In this paper, we address these limitations by modeling heterogeneous nodes within dynamically changing network environments and leveraging simulation-based strategies to determine optimal parallel configurations. Our approach enables fine-grained workload allocation tailored for heterogeneous nodes and complex network scenarios, achieving performance competitive with state-of-the-art methods under regular and stable network conditions. Additionally, we introduce a strategy pruning technique to rapidly discard infeasible parallel configurations, substantially reducing the search space and accelerating the search process through parallel execution within the simulator. Preliminary evaluations confirm that our method notably enhances training performance on heterogeneous nodes and demonstrates improved adaptability in complex, dynamic scenarios such as cloud computing environments.
Authors: Zhen Yang, Hongyi Lin, Yifan He, Jie Xu, Zeyu Sun, Shuo Liu, Pengpeng Wang, Zhongxing Yu, Qingyuan Liang
Abstract: In recent years, code intelligence has gained increasing importance in the field of automated software engineering. Meanwhile, the widespread adoption of Pretrained Language Models (PLMs) and Large Language Models (LLMs) has raised concerns regarding data contamination and its potential impact on model performance evaluation. This paper presents a systematic empirical study to investigate the fine-grained data contamination on code intelligence tasks. Our study involves diverse representative PLMs, namely RoBERTa and GPT-2, and LLMs, namely LLaMA and StarCoder, covering three major tasks: code translation, code generation, and code summarization. We categorize contamination scenarios into four types according to the code intelligence practice, namely input-only, output-only, unpaired, and paired contamination settings, and construct corresponding experimental and control groups for exploration. Experimental results show that, under the pre-training, fine-tuning, and inference paradigm adopted by PLMs, even deliberately injecting paired contamination does not lead to significant performance overestimation. But direct inference or small-scale fine-tuning uncovers the contamination effects. In contrast, LLMs with pre-training and inference paradigm are significantly affected by the paired contamination. Apart from the above, other contamination scenarios have no impact on both PLMs and LLMs. Our findings challenge the conventional belief that contamination inevitably leads to performance overestimation, providing new insights into the evaluation and deployment of code intelligence models.
Authors: Mijeong Kim, Gunhee Kim, Jungyoon Choi, Wonjae Roh, Bohyung Han
Abstract: We introduce PhysGaia, a novel physics-aware dataset specifically designed for Dynamic Novel View Synthesis (DyNVS), encompassing both structured objects and unstructured physical phenomena. Unlike existing datasets that primarily focus on photorealistic reconstruction, PhysGaia is created to actively support physics-aware dynamic scene modeling. Our dataset provides complex dynamic scenarios with rich interactions among multiple objects, where they realistically collide with each other and exchange forces. Furthermore, it contains a diverse range of physical materials, such as liquid, gas, viscoelastic substance, and textile, which moves beyond the rigid bodies prevalent in existing datasets. All scenes in PhysGaia are faithfully generated to strictly adhere to physical laws, leveraging carefully selected material-specific physics solvers. To enable quantitative evaluation of physical modeling, our dataset provides essential ground-truth information, including 3D particle trajectories and physics parameters, e.g., viscosity. To facilitate research adoption, we also provide essential integration pipelines for using state-of-the-art DyNVS models with our dataset and report their results. By addressing the critical lack of datasets for physics-aware modeling, PhysGaia will significantly advance research in dynamic view synthesis, physics-based scene understanding, and deep learning models integrated with physical simulation -- ultimately enabling more faithful reconstruction and interpretation of complex dynamic scenes. Our datasets and codes are available in the project website, http://cvlab.snu.ac.kr/research/PhysGaia.
Authors: Haoyuan Wang, Chen Liu, Minh-Ngoc Tran, Chao Wang
Abstract: This paper introduces a novel multivariate volatility modeling framework, named Long Short-Term Memory enhanced BEKK (LSTM-BEKK), that integrates deep learning into multivariate GARCH processes. By combining the flexibility of recurrent neural networks with the econometric structure of BEKK models, our approach is designed to better capture nonlinear, dynamic, and high-dimensional dependence structures in financial return data. The proposed model addresses key limitations of traditional multivariate GARCH-based methods, particularly in capturing persistent volatility clustering and asymmetric co-movement across assets. Leveraging the data-driven nature of LSTMs, the framework adapts effectively to time-varying market conditions, offering improved robustness and forecasting performance. Empirical results across multiple equity markets confirm that the LSTM-BEKK model achieves superior performance in terms of out-of-sample portfolio risk forecast, while maintaining the interpretability from the BEKK models. These findings highlight the potential of hybrid econometric-deep learning models in advancing financial risk management and multivariate volatility forecasting.
Authors: Yougang Lyu, Xiaoyu Zhang, Lingyong Yan, Maarten de Rijke, Zhaochun Ren, Xiuying Chen
Abstract: Web agents for online shopping have shown great promise in automating user interactions across e-commerce platforms. Benchmarks for assessing such agents do not reflect the complexity of real-world shopping scenarios, as they often consist of overly simple queries with deterministic paths, such as "Find iPhone 15." Real shopping scenarios are inherently more layered, involving multi-dimensional product attributes, search filters, and user-specific sorting preferences. To address this gap, we introduce DeepShop, a benchmark designed to evaluate web agents in complex and realistic online shopping environments. DeepShop comprises three key components. (1) Query diversity evolution: Starting from real user queries, we generate diverse queries across five popular online shopping domains. (2) Query complexity evolution: We further evolve these queries to increase complexity, considering product attributes, search filters, and sorting preferences, and classify them into three levels: easy, medium, and hard, based on the number of evolutions. (3) Fine-grained and holistic evaluation: We propose an automated evaluation framework that assesses agent performance in terms of fine-grained aspects (product attributes, search filters, and sorting preferences) and reports the overall success rate through holistic evaluation. We conduct a systematic evaluation of retrieval-augmented generation (RAG) methods, web agents, and deep research systems. Results show that RAG struggles with complex queries due to its lack of web interaction, while other methods face significant challenges with filters and sorting preferences, leading to low overall success rates. We also perform cross-category, complexity-based evaluations and error analyses to support the advancement of deep research shopping agents.
Authors: Stefano Fiorini, Hakan Aktas, Iulia Duta, Stefano Coniglio, Pietro Morerio, Alessio Del Bue, Pietro Li\`o
Abstract: Sheaf Neural Networks (SNNs) represent a powerful generalization of Graph Neural Networks (GNNs) that significantly improve our ability to model complex relational data. While directionality has been shown to substantially boost performance in graph learning tasks and is key to many real-world applications, existing SNNs fall short in representing it. To address this limitation, we introduce the Directed Cellular Sheaf, a special type of cellular sheaf designed to explicitly account for edge orientation. Building on this structure, we define a new sheaf Laplacian, the Directed Sheaf Laplacian, which captures both the graph's topology and its directional information. This operator serves as the backbone of the Directed Sheaf Neural Network (DSNN), the first SNN model to embed a directional bias into its architecture. Extensive experiments on nine real-world benchmarks show that DSNN consistently outperforms baseline methods.
Authors: Geonyoung Lee, Geonhee Han, Paul Hongsuck Seo
Abstract: Language-queried Audio Source Separation (LASS) enables open-vocabulary sound separation via natural language queries. While existing methods rely on task-specific training, we explore whether pretrained diffusion models, originally designed for audio generation, can inherently perform separation without further training. In this study, we introduce a training-free framework leveraging generative priors for zero-shot LASS. Analyzing na\"ive adaptations, we identify key limitations arising from modality-specific challenges.To address these issues, we propose Diffusion-Guided Mask Optimization (DGMO), a test-time optimization framework that refines spectrogram masks for precise, input-aligned separation. Our approach effectively repurposes pretrained diffusion models for source separation, achieving competitive performance without task-specific supervision. This work expands the application of diffusion models beyond generation, establishing a new paradigm for zero-shot audio separation. The code is available at: https://wltschmrz.github.io/DGMO/
Authors: Parth Atulbhai Gandhi, Akansha Shukla, David Tayouri, Beni Ifland, Yuval Elovici, Rami Puzis, Asaf Shabtai
Abstract: Evaluating the security of multi-agent systems (MASs) powered by large language models (LLMs) is challenging, primarily because of the systems' complex internal dynamics and the evolving nature of LLM vulnerabilities. Traditional attack graph (AG) methods often lack the specific capabilities to model attacks on LLMs. This paper introduces AI-agent application Threat assessment with Attack Graphs (ATAG), a novel framework designed to systematically analyze the security risks associated with AI-agent applications. ATAG extends the MulVAL logic-based AG generation tool with custom facts and interaction rules to accurately represent AI-agent topologies, vulnerabilities, and attack scenarios. As part of this research, we also created the LLM vulnerability database (LVD) to initiate the process of standardizing LLM vulnerabilities documentation. To demonstrate ATAG's efficacy, we applied it to two multi-agent applications. Our case studies demonstrated the framework's ability to model and generate AGs for sophisticated, multi-step attack scenarios exploiting vulnerabilities such as prompt injection, excessive agency, sensitive information disclosure, and insecure output handling across interconnected agents. ATAG is an important step toward a robust methodology and toolset to help understand, visualize, and prioritize complex attack paths in multi-agent AI systems (MAASs). It facilitates proactive identification and mitigation of AI-agent threats in multi-agent applications.
Authors: Wenjing Tang, Xinyu He, Yongxi Huang, Yunxiao Xiao, Cewu Lu, Panpan Cai
Abstract: Task planning under uncertainty is essential for home-service robots operating in the real world. Tasks involve ambiguous human instructions, hidden or unknown object locations, and open-vocabulary object types, leading to significant open-ended uncertainty and a boundlessly large planning space. To address these challenges, we propose Tru-POMDP, a planner that combines structured belief generation using Large Language Models (LLMs) with principled POMDP planning. Tru-POMDP introduces a hierarchical Tree of Hypotheses (TOH), which systematically queries an LLM to construct high-quality particle beliefs over possible world states and human goals. We further formulate an open-ended POMDP model that enables rigorous Bayesian belief tracking and efficient belief-space planning over these LLM-generated hypotheses. Experiments on complex object rearrangement tasks across diverse kitchen environments show that Tru-POMDP significantly outperforms state-of-the-art LLM-based and LLM-tree-search hybrid planners, achieving higher success rates with significantly better plans, stronger robustness to ambiguity and occlusion, and greater planning efficiency.
Authors: Helin Wang, Jiarui Hai, Dading Chong, Karan Thakkar, Tiantian Feng, Dongchao Yang, Junhyeok Lee, Laureano Moro Velazquez, Jesus Villalba, Zengyi Qin, Shrikanth Narayanan, Mounya Elhiali, Najim Dehak
Abstract: Recent advancements in generative artificial intelligence have significantly transformed the field of style-captioned text-to-speech synthesis (CapTTS). However, adapting CapTTS to real-world applications remains challenging due to the lack of standardized, comprehensive datasets and limited research on downstream tasks built upon CapTTS. To address these gaps, we introduce CapSpeech, a new benchmark designed for a series of CapTTS-related tasks, including style-captioned text-to-speech synthesis with sound events (CapTTS-SE), accent-captioned TTS (AccCapTTS), emotion-captioned TTS (EmoCapTTS), and text-to-speech synthesis for chat agent (AgentTTS). CapSpeech comprises over 10 million machine-annotated audio-caption pairs and nearly 0.36 million human-annotated audio-caption pairs. In addition, we introduce two new datasets collected and recorded by a professional voice actor and experienced audio engineers, specifically for the AgentTTS and CapTTS-SE tasks. Alongside the datasets, we conduct comprehensive experiments using both autoregressive and non-autoregressive models on CapSpeech. Our results demonstrate high-fidelity and highly intelligible speech synthesis across a diverse range of speaking styles. To the best of our knowledge, CapSpeech is the largest available dataset offering comprehensive annotations for CapTTS-related tasks. The experiments and findings further provide valuable insights into the challenges of developing CapTTS systems.
Authors: Changyi Xiao, Mengdi Zhang, Yixin Cao
Abstract: Recent studies, including DeepSeek-R1 and Kimi-k1.5, have demonstrated that reinforcement learning with rule-based, binary-valued reward functions can significantly enhance the reasoning capabilities of large language models. These models primarily utilize REINFORCE-based policy optimization techniques, such as REINFORCE with baseline and group relative policy optimization (GRPO). However, a key limitation remains: current policy optimization methods either neglect reward normalization or employ static normalization strategies, which fail to adapt to the dynamic nature of policy updates during training. This may result in unstable gradient estimates and hinder training stability. To address this issue, we propose Beta Normalization Policy Optimization (BNPO), a novel policy optimization method that adaptively normalizes rewards using a Beta distribution with dynamically updated parameters. BNPO aligns the normalization with the changing policy distribution, enabling more precise and lower-variance gradient estimation, which in turn promotes stable training dynamics. We provide theoretical analysis demonstrating BNPO's variance-reducing properties and show that it generalizes both REINFORCE and GRPO under binary-valued reward settings. Furthermore, we introduce an advantage decomposition mechanism to extend BNPO's applicability to more complex reward systems. Experimental results confirm that BNPO achieves state-of-the-art performance among policy optimization methods on reasoning tasks. The code is available at https://github.com/changyi7231/BNPO.
Authors: Jintian Shao, Yiming Cheng
Abstract: Chain-of-Thought (CoT) prompting has demonstrably enhanced the performance of Large Language Models on tasks requiring multi-step inference. This success has led to widespread claims of emergent reasoning capabilities in these models. In this paper, we present a theoretical counter-perspective: Chain-of-Thought (CoT) does not elicit genuine, abstract reasoning. Instead, we argue that Chain-of-Thought functions as a powerful structural constraint that guides Large Language Models to imitate the form of reasoning. By forcing the generation of intermediate steps, Chain-of-Thought leverages the model immense capacity for sequence prediction and pattern matching, effectively constraining its output to sequences that resemble coherent thought processes. Chain-of-Thought (CoT) prompting has demonstrably enhanced the performance of Large Language Models on tasks requiring multi-step inference. This success has led to widespread claims of emergent reasoning capabilities in these models. In this paper, we present a theoretical counter-perspective: Chain-of-Thought (CoT) does not elicit genuine, abstract reasoning. Instead, we argue that Chain-of-Thought functions as a powerful structural constraint that guides Large Language Models to imitate the form of reasoning. By forcing the generation of intermediate steps, Chain-of-Thought leverages the model immense capacity for sequence prediction and pattern matching, effectively constraining its output to sequences that resemble coherent thought processes.
Authors: Jakub Krajewski, Marcin Chochowski, Daniel Korzekwa
Abstract: Mixture of Experts (MoE) architectures have emerged as pivotal for scaling Large Language Models (LLMs) efficiently. Fine-grained MoE approaches - utilizing more numerous, smaller experts - have demonstrated potential in improving model convergence and quality. This work proposes a set of training recipes and provides a comprehensive empirical evaluation of fine-grained MoE, directly comparing its scaling properties against standard MoE configurations for models with up to 56B total (17B active) parameters. We investigate convergence speed, model performance on downstream benchmarks, and practical training considerations across various setups. Overall, at the largest scale we show that fine-grained MoE achieves better validation loss and higher accuracy across a set of downstream benchmarks. This study offers empirical grounding and practical insights for leveraging fine-grained MoE in the development of future large-scale models.
Authors: Yusuke Sakai, Takumi Goto, Taro Watanabe
Abstract: We propose IMPARA-GED, a novel reference-free automatic grammatical error correction (GEC) evaluation method with grammatical error detection (GED) capabilities. We focus on the quality estimator of IMPARA, an existing automatic GEC evaluation method, and construct that of IMPARA-GED using a pre-trained language model with enhanced GED capabilities. Experimental results on SEEDA, a meta-evaluation dataset for automatic GEC evaluation methods, demonstrate that IMPARA-GED achieves the highest correlation with human sentence-level evaluations.
Authors: Yin Fang, Qiao Jin, Guangzhi Xiong, Bowen Jin, Xianrui Zhong, Siru Ouyang, Aidong Zhang, Jiawei Han, Zhiyong Lu
Abstract: Cell type annotation is a key task in analyzing the heterogeneity of single-cell RNA sequencing data. Although recent foundation models automate this process, they typically annotate cells independently, without considering batch-level cellular context or providing explanatory reasoning. In contrast, human experts often annotate distinct cell types for different cell clusters based on their domain knowledge. To mimic this workflow, we introduce the CellPuzzles task, where the objective is to assign unique cell types to a batch of cells. This benchmark spans diverse tissues, diseases, and donor conditions, and requires reasoning across the batch-level cellular context to ensure label uniqueness. We find that off-the-shelf large language models (LLMs) struggle on CellPuzzles, with the best baseline (OpenAI's o1) achieving only 19.0% batch-level accuracy. To fill this gap, we propose Cell-o1, a 7B LLM trained via supervised fine-tuning on distilled reasoning traces, followed by reinforcement learning with batch-level rewards. Cell-o1 achieves state-of-the-art performance, outperforming o1 by over 73% and generalizing well across contexts. Further analysis of training dynamics and reasoning behaviors provides insights into batch-level annotation performance and emergent expert-like reasoning. Code and data are available at https://github.com/ncbi-nlp/cell-o1.
Authors: Praneet Sai Madhu Surabhi, Dheeraj Reddy Mudireddy, Jian Tao
Abstract: This paper presents ThinkTank, a comprehensive and scalable framework designed to transform specialized AI agent systems into versatile collaborative intelligence platforms capable of supporting complex problem-solving across diverse domains. ThinkTank systematically generalizes agent roles, meeting structures, and knowledge integration mechanisms by adapting proven scientific collaboration methodologies. Through role abstraction, generalization of meeting types for iterative collaboration, and the integration of Retrieval-Augmented Generation with advanced knowledge storage, the framework facilitates expertise creation and robust knowledge sharing. ThinkTank enables organizations to leverage collaborative AI for knowledge-intensive tasks while ensuring data privacy and security through local deployment, utilizing frameworks like Ollama with models such as Llama3.1. The ThinkTank framework is designed to deliver significant advantages in cost-effectiveness, data security, scalability, and competitive positioning compared to cloud-based alternatives, establishing it as a universal platform for AI-driven collaborative problem-solving. The ThinkTank code is available at https://github.com/taugroup/ThinkTank
Authors: Stepan I. Manukhov, Alexander Kolesov, Vladimir V. Palyulin, Alexander Korotin
Abstract: Electrostatic field matching (EFM) has recently appeared as a novel physics-inspired paradigm for data generation and transfer using the idea of an electric capacitor. However, it requires modeling electrostatic fields using neural networks, which is non-trivial because of the necessity to take into account the complex field outside the capacitor plates. In this paper, we propose Interaction Field Matching (IFM), a generalization of EFM which allows using general interaction fields beyond the electrostatic one. Furthermore, inspired by strong interactions between quarks and antiquarks in physics, we design a particular interaction field realization which solves the problems which arise when modeling electrostatic fields in EFM. We show the performance on a series of toy and image data transfer problems.
Authors: Zewen Yang, Xiaobing Dai, Dian Yu, Qianru Li, Yu Li, Valentin Le Mesle
Abstract: Generative models have become increasingly powerful tools for robot motion generation, enabling flexible and multimodal trajectory generation across various tasks. Yet, most existing approaches remain limited in handling multiple types of constraints, such as collision avoidance and dynamic consistency, which are often treated separately or only partially considered. This paper proposes UniConFlow, a unified flow matching (FM) based framework for trajectory generation that systematically incorporates both equality and inequality constraints. UniConFlow introduces a novel prescribed-time zeroing function to enhance flexibility during the inference process, allowing the model to adapt to varying task requirements. To ensure constraint satisfaction, particularly with respect to obstacle avoidance, admissible action range, and kinodynamic consistency, the guidance inputs to the FM model are derived through a quadratic programming formulation, which enables constraint-aware generation without requiring retraining or auxiliary controllers. We conduct mobile navigation and high-dimensional manipulation tasks, demonstrating improved safety and feasibility compared to state-of-the-art constrained generative planners. Project page is available at https://uniconflow.github.io.
Authors: Zhixiong Su, Yichen Wang, Herun Wan, Zhaohan Zhang, Minnan Luo
Abstract: The misuse of large language models (LLMs) poses potential risks, motivating the development of machine-generated text (MGT) detection. Existing literature primarily concentrates on binary, document-level detection, thereby neglecting texts that are composed jointly by human and LLM contributions. Hence, this paper explores the possibility of fine-grained MGT detection under human-AI coauthoring. We suggest fine-grained detectors can pave pathways toward coauthored text detection with a numeric AI ratio. Specifically, we propose a dataset, HACo-Det, which produces human-AI coauthored texts via an automatic pipeline with word-level attribution labels. We retrofit seven prevailing document-level detectors to generalize them to word-level detection. Then we evaluate these detectors on HACo-Det on both word- and sentence-level detection tasks. Empirical results show that metric-based methods struggle to conduct fine-grained detection with a 0.462 average F1 score, while finetuned models show superior performance and better generalization across domains. However, we argue that fine-grained co-authored text detection is far from solved. We further analyze factors influencing performance, e.g., context window, and highlight the limitations of current methods, pointing to potential avenues for improvement.
Authors: Yicheng Xiao, Lin Song, Rui Yang, Cheng Cheng, Zunnan Xu, Zhaoyang Zhang, Yixiao Ge, Xiu Li, Ying Shan
Abstract: With the advancement of language models, unified multimodal understanding and generation have made significant strides, with model architectures evolving from separated components to unified single-model frameworks. This paper explores an efficient training paradigm to build a single transformer for unified multimodal understanding and generation. Specifically, we propose a multimodal warmup strategy utilizing prior knowledge to extend capabilities. To address cross-modal compatibility challenges, we introduce feature pre-scaling and multimodal AdaLN techniques. Integrating the proposed technologies, we present the HaploOmni, a new single multimodal transformer. With limited training costs, HaploOmni achieves competitive performance across multiple image and video understanding and generation benchmarks over advanced unified models. All codes will be made public at https://github.com/Tencent/HaploVLM.
Authors: Rachid Zeghlache, Ikram Brahim, Pierre-Henri Conze, Mathieu Lamard, Mohammed El Amine Lazouni, Zineb Aziza Elaouaber, Leila Ryma Lazouni, Christopher Nielsen, Ahmad O. Ahsan, Matthias Wilms, Nils D. Forkert, Lovre Antonio Budimir, Ivana Matovinovi\'c, Donik Vr\v{s}nak, Sven Lon\v{c}ari\'c, Philippe Zhang, Weili Jiang, Yihao Li, Yiding Hao, Markus Frohmann, Patrick Binder, Marcel Huber, Taha Emre, Teresa Finisterra Ara\'ujo, Marzieh Oghbaie, Hrvoje Bogunovi\'c, Amerens A. Bekkers, Nina M. van Liebergen, Hugo J. Kuijf, Abdul Qayyum, Moona Mazher, Steven A. Niederer, Alberto J. Beltr\'an-Carrero, Juan J. G\'omez-Valverde, Javier Torresano-Rodr\'iquez, \'Alvaro Caballero-Sastre, Mar\'ia J. Ledesma Carbayo, Yosuke Yamagishi, Yi Ding, Robin Peretzke, Alexandra Ertl, Maximilian Fischer, Jessica K\"achele, Sofiane Zehar, Karim Boukli Hacene, Thomas Monfort, B\'eatrice Cochener, Mostafa El Habib Daho, Anas-Alexis Benyoussef, Gwenol\'e Quellec
Abstract: The MARIO challenge, held at MICCAI 2024, focused on advancing the automated detection and monitoring of age-related macular degeneration (AMD) through the analysis of optical coherence tomography (OCT) images. Designed to evaluate algorithmic performance in detecting neovascular activity changes within AMD, the challenge incorporated unique multi-modal datasets. The primary dataset, sourced from Brest, France, was used by participating teams to train and test their models. The final ranking was determined based on performance on this dataset. An auxiliary dataset from Algeria was used post-challenge to evaluate population and device shifts from submitted solutions. Two tasks were involved in the MARIO challenge. The first one was the classification of evolution between two consecutive 2D OCT B-scans. The second one was the prediction of future AMD evolution over three months for patients undergoing anti-vascular endothelial growth factor (VEGF) therapy. Thirty-five teams participated, with the top 12 finalists presenting their methods. This paper outlines the challenge's structure, tasks, data characteristics, and winning methodologies, setting a benchmark for AMD monitoring using OCT, infrared imaging, and clinical data (such as the number of visits, age, gender, etc.). The results of this challenge indicate that artificial intelligence (AI) performs as well as a physician in measuring AMD progression (Task 1) but is not yet able of predicting future evolution (Task 2).
Authors: Richard Armitage
Abstract: Background: Large language models (LLMs) have demonstrated substantial potential to support clinical practice. Other than Chat GPT4 and its predecessors, few LLMs, especially those of the leading and more powerful reasoning model class, have been subjected to medical specialty examination questions, including in the domain of primary care. This paper aimed to test the capabilities of leading LLMs as of May 2025 (o3, Claude Opus 4, Grok3, and Gemini 2.5 Pro) in primary care education, specifically in answering Member of the Royal College of General Practitioners (MRCGP) style examination questions. Methods: o3, Claude Opus 4, Grok3, and Gemini 2.5 Pro were tasked to answer 100 randomly chosen multiple choice questions from the Royal College of General Practitioners GP SelfTest on 25 May 2025. Questions included textual information, laboratory results, and clinical images. Each model was prompted to answer as a GP in the UK and was provided with full question information. Each question was attempted once by each model. Responses were scored against correct answers provided by GP SelfTest. Results: The total score of o3, Claude Opus 4, Grok3, and Gemini 2.5 Pro was 99.0%, 95.0%, 95.0%, and 95.0%, respectively. The average peer score for the same questions was 73.0%. Discussion: All models performed remarkably well, and all substantially exceeded the average performance of GPs and GP registrars who had answered the same questions. o3 demonstrated the best performance, while the performances of the other leading models were comparable with each other and were not substantially lower than that of o3. These findings strengthen the case for LLMs, particularly reasoning models, to support the delivery of primary care, especially those that have been specifically trained on primary care clinical data.
Authors: Florian Ludwig, Torsten Zesch, Frederike Zufall
Abstract: The assessment of legal problems requires the consideration of a specific legal system and its levels of abstraction, from constitutional law to statutory law to case law. The extent to which Large Language Models (LLMs) internalize such legal systems is unknown. In this paper, we propose and investigate different approaches to condition LLMs at different levels of abstraction in legal systems. This paper examines different approaches to conditioning LLMs at multiple levels of abstraction in legal systems to detect potentially punishable hate speech. We focus on the task of classifying whether a specific social media posts falls under the criminal offense of incitement to hatred as prescribed by the German Criminal Code. The results show that there is still a significant performance gap between models and legal experts in the legal assessment of hate speech, regardless of the level of abstraction with which the models were conditioned. Our analysis revealed, that models conditioned on abstract legal knowledge lacked deep task understanding, often contradicting themselves and hallucinating answers, while models using concrete legal knowledge performed reasonably well in identifying relevant target groups, but struggled with classifying target conducts.
Authors: David McVicar, Brian Avant, Adrian Gould, Diego Torrejon, Charles Della Porta, Ryan Mukherjee
Abstract: BlackSky introduces Smartflow, a cloud-based framework enabling scalable spatiotemporal geospatial research built on open-source tools and technologies. Using STAC-compliant catalogs as a common input, heterogeneous geospatial data can be processed into standardized datacubes for analysis and model training. Model experimentation is managed using a combination of tools, including ClearML, Tensorboard, and Apache Superset. Underpinning Smartflow is Kubernetes, which orchestrates the provisioning and execution of workflows to support both horizontal and vertical scalability. This combination of features makes Smartflow well-suited for geospatial model development and analysis over large geographic areas, time scales, and expansive image archives. We also present a novel neural architecture, built using Smartflow, to monitor large geographic areas for heavy construction. Qualitative results based on data from the IARPA Space-based Machine Automated Recognition Technique (SMART) program are presented that show the model is capable of detecting heavy construction throughout all major phases of development.
Authors: Pierre Lepagnol, Sahar Ghannay, Thomas Gerald, Christophe Servan, Sophie Rosset
Abstract: Understanding user queries is fundamental in many applications, such as home assistants, booking systems, or recommendations. Accordingly, it is crucial to develop accurate Spoken Language Understanding (SLU) approaches to ensure the reliability of the considered system. Current State-of-the-Art SLU techniques rely on large amounts of training data; however, only limited annotated examples are available for specific tasks or languages. In the meantime, instruction-tuned large language models (LLMs) have shown exceptional performance on unseen tasks in a few-shot setting when provided with adequate prompts. In this work, we propose to explore example selection by leveraging Information retrieval (IR) approaches to build an enhanced prompt that is applied to an SLU task. We evaluate the effectiveness of the proposed method on several SLU benchmarks. Experimental results show that lexical IR methods significantly enhance performance without increasing prompt length.
Authors: Mikolaj Walczak, Romina Aalishah, Wyatt Mackey, Brittany Story, David L. Boothe Jr., Nicholas Waytowich, Xiaomin Lin, Tinoosh Mohsenin
Abstract: Deep reinforcement learning agents are often fragile while humans remain adaptive and flexible to varying scenarios. To bridge this gap, we present EDEN, a biologically inspired navigation framework that integrates learned entorhinal-like grid cell representations and reinforcement learning to enable autonomous navigation. Inspired by the mammalian entorhinal-hippocampal system, EDEN allows agents to perform path integration and vector-based navigation using visual and motion sensor data. At the core of EDEN is a grid cell encoder that transforms egocentric motion into periodic spatial codes, producing low-dimensional, interpretable embeddings of position. To generate these activations from raw sensory input, we combine fiducial marker detections in the lightweight MiniWorld simulator and DINO-based visual features in the high-fidelity Gazebo simulator. These spatial representations serve as input to a policy trained with Proximal Policy Optimization (PPO), enabling dynamic, goal-directed navigation. We evaluate EDEN in both MiniWorld, for rapid prototyping, and Gazebo, which offers realistic physics and perception noise. Compared to baseline agents using raw state inputs (e.g., position, velocity) or standard convolutional image encoders, EDEN achieves a 99% success rate, within the simple scenarios, and >94% within complex floorplans with occluded paths with more efficient and reliable step-wise navigation. In addition, as a replacement of ground truth activations, we present a trainable Grid Cell encoder enabling the development of periodic grid-like patterns from vision and motion sensor data, emulating the development of such patterns within biological mammals. This work represents a step toward biologically grounded spatial intelligence in robotics, bridging neural navigation principles with reinforcement learning for scalable deployment.
Authors: Yuval Kansal, Shmuel Berman, Lydia Liu
Abstract: Factuality is a necessary precursor to useful educational tools. As adoption of Large Language Models (LLMs) in education continues of grow, ensuring correctness in all settings is paramount. Despite their strong English capabilities, LLM performance in other languages is largely untested. In this work, we evaluate the correctness of the Llama3.1 family of models in answering factual questions appropriate for middle and high school students. We demonstrate that LLMs not only provide extraneous and less truthful information, but also exacerbate existing biases against rare languages.
Authors: Sinem Erisken (Independent Researcher), Timothy Gothard (Independent Researcher), Martin Leitgab (Independent Researcher), Ram Potham (Independent Researcher)
Abstract: Traditional AI safety evaluations on isolated LLMs are insufficient as multi-agent AI ensembles become prevalent, introducing novel emergent risks. This paper introduces the Multi-Agent Emergent Behavior Evaluation (MAEBE) framework to systematically assess such risks. Using MAEBE with the Greatest Good Benchmark (and a novel double-inversion question technique), we demonstrate that: (1) LLM moral preferences, particularly for Instrumental Harm, are surprisingly brittle and shift significantly with question framing, both in single agents and ensembles. (2) The moral reasoning of LLM ensembles is not directly predictable from isolated agent behavior due to emergent group dynamics. (3) Specifically, ensembles exhibit phenomena like peer pressure influencing convergence, even when guided by a supervisor, highlighting distinct safety and alignment challenges. Our findings underscore the necessity of evaluating AI systems in their interactive, multi-agent contexts.
Authors: Pengtao Chen, Xianfang Zeng, Maosen Zhao, Peng Ye, Mingzhu Shen, Wei Cheng, Gang Yu, Tao Chen
Abstract: While Diffusion Transformers (DiTs) have achieved breakthroughs in video generation, this long sequence generation task remains constrained by the quadratic complexity of attention mechanisms, resulting in significant inference latency. Through detailed analysis of attention maps in Video Diffusion Transformer (vDiT), we identify three recurring sparsity patterns: diagonal, multi-diagonal, and vertical-stripe structures. And even 3-6\% attention heads can be skipped. Crucially, these patterns exhibit strong layer-depth and head-position correlations but show limited dependence on the input content. Leveraging these findings, we propose Sparse-vDiT, a sparsity acceleration framework for vDiT comprising: 1) Pattern-optimized sparse kernels that replace dense attention with computationally efficient implementations for each identified sparsity pattern. 2) An offline sparse diffusion search algorithm that selects the optimal sparse computation strategy per layer and head via hardware-aware cost modeling. After determining the optimal configuration, we fuse heads within the same layer that share the same attention strategy, enhancing inference efficiency. Integrated into state-of-the-art vDiT models (CogVideoX1.5, HunyuanVideo, and Wan2.1), Sparse-vDiT achieves 2.09$\times$, 2.38$\times$, and 1.67$\times$ theoretical FLOP reduction, and actual inference speedups of 1.76$\times$, 1.85$\times$, and 1.58$\times$, respectively, while maintaining high visual fidelity, with PSNR values reaching 24.13, 27.09, and 22.59. Our work demonstrates that latent structural sparsity in vDiTs can be systematically exploited for long video synthesis.
Authors: Qijun Luo, Mengqi Li, Lei Zhao, Xiao Li
Abstract: Training language models on long sequence data is a demanding requirement for enhancing the model's capability on complex tasks, e.g., long-chain reasoning. However, as the sequence length scales up, the memory cost for storing activation values becomes huge during the Backpropagation (BP) process, even with the application of gradient checkpointing technique. To tackle this challenge, we propose a memory-efficient and exact BP method called StreamBP, which performs a linear decomposition of the chain rule along the sequence dimension in a layer-wise manner, significantly reducing the memory cost of activation values and logits. The proposed method is applicable to common objectives such as SFT, GRPO, and DPO. From an implementation perspective, StreamBP achieves less computational FLOPs and faster BP speed by leveraging the causal structure of the language model. Compared to gradient checkpointing, StreamBP scales up the maximum sequence length of BP by 2.8-5.5 times larger, while using comparable or even less BP time. Note that StreamBP's sequence length scaling ability can be directly transferred to batch size scaling for accelerating training. We further develop a communication-efficient distributed StreamBP to effectively support multi-GPU training and broaden its applicability. Our code can be easily integrated into the training pipeline of any transformer models and is available at https://github.com/Ledzy/StreamBP.
Authors: Adrian de Wynter
Abstract: An evaluator is trustworthy when there exists some agreed-upon way to measure its performance as a labeller. The two ways to establish trustworthiness are either by testing it, or by assuming the evaluator `knows' somehow the way to label the corpus. However, if labelled references (e.g., a development set) are unavailable, neither of these approaches work: the former requires the data, and the latter is an assumption, not evidence. To address this, we introduce an algorithm (the `No-Data Algorithm') by which to establish trust in an evaluator without any existing references. Our algorithm works by successively posing challenges to said evaluator. We show that this is sufficient to establish trustworthiness w.h.p., in such a way that when the evaluator actually knows the way to label the corpus, the No-Data Algorithm accepts its output; and, conversely, flags untrustworthy evaluators when these are unable to prove it. We present formal proofs of correctness and limited experiments.
Authors: Bin Ma, Yuyuan Feng, Minhua Lin, Enyan Dai
Abstract: Graph Neural Networks (GNNs) have become essential tools for analyzing graph-structured data in domains such as drug discovery and financial analysis, leading to growing demands for model transparency. Recent advances in explainable GNNs have addressed this need by revealing important subgraphs that influence predictions, but these explanation mechanisms may inadvertently expose models to security risks. This paper investigates how such explanations potentially leak critical decision logic that can be exploited for model stealing. We propose {\method}, a novel stealing framework that integrates explanation alignment for capturing decision logic with guided data augmentation for efficient training under limited queries, enabling effective replication of both the predictive behavior and underlying reasoning patterns of target models. Experiments on molecular graph datasets demonstrate that our approach shows advantages over conventional methods in model stealing. This work highlights important security considerations for the deployment of explainable GNNs in sensitive domains and suggests the need for protective measures against explanation-based attacks. Our code is available at https://github.com/beanmah/EGSteal.
Authors: Lloyd Pellatt, Fotios Drakopoulos, Shievanie Sabesan, Nicholas A. Lesica
Abstract: The mapping from sound to neural activity that underlies hearing is highly non-linear. The first few stages of this mapping in the cochlea have been modelled successfully, with biophysical models built by hand and, more recently, with DNN models trained on datasets simulated by biophysical models. Modelling the auditory brain has been a challenge because central auditory processing is too complex for models to be built by hand, and datasets for training DNN models directly have not been available. Recent work has taken advantage of large-scale high resolution neural recordings from the auditory midbrain to build a DNN model of normal hearing with great success. But this model assumes that auditory processing is the same in all brains, and therefore it cannot capture the widely varying effects of hearing loss. We propose a novel variational-conditional model to learn to encode the space of hearing loss directly from recordings of neural activity in the auditory midbrain of healthy and noise exposed animals. With hearing loss parametrised by only 6 free parameters per animal, our model accurately predicts 62\% of the explainable variance in neural responses from normal hearing animals and 68% for hearing impaired animals, within a few percentage points of state of the art animal specific models. We demonstrate that the model can be used to simulate realistic activity from out of sample animals by fitting only the learned conditioning parameters with Bayesian optimisation, achieving crossentropy loss within 2% of the optimum in 15-30 iterations. Including more animals in the training data slightly improved the performance on unseen animals. This model will enable future development of parametrised hearing loss compensation models trained to directly restore normal neural coding in hearing impaired brains, which can be quickly fitted for a new user by human in the loop optimisation.
Authors: Ashwin Vinod, Shrey Pandit, Aditya Vavre, Linshen Liu
Abstract: Emerging embodied AI applications, such as wearable cameras and autonomous agents, have underscored the need for robust reasoning from first person video streams. We introduce EgoVLM, a vision-language model specifically designed to integrate visual comprehension and spatial-temporal reasoning within egocentric video contexts. EgoVLM is fine-tuned via Group Relative Policy Optimization (GRPO), a reinforcement learning method adapted to align model outputs with human-like reasoning steps. Following DeepSeek R1-Zero's approach, we directly tune using RL without any supervised fine-tuning phase on chain-of-thought (CoT) data. We evaluate EgoVLM on egocentric video question answering benchmarks and show that domain-specific training substantially improves performance over general-purpose VLMs. Our EgoVLM-3B, trained exclusively on non-CoT egocentric data, outperforms the base Qwen2.5-VL 3B and 7B models by 14.33 and 13.87 accuracy points on the EgoSchema benchmark, respectively. By explicitly generating reasoning traces, EgoVLM enhances interpretability, making it well-suited for downstream applications. Furthermore, we introduce a novel keyframe-based reward that incorporates salient frame selection to guide reinforcement learning optimization. This reward formulation opens a promising avenue for future exploration in temporally grounded egocentric reasoning.
Authors: Chetwin Low, Weimin Wang
Abstract: In this paper, we present TalkingMachines -- an efficient framework that transforms pretrained video generation models into real-time, audio-driven character animators. TalkingMachines enables natural conversational experiences by integrating an audio large language model (LLM) with our video generation foundation model. Our primary contributions include: (1) We adapt a pretrained SOTA image-to-video DiT into an audio-driven avatar generation model of 18 billion parameters; (2) We enable infinite video streaming without error accumulation through asymmetric knowledge distillation from a bidirectional teacher model into a sparse causal, autoregressive student model; (3) We design a high-throughput, low-latency inference pipeline incorporating several key engineering optimizations such as: (a) disaggregation of the DiT and VAE decoder across separate devices, (b) efficient overlap of inter-device communication and computation using CUDA streams, (c) elimination of redundant recomputations to maximize frame-generation throughput. Please see demo videos here - https://aaxwaz.github.io/TalkingMachines/
Authors: Yang Guo, Yutian Tao, Yifei Ming, Robert D. Nowak, Yingyu Liang
Abstract: Retrieval-augmented generation (RAG) has seen many empirical successes in recent years by aiding the LLM with external knowledge. However, its theoretical aspect has remained mostly unexplored. In this paper, we propose the first finite-sample generalization bound for RAG in in-context linear regression and derive an exact bias-variance tradeoff. Our framework views the retrieved texts as query-dependent noisy in-context examples and recovers the classical in-context learning (ICL) and standard RAG as the limit cases. Our analysis suggests that an intrinsic ceiling on generalization error exists on RAG as opposed to the ICL. Furthermore, our framework is able to model retrieval both from the training data and from external corpora by introducing uniform and non-uniform RAG noise. In line with our theory, we show the sample efficiency of ICL and RAG empirically with experiments on common QA benchmarks, such as Natural Questions and TriviaQA.
Authors: Sophie Greenwood, Karen Levy, Solon Barocas, Hoda Heidari, Jon Kleinberg
Abstract: As AI technologies improve, people are increasingly willing to delegate tasks to AI agents. In many cases, the human decision-maker chooses whether to delegate to an AI agent based on properties of the specific instance of the decision-making problem they are facing. Since humans typically lack full awareness of all the factors relevant to this choice for a given decision-making instance, they perform a kind of categorization by treating indistinguishable instances -- those that have the same observable features -- as the same. In this paper, we define the problem of designing the optimal algorithmic delegate in the presence of categories. This is an important dimension in the design of algorithms to work with humans, since we show that the optimal delegate can be an arbitrarily better teammate than the optimal standalone algorithmic agent. The solution to this optimal delegation problem is not obvious: we discover that this problem is fundamentally combinatorial, and illustrate the complex relationship between the optimal design and the properties of the decision-making task even in simple settings. Indeed, we show that finding the optimal delegate is computationally hard in general. However, we are able to find efficient algorithms for producing the optimal delegate in several broad cases of the problem, including when the optimal action may be decomposed into functions of features observed by the human and the algorithm. Finally, we run computational experiments to simulate a designer updating an algorithmic delegate over time to be optimized for when it is actually adopted by users, and show that while this process does not recover the optimal delegate in general, the resulting delegate often performs quite well.
Authors: Xiaoying Zhang, Hao Sun, Yipeng Zhang, Kaituo Feng, Chaochao Lu, Chao Yang, Helen Meng
Abstract: Recent advances in reinforcement learning (RL) with numerical feedback, such as scalar rewards, have significantly enhanced the complex reasoning capabilities of large language models (LLMs). Despite this success, we identify three key challenges encountered by RL with solely numerical feedback: performance plateaus, limited effectiveness of self-reflection, and persistent failures. We then demonstrate that RL-finetuned models, even after exhibiting performance plateaus, can generate correct refinements on persistently failed problems by leveraging natural language feedback in the form of critiques. Building on this insight, we propose Critique-GRPO, an online RL framework that integrates both natural language and numerical feedback for effective policy optimization. Critique-GRPO enables LLMs to learn from initial responses and critique-guided refinements simultaneously while maintaining exploration. Extensive experiments using Qwen2.5-7B-Base and Qwen3-8B-Base show that Critique-GRPO consistently outperforms supervised learning-based and RL-based fine-tuning approaches across eight challenging mathematical, STEM, and general reasoning tasks, improving average pass@1 scores by approximately 4.5% and 5%, respectively. Notably, Critique-GRPO surpasses a strong baseline that incorporates expert demonstrations within online RL. Further analysis reveals two critical insights about policy exploration: (1) higher entropy does not always guarantee efficient learning from exploration, and (2) longer responses do not necessarily lead to more effective exploration.
Authors: Kai Lion, Liang Zhang, Bingcong Li, Niao He
Abstract: We show that low-rank adaptation of large-scale models suffers from a low stable rank that is well below the linear algebraic rank of the subspace, degrading fine-tuning performance. To mitigate the underutilization of the allocated subspace, we propose PoLAR, a parameterization inspired by the polar decomposition that factorizes the low-rank update into two direction matrices constrained to Stiefel manifolds and an unconstrained scale matrix. Our theory shows that PoLAR yields an exponentially faster convergence rate on a canonical low-rank adaptation problem. Pairing the parameterization with Riemannian optimization leads to consistent gains on three different benchmarks testing general language understanding, commonsense reasoning, and mathematical problem solving with base model sizes ranging from 350M to 27B.
Authors: Mengdi Jia, Zekun Qi, Shaochen Zhang, Wenyao Zhang, Xinqiang Yu, Jiawei He, He Wang, Li Yi
Abstract: Spatial reasoning is a key aspect of cognitive psychology and remains a major bottleneck for current vision-language models (VLMs). While extensive research has aimed to evaluate or improve VLMs' understanding of basic spatial relations, such as distinguishing left from right, near from far, and object counting, these tasks represent only the most fundamental level of spatial reasoning. In this work, we introduce OmniSpatial, a comprehensive and challenging benchmark for spatial reasoning, grounded in cognitive psychology. OmniSpatial covers four major categories: dynamic reasoning, complex spatial logic, spatial interaction, and perspective-taking, with 50 fine-grained subcategories. Through Internet data crawling and careful manual annotation, we construct over 1.5K question-answer pairs. Extensive experiments show that both open- and closed-source VLMs, as well as existing reasoning and spatial understanding models, exhibit significant limitations in comprehensive spatial understanding. We further analyze failure cases and propose potential directions for future research.
Authors: Siqi Chen, Xinyu Dong, Haolei Xu, Xingyu Wu, Fei Tang, Hang Zhang, Yuchen Yan, Linjuan Wu, Wenqi Zhang, Guiyang Hou, Yongliang Shen, Weiming Lu, Yueting Zhuang
Abstract: Large Language Models (LLMs) and Multimodal LLMs have shown promising capabilities for SVG processing, yet existing benchmarks suffer from limited real-world coverage, lack of complexity stratification, and fragmented evaluation paradigms. We introduce SVGenius, a comprehensive benchmark comprising 2,377 queries across three progressive dimensions: understanding, editing, and generation. Built on real-world data from 24 application domains with systematic complexity stratification, SVGenius evaluates models through 8 task categories and 18 metrics. We assess 22 mainstream models spanning different scales, architectures, training paradigms, and accessibility levels. Our analysis reveals that while proprietary models significantly outperform open-source counterparts, all models exhibit systematic performance degradation with increasing complexity, indicating fundamental limitations in current approaches; however, reasoning-enhanced training proves more effective than pure scaling for overcoming these limitations, though style transfer remains the most challenging capability across all model types. SVGenius establishes the first systematic evaluation framework for SVG processing, providing crucial insights for developing more capable vector graphics models and advancing automated graphic design applications. Appendix and supplementary materials (including all data and code) are available at https://zju-real.github.io/SVGenius.
Authors: Qianhui Wu, Kanzhi Cheng, Rui Yang, Chaoyun Zhang, Jianwei Yang, Huiqiang Jiang, Jian Mu, Baolin Peng, Bo Qiao, Reuben Tan, Si Qin, Lars Liden, Qingwei Lin, Huan Zhang, Tong Zhang, Jianbing Zhang, Dongmei Zhang, Jianfeng Gao
Abstract: One of the principal challenges in building VLM-powered GUI agents is visual grounding, i.e., localizing the appropriate screen region for action execution based on both the visual content and the textual plans. Most existing work formulates this as a text-based coordinate generation task. However, these approaches suffer from several limitations: weak spatial-semantic alignment, inability to handle ambiguous supervision targets, and a mismatch between the dense nature of screen coordinates and the coarse, patch-level granularity of visual features extracted by models like Vision Transformers. In this paper, we propose GUI-Actor, a VLM-based method for coordinate-free GUI grounding. At its core, GUI-Actor introduces an attention-based action head that learns to align a dedicated
Authors: Pralaypati Ta, Sriram Venkatesaperumal, Keerthi Ram, Mohanasankar Sivaprakasam
Abstract: Neuroscience research publications encompass a vast wealth of knowledge. Accurately retrieving existing information and discovering new insights from this extensive literature is essential for advancing the field. However, when knowledge is dispersed across multiple sources, current state-of-the-art retrieval methods often struggle to extract the necessary information. A knowledge graph (KG) can integrate and link knowledge from multiple sources, but existing methods for constructing KGs in neuroscience often rely on labeled data and require domain expertise. Acquiring large-scale, labeled data for a specialized area like neuroscience presents significant challenges. This work proposes novel methods for constructing KG from unlabeled large-scale neuroscience research corpus utilizing large language models (LLM), neuroscience ontology, and text embeddings. We analyze the semantic relevance of neuroscience text segments identified by LLM for building the knowledge graph. We also introduce an entity-augmented information retrieval algorithm to extract knowledge from the KG. Several experiments were conducted to evaluate the proposed approaches, and the results demonstrate that our methods significantly enhance knowledge discovery from the unlabeled neuroscience research corpus. It achieves an F1 score of 0.84 for entity extraction, and the knowledge obtained from the KG improves answers to over 54% of the questions.
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, 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 achieves impressive performance across diverse tasks, including image understanding, generation, manipulation, and perception. We fully open-source the UniWorld framework, including model weights, training and evaluation scripts, and datasets to promote reproducibility and further research.
Authors: Pietro Lesci, Clara Meister, Thomas Hofmann, Andreas Vlachos, Tiago Pimentel
Abstract: Modern language models are typically trained over subword sequences, but ultimately define probabilities over character-strings. Ideally, the choice of the tokeniser -- which maps character-strings to subwords -- should not affect the probability assigned to the underlying character-string; in practice, it does. We define this mismatch as tokenisation bias. In this work, we quantify one particular type of tokenisation bias: the effect of including or not a subword (e.g., $\langle hello \rangle$) in a tokeniser's vocabulary on the probability a trained model assigns to the corresponding characters (i.e., \textit{``hello''}). Estimating this effect is challenging because each model is trained with only one tokeniser. We address this by framing tokenisation bias as a causal effect and estimating it using the regression discontinuity design. Specifically, we exploit the fact that tokenisation algorithms rank subwords and add the first $K$ to a tokeniser's vocabulary, where $K$ is an arbitrary cutoff point. As such, we can estimate a causal effect by comparing similar subwords around this cutoff. Experimentally, we find that tokenisation consistently affects models' outputs across scales, vocabularies, and tokenisers. Notably, a subword's presence in a small model's vocabulary may increase its characters' probability by up to 17 times, highlighting tokenisation as a key design choice in language modelling.
Authors: Yuanze Lin, Yi-Wen Chen, Yi-Hsuan Tsai, Ronald Clark, Ming-Hsuan Yang
Abstract: Although diffusion-based models can generate high-quality and high-resolution video sequences from textual or image inputs, they lack explicit integration of geometric cues when controlling scene lighting and visual appearance across frames. To address this limitation, we propose IllumiCraft, an end-to-end diffusion framework accepting three complementary inputs: (1) high-dynamic-range (HDR) video maps for detailed lighting control; (2) synthetically relit frames with randomized illumination changes (optionally paired with a static background reference image) to provide appearance cues; and (3) 3D point tracks that capture precise 3D geometry information. By integrating the lighting, appearance, and geometry cues within a unified diffusion architecture, IllumiCraft generates temporally coherent videos aligned with user-defined prompts. It supports background-conditioned and text-conditioned video relighting and provides better fidelity than existing controllable video generation methods. Project Page: https://yuanze-lin.me/IllumiCraft_page
Authors: Xinyue Zheng, Haowei Lin, Kaichen He, Zihao Wang, Zilong Zheng, Yitao Liang
Abstract: Developing AI agents capable of interacting with open-world environments to solve diverse tasks is a compelling challenge. However, evaluating such open-ended agents remains difficult, with current benchmarks facing scalability limitations. To address this, we introduce Minecraft Universe (MCU), a comprehensive evaluation framework set within the open-world video game Minecraft. MCU incorporates three key components: (1) an expanding collection of 3,452 composable atomic tasks that encompasses 11 major categories and 41 subcategories of challenges; (2) a task composition mechanism capable of generating infinite diverse tasks with varying difficulty; and (3) a general evaluation framework that achieves 91.5\% alignment with human ratings for open-ended task assessment. Empirical results reveal that even state-of-the-art foundation agents struggle with the increasing diversity and complexity of tasks. These findings highlight the necessity of MCU as a robust benchmark to drive progress in AI agent development within open-ended environments. Our evaluation code and scripts are available at https://github.com/CraftJarvis/MCU.
Authors: David Herrera-Poyatos, Cristina Zuheros, Rosana Montes, Francisco Herrera
Abstract: Social Media and Internet have the potential to be exploited as a source of opinion to enrich Decision Making solutions. Crowd Decision Making (CDM) is a methodology able to infer opinions and decisions from plain texts, such as reviews published in social media platforms, by means of Sentiment Analysis. Currently, the emergence and potential of Large Language Models (LLMs) lead us to explore new scenarios of automatically understand written texts, also known as natural language processing. This paper analyzes the use of ChatGPT based on prompt design strategies to assist in CDM processes to extract opinions and make decisions. We integrate ChatGPT in CDM processes as a flexible tool that infer the opinions expressed in texts, providing numerical or linguistic evaluations where the decision making models are based on the prompt design strategies. We include a multi-criteria decision making scenario with a category ontology for criteria. We also consider ChatGPT as an end-to-end CDM model able to provide a general opinion and score on the alternatives. We conduct empirical experiments on real data extracted from TripAdvisor, the TripR-2020Large dataset. The analysis of results show a promising branch for developing quality decision making models using ChatGPT. Finally, we discuss the challenges of consistency, sensitivity and explainability associated to the use of LLMs in CDM processes, raising open questions for future studies.
Authors: Robin Bloomfield, John Rushby
Abstract: We outline the principles of classical assurance for computer-based systems that pose significant risks. We then consider application of these principles to systems that employ Artificial Intelligence (AI) and Machine Learning (ML). A key element in this "dependability" perspective is a requirement for thorough understanding of the behavior of critical components, and this is considered infeasible for AI and ML. Hence the dependability perspective aims to minimize trust in AI and ML elements by using "defense in depth" with a hierarchy of less complex systems, some of which may be highly assured conventionally engineered components, to "guard" them. This may be contrasted with the "trustworthy" perspective that seeks to apply assurance to the AI and ML elements themselves. In cyber-physical and many other systems, it is difficult to provide guards that do not depend on AI and ML to perceive their environment (e.g., vehicles sharing the road with a self-driving car), so both perspectives are needed and there is a continuum or spectrum between them. We focus on architectures toward the dependability end of the continuum and invite others to consider additional points along the spectrum. For guards that require perception using AI and ML, we examine ways to minimize the trust placed in these elements; they include diversity, defense in depth, explanations, and micro-ODDs. We also examine methods to enforce acceptable behavior, given a model of the world. These include classical cyber-physical calculations and envelopes, and normative rules based on overarching principles, constitutions, ethics, or reputation. We apply our perspective to autonomous systems, AI systems for specific functions, general-purpose AI such as Large Language Models (LLMs), and Artificial General Intelligence (AGI), and we propose current best practice and conclude with a fourfold agenda for research.
Authors: Tianyang Zhang, Zhuoxuan Jiang, Shengguang Bai, Tianrui Zhang, Lin Lin, Yang Liu, Jiawei Ren
Abstract: With the ever-increasing demands on Question Answering (QA) systems for IT operations and maintenance, an efficient and supervised fine-tunable framework is necessary to ensure the data security, private deployment and continuous upgrading. Although Large Language Models (LLMs) have notably improved the open-domain QA's performance, how to efficiently handle enterprise-exclusive corpora and build domain-specific QA systems are still less-studied for industrial applications. In this paper, we propose a general and comprehensive framework based on Retrieval Augmented Generation (RAG) and facilitate the whole business process of establishing QA systems for IT operations and maintenance. In accordance with the prevailing RAG method, our proposed framework, named with RAG4ITOps, composes of two major stages: (1) Models Fine-tuning \& Data Vectorization, and (2) Online QA System Process. At the Stage 1, we leverage a contrastive learning method with two negative sampling strategies to fine-tune the embedding model, and design the instruction templates to fine-tune the LLM with a Retrieval Augmented Fine-Tuning method. At the Stage 2, an efficient process of QA system is built for serving. We collect enterprise-exclusive corpora from the domain of cloud computing, and the extensive experiments show that our method achieves superior results than counterparts on two kinds of QA tasks. Our experiment also provide a case for applying the RAG4ITOps to real-world enterprise-level applications.
Authors: Chen Qian, Dongrui Liu, Jie Zhang, Yong Liu, Jing Shao
Abstract: Ensuring awareness of fairness and privacy in Large Language Models (LLMs) is critical. Interestingly, we discover a counter-intuitive trade-off phenomenon that enhancing an LLM's privacy awareness through Supervised Fine-Tuning (SFT) methods significantly decreases its fairness awareness with thousands of samples. To address this issue, inspired by the information theory, we introduce a training-free method to \textbf{S}uppress the \textbf{P}rivacy and fa\textbf{I}rness coupled \textbf{N}eurons (\textbf{SPIN}), which theoretically and empirically decrease the mutual information between fairness and privacy awareness. Extensive experimental results demonstrate that SPIN eliminates the trade-off phenomenon and significantly improves LLMs' fairness and privacy awareness simultaneously without compromising general capabilities, \eg improving Qwen-2-7B-Instruct's fairness awareness by 12.2\% and privacy awareness by 14.0\%. More crucially, SPIN remains robust and effective with limited annotated data or even when only malicious fine-tuning data is available, whereas SFT methods may fail to perform properly in such scenarios. Furthermore, we show that SPIN could generalize to other potential trade-off dimensions. We hope this study provides valuable insights into concurrently addressing fairness and privacy concerns in LLMs and can be integrated into comprehensive frameworks to develop more ethical and responsible AI systems. Our code is available at https://github.com/ChnQ/SPIN.
Authors: Zerui Cheng, Edoardo Contente, Ben Finch, Oleg Golev, Jonathan Hayase, Andrew Miller, Niusha Moshrefi, Anshul Nasery, Sandeep Nailwal, Sewoong Oh, Himanshu Tyagi, Pramod Viswanath
Abstract: The rapid rise of AI has split model serving between open-weight distribution, which often lacks owner control and monetization, and opaque API-based approaches that risk user privacy and model transparency, forming a dichotomy that hinders an equitable AI ecosystem. This position paper introduces, rigorously formulates, and champions the Open-access, Monetizable, and Loyal (OML) paradigm for AI model serving: a foundational shift to securely distribute and serve AI models by synthesizing transparency with granular monetization and critical safety controls. We survey diverse OML constructions from theory and practice, analyze their security, performance, and practical trade-offs, outline a conceptual OML deployment protocol, and discuss market and policy implications. We assert that OML can foster a democratized, self-sustaining, and innovative AI landscape, mitigating centralized power risks. Finally, we call on the research community to further explore the broad design space of OML, spanning cryptographic, AI-native, and socio-economic mechanisms, to realize its full potential for a collaborative, accountable, and resilient AI future.
Authors: Seongmin Lee (Polo), Hsiang Hsu (Polo), Chun-Fu Chen (Polo), Duen Horng (Polo), Chau
Abstract: LLM hallucination, where unfaithful text is generated, presents a critical challenge for LLMs' practical applications. Current detection methods often resort to external knowledge, LLM fine-tuning, or supervised training with large hallucination-labeled datasets. Moreover, these approaches do not distinguish between different types of hallucinations, which is crucial for enhancing detection performance. To address such limitations, we introduce hallucination probing, a new task that classifies LLM-generated text into three categories: aligned, misaligned, and fabricated. Driven by our novel discovery that perturbing key entities in prompts affects LLM's generation of these three types of text differently, we propose SHINE, a novel hallucination probing method that does not require external knowledge, supervised training, or LLM fine-tuning. SHINE is effective in hallucination probing across three modern LLMs, and achieves state-of-the-art performance in hallucination detection, outperforming seven competing methods across four datasets and four LLMs, underscoring the importance of probing for accurate detection.
Authors: Peter G. Brodeur, Thomas A. Buckley, Zahir Kanjee, Ethan Goh, Evelyn Bin Ling, Priyank Jain, Stephanie Cabral, Raja-Elie Abdulnour, Adrian D. Haimovich, Jason A. Freed, Andrew Olson, Daniel J. Morgan, Jason Hom, Robert Gallo, Liam G. McCoy, Haadi Mombini, Christopher Lucas, Misha Fotoohi, Matthew Gwiazdon, Daniele Restifo, Daniel Restrepo, Eric Horvitz, Jonathan Chen, Arjun K. Manrai, Adam Rodman
Abstract: A seminal paper published by Ledley and Lusted in 1959 introduced complex clinical diagnostic reasoning cases as the gold standard for the evaluation of expert medical computing systems, a standard that has held ever since. Here, we report the results of a physician evaluation of a large language model (LLM) on challenging clinical cases against a baseline of hundreds of physicians. We conduct five experiments to measure clinical reasoning across differential diagnosis generation, display of diagnostic reasoning, triage differential diagnosis, probabilistic reasoning, and management reasoning, all adjudicated by physician experts with validated psychometrics. We then report a real-world study comparing human expert and AI second opinions in randomly-selected patients in the emergency room of a major tertiary academic medical center in Boston, MA. We compared LLMs and board-certified physicians at three predefined diagnostic touchpoints: triage in the emergency room, initial evaluation by a physician, and admission to the hospital or intensive care unit. In all experiments--both vignettes and emergency room second opinions--the LLM displayed superhuman diagnostic and reasoning abilities, as well as continued improvement from prior generations of AI clinical decision support. Our study suggests that LLMs have achieved superhuman performance on general medical diagnostic and management reasoning, fulfilling the vision put forth by Ledley and Lusted, and motivating the urgent need for prospective trials.
Authors: Yewen Li, Shuai Mao, Jingtong Gao, Nan Jiang, Yunjian Xu, Qingpeng Cai, Fei Pan, Peng Jiang, Bo An
Abstract: Auto-bidding is essential in facilitating online advertising by automatically placing bids on behalf of advertisers. Generative auto-bidding, which generates bids based on an adjustable condition using models like transformers and diffusers, has recently emerged as a new trend due to its potential to learn optimal strategies directly from data and adjust flexibly to preferences. However, generative models suffer from low-quality data leading to a mismatch between the condition, like return to go, and true action value, especially in long sequential decision-making. Besides, the majority preference in the dataset may hinder models' generalization ability on minority advertisers' preferences. While it is possible to collect high-quality data and retrain multiple models for different preferences, the high cost makes it unaffordable, hindering the advancement of auto-bidding into the era of large foundation models. To address this, we propose a flexible and practical Generative Auto-bidding scheme using post-training Search, termed GAS, to refine a base policy model's output and adapt to various preferences. We use weak-to-strong search alignment by training small critics for different preferences and an MCTS-inspired search to refine the model's output. Specifically, a novel voting mechanism with transformer-based critics trained with policy indications could enhance search alignment performance. Additionally, utilizing the search, we provide a fine-tuning method for high-frequency preference scenarios considering computational efficiency. Extensive experiments conducted on the real-world dataset and online A/B test on the Kuaishou advertising platform demonstrate the effectiveness of GAS, achieving significant improvements, e.g., 4.60% increment of target cost.
Authors: Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, Chuning Tang, Congcong Wang, Dehao Zhang, Enming Yuan, Enzhe Lu, Fengxiang Tang, Flood Sung, Guangda Wei, Guokun Lai, Haiqing Guo, Han Zhu, Hao Ding, Hao Hu, Hao Yang, Hao Zhang, Haotian Yao, Haotian Zhao, Haoyu Lu, Haoze Li, Haozhen Yu, Hongcheng Gao, Huabin Zheng, Huan Yuan, Jia Chen, Jianhang Guo, Jianlin Su, Jianzhou Wang, Jie Zhao, Jin Zhang, Jingyuan Liu, Junjie Yan, Junyan Wu, Lidong Shi, Ling Ye, Longhui Yu, Mengnan Dong, Neo Zhang, Ningchen Ma, Qiwei Pan, Qucheng Gong, Shaowei Liu, Shengling Ma, Shupeng Wei, Sihan Cao, Siying Huang, Tao Jiang, Weihao Gao, Weimin Xiong, Weiran He, Weixiao Huang, Weixin Xu, Wenhao Wu, Wenyang He, Xianghui Wei, Xianqing Jia, Xingzhe Wu, Xinran Xu, Xinxing Zu, Xinyu Zhou, Xuehai Pan, Y. Charles, Yang Li, Yangyang Hu, Yangyang Liu, Yanru Chen, Yejie Wang, Yibo Liu, Yidao Qin, Yifeng Liu, Ying Yang, Yiping Bao, Yulun Du, Yuxin Wu, Yuzhi Wang, Zaida Zhou, Zhaoji Wang, Zhaowei Li, Zhen Zhu, Zheng Zhang, Zhexu Wang, Zhilin Yang, Zhiqi Huang, Zihao Huang, Ziyao Xu, Zonghan Yang, Zongyu Lin
Abstract: Language model pretraining with next token prediction has proved effective for scaling compute but is limited to the amount of available training data. Scaling reinforcement learning (RL) unlocks a new axis for the continued improvement of artificial intelligence, with the promise that large language models (LLMs) can scale their training data by learning to explore with rewards. However, prior published work has not produced competitive results. In light of this, we report on the training practice of Kimi k1.5, our latest multi-modal LLM trained with RL, including its RL training techniques, multi-modal data recipes, and infrastructure optimization. Long context scaling and improved policy optimization methods are key ingredients of our approach, which establishes a simplistic, effective RL framework without relying on more complex techniques such as Monte Carlo tree search, value functions, and process reward models. Notably, our system achieves state-of-the-art reasoning performance across multiple benchmarks and modalities -- e.g., 77.5 on AIME, 96.2 on MATH 500, 94-th percentile on Codeforces, 74.9 on MathVista -- matching OpenAI's o1. Moreover, we present effective long2short methods that use long-CoT techniques to improve short-CoT models, yielding state-of-the-art short-CoT reasoning results -- e.g., 60.8 on AIME, 94.6 on MATH500, 47.3 on LiveCodeBench -- outperforming existing short-CoT models such as GPT-4o and Claude Sonnet 3.5 by a large margin (up to +550%).
Authors: Anirudh Chari, Suraj Reddy, Aditya Tiwari, Richard Lian, Brian Zhou
Abstract: While large language models (LLMs) have shown promising capabilities as zero-shot planners for embodied agents, their inability to learn from experience and build persistent mental models limits their robustness in complex open-world environments like Minecraft. We introduce MINDSTORES, an experience-augmented planning framework that enables embodied agents to build and leverage mental models through natural interaction with their environment. Drawing inspiration from how humans construct and refine cognitive mental models, our approach extends existing zero-shot LLM planning by maintaining a database of past experiences that informs future planning iterations. The key innovation is representing accumulated experiences as natural language embeddings of (state, task, plan, outcome) tuples, which can then be efficiently retrieved and reasoned over by an LLM planner to generate insights and guide plan refinement for novel states and tasks. Through extensive experiments in the MineDojo environment, a simulation environment for agents in Minecraft that provides low-level controls for Minecraft, we find that MINDSTORES learns and applies its knowledge significantly better than existing memory-based LLM planners while maintaining the flexibility and generalization benefits of zero-shot approaches, representing an important step toward more capable embodied AI systems that can learn continuously through natural experience.
Authors: Jibang Wu, Chenghao Yang, Simon Mahns, Chaoqi Wang, Hao Zhu, Fei Fang, Haifeng Xu
Abstract: This paper develops an agentic framework that employs large language models (LLMs) to automate the generation of persuasive and grounded marketing content, using real estate listing descriptions as our focal application domain. Our method is designed to align the generated content with user preferences while highlighting useful factual attributes. This agent consists of three key modules: (1) Grounding Module, mimicking expert human behavior to predict marketable features; (2) Personalization Module, aligning content with user preferences; (3) Marketing Module, ensuring factual accuracy and the inclusion of localized features. We conduct systematic human-subject experiments in the domain of real estate marketing, with a focus group of potential house buyers. The results demonstrate that marketing descriptions generated by our approach are preferred over those written by human experts by a clear margin while maintaining the same level of factual accuracy. Our findings suggest a promising agentic approach to automate large-scale targeted marketing while ensuring factuality of content generation.
Authors: Maximilian Holsman, Yukun Huang, Bhuwan Dhingra
Abstract: Speculative Decoding (SD) enforces strict distributional equivalence to the target model when accepting candidate tokens. While it maintains the target model's generation quality, this strict equivalence limits the speedup achievable by SD and prevents users from trading deviations from the target distribution in exchange for further inference speed gains. To address these limitations, we introduce Fuzzy Speculative Decoding (FSD) - a decoding algorithm that generalizes SD by accepting candidate tokens based on the divergences between the target and draft model distributions. By allowing for controlled divergence from the target model, FSD enables users to flexibly trade generation quality for inference speed. Across several benchmarks, our method is able to achieve significant runtime improvements of over 5 tokens per second faster than SD at only an approximate 2% absolute reduction in benchmark accuracy. In many cases, FSD is even able to match SD benchmark accuracy at over 2 tokens per second faster, demonstrating that distributional equivalence is not necessary to maintain target model performance. Furthermore, FSD can be seamlessly integrated into existing SD extensions; we demonstrate this by applying FSD to EAGLE-2, greatly enhancing this existing extension's efficiency while allowing it to leverage FSD's tunable quality-speed trade-off.
Authors: Tianyang Zhang, Zhuoxuan Jiang, Haotian Zhang, Lin Lin, Shaohua Zhang
Abstract: We propose a novel system, MathMistake Checker, designed to automate step-by-step mistake finding in mathematical problems with lengthy answers through a two-stage process. The system aims to simplify grading, increase efficiency, and enhance learning experiences from a pedagogical perspective. It integrates advanced technologies, including computer vision and the chain-of-thought capabilities of the latest large language models (LLMs). Our system supports open-ended grading without reference answers and promotes personalized learning by providing targeted feedback. We demonstrate its effectiveness across various types of math problems, such as calculation and word problems.
Authors: Zhaoxin Li, Zhang Xi-Jia, Batuhan Altundas, Letian Chen, Rohan Paleja, Matthew Gombolay
Abstract: Semantic interpretability in Reinforcement Learning (RL) enables transparency and verifiability by making the agent's decisions understandable and verifiable. Achieving this, however, requires a feature space composed of human-understandable concepts, which traditionally rely on human specification and may fail to generalize to unseen environments. We introduce interpretable Tree-based Reinforcement learning via Automated Concept Extraction (iTRACE), an automated framework that leverages pre-trained vision-language models (VLM) for semantic feature extraction and interpretable tree-based models for policy optimization. iTRACE first extracts semantically meaningful features, then maps them to policies via interpretable trees. To address the impracticality of running VLMs in RL loops, we distill their outputs into a lightweight model. By leveraging Vision-Language Models (VLMs) to automate tree-based reinforcement learning, iTRACE eliminates the need for human annotation traditionally required by interpretable models, while also addressing the limitations of VLMs alone, such as their lack of grounding in action spaces and inability to directly optimize policies. iTRACE outperforms MLP baselines that use the same interpretable features and matches the performance of CNN-based policies, producing verifiable, semantically interpretable, and human-aligned behaviors without requiring human annotation.
Authors: Lo Pang-Yun Ting, Yu-Hao Chiang, Yi-Tung Tsai, Hsu-Chao Lai, Kun-Ta Chuang
Abstract: In the semiconductor industry, integrated circuit (IC) processes play a vital role, as the rising complexity and market expectations necessitate improvements in yield. Identifying IC defects and assigning IC testing tasks to the right engineers improves efficiency and reduces losses. While current studies emphasize fault localization or defect classification, they overlook the integration of defect characteristics, historical failures, and the insights from engineer expertise, which restrains their effectiveness in improving IC handling. To leverage AI for these challenges, we propose DeCo, an innovative approach for optimizing task assignment in IC testing. DeCo constructs a novel defect-aware graph from IC testing reports, capturing co-failure relationships to enhance defect differentiation, even with scarce defect data. Additionally, it formulates defect-aware representations for engineers and tasks, reinforced by local and global structure modeling on the defect-aware graph. Finally, a contrasting-based assignment mechanism pairs testing tasks with QA engineers by considering their skill level and current workload, thus promoting an equitable and efficient job dispatch. Experiments on a real-world dataset demonstrate that DeCo achieves the highest task-handling success rates in different scenarios, exceeding 80\%, while also maintaining balanced workloads on both scarce or expanded defect data. Moreover, case studies reveal that DeCo can assign tasks to potentially capable engineers, even for their unfamiliar defects, highlighting its potential as an AI-driven solution for the real-world IC failure analysis and task handling.
Authors: Lo Pang-Yun Ting, Hong-Pei Chen, An-Shan Liu, Chun-Yin Yeh, Po-Lin Chen, Kun-Ta Chuang
Abstract: Early detection of patient deterioration is crucial for reducing mortality rates. Heart rate data has shown promise in assessing patient health, and wearable devices offer a cost-effective solution for real-time monitoring. However, extracting meaningful insights from diverse heart rate data and handling missing values in wearable device data remain key challenges. To address these challenges, we propose TARL, an innovative approach that models the structural relationships of representative subsequences, known as shapelets, in heart rate time series. TARL creates a shapelet-transition knowledge graph to model shapelet dynamics in heart rate time series, indicating illness progression and potential future changes. We further introduce a transition-aware knowledge embedding to reinforce relationships among shapelets and quantify the impact of missing values, enabling the formulation of comprehensive heart rate representations. These representations capture explanatory structures and predict future heart rate trends, aiding early illness detection. We collaborate with physicians and nurses to gather ICU patient heart rate data from wearables and diagnostic metrics assessing illness severity for evaluating deterioration. Experiments on real-world ICU data demonstrate that TARL achieves both high reliability and early detection. A case study further showcases TARL's explainable detection process, highlighting its potential as an AI-driven tool to assist clinicians in recognizing early signs of patient deterioration.
Authors: Brandon Lepine, Gawesha Weerantunga, Juho Kim, Pamela Mishkin, Matthew Beane
Abstract: Current AI benchmarks miss the messy, multi-turn nature of human-AI collaboration. We present an evaluation framework that decomposes real-world tasks into interdependent subtasks, letting us track both LLM performance and users' strategies across a dialogue. Complementing this framework, we develop a suite of metrics, including a composite usage derived from semantic similarity, word overlap, and numerical matches; structural coherence; intra-turn diversity; and a novel measure of the "information frontier" reflecting the alignment between AI outputs and users' working knowledge. We demonstrate our methodology in a financial valuation task that mirrors real-world complexity. Our empirical findings reveal that while greater integration of LLM-generated content generally enhances output quality, its benefits are moderated by factors such as response incoherence, excessive subtask diversity, and the distance of provided information from users' existing knowledge. These results suggest that proactive dialogue strategies designed to inject novelty may inadvertently undermine task performance. Our work thus advances a more holistic evaluation of human-AI collaboration, offering both a robust methodological framework and actionable insights for developing more effective AI-augmented work processes.
Authors: Lee Harris
Abstract: Dates often contribute towards highly impactful medical decisions, but it is rarely clear how to extract this data. AI has only just begun to be used transcribe such documents, and common methods are either to trust that the output produced by a complex AI model, or to parse the text using regular expressions. Recent work has established that regular expressions are an explainable form of logic, but it is difficult to decompose these into the component parts that are required to construct precise UNIX timestamps. First, we test publicly-available regular expressions, and we found that these were unable to capture a significant number of our dates. Next, we manually created easily-decomposable regular expressions, and we found that these were able to detect the majority of real dates, but also a lot of sequences of text that look like dates. Finally, we used regular expression synthesis to automatically identify regular expressions from the reverse-engineered UNIX timestamps that we created. We find that regular expressions created by regular expression synthesis detect far fewer sequences of text that look like dates than those that were manually created, at the cost of a slight increase to the number of missed dates. Overall, our results show that regular expressions can be created through regular expression synthesis to identify complex dates and date ranges in text transcriptions. To our knowledge, our proposed way of learning deterministic logic by reverse-engineering several many-one mappings and feeding these into a regular expression synthesiser is a new approach.
Authors: Junyang Wang, Haiyang Xu, Xi Zhang, Ming Yan, Ji Zhang, Fei Huang, Jitao Sang
Abstract: The exponential rise in mobile device usage necessitates streamlined automation for effective task management, yet many AI frameworks fall short due to inadequate operational expertise. While manually written knowledge can bridge this gap, it is often burdensome and inefficient. We introduce Mobile-Agent-V, an innovative framework that utilizes video as a guiding tool to effortlessly and efficiently inject operational knowledge into mobile automation processes. By deriving knowledge directly from video content, Mobile-Agent-V eliminates manual intervention, significantly reducing the effort and time required for knowledge acquisition. To rigorously evaluate this approach, we propose Mobile-Knowledge, a benchmark tailored to assess the impact of external knowledge on mobile agent performance. Our experimental findings demonstrate that Mobile-Agent-V enhances performance by 36% compared to existing methods, underscoring its effortless and efficient advantages in mobile automation.
Authors: Lanxiang Hu, Mingjia Huo, Yuxuan Zhang, Haoyang Yu, Eric P. Xing, Ion Stoica, Tajana Rosing, Haojian Jin, Hao Zhang
Abstract: Playing video games requires perception, memory, and planning, exactly the faculties modern large language model (LLM) agents are expected to master. We study the major challenges in using popular video games to evaluate modern LLMs and find that directly dropping LLMs into games cannot make an effective evaluation, for three reasons -- brittle vision perception, prompt sensitivity, and potential data contamination. We introduce lmgame-Bench to turn games into reliable evaluations. lmgame-Bench features a suite of platformer, puzzle, and narrative games delivered through a unified Gym-style API and paired with lightweight perception and memory scaffolds, and is designed to stabilize prompt variance and remove contamination. Across 13 leading models, we show lmgame-Bench is challenging while still separating models well. Correlation analysis shows that every game probes a unique blend of capabilities often tested in isolation elsewhere. More interestingly, performing reinforcement learning on a single game from lmgame-Bench transfers both to unseen games and to external planning tasks. Our evaluation code is available at https://github.com/lmgame-org/GamingAgent/lmgame-bench.
URLs: https://github.com/lmgame-org/GamingAgent/lmgame-bench.
Authors: Jialiang Sun, Yuzhi Tang, Ao Li, Chris J. Maddison, Kuldeep S. Meel
Abstract: Mathematical reasoning lies at the heart of artificial intelligence, underpinning applications in education, program verification, and research-level mathematical discovery. Mathematical competitions, in particular, present two challenging problem types: theorem proving, which requires rigorous proofs of stated conclusions, and answer construction, which involves hypothesizing and formally verifying mathematical objects. Large Language Models (LLMs) effectively generate creative candidate answers but struggle with formal verification, while symbolic provers ensure rigor but cannot efficiently handle creative conjecture generation. We introduce the Enumerate-Conjecture-Prove (ECP) framework, a modular neuro-symbolic method integrating LLM-based enumeration and pattern-driven conjecturing with formal theorem proving. We present ConstructiveBench, a dataset of 3,431 answer-construction problems in various math competitions with verified Lean formalizations. On the ConstructiveBench dataset, ECP improves the accuracy of answer construction from a Chain-of-Thought (CoT) baseline of 14.54% to 45.06% with the gpt-4.1-mini model. Moreover, combined with ECP's constructed answers, the state-of-the-art DeepSeek-Prover-V2-7B model generates correct proofs for 858 of the 3,431 constructive problems in Lean, achieving 25.01% accuracy compared to 9.86% for symbolic-only baselines. Our code and dataset are publicly available at https://github.com/JackSun200312/ECP.
Authors: Anqing Jiang, Yu Gao, Zhigang Sun, Yiru Wang, Jijun Wang, Jinghao Chai, Qian Cao, Yuweng Heng, Hao Jiang, Yunda Dong, Zongzheng Zhang, Xianda Guo, Hao Sun, Hao Zhao
Abstract: Research interest in end-to-end autonomous driving has surged owing to its fully differentiable design integrating modular tasks, i.e. perception, prediction and planing, which enables optimization in pursuit of the ultimate goal. Despite the great potential of the end-to-end paradigm, existing methods suffer from several aspects including expensive BEV (bird's eye view) computation, action diversity, and sub-optimal decision in complex real-world scenarios. To address these challenges, we propose a novel hybrid sparse-dense diffusion policy, empowered by a Vision-Language Model (VLM), called Diff-VLA. We explore the sparse diffusion representation for efficient multi-modal driving behavior. Moreover, we rethink the effectiveness of VLM driving decision and improve the trajectory generation guidance through deep interaction across agent, map instances and VLM output. Our method shows superior performance in Autonomous Grand Challenge 2025 which contains challenging real and reactive synthetic scenarios. Our methods achieves 45.0 PDMS.
Authors: Ming Yin, Yuanhao Qu, Ling Yang, Le Cong, Mengdi Wang
Abstract: We investigate how to teach large language models (LLMs) to perform scientific reasoning by leveraging expert discussions as a learning signal. Focusing on the genomics domain, we develop an automated pipeline to extract trainable data and introduce Genome-Bench, a new benchmark constructed from over a decade of scientific forum discussions on genome engineering. Our pipeline transforms raw interactions into a reinforcement learning-friendly multiple-choice questions format, supported by 3000+ high-quality question-answer pairs spanning foundational biology, experimental troubleshooting, tool usage, and beyond. We fine-tune an LLM using RL with a rule-based reward signal derived from the synthetic MCQ dataset to enhance domain-specific reasoning. Our results show that reinforcement learning from scientific discussions improves model performance by over 15% compared to the base model on Genome-Bench, narrowing the gap between open-source LLMs and expert-level reasoning. To our knowledge, this is the first end-to-end pipeline for teaching LLMs to reason from scientific discussions, with promising potential for generalization across scientific domains beyond biology.
Authors: Linqiang Guo (Peter), Wei Liu (Peter), Yi Wen Heng (Peter), Tse-Hsun (Peter), Chen, Yang Wang
Abstract: Mobile GUI agents aim to autonomously complete user-instructed tasks across mobile apps. Recent advances in Multimodal Large Language Models (MLLMs) enable these agents to interpret UI screens, identify actionable elements, and perform interactions such as tapping or typing. However, existing agents remain reactive: they reason only over the current screen and lack a structured model of app navigation flow, limiting their ability to understand context, detect unexpected outcomes, and recover from errors. We present MAPLE, a state-aware multi-agent framework that abstracts app interactions as a Finite State Machine (FSM). We computationally model each UI screen as a discrete state and user actions as transitions, allowing the FSM to provide a structured representation of the app execution. MAPLE consists of specialized agents responsible for four phases of task execution: planning, execution, verification, error recovery, and knowledge retention. These agents collaborate to dynamically construct FSMs in real time based on perception data extracted from the UI screen, allowing the GUI agents to track navigation progress and flow, validate action outcomes through pre- and post-conditions of the states, and recover from errors by rolling back to previously stable states. Our evaluation results on two challenging cross-app benchmarks, Mobile-Eval-E and SPA-Bench, show that MAPLE outperforms the state-of-the-art baseline, improving task success rate by up to 12%, recovery success by 13.8%, and action accuracy by 6.5%. Our results highlight the importance of structured state modeling in guiding mobile GUI agents during task execution. Moreover, our FSM representation can be integrated into future GUI agent architectures as a lightweight, model-agnostic memory layer to support structured planning, execution verification, and error recovery.
Authors: Ya Wen, Jixuan Cai, Qiyao Ma, Linyan Li, Xinhua Chen, Chris Webster, Yulun Zhou
Abstract: Representation learning of geospatial locations remains a core challenge in achieving general geospatial intelligence. Current embedding methods often lack versatility, limiting their utility across diverse tasks in both human and natural domains. We present MobCLIP, the first nationwide general-purpose location encoder, integrating an unprecedented diversity of data modalities through effective and scalable multimodal fusion. Adopting a novel CLIP-based architecture, our framework aligns 100M+ POIs, nationwide remote sensing imagery, and structured demographic statistics with a billion-edge mobility graph. By tokenizing spatial locations into grid cells inspired by Vision Transformers, we establish a unified representation space bridging mobility patterns and multimodal features. To rigorously evaluate the general-purpose effectiveness of MobCLIP, we construct a benchmark dataset composed of 11 downstream prediction tasks across social, economic, and natural domains. Experiments show that MobCLIP, with four input modalities and a compact 128-dimensional representation space, achieves significantly superior general-purpose predictive performances than state-of-the-art models by an average of 35%. Thanks to the effective integration of human-centric modalities, the performance gain is particularly profound in human-centric tasks, such as energy consumption (+260%), offline retail consumption amount (+98%), and crime cases (+95%) predictions. Echoing LLM scaling laws, we further demonstrate the scaling behavior in geospatial representation learning. We open-source code and pretrained models at: https://github.com/ylzhouchris/MobCLIP.
Authors: Jisu Shin, Seunghyun Shin, Hae-Gon Jeon
Abstract: Understanding the informative structures of scenes is essential for low-level vision tasks. Unfortunately, it is difficult to obtain a concrete visual definition of the informative structures because influences of visual features are task-specific. In this paper, we propose a single general neural network architecture for extracting task-specific structure guidance for scenes. To do this, we first analyze traditional spectral clustering methods, which computes a set of eigenvectors to model a segmented graph forming small compact structures on image domains. We then unfold the traditional graph-partitioning problem into a learnable network, named \textit{Scene Structure Guidance Network (SSGNet)}, to represent the task-specific informative structures. The SSGNet yields a set of coefficients of eigenvectors that produces explicit feature representations of image structures. In addition, our SSGNet is light-weight ($\sim$ 56K parameters), and can be used as a plug-and-play module for off-the-shelf architectures. We optimize the SSGNet without any supervision by proposing two novel training losses that enforce task-specific scene structure generation during training. Our main contribution is to show that such a simple network can achieve state-of-the-art results for several low-level vision applications. We also demonstrate that our network generalizes well on unseen datasets, compared to existing methods which use structural embedding frameworks. We further propose a lighter version of SSGNet ($\sim$ 29K parameters) for depth computation, SSGNet-D, and successfully execute it on edge computing devices like Jetson AGX Orin, improving the performance of baseline network, even in the wild, with little computational delay.
Authors: Yehya Farhat, Hamza ElMokhtar Shili, Fangshuo Liao, Chen Dun, Mirian Hipolito Garcia, Guoqing Zheng, Ahmed Hassan Awadallah, Robert Sim, Dimitrios Dimitriadis, Anastasios Kyrillidis
Abstract: Mixture-of-Experts (MoEs) achieve scalability by dynamically activating subsets of their components. Yet, understanding how expertise emerges through joint training of gating mechanisms and experts remains incomplete, especially in scenarios without clear task partitions. Motivated by inference costs and data heterogeneity, we study how joint training of gating functions and experts can dynamically allocate domain-specific expertise across multiple underlying data distributions. As an outcome of our framework, we develop an instance tailored specifically to decentralized training scenarios, introducing \textit{Dynamically Decentralized Orchestration of MoEs} or \texttt{DDOME}. \texttt{DDOME} leverages heterogeneity emerging from distributional shifts across decentralized data sources to specialize experts dynamically. By integrating a pretrained common expert to inform a gating function, \texttt{DDOME} achieves personalized expert subset selection on-the-fly, facilitating just-in-time personalization. We empirically validate \texttt{DDOME} within a Federated Learning (FL) context: \texttt{DDOME} attains from 4\% up to an 24\% accuracy improvement over state-of-the-art FL baselines in image and text classification tasks, while maintaining competitive zero-shot generalization capabilities. Furthermore, we provide theoretical insights confirming that the joint gating-experts training is critical for achieving meaningful expert specialization.
Authors: Zhihao Shi, Jie Wang, Fanghua Lu, Hanzhu Chen, Defu Lian, Zheng Wang, Jieping Ye, Feng Wu
Abstract: Node representation learning on attributed graphs -- whose nodes are associated with rich attributes (e.g., texts and protein sequences) -- plays a crucial role in many important downstream tasks. To encode the attributes and graph structures simultaneously, recent studies integrate pre-trained models with graph neural networks (GNNs), where pre-trained models serve as node encoders (NEs) to encode the attributes. As jointly training large NEs and GNNs on large-scale graphs suffers from severe scalability issues, many methods propose to train NEs and GNNs separately. Consequently, they do not take feature convolutions in GNNs into consideration in the training phase of NEs, leading to a significant learning bias relative to the joint training. To address this challenge, we propose an efficient label regularization technique, namely Label Deconvolution (LD), to alleviate the learning bias by a novel and highly scalable approximation to the inverse mapping of GNNs. The inverse mapping leads to an objective function that is equivalent to that by the joint training, while it can effectively incorporate GNNs in the training phase of NEs against the learning bias. More importantly, we show that LD converges to the optimal objective function values by the joint training under mild assumptions. Experiments demonstrate LD significantly outperforms state-of-the-art methods on Open Graph Benchmark datasets.
Authors: Zhiyuan Zhao, Haoxin Liu, Alexander Rodriguez, B. Aditya Prakash
Abstract: Time-series forecasting is a critical challenge in various domains and has witnessed substantial progress in recent years. Many real-life scenarios, such as public health, economics, and social applications, involve feedback loops where predictions can influence the predicted outcome, subsequently altering the target variable's distribution. This phenomenon, known as performativity, introduces the potential for 'self-negating' or 'self-fulfilling' predictions. Despite extensive studies in classification problems across domains, performativity remains largely unexplored in the context of time-series forecasting from a machine-learning perspective. In this paper, we formalize performative time-series forecasting (PeTS), addressing the challenge of accurate predictions when performativity-induced distribution shifts are possible. We propose a novel approach, Feature Performative-Shifting (FPS), which leverages the concept of delayed response to anticipate distribution shifts and subsequently predicts targets accordingly. We provide theoretical insights suggesting that FPS can potentially lead to reduced generalization error. We conduct comprehensive experiments using multiple time-series models on COVID-19 and traffic forecasting tasks. The results demonstrate that FPS consistently outperforms conventional time-series forecasting methods, highlighting its efficacy in handling performativity-induced challenges.
Authors: Daniel Jarne Ornia, Giannis Delimpaltadakis, Jens Kober, Javier Alonso-Mora
Abstract: In Reinforcement Learning (RL), agents have no incentive to exhibit predictable behaviors, and are often pushed (through e.g. policy entropy regularisation) to randomise their actions in favor of exploration. This often makes it challenging for other agents and humans to predict an agent's behavior, triggering unsafe scenarios (e.g. in human-robot interaction). We propose a novel method to induce predictable behavior in RL agents, termed Predictability-Aware RL (PARL), employing the agent's trajectory entropy rate to quantify predictability. Our method maximizes a linear combination of a standard discounted reward and the negative entropy rate, thus trading off optimality with predictability. We show how the entropy rate can be formally cast as an average reward, how entropy-rate value functions can be estimated from a learned model and incorporate this in policy-gradient algorithms, and demonstrate how this approach produces predictable (near-optimal) policies in tasks inspired by human-robot use-cases.
Authors: Yixin Wan, Kai-Wei Chang
Abstract: Recent large-scale T2I models like DALLE-3 have made progress in reducing gender stereotypes when generating single-person images. However, significant biases remain when generating images with more than one person. To systematically evaluate this, we propose the Paired Stereotype Test (PST) framework, which queries T2I models to depict two individuals assigned with male-stereotyped and female-stereotyped social identities, respectively (e.g. "a CEO" and "an Assistant"). This contrastive setting often triggers T2I models to generate gender-stereotyped images. Using PST, we evaluate two aspects of gender biases -- the well-known bias in gendered occupation and a novel aspect: bias in organizational power. Experiments show that over 74\% images generated by DALLE-3 display gender-occupational biases. Additionally, compared to single-person settings, DALLE-3 is more likely to perpetuate male-associated stereotypes under PST. We further propose FairCritic, a novel and interpretable framework that leverages an LLM-based critic model to i) detect bias in generated images, and ii) adaptively provide feedback to T2I models for improving fairness. FairCritic achieves near-perfect fairness on PST, overcoming the limitations of previous prompt-based intervention approaches.
Authors: Naman Shah, Jayesh Nagpal, Siddharth Srivastava
Abstract: Humans efficiently generalize from limited demonstrations, but robots still struggle to transfer learned knowledge to complex, unseen tasks with longer horizons and increased complexity. We propose the first known method enabling robots to autonomously invent relational concepts directly from small sets of unannotated, unsegmented demonstrations. The learned symbolic concepts are grounded into logic-based world models, facilitating efficient zero-shot generalization to significantly more complex tasks. Empirical results demonstrate that our approach achieves performance comparable to hand-crafted models, successfully scaling execution horizons and handling up to 18 times more objects than seen in training, providing the first autonomous framework for learning transferable symbolic abstractions from raw robot trajectories.
Authors: Mariano V. Ntrougkas, Nikolaos Gkalelis, Vasileios Mezaris
Abstract: The development and adoption of Vision Transformers and other deep-learning architectures for image classification tasks has been rapid. However, the "black box" nature of neural networks is a barrier to adoption in applications where explainability is essential. While some techniques for generating explanations have been proposed, primarily for Convolutional Neural Networks, adapting such techniques to the new paradigm of Vision Transformers is non-trivial. This paper presents T-TAME, Transformer-compatible Trainable Attention Mechanism for Explanations, a general methodology for explaining deep neural networks used in image classification tasks. The proposed architecture and training technique can be easily applied to any convolutional or Vision Transformer-like neural network, using a streamlined training approach. After training, explanation maps can be computed in a single forward pass; these explanation maps are comparable to or outperform the outputs of computationally expensive perturbation-based explainability techniques, achieving SOTA performance. We apply T-TAME to three popular deep learning classifier architectures, VGG-16, ResNet-50, and ViT-B-16, trained on the ImageNet dataset, and we demonstrate improvements over existing state-of-the-art explainability methods. A detailed analysis of the results and an ablation study provide insights into how the T-TAME design choices affect the quality of the generated explanation maps.
Authors: Yanglin Feng, Yang Qin, Dezhong Peng, Hongyuan Zhu, Xi Peng, Peng Hu
Abstract: In this paper, we present and study a new instance-level retrieval task: PointCloud-Text Matching (PTM), which aims to identify the exact cross-modal instance that matches a given point-cloud query or text query. PTM has potential applications in various scenarios, such as indoor/urban-canyon localization and scene retrieval. However, there is a lack of suitable and targeted datasets for PTM in practice. To address this issue, we present a new PTM benchmark dataset, namely SceneDepict-3D2T. We observe that the data poses significant challenges due to its inherent characteristics, such as the sparsity, noise, or disorder of point clouds and the ambiguity, vagueness, or incompleteness of texts, which render existing cross-modal matching methods ineffective for PTM. To overcome these challenges, we propose a PTM baseline, named Robust PointCloud-Text Matching method (RoMa). RoMa consists of two key modules: a Dual Attention Perception module (DAP) and a Robust Negative Contrastive Learning module (RNCL). Specifically, DAP leverages token-level and feature-level attention mechanisms to adaptively focus on useful local and global features, and aggregate them into common representations, thereby reducing the adverse impact of noise and ambiguity. To handle noisy correspondence, RNCL enhances robustness against mismatching by dividing negative pairs into clean and noisy subsets and assigning them forward and reverse optimization directions, respectively. We conduct extensive experiments on our benchmarks and demonstrate the superiority of our RoMa.
Authors: Dongjing Shan, guiqiang chen
Abstract: Vision Transformers (ViTs) have achieved impressive results in large-scale image classification. However, when training from scratch on small datasets, there is still a significant performance gap between ViTs and Convolutional Neural Networks (CNNs), which is attributed to the lack of inductive bias. To address this issue, we propose a Graph-based Vision Transformer (GvT) that utilizes graph convolutional projection and graph-pooling. In each block, queries and keys are calculated through graph convolutional projection based on the spatial adjacency matrix, while dot-product attention is used in another graph convolution to generate values. When using more attention heads, the queries and keys become lower-dimensional, making their dot product an uninformative matching function. To overcome this low-rank bottleneck in attention heads, we employ talking-heads technology based on bilinear pooled features and sparse selection of attention tensors. This allows interaction among filtered attention scores and enables each attention mechanism to depend on all queries and keys. Additionally, we apply graph-pooling between two intermediate blocks to reduce the number of tokens and aggregate semantic information more effectively. Our experimental results show that GvT produces comparable or superior outcomes to deep convolutional networks and surpasses vision transformers without pre-training on large datasets. The code for our proposed model is publicly available on the website.
Authors: Anselm Paulus, Arman Zharmagambetov, Chuan Guo, Brandon Amos, Yuandong Tian
Abstract: Large Language Models (LLMs) are vulnerable to jailbreaking attacks that lead to generation of inappropriate or harmful content. Manual red-teaming requires a time-consuming search for adversarial prompts, whereas automatic adversarial prompt generation often leads to semantically meaningless attacks that do not scale well. In this paper, we present a novel method that uses another LLM, called AdvPrompter, to generate human-readable adversarial prompts in seconds. AdvPrompter, which is trained using an alternating optimization algorithm, generates suffixes that veil the input instruction without changing its meaning, such that the TargetLLM is lured to give a harmful response. Experimental results on popular open source TargetLLMs show highly competitive results on the AdvBench and HarmBench datasets, that also transfer to closed-source black-box LLMs. We also show that training on adversarial suffixes generated by AdvPrompter is a promising strategy for improving the robustness of LLMs to jailbreaking attacks.
Authors: Zhuoxuan Jiang, Haoyuan Peng, Shanshan Feng, Fan Li, Dongsheng Li
Abstract: Self-correction is emerging as a promising approach to mitigate the issue of hallucination in Large Language Models (LLMs). To facilitate effective self-correction, recent research has proposed mistake detection as its initial step. However, current literature suggests that LLMs often struggle with reliably identifying reasoning mistakes when using simplistic prompting strategies. To address this challenge, we introduce a unique prompting strategy, termed the Pedagogical Chain-of-Thought (PedCoT), which is specifically designed to guide the identification of reasoning mistakes, particularly mathematical reasoning mistakes. PedCoT consists of pedagogical principles for prompts (PPP) design, two-stage interaction process (TIP) and grounded PedCoT prompts, all inspired by the educational theory of the Bloom Cognitive Model (BCM). We evaluate our approach on two public datasets featuring math problems of varying difficulty levels. The experiments demonstrate that our zero-shot prompting strategy significantly outperforms strong baselines. The proposed method can achieve the goal of reliable mathematical mistake identification and provide a foundation for automatic math answer grading. The results underscore the significance of educational theory, serving as domain knowledge, in guiding prompting strategy design for addressing challenging tasks with LLMs effectively.
Authors: Xuesong Wang, He Zhao, Edwin V. Bonilla
Abstract: Neural Processes (NPs) are deep probabilistic models that represent stochastic processes by conditioning their prior distributions on a set of context points. Despite their advantages in uncertainty estimation for complex distributions, NPs enforce parameterization coupling between the conditional prior model and the posterior model. We show that this coupling amounts to prior misspecification and revisit the NP objective to address this issue. More specifically, we propose R\'enyi Neural Processes (RNP), a method that replaces the standard KL divergence with the R\'enyi divergence, dampening the effects of the misspecified prior during posterior updates. We validate our approach across multiple benchmarks including regression and image inpainting tasks, and show significant performance improvements of RNPs in real-world problems. Our extensive experiments show consistently better log-likelihoods over state-of-the-art NP models.
Authors: Devichand Budagam, Azamat Zhanatuly Imanbayev, Iskander Rafailovich Akhmetov, Aleksandr Sinitca, Sergey Antonov, Dmitrii Kaplun
Abstract: Teeth segmentation and recognition play a vital role in a variety of dental applications and diagnostic procedures. The integration of deep learning models has facilitated the development of precise and automated segmentation methods. Although prior research has explored teeth segmentation, not many methods have successfully performed tooth segmentation and detection simultaneously. This study presents UFBA-425, a dental dataset derived from the UFBA-UESC dataset, featuring bounding box and polygon annotations for 425 panoramic dental X-rays. Additionally, this work introduces OralBBNet, an architecture featuring distinct segmentation and detection heads as U-Net and YOLOv8, respectively. OralBBNet is designed to improve the accuracy and robustness of tooth classification and segmentation on panoramic X-rays by leveraging the complementary strengths of U-Net and YOLOv8. Our approach achieved a 1-3% improvement in mean average precision (mAP) for teeth detection compared to existing techniques and a 15-20% improvement in the dice score for teeth segmentation over U-Net over various tooth categories and 2-4% improvement in the dice score when compared with other segmentation architectures. The results of this study establish a foundation for the wider implementation of object detection models in dental diagnostics.
Authors: Shester Gueuwou, Xiaodan Du, Greg Shakhnarovich, Karen Livescu
Abstract: A persistent challenge in sign language video processing, including the task of sign to written language translation, is how we learn representations of sign language in an effective and efficient way that preserves the important attributes of these languages, while remaining invariant to irrelevant visual differences. Informed by the nature and linguistics of signed languages, our proposed method focuses on just the most relevant parts in a signing video: the face, hands and body pose of the signer. However, instead of fully relying on pose estimation from off-the-shelf pose tracking models, which have inconsistent performance for hands and faces, we propose to learn a representation of the complex handshapes and facial expressions of sign languages in a self-supervised fashion. Our approach is based on learning from individual frames (rather than video sequences) and is therefore much more efficient than prior work on sign language pre-training. Compared to a recent model that established a new state of the art in sign language translation on the How2Sign dataset, our approach yields similar translation performance, using less than 3\% of the compute.
Authors: Zixiang Xian, Chenhui Cui, Rubing Huang, Chunrong Fang, Zhenyu Chen
Abstract: The advent of large language models (LLMs) has significantly advanced artificial intelligence (AI) in software engineering (SE), with source code embeddings playing a crucial role in tasks such as source code clone detection and source code clustering. However, existing methods for source code embedding, including those based on LLMs, often rely on costly supervised training or fine-tuning for domain adaptation. This paper proposes a novel approach to embedding source code by combining large language and sentence embedding models. This approach attempts to eliminate the need for task-specific training or fine-tuning and to effectively address the issue of erroneous information commonly found in LLM-generated outputs. To evaluate the performance of our proposed approach, we conducted a series of experiments on three datasets with different programming languages by considering various LLMs and sentence embedding models. The experimental results have demonstrated the effectiveness and superiority of our approach over the state-of-the-art unsupervised approaches, such as SourcererCC, Code2vec, InferCode, TransformCode, and LLM2Vec. Our findings highlight the potential of our approach to advance the field of SE by providing robust and efficient solutions for source code embedding tasks.
Authors: Francesco Verdini, Pierfrancesco Melucci, Stefano Perna, Francesco Cariaggi, Marco Gaido, Sara Papi, Szymon Mazurek, Marek Kasztelnik, Luisa Bentivogli, S\'ebastien Brati\`eres, Paolo Merialdo, Simone Scardapane
Abstract: The remarkable performance achieved by Large Language Models (LLM) has driven research efforts to leverage them for a wide range of tasks and input modalities. In speech-to-text (S2T) tasks, the emerging solution consists of projecting the output of the encoder of a Speech Foundational Model (SFM) into the LLM embedding space through an adapter module. However, no work has yet investigated how much the downstream-task performance depends on each component (SFM, adapter, LLM) nor whether the best design of the adapter depends on the chosen SFM and LLM. To fill this gap, we evaluate the combination of 5 adapter modules, 2 LLMs (Mistral and Llama), and 2 SFMs (Whisper and SeamlessM4T) on two widespread S2T tasks, namely Automatic Speech Recognition and Speech Translation. Our results demonstrate that the SFM plays a pivotal role in downstream performance, while the adapter choice has moderate impact and depends on the SFM and LLM.
Authors: Yu Ying Chiu, Liwei Jiang, Bill Yuchen Lin, Chan Young Park, Shuyue Stella Li, Sahithya Ravi, Mehar Bhatia, Maria Antoniak, Yulia Tsvetkov, Vered Shwartz, Yejin Choi
Abstract: Robust, diverse, and challenging cultural knowledge benchmarks are essential for measuring our progress towards making LMs that are helpful across diverse cultures. We introduce CulturalBench: a set of 1,696 human-written and human-verified questions to assess LMs' cultural knowledge, covering 45 global regions including underrepresented ones like Bangladesh, Zimbabwe, and Peru. Questions are each verified by five independent annotators and span 17 diverse topics ranging from food preferences to greeting etiquette. We construct CulturalBench using methods inspired by Human-AI Red-Teaming. Compared to human performance (92.4% accuracy), the hard version of CulturalBench is challenging even for the best-performing frontier LMs, ranging from 28.7% to 61.5% in accuracy. We find that LMs often struggle with tricky questions that have multiple correct answers (e.g., What utensils do the Chinese usually use?), revealing a tendency to overfit to a single answer. Our results indicate that GPT-4o substantially outperform other models across cultures, besting local providers (e.g., Mistral on European culture and DeepSeek on Chinese culture). Across the board, models under-perform on questions related to North Africa, South America and Middle East.
Authors: Wenxuan Wang, Kuiyi Gao, Youliang Yuan, Jen-tse Huang, Qiuzhi Liu, Shuai Wang, Wenxiang Jiao, Zhaopeng Tu
Abstract: Text-based image generation models, such as Stable Diffusion and DALL-E 3, hold significant potential in content creation and publishing workflows, making them the focus in recent years. Despite their remarkable capability to generate diverse and vivid images, considerable efforts are being made to prevent the generation of harmful content, such as abusive, violent, or pornographic material. To assess the safety of existing models, we introduce a novel jailbreaking method called Chain-of-Jailbreak (CoJ) attack, which compromises image generation models through a step-by-step editing process. Specifically, for malicious queries that cannot bypass the safeguards with a single prompt, we intentionally decompose the query into multiple sub-queries. The image generation models are then prompted to generate and iteratively edit images based on these sub-queries. To evaluate the effectiveness of our CoJ attack method, we constructed a comprehensive dataset, CoJ-Bench, encompassing nine safety scenarios, three types of editing operations, and three editing elements. Experiments on four widely-used image generation services provided by GPT-4V, GPT-4o, Gemini 1.5 and Gemini 1.5 Pro, demonstrate that our CoJ attack method can successfully bypass the safeguards of models for over 60% cases, which significantly outperforms other jailbreaking methods (i.e., 14%). Further, to enhance these models' safety against our CoJ attack method, we also propose an effective prompting-based method, Think Twice Prompting, that can successfully defend over 95% of CoJ attack. We release our dataset and code to facilitate the AI safety research.
Authors: Jianqing Zhang, Yang Liu, Yang Hua, Jian Cao, Qiang Yang
Abstract: Model heterogeneity poses a significant challenge in Heterogeneous Federated Learning (HtFL). In scenarios with diverse model architectures, directly aggregating model parameters is impractical, leading HtFL methods to incorporate an extra objective alongside the original local objective on each client to facilitate collaboration. However, this often results in a mismatch between the extra and local objectives. To resolve this, we propose Federated Learning-to-Guide (FedL2G), a method that adaptively learns to guide local training in a federated manner, ensuring the added objective aligns with each client's original goal. With theoretical guarantees, FedL2G utilizes only first-order derivatives w.r.t. model parameters, achieving a non-convex convergence rate of O(1/T). We conduct extensive experiments across two data heterogeneity and six model heterogeneity settings, using 14 heterogeneous model architectures (e.g., CNNs and ViTs). The results show that FedL2G significantly outperforms seven state-of-the-art methods.
Authors: Youngjae Min, Navid Azizan
Abstract: Incorporating prior knowledge or specifications of input-output relationships into machine learning models has attracted significant attention, as it enhances generalization from limited data and leads to conforming outputs. However, most existing approaches use soft constraints by penalizing violations through regularization, which offers no guarantee of constraint satisfaction, especially on inputs far from the training distribution -- an essential requirement in safety-critical applications. On the other hand, imposing hard constraints on neural networks may hinder their representational power, adversely affecting performance. To address this, we propose HardNet, a practical framework for constructing neural networks that inherently satisfy hard constraints without sacrificing model capacity. Unlike approaches that modify outputs only at inference time, HardNet enables end-to-end training with hard constraint guarantees, leading to improved performance. To the best of our knowledge, HardNet is the first method with an efficient forward pass to enforce more than one input-dependent inequality constraint. It allows unconstrained optimization of the network parameters using standard algorithms by appending a differentiable closed-form enforcement layer to the network's output. Furthermore, we show that HardNet is expressive and retains the universal approximation capabilities of neural networks. We demonstrate the versatility and effectiveness of HardNet across various applications: learning with piecewise constraints, learning optimization solvers with guaranteed feasibility, and optimizing control policies in safety-critical systems.
Authors: Bokai Hu, Sai Ashish Somayajula, Xin Pan, Pengtao Xie
Abstract: Instruction-fine-tuned large language models (LLMs) under 14B parameters continue to underperform on natural language understanding (NLU) tasks, often trailing smaller models like BERT-base on benchmarks such as GLUE and SuperGLUE. Motivated by the success of reinforcement learning in reasoning tasks (e.g., DeepSeek), we explore Proximal Policy Optimization (PPO) as a framework to improve the NLU capabilities of LLMs. We frame NLU as a reinforcement learning environment, treating token generation as a sequence of actions and optimizing for reward signals based on alignment with ground-truth labels. PPO consistently outperforms supervised fine-tuning, yielding an average improvement of 6.3 points on GLUE, and surpasses zero-shot and few-shot prompting by 38.7 and 26.1 points, respectively. Notably, PPO-tuned models outperform GPT-4o by over 4\% on average across sentiment and natural language inference tasks, including gains of 7.3\% on the Mental Health dataset and 10.9\% on SIGA-nli. This work highlights a promising direction for adapting LLMs to new tasks by reframing them as reinforcement learning problems, enabling learning through simple end-task rewards rather than extensive data curation.
Authors: Leshem Choshen, Yang Zhang, Jacob Andreas
Abstract: Scaling laws predict the loss of a target machine learning model by extrapolating from easier-to-train models with fewer parameters or smaller training sets. This provides an efficient way for practitioners and researchers alike to compare pretraining decisions involving optimizers, datasets, and model architectures. Despite the widespread use of scaling laws to model the dynamics of language model training, there has been little work on understanding how to best estimate and interpret them. We collect (and release) a large-scale dataset containing losses and downstream evaluations for 485 previously published pretrained models. We use these to estimate more than 1000 scaling laws, then derive a set of best practices for estimating scaling laws in new model families. We find that fitting scaling laws to intermediate checkpoints of training runs (and not just their final losses) substantially improves accuracy, and that -- all else equal -- estimates of performance are generally most accurate when derived from other models of similar sizes. However, because there is a significant degree of variability across model seeds, training multiple small models is sometimes more useful than training a single large one. Moreover, while different model families differ scaling behavior, they are often similar enough that a target model's behavior can be predicted from a single model with the same architecture, along with scaling parameter estimates derived from other model families.
Authors: Eric Elmoznino, Tom Marty, Tejas Kasetty, Leo Gagnon, Sarthak Mittal, Mahan Fathi, Dhanya Sridhar, Guillaume Lajoie
Abstract: A central goal of machine learning is generalization. While the No Free Lunch Theorem states that we cannot obtain theoretical guarantees for generalization without further assumptions, in practice we observe that simple models which explain the training data generalize best: a principle called Occam's razor. Despite the need for simple models, most current approaches in machine learning only minimize the training error, and at best indirectly promote simplicity through regularization or architecture design. Here, we draw a connection between Occam's razor and in-context learning: an emergent ability of certain sequence models like Transformers to learn at inference time from past observations in a sequence. In particular, we show that the next-token prediction loss used to train in-context learners is directly equivalent to a data compression technique called prequential coding, and that minimizing this loss amounts to jointly minimizing both the training error and the complexity of the model that was implicitly learned from context. Our theory and the empirical experiments we use to support it not only provide a normative account of in-context learning, but also elucidate the shortcomings of current in-context learning methods, suggesting ways in which they can be improved. We make our code available at https://github.com/3rdCore/PrequentialCode.
Authors: Eric Elmoznino, Thomas Jiralerspong, Yoshua Bengio, Guillaume Lajoie
Abstract: Compositionality is believed to be fundamental to intelligence. In humans, it underlies the structure of thought, language, and higher-level reasoning. In AI, compositional representations can enable a powerful form of out-of-distribution generalization, in which a model systematically adapts to novel combinations of known concepts. However, while we have strong intuitions about what compositionality is, we lack satisfying formal definitions for it that are measurable and mathematical. Here, we propose such a definition, which we call representational compositionality, that accounts for and extends our intuitions about compositionality. The definition is conceptually simple, quantitative, grounded in algorithmic information theory, and applicable to any representation. Intuitively, representational compositionality states that a compositional representation satisfies three properties. First, it must be expressive. Second, it must be possible to re-describe the representation as a function of discrete symbolic sequences with re-combinable parts, analogous to sentences in natural language. Third, the function that relates these symbolic sequences to the representation, analogous to semantics in natural language, must be simple. Through experiments on both synthetic and real world data, we validate our definition of compositionality and show how it unifies disparate intuitions from across the literature in both AI and cognitive science. We also show that representational compositionality, while theoretically intractable, can be readily estimated using standard deep learning tools. We hope that our definition can inspire the design of novel, theoretically-driven models that better capture the mechanisms of compositional thought. We make our code available at https://github.com/EricElmoznino/complexity_compositionality.
URLs: https://github.com/EricElmoznino/complexity_compositionality.
Authors: Kang Zhao, Tao Yuan, Han Bao, Zhenfeng Su, Chang Gao, Zhaofeng Sun, Zichen Liang, Liping Jing, Jianfei Chen
Abstract: To date, 2:4 sparsity has stood as the only sparse pattern that can be accelerated using sparse tensor cores on GPUs. In practice, 2:4 sparsity often possesses low actual speedups ($\leq 1.3$) and requires fixed sparse ratios, meaning that other ratios, such as 4:8, 8:16, or those exceeding 50% sparsity, do not incur any speedups on GPUs. Recent studies suggest that V:N:M sparsity is promising in addressing these limitations of 2:4 sparsity. However, regarding accuracy, the effects of V:N:M sparsity on broader Transformer models, such as vision Transformers and large language models (LLMs), are largely unexamined. Moreover, Some specific issues related to V:N:M sparsity, such as how to select appropriate V and M values, remain unresolved. In this study, we thoroughly investigate the application of V:N:M sparsity in vision models and LLMs across multiple tasks, from pertaining to downstream tasks. We propose three key approaches to enhance the applicability and accuracy of V:N:M-sparse Transformers, including heuristic V and M selection, V:N:M-specific channel permutation, and three-staged LoRA training techniques. Experimental results show that, with our methods, the DeiT-small achieves lossless accuracy at 64:2:5 sparsity, while the DeiT-base maintains accuracy even at 64:2:8 sparsity. In addition, the fine-tuned LLama2-7B at 64:2:5 sparsity performs comparably or better than training-free 2:4 sparse alternatives on downstream tasks. More importantly, V:N:M-sparse Transformers offer a wider range of speedup-accuracy trade-offs compared to 2:4 sparsity. Overall, our exploration largely facilitates the V:N:M sparsity to act as a truly effective acceleration solution for Transformers in cost-sensitive inference scenarios.
Authors: Shih-Yang Liu, Maksim Khadkevich, Nai Chit Fung, Charbel Sakr, Chao-Han Huck Yang, Chien-Yi Wang, Saurav Muralidharan, Hongxu Yin, Kwang-Ting Cheng, Jan Kautz, Yu-Chiang Frank Wang, Pavlo Molchanov, Min-Hung Chen
Abstract: While post-training compression techniques effectively reduce the memory footprint, latency, and power consumption of Large Language Models (LLMs), they often result in noticeable accuracy degradation and remain limited by hardware and kernel constraints that restrict supported compression formats ultimately reducing flexibility across a wide range of deployment scenarios. In this work, we propose EoRA, a novel fine-tuning-free method that augments compressed LLMs with low-rank matrices, allowing users to rapidly enhance task-specific performance and freely balance the trade-off between accuracy and computational overhead beyond the constraints of compression formats. EoRA consistently outperforms prior training-free low rank methods in recovering the accuracy of compressed LLMs, achieving notable accuracy improvements (e.g., $\mathbf{10.84\%}$ on ARC-Challenge, $\mathbf{6.74\%}$ on MathQA, and $\mathbf{6.74\%}$ on GSM8K) for LLaMA3-8B compressed to 3-bit. We also introduce an optimized CUDA kernel, accelerating inference by up to 1.4x and reducing memory overhead through quantizing EoRA. Overall, EoRA offers a prompt solution for improving the accuracy of compressed models under varying user requirements, enabling more efficient and flexible deployment of LLMs. Code is available at https://github.com/NVlabs/EoRA.
Authors: Chenghua Huang, Zhizhen Fan, Lu Wang, Fangkai Yang, Pu Zhao, Zeqi Lin, Qingwei Lin, Dongmei Zhang, Saravan Rajmohan, Qi Zhang
Abstract: Reinforcement Learning from Human Feedback (RLHF) is a crucial technique for aligning language models with human preferences, playing a pivotal role in the success of conversational models like GPT-4, ChatGPT, and Llama 2. A core challenge in employing RLHF lies in training a reliable reward model (RM), which relies on high-quality labels typically provided by human experts or advanced AI system. These methods can be costly and may introduce biases that affect the language model's responses. As language models improve, human input may become less effective in further enhancing their performance. In this paper, we propose Self-Evolved Reward Learning (SER), a novel approach where the RM generates additional training data to iteratively improve itself. We conducted extensive experiments on multiple datasets such as HH-RLHF and UltraFeedback, using models like Mistral and Llama 3, and compare SER against various baselines. Our results demonstrate that even with limited human-annotated data, learning from self-feedback can robustly enhance RM performance, thereby boosting the capabilities of large language models (LLMs). Resources of this paper can be found at https://aka.ms/ser
URLs: https://aka.ms/ser
Authors: Lin Wang, Xiaocui Yang, Shi Feng, Daling Wang, Yifei Zhang, Zhitao Zhang
Abstract: Multimodal conversation, a crucial form of human communication, carries rich emotional content, making the exploration of the causes of emotions within it a research endeavor of significant importance. However, existing research on the causes of emotions typically employs an utterance selection method within a single textual modality to locate causal utterances. This approach remains limited to coarse-grained assessments, lacks nuanced explanations of emotional causation, and demonstrates inadequate capability in identifying multimodal emotional triggers. Therefore, we introduce a task-\textbf{Multimodal Emotion Cause Explanation in Conversation (MECEC)}. This task aims to generate a summary based on the multimodal context of conversations, clearly and intuitively describing the reasons that trigger a given emotion. To adapt to this task, we develop a new dataset (ECEM) based on the MELD dataset. ECEM combines video clips with detailed explanations of character emotions, helping to explore the causal factors behind emotional expression in multimodal conversations. A novel approach, FAME-Net, is further proposed, that harnesses the power of Large Language Models (LLMs) to analyze visual data and accurately interpret the emotions conveyed through facial expressions in videos. By exploiting the contagion effect of facial emotions, FAME-Net effectively captures the emotional causes of individuals engaged in conversations. Our experimental results on the newly constructed dataset show that FAME-Net outperforms several excellent baselines. Code and dataset are available at https://github.com/3222345200/FAME-Net.
Authors: Gabriele Oliaro, Zhihao Jia, Daniel Campos, Aurick Qiao
Abstract: Speculative decoding is widely adopted to reduce latency in large language model (LLM) inference by leveraging smaller draft models capable of handling diverse user tasks. However, emerging AI applications, such as LLM-based agents, present unique workload characteristics: instead of diverse independent requests, agentic frameworks typically submit repetitive inference requests, such as multi-agent pipelines performing similar subtasks or self-refinement loops iteratively enhancing outputs. These workloads result in long and highly predictable sequences, which current speculative decoding methods do not effectively exploit. To address this gap, we introduce \emph{SuffixDecoding}, a novel method that utilizes efficient suffix trees to cache long token sequences from prompts and previous outputs. By adaptively speculating more tokens when acceptance likelihood is high and fewer when it is low, SuffixDecoding effectively exploits opportunities for longer speculations while conserving computation when those opportunities are limited. Evaluations on agentic benchmarks, including SWE-Bench and Text-to-SQL, demonstrate that SuffixDecoding achieves speedups of up to 5.3$\times$, outperforming state-of-the-art methods -- 2.8$\times$ faster than model-based approaches like EAGLE-2/3 and 1.9$\times$ faster than model-free approaches such as Token Recycling. SuffixDecoding is open-sourced at https://github.com/snowflakedb/ArcticInference.
Authors: Antonio De Santis, Riccardo Campi, Matteo Bianchi, Marco Brambilla
Abstract: Convolutional Neural Networks (CNNs) have seen significant performance improvements in recent years. However, due to their size and complexity, they function as black-boxes, leading to transparency concerns. State-of-the-art saliency methods generate local explanations that highlight the area in the input image where a class is identified but cannot explain how a concept of interest contributes to the prediction, which is essential for bias mitigation. On the other hand, concept-based methods, such as TCAV (Testing with Concept Activation Vectors), provide insights into how sensitive is the network to a concept, but cannot compute its attribution in a specific prediction nor show its location within the input image. This paper introduces a novel post-hoc explainability framework, Visual-TCAV, which aims to bridge the gap between these methods by providing both local and global explanations for CNN-based image classification. Visual-TCAV uses Concept Activation Vectors (CAVs) to generate saliency maps that show where concepts are recognized by the network. Moreover, it can estimate the attribution of these concepts to the output of any class using a generalization of Integrated Gradients. This framework is evaluated on popular CNN architectures, with its validity further confirmed via experiments where ground truth for explanations is known, and a comparison with TCAV. Our code is available at https://github.com/DataSciencePolimi/Visual-TCAV.
Authors: Reece O'Mahoney, Alexander L. Mitchell, Wanming Yu, Ingmar Posner, Ioannis Havoutis
Abstract: We present a diffusion-based approach to quadrupedal locomotion that simultaneously addresses the limitations of learning and interpolating between multiple skills and of (modes) offline adapting to new locomotion behaviours after training. This is the first framework to apply classifier-free guided diffusion to quadruped locomotion and demonstrate its efficacy by extracting goal-conditioned behaviour from an originally unlabelled dataset. We show that these capabilities are compatible with a multi-skill policy and can be applied with little modification and minimal compute overhead, i.e., running entirely on the robots onboard CPU. We verify the validity of our approach with hardware experiments on the ANYmal quadruped platform.
Authors: Minglu Zhao, Dehong Xu, Deqian Kong, Wen-Hao Zhang, Ying Nian Wu
Abstract: We present a minimalistic representation model for the head direction (HD) system, aiming to learn a high-dimensional representation of head direction that captures essential properties of HD cells. Our model is a representation of rotation group $U(1)$, and we study both the fully connected version and convolutional version. We demonstrate the emergence of Gaussian-like tuning profiles and a 2D circle geometry in both versions of the model. We also demonstrate that the learned model is capable of accurate path integration.
Authors: Jubin Abhishek Soni, Amit Anand, Rajesh Kumar Pandey, Aniket Abhishek Soni
Abstract: While the Web has become a global platform for communication, malicious actors, including hackers and hacktivist groups, often disseminate ideological content and coordinate activities through the "Dark Web", an obscure counterpart of the conventional web. Presently, challenges such as information overload and the fragmented nature of cyber threat data impede comprehensive profiling of these actors, thereby limiting the efficacy of predictive analyses of their online activities. Concurrently, the proliferation of internet-connected devices has surpassed the global human population, with this disparity projected to widen as the Internet of Things (IoT) expands. Technical communities are actively advancing IoT-related research to address its growing societal integration. This paper proposes a novel predictive threat intelligence framework designed to systematically collect, analyze, and visualize Dark Web data to identify malicious websites and correlate this information with potential IoT vulnerabilities. The methodology integrates automated data harvesting, analytical techniques, and visual mapping tools, while also examining vulnerabilities in IoT devices to assess exploitability. By bridging gaps in cybersecurity research, this study aims to enhance predictive threat modeling and inform policy development, thereby contributing to intelligence research initiatives focused on mitigating cyber risks in an increasingly interconnected digital ecosystem.
Authors: Lennart Schneider, Martin Wistuba, Aaron Klein, Jacek Golebiowski, Giovanni Zappella, Felice Antonio Merra
Abstract: Optimal prompt selection is crucial for maximizing large language model (LLM) performance on downstream tasks, especially in black-box settings where models are only accessible via APIs. Black-box prompt selection is challenging due to potentially large, combinatorial search spaces, absence of gradient information, and high evaluation cost of prompts on a validation set. We propose HbBoPs, a novel method that combines a structural-aware deep kernel Gaussian Process with Hyperband as a multi-fidelity scheduler to efficiently select prompts. HbBoPs uses embeddings of instructions and few-shot exemplars, treating them as modular components within prompts. This enhances the surrogate model's ability to predict which prompt to evaluate next in a sample-efficient manner. Hyperband improves query-efficiency by adaptively allocating resources across different fidelity levels, reducing the number of validation instances required for evaluating prompts. Extensive experiments across ten diverse benchmarks and three LLMs demonstrate that HbBoPs outperforms state-of-the-art methods in both performance and efficiency.
Authors: Xiaoning Dong, Wenbo Hu, Wei Xu, Tianxing He
Abstract: Large language models (LLMs) have made significant advancements across various tasks, but their safety alignment remain a major concern. Exploring jailbreak prompts can expose LLMs' vulnerabilities and guide efforts to secure them. Existing methods primarily design sophisticated instructions for the LLM to follow, or rely on multiple iterations, which could hinder the performance and efficiency of jailbreaks. In this work, we propose a novel jailbreak paradigm, Simple Assistive Task Linkage (SATA), which can effectively circumvent LLM safeguards and elicit harmful responses. Specifically, SATA first masks harmful keywords within a malicious query to generate a relatively benign query containing one or multiple [MASK] special tokens. It then employs a simple assistive task such as a masked language model task or an element lookup by position task to encode the semantics of the masked keywords. Finally, SATA links the assistive task with the masked query to jointly perform the jailbreak. Extensive experiments show that SATA achieves state-of-the-art performance and outperforms baselines by a large margin. Specifically, on AdvBench dataset, with mask language model (MLM) assistive task, SATA achieves an overall attack success rate (ASR) of 85% and harmful score (HS) of 4.57, and with element lookup by position (ELP) assistive task, SATA attains an overall ASR of 76% and HS of 4.43.
Authors: Meng-Chieh Lee, Qi Zhu, Costas Mavromatis, Zhen Han, Soji Adeshina, Vassilis N. Ioannidis, Huzefa Rangwala, Christos Faloutsos
Abstract: Given a semi-structured knowledge base (SKB), where text documents are interconnected by relations, how can we effectively retrieve relevant information to answer user questions? Retrieval-Augmented Generation (RAG) retrieves documents to assist large language models (LLMs) in question answering; while Graph RAG (GRAG) uses structured knowledge bases as its knowledge source. However, many questions require both textual and relational information from SKB - referred to as "hybrid" questions - which complicates the retrieval process and underscores the need for a hybrid retrieval method that leverages both information. In this paper, through our empirical analysis, we identify key insights that show why existing methods may struggle with hybrid question answering (HQA) over SKB. Based on these insights, we propose HybGRAG for HQA consisting of a retriever bank and a critic module, with the following advantages: (1) Agentic, it automatically refines the output by incorporating feedback from the critic module, (2) Adaptive, it solves hybrid questions requiring both textual and relational information with the retriever bank, (3) Interpretable, it justifies decision making with intuitive refinement path, and (4) Effective, it surpasses all baselines on HQA benchmarks. In experiments on the STaRK benchmark, HybGRAG achieves significant performance gains, with an average relative improvement in Hit@1 of 51%.
Authors: Wei Liu, Junlong Li, Xiwen Zhang, Fan Zhou, Yu Cheng, Junxian He
Abstract: Self-evolving trainin--where models iteratively learn from their own outputs--has emerged as a key approach for complex reasoning tasks, addressing the scarcity of high-quality chain-of-thought data. However, its effectiveness in multimodal reasoning, a domain more intricate than text-only reasoning, remains underexplored, and the understanding of critical factors in this training paradigm remains limited. Furthermore, a central challenge for this training method is performance saturation, which impedes further improvements and scalability. Inspired by reinforcement learning (RL), in this paper, we reframe self-evolving training for multimodal reasoning through the lens of RL, identifying three pivotal factors: Training Method, Reward Model, and Prompt Variation. Through systematic analysis, we establish relatively optimal design principles that significantly enhance multimodal reasoning capabilities. Moreover, delving deeper into training dynamics, we uncover the roots of saturation and propose a new automatic balancing mechanism to mitigate this limitation. Building on these insights, we propose M-STAR (Multimodal Self-evolving Training for Reasoning), a framework that achieves consistent performance gains across models of varying sizes and diverse benchmarks. All resources are made publicly available at https://mstar-lmm.github.io.
Authors: Rui Sun, Yumin Zhang, Varun Ojha, Tejal Shah, Haoran Duan, Bo Wei, Rajiv Ranjan
Abstract: We propose Exemplar-Condensed federated class-incremental learning (ECoral) to distil the training characteristics of real images from streaming data into informative rehearsal exemplars. The proposed method eliminates the limitations of exemplar selection in replay-based approaches for mitigating catastrophic forgetting in federated continual learning (FCL). The limitations particularly related to the heterogeneity of information density of each summarized data. Our approach maintains the consistency of training gradients and the relationship to past tasks for the summarized exemplars to represent the streaming data compared to the original images effectively. Additionally, our approach reduces the information-level heterogeneity of the summarized data by inter-client sharing of the disentanglement generative model. Extensive experiments show that our ECoral outperforms several state-of-the-art methods and can be seamlessly integrated with many existing approaches to enhance performance.
Authors: Xiaohui Chen, Yinkai Wang, Jiaxing He, Yuanqi Du, Soha Hassoun, Xiaolin Xu, Li-Ping Liu
Abstract: Graph generation is a critical task in numerous domains, including molecular design and social network analysis, due to its ability to model complex relationships and structured data. While most modern graph generative models utilize adjacency matrix representations, this work revisits an alternative approach that represents graphs as sequences of node set and edge set. We advocate for this approach due to its efficient encoding of graphs and propose a novel representation. Based on this representation, we introduce the Graph Generative Pre-trained Transformer (G2PT), an auto-regressive model that learns graph structures via next-token prediction. To further exploit G2PT's capabilities as a general-purpose foundation model, we explore fine-tuning strategies for two downstream applications: goal-oriented generation and graph property prediction. We conduct extensive experiments across multiple datasets. Results indicate that G2PT achieves superior generative performance on both generic graph and molecule datasets. Furthermore, G2PT exhibits strong adaptability and versatility in downstream tasks from molecular design to property prediction. Code available at https://github.com/tufts-ml/G2PT,
Authors: Yindu Su, Huike Zou, Lin Sun, Ting Zhang, Haiyang Yang, Liyu Chen, David Lo, Qingheng Zhang, Shuguang Han, Jufeng Chen
Abstract: Product Attribute Value Identification (PAVI) involves identifying attribute values from product profiles, a key task for improving product search, recommendation, and business analytics on e-commerce platforms. However, existing PAVI methods face critical challenges, such as inferring implicit values, handling out-of-distribution (OOD) values, and producing normalized outputs. To address these limitations, we introduce Taxonomy-Aware Contrastive Learning Retrieval (TACLR), the first retrieval-based method for PAVI. TACLR formulates PAVI as an information retrieval task by encoding product profiles and candidate values into embeddings and retrieving values based on their similarity. It leverages contrastive training with taxonomy-aware hard negative sampling and employs adaptive inference with dynamic thresholds. TACLR offers three key advantages: (1) it effectively handles implicit and OOD values while producing normalized outputs; (2) it scales to thousands of categories, tens of thousands of attributes, and millions of values; and (3) it supports efficient inference for high-load industrial deployment. Extensive experiments on proprietary and public datasets validate the effectiveness and efficiency of TACLR. Further, it has been successfully deployed on the real-world e-commerce platform Xianyu, processing millions of product listings daily with frequently updated, large-scale attribute taxonomies. We release the code to facilitate reproducibility and future research at https://github.com/SuYindu/TACLR.
Authors: Tong Liu, Xiao Yu, Wenxuan Zhou, Jindong Gu, Volker Tresp
Abstract: Efficient preference optimization algorithms such as Direct Preference Optimization (DPO) have become a popular approach in aligning large language models (LLMs) with human preferences. These algorithms implicitly treat the LLM as a reward model, and focus on training it to correct misranked preference pairs. However, recent work~\citep{chen2024preference} empirically finds that DPO training \textit{rarely improves these misranked preference pairs}, despite its gradient emphasizing on these cases. We introduce FocalPO, a DPO variant that instead \textit{down-weighs} misranked preference pairs and prioritizes enhancing the model's understanding of pairs that it can already rank correctly. Inspired by Focal Loss used in vision tasks, FocalPO achieves this by adding a modulating factor to dynamically scale DPO loss. Our experiment demonstrates that FocalPO surpasses DPO and its variants on popular benchmarks like Alpaca Eval 2.0 using Mistral-Base-7B and Llama-3-Instruct-8B, with the introduced hyperparameter fixed. Additionally, we empirically reveals how FocalPO affects training on correct and incorrect sample groups, further underscoring its effectiveness.
Authors: Xiaoyu Zhang, Juan Zhai, Shiqing Ma, Qingshuang Bao, Weipeng Jiang, Qian Wang, Chao Shen, Yang Liu
Abstract: Large Language Models (LLMs) have emerged as the new recommendation engines, surpassing traditional methods in both capability and scope, particularly in code generation. In this paper, we reveal a novel provider bias in LLMs: without explicit directives, these models show systematic preferences for services from specific providers in their recommendations (e.g., favoring Google Cloud over Microsoft Azure). To systematically investigate this bias, we develop an automated pipeline to construct the dataset, incorporating 6 distinct coding task categories and 30 real-world application scenarios. Leveraging this dataset, we conduct the first comprehensive empirical study of provider bias in LLM code generation across seven state-of-the-art LLMs, utilizing approximately 500 million tokens (equivalent to $5,000+ in computational costs). Our findings reveal that LLMs exhibit significant provider preferences, predominantly favoring services from Google and Amazon, and can autonomously modify input code to incorporate their preferred providers without users' requests. Such a bias holds far-reaching implications for market dynamics and societal equilibrium, potentially contributing to digital monopolies. It may also deceive users and violate their expectations, leading to various consequences. We call on the academic community to recognize this emerging issue and develop effective evaluation and mitigation methods to uphold AI security and fairness.
Authors: Jingyang Ke, Feiyang Wu, Jiyi Wang, Jeffrey Markowitz, Anqi Wu
Abstract: Traditional approaches to studying decision-making in neuroscience focus on simplified behavioral tasks where animals perform repetitive, stereotyped actions to receive explicit rewards. While informative, these methods constrain our understanding of decision-making to short timescale behaviors driven by explicit goals. In natural environments, animals exhibit more complex, long-term behaviors driven by intrinsic motivations that are often unobservable. Recent works in time-varying inverse reinforcement learning (IRL) aim to capture shifting motivations in long-term, freely moving behaviors. However, a crucial challenge remains: animals make decisions based on their history, not just their current state. To address this, we introduce SWIRL (SWitching IRL), a novel framework that extends traditional IRL by incorporating time-varying, history-dependent reward functions. SWIRL models long behavioral sequences as transitions between short-term decision-making processes, each governed by a unique reward function. SWIRL incorporates biologically plausible history dependency to capture how past decisions and environmental contexts shape behavior, offering a more accurate description of animal decision-making. We apply SWIRL to simulated and real-world animal behavior datasets and show that it outperforms models lacking history dependency, both quantitatively and qualitatively. This work presents the first IRL model to incorporate history-dependent policies and rewards to advance our understanding of complex, naturalistic decision-making in animals.
Authors: Tanya Rodchenko, Natasha Noy, Nino Scherrer
Abstract: While Large Language Models require more and more data to train and scale, rather than looking for any data to acquire, we should consider what types of tasks are more likely to benefit from data scaling. We should be intentional in our data acquisition. We argue that the shape of the data itself, such as its compositional and structural patterns, informs which tasks to prioritize in data scaling, and shapes the development of the next generation of compute paradigms for tasks where data scaling is inefficient, or even insufficient.
Authors: Edoardo Cetin, Tianyu Zhao, Yujin Tang
Abstract: We propose a new finetuning method to provide pre-trained large language models (LMs) the ability to scale test-time compute through the diffusion framework. By increasing the number of diffusion steps, we show our finetuned models achieve monotonically increasing accuracy, directly translating to improved performance across downstream tasks. Furthermore, our finetuned models can expertly answer questions on specific topics by integrating powerful guidance techniques, and autonomously determine the compute required for a given problem by leveraging adaptive ODE solvers. Our method is universally applicable to any foundation model pre-trained with a cross-entropy loss and does not modify any of its original weights, fully preserving its strong single-step generation capabilities. We show our method is more effective and fully compatible with traditional finetuning approaches, introducing an orthogonal new direction to unify the strengths of the autoregressive and diffusion frameworks.
Authors: Lo Pang-Yun Ting, Zhen Tan, Hong-Pei Chen, Cheng-Te Li, Po-Lin Chen, Kun-Ta Chuang, Huan Liu
Abstract: Early detection of patient deterioration is essential for timely treatment, with vital signs like heart rates being key health indicators. Existing methods tend to solely analyze vital sign waveforms, ignoring transition relationships of waveforms within each vital sign and the correlation strengths among various vital signs. Such studies often overlook nuanced illness deterioration, which is the early sign of worsening health but is difficult to detect. In this paper, we introduce CAND, a novel method that organizes the transition relationships and the correlations within and among vital signs as domain-specific and cross-domain knowledge. CAND jointly models these knowledge in a unified representation space, considerably enhancing the early detection of nuanced illness deterioration. In addition, CAND integrates a Bayesian inference method that utilizes augmented knowledge from domain-specific and cross-domain knowledge to address the ambiguities in correlation strengths. With this architecture, the correlation strengths can be effectively inferred to guide joint modeling and enhance representations of vital signs. This allows a more holistic and accurate interpretation of patient health. Our experiments on a real-world ICU dataset demonstrate that CAND significantly outperforms existing methods in both effectiveness and earliness in detecting nuanced illness deterioration. Moreover, we conduct a case study for the interpretable detection process to showcase the practicality of CAND.
Authors: Mariano V. Ntrougkas, Vasileios Mezaris, Ioannis Patras
Abstract: The adoption of Deep Neural Networks (DNNs) in critical fields where predictions need to be accompanied by justifications is hindered by their inherent black-box nature. In this paper, we introduce P-TAME (Perturbation-based Trainable Attention Mechanism for Explanations), a model-agnostic method for explaining DNN-based image classifiers. P-TAME employs an auxiliary image classifier to extract features from the input image, bypassing the need to tailor the explanation method to the internal architecture of the backbone classifier being explained. Unlike traditional perturbation-based methods, which have high computational requirements, P-TAME offers an efficient alternative by generating high-resolution explanations in a single forward pass during inference. We apply P-TAME to explain the decisions of VGG-16, ResNet-50, and ViT-B-16, three distinct and widely used image classifiers. Quantitative and qualitative results show that our method matches or outperforms previous explainability methods, including model-specific approaches. Code and trained models will be released upon acceptance.
Authors: Nhan Phan, Thu Nguyen, P{\aa}l Halvorsen, Michael A. Riegler
Abstract: Principal Component Analysis (PCA) is a commonly used tool for dimension reduction and denoising. Therefore, it is also widely used on the data prior to training a neural network. However, this approach can complicate the explanation of explainable AI (XAI) methods for the decision of the model. In this work, we analyze the potential issues with this approach and propose Principal Components-based Initialization (PCsInit), a strategy to incorporate PCA into the first layer of a neural network via initialization of the first layer in the network with the principal components, and its two variants PCsInit-Act and PCsInit-Sub. Explanations using these strategies are as direct and straightforward as for neural networks and are simpler than using PCA prior to training a neural network on the principal components. Moreover, as will be illustrated in the experiments, such training strategies can also allow further improvement of training via backpropagation.
Authors: Jungwon Seo, Ferhat Ozgur Catak, Chunming Rong
Abstract: As privacy concerns and data regulations grow, federated learning (FL) has emerged as a promising approach for training machine learning models across decentralized data sources without sharing raw data. However, a significant challenge in FL is that client data are often non-IID (non-independent and identically distributed), leading to reduced performance compared to centralized learning. While many methods have been proposed to address this issue, their underlying mechanisms are often viewed from different perspectives. Through a comprehensive investigation from gradient descent to FL, and from IID to non-IID data settings, we find that inconsistencies in client loss landscapes primarily cause performance degradation in non-IID scenarios. From this understanding, we observe that existing methods can be grouped into two main strategies: (i) adjusting parameter update paths and (ii) modifying client loss landscapes. These findings offer a clear perspective on addressing non-IID challenges in FL and help guide future research in the field.
Authors: Xin Xu, Qiyun Xu, Tong Xiao, Tianhao Chen, Yuchen Yan, Jiaxin Zhang, Shizhe Diao, Can Yang, Yang Wang
Abstract: Large language models (LLMs) have demonstrated remarkable capabilities in solving complex reasoning tasks, particularly in mathematics. However, the domain of physics reasoning presents unique challenges that have received significantly less attention. Existing benchmarks often fall short in evaluating LLMs' abilities on the breadth and depth of undergraduate-level physics, underscoring the need for a comprehensive evaluation. To fill this gap, we introduce UGPhysics, a large-scale and comprehensive benchmark specifically designed to evaluate UnderGraduate-level Physics (UGPhysics) reasoning with LLMs. UGPhysics includes 5,520 undergraduate-level physics problems in both English and Chinese, covering 13 subjects with seven different answer types and four distinct physics reasoning skills, all rigorously screened for data leakage. Additionally, we develop a Model-Assistant Rule-based Judgment (MARJ) pipeline specifically tailored for assessing answer correctness of physics problems, ensuring accurate evaluation. Our evaluation of 31 leading LLMs shows that the highest overall accuracy, 49.8% (achieved by OpenAI-o1-mini), emphasizes the necessity for models with stronger physics reasoning skills, beyond math abilities. We hope UGPhysics, along with MARJ, will drive future advancements in AI for physics reasoning. Codes and data are available at https://github.com/YangLabHKUST/UGPhysics .
Authors: Gaurush Hiranandani, Haolun Wu, Subhojyoti Mukherjee, Sanmi Koyejo
Abstract: Many commercial Large Language Models (LLMs) are often closed-source, limiting developers to prompt tuning for aligning content generation with specific applications. While these models currently do not provide access to token logits, we argue that if such access were available, it would enable more powerful adaptation techniques beyond prompt engineering. In this paper, we propose a token-level probability reweighting framework that, given access to logits and a small amount of task-specific data, can effectively steer black-box LLMs toward application-specific content generation. Our approach views next-token prediction through the lens of supervised classification. We show that aligning black-box LLMs with task-specific data can be formulated as a label noise correction problem, leading to Plugin model -- an autoregressive probability reweighting model that operates solely on logits. We provide theoretical justification for why reweighting logits alone is sufficient for task adaptation. Extensive experiments with multiple datasets, LLMs, and reweighting models demonstrate the effectiveness of our method, advocating for broader access to token logits in closed-source models.
Authors: Shu Wang, Yixiang Fang, Yingli Zhou, Xilin Liu, Yuchi Ma
Abstract: Retrieval-Augmented Generation (RAG) has proven effective in integrating external knowledge into large language models (LLMs) for solving question-answer (QA) tasks. The state-of-the-art RAG approaches often use the graph data as the external data since they capture the rich semantic information and link relationships between entities. However, existing graph-based RAG approaches cannot accurately identify the relevant information from the graph and also consume large numbers of tokens in the online retrieval process. To address these issues, we introduce a novel graph-based RAG approach, called Attributed Community-based Hierarchical RAG (ArchRAG), by augmenting the question using attributed communities, and also introducing a novel LLM-based hierarchical clustering method. To retrieve the most relevant information from the graph for the question, we build a novel hierarchical index structure for the attributed communities and develop an effective online retrieval method. Experimental results demonstrate that ArchRAG outperforms existing methods in both accuracy and token cost. Moreover, ArchRAG has been successfully applied to domain knowledge QA in Huawei Cloud Computing.
Authors: Haoyang Li, Xuejia Chen, Zhanchao XU, Darian Li, Nicole Hu, Fei Teng, Yiming Li, Luyu Qiu, Chen Jason Zhang, Qing Li, Lei Chen
Abstract: Large Language Models (LLMs) have demonstrated impressive capabilities in natural language processing tasks, such as text generation and semantic understanding. However, their performance on numerical reasoning tasks, such as basic arithmetic, numerical retrieval, and magnitude comparison, remains surprisingly poor. This gap arises from their reliance on surface-level statistical patterns rather than understanding numbers as continuous magnitudes. Existing benchmarks primarily focus on either linguistic competence or structured mathematical problem-solving, neglecting fundamental numerical reasoning required in real-world scenarios. To bridge this gap, we propose NumericBench, a comprehensive benchmark to evaluate six fundamental numerical capabilities: number recognition, arithmetic operations, contextual retrieval, comparison, summary, and logical reasoning. NumericBench includes datasets ranging from synthetic number lists to the crawled real-world data, addressing challenges like long contexts, noise, and multi-step reasoning. Extensive experiments on state-of-the-art LLMs, including GPT-4 and DeepSeek, reveal persistent weaknesses in numerical reasoning, highlighting the urgent need to improve numerically-aware language modeling. The benchmark is released in: https://github.com/TreeAI-Lab/NumericBench.
Authors: Wenxuan Wang, Xiaoyuan Liu, Kuiyi Gao, Jen-tse Huang, Youliang Yuan, Pinjia He, Shuai Wang, Zhaopeng Tu
Abstract: Multimodal Large Language Models (MLLMs) have expanded the capabilities of traditional language models by enabling interaction through both text and images. However, ensuring the safety of these models remains a significant challenge, particularly in accurately identifying whether multimodal content is safe or unsafe-a capability we term safety awareness. In this paper, we introduce MMSafeAware, the first comprehensive multimodal safety awareness benchmark designed to evaluate MLLMs across 29 safety scenarios with 1500 carefully curated image-prompt pairs. MMSafeAware includes both unsafe and over-safety subsets to assess models abilities to correctly identify unsafe content and avoid over-sensitivity that can hinder helpfulness. Evaluating nine widely used MLLMs using MMSafeAware reveals that current models are not sufficiently safe and often overly sensitive; for example, GPT-4V misclassifies 36.1% of unsafe inputs as safe and 59.9% of benign inputs as unsafe. We further explore three methods to improve safety awareness-prompting-based approaches, visual contrastive decoding, and vision-centric reasoning fine-tuning-but find that none achieve satisfactory performance. Our findings highlight the profound challenges in developing MLLMs with robust safety awareness, underscoring the need for further research in this area. All the code and data will be publicly available to facilitate future research.
Authors: Shuaiqun Pan, Yash J. Patel, Aneta Neumann, Frank Neumann, Thomas B\"ack, Hao Wang
Abstract: Variational quantum algorithms, such as the Recursive Quantum Approximate Optimization Algorithm (RQAOA), have become increasingly popular, offering promising avenues for employing Noisy Intermediate-Scale Quantum devices to address challenging combinatorial optimization tasks like the maximum cut problem. In this study, we utilize an evolutionary algorithm equipped with a unique fitness function. This approach targets hard maximum cut instances within the latent space of a Graph Autoencoder, identifying those that pose significant challenges or are particularly tractable for RQAOA, in contrast to the classic Goemans and Williamson algorithm. Our findings not only delineate the distinct capabilities and limitations of each algorithm but also expand our understanding of RQAOA's operational limits. Furthermore, the diverse set of graphs we have generated serves as a crucial benchmarking asset, emphasizing the need for more advanced algorithms to tackle combinatorial optimization challenges. Additionally, our results pave the way for new avenues in graph generation research, offering exciting opportunities for future explorations.
Authors: Bo Wang, Weiyi He, Shenglai Zeng, Zhen Xiang, Yue Xing, Jiliang Tang, Pengfei He
Abstract: Large Language Model (LLM) agents have become increasingly prevalent across various real-world applications. They enhance decision-making by storing private user-agent interactions in the memory module for demonstrations, introducing new privacy risks for LLM agents. In this work, we systematically investigate the vulnerability of LLM agents to our proposed Memory EXTRaction Attack (MEXTRA) under a black-box setting. To extract private information from memory, we propose an effective attacking prompt design and an automated prompt generation method based on different levels of knowledge about the LLM agent. Experiments on two representative agents demonstrate the effectiveness of MEXTRA. Moreover, we explore key factors influencing memory leakage from both the agent designer's and the attacker's perspectives. Our findings highlight the urgent need for effective memory safeguards in LLM agent design and deployment.
Authors: Shengguang Wu, Fan-Yun Sun, Kaiyue Wen, Nick Haber
Abstract: Recent studies have shown that Large Vision-Language Models (VLMs) tend to neglect image content and over-rely on language-model priors, resulting in errors in visually grounded tasks and hallucinations. We hypothesize that this issue arises because existing VLMs are not explicitly trained to generate texts that are accurately grounded in fine-grained image details. To enhance visual feedback during VLM training, we propose S-VCO (Symmetrical Visual Contrastive Optimization), a novel finetuning objective that steers the model toward capturing important visual details and aligning them with corresponding text tokens. To further facilitate this detailed alignment, we introduce MVC, a paired image-text dataset built by automatically filtering and augmenting visual counterfactual data to challenge the model with hard contrastive cases involving Minimal Visual Contrasts. Experiments show that our method consistently improves VLM performance across diverse benchmarks covering various abilities and domains, achieving up to a 22% reduction in hallucinations, and significant gains in vision-centric and general tasks. Notably, these improvements become increasingly pronounced in benchmarks with higher visual dependency. In short, S-VCO offers a significant enhancement of VLM's visually-dependent task performance while retaining or even improving the model's general abilities. We opensource our code at https://s-vco.github.io/
URLs: https://s-vco.github.io/
Authors: Maya Varma, Ashwin Kumar, Rogier van der Sluijs, Sophie Ostmeier, Louis Blankemeier, Pierre Chambon, Christian Bluethgen, Jip Prince, Curtis Langlotz, Akshay Chaudhari
Abstract: Medical images are acquired at high resolutions with large fields of view in order to capture fine-grained features necessary for clinical decision-making. Consequently, training deep learning models on medical images can incur large computational costs. In this work, we address the challenge of downsizing medical images in order to improve downstream computational efficiency while preserving clinically-relevant features. We introduce MedVAE, a family of six large-scale 2D and 3D autoencoders capable of encoding medical images as downsized latent representations and decoding latent representations back to high-resolution images. We train MedVAE autoencoders using a novel two-stage training approach with 1,052,730 medical images. Across diverse tasks obtained from 20 medical image datasets, we demonstrate that (1) utilizing MedVAE latent representations in place of high-resolution images when training downstream models can lead to efficiency benefits (up to 70x improvement in throughput) while simultaneously preserving clinically-relevant features and (2) MedVAE can decode latent representations back to high-resolution images with high fidelity. Our work demonstrates that large-scale, generalizable autoencoders can help address critical efficiency challenges in the medical domain. Our code is available at https://github.com/StanfordMIMI/MedVAE.
Authors: Yang Yao, Xuan Tong, Ruofan Wang, Yixu Wang, Lujundong Li, Liang Liu, Yan Teng, Yingchun Wang
Abstract: Large Reasoning Models (LRMs) have significantly advanced beyond traditional Large Language Models (LLMs) with their exceptional logical reasoning capabilities, yet these improvements introduce heightened safety risks. When subjected to jailbreak attacks, their ability to generate more targeted and organized content can lead to greater harm. Although some studies claim that reasoning enables safer LRMs against existing LLM attacks, they overlook the inherent flaws within the reasoning process itself. To address this gap, we propose the first jailbreak attack targeting LRMs, exploiting their unique vulnerabilities stemming from the advanced reasoning capabilities. Specifically, we introduce a Chaos Machine, a novel component to transform attack prompts with diverse one-to-one mappings. The chaos mappings iteratively generated by the machine are embedded into the reasoning chain, which strengthens the variability and complexity and also promotes a more robust attack. Based on this, we construct the Mousetrap framework, which makes attacks projected into nonlinear-like low sample spaces with mismatched generalization enhanced. Also, due to the more competing objectives, LRMs gradually maintain the inertia of unpredictable iterative reasoning and fall into our trap. Success rates of the Mousetrap attacking o1-mini, Claude-Sonnet and Gemini-Thinking are as high as 96%, 86% and 98% respectively on our toxic dataset Trotter. On benchmarks such as AdvBench, StrongREJECT, and HarmBench, attacking Claude-Sonnet, well-known for its safety, Mousetrap can astonishingly achieve success rates of 87.5%, 86.58% and 93.13% respectively. Attention: This paper contains inappropriate, offensive and harmful content.
Authors: Qiuchen Wang, Ruixue Ding, Zehui Chen, Weiqi Wu, Shihang Wang, Pengjun Xie, Feng Zhao
Abstract: Understanding information from visually rich documents remains a significant challenge for traditional Retrieval-Augmented Generation (RAG) methods. Existing benchmarks predominantly focus on image-based question answering (QA), overlooking the fundamental challenges of efficient retrieval, comprehension, and reasoning within dense visual documents. To bridge this gap, we introduce ViDoSeek, a novel dataset designed to evaluate RAG performance on visually rich documents requiring complex reasoning. Based on it, we identify key limitations in current RAG approaches: (i) purely visual retrieval methods struggle to effectively integrate both textual and visual features, and (ii) previous approaches often allocate insufficient reasoning tokens, limiting their effectiveness. To address these challenges, we propose ViDoRAG, a novel multi-agent RAG framework tailored for complex reasoning across visual documents. ViDoRAG employs a Gaussian Mixture Model (GMM)-based hybrid strategy to effectively handle multi-modal retrieval. To further elicit the model's reasoning capabilities, we introduce an iterative agent workflow incorporating exploration, summarization, and reflection, providing a framework for investigating test-time scaling in RAG domains. Extensive experiments on ViDoSeek validate the effectiveness and generalization of our approach. Notably, ViDoRAG outperforms existing methods by over 10% on the competitive ViDoSeek benchmark. The code is available at https://github.com/Alibaba-NLP/ViDoRAG.
Authors: Gengyuan Hu, Wanli Ouyang, Chao-Yang Lu, Chen Lin, Han-Sen Zhong
Abstract: Scaling quantum computing to practical applications necessitates reliable quantum error correction. Although numerous correction codes have been proposed, the overall correction efficiency critically limited by the decode algorithms. We introduce GraphQEC, a code-agnostic decoder leveraging machine-learning on the graph structure of stabilizer codes with linear time complexity. GraphQEC demonstrates unprecedented accuracy and efficiency across all tested code families, including surface codes, color codes, and quantum low-density parity-check (QLDPC) codes. For instance, on a distance-12 QLDPC code, GraphQEC achieves a logical error rate of $9.55 \times 10^{-5}$, an 18-fold improvement over the previous best specialized decoder's $1.74 \times 10^{-3}$ under $p=0.005$ physical error rates, while maintaining $157\mu$s/cycle decoding speed. Our approach represents the first universal solution for real-time quantum error correction across arbitrary stabilizer codes.
Authors: Keyu Duan, Yiran Zhao, Zhili Feng, Jinjie Ni, Tianyu Pang, Qian Liu, Tianle Cai, Longxu Dou, Kenji Kawaguchi, Anirudh Goyal, J. Zico Kolter, Michael Qizhe Shieh
Abstract: Large Language Models (LLMs) have been observed to process non-human-readable text sequences, such as jailbreak prompts, often viewed as a bug for aligned LLMs. In this work, we present a systematic investigation challenging this perception, demonstrating that unnatural languages - strings that appear incomprehensible to humans but maintain semantic meanings for LLMs - contain latent features usable by models. Notably, unnatural languages possess latent features that can be generalized across different models and tasks during inference. Furthermore, models fine-tuned on unnatural versions of instruction datasets perform on-par with those trained on natural language, achieving 49.71 win rates in Length-controlled AlpacaEval 2.0 in average across various base models. In addition, through comprehensive analysis, we demonstrate that LLMs process unnatural languages by filtering noise and inferring contextual meaning from filtered words.
Authors: Xiner Li, Masatoshi Uehara, Xingyu Su, Gabriele Scalia, Tommaso Biancalani, Aviv Regev, Sergey Levine, Shuiwang Ji
Abstract: Diffusion models have shown promising generative capabilities across diverse domains, yet aligning their outputs with desired reward functions remains a challenge, particularly in cases where reward functions are non-differentiable. Some gradient-free guidance methods have been developed, but they often struggle to achieve optimal inference-time alignment. In this work, we newly frame inference-time alignment in diffusion as a search problem and propose Dynamic Search for Diffusion (DSearch), which subsamples from denoising processes and approximates intermediate node rewards. It also dynamically adjusts beam width and tree expansion to efficiently explore high-reward generations. To refine intermediate decisions, DSearch incorporates adaptive scheduling based on noise levels and a lookahead heuristic function. We validate DSearch across multiple domains, including biological sequence design, molecular optimization, and image generation, demonstrating superior reward optimization compared to existing approaches.
Authors: Yi Shen, Jian Zhang, Jieyun Huang, Shuming Shi, Wenjing Zhang, Jiangze Yan, Ning Wang, Kai Wang, Zhaoxiang Liu, Shiguo Lian
Abstract: Recent advancements in slow thinking reasoning models have shown exceptional performance in complex reasoning tasks. However, these models often exhibit overthinking (generating redundant reasoning steps for simple problems), leading to excessive computational resource usage. While current mitigation strategies uniformly reduce reasoning tokens, they risk degrading performance on challenging tasks that require extended reasoning. This paper introduces Difficulty-Adaptive Slow Thinking (DAST), a novel framework that enables models to autonomously adjust the length of Chain-of-Thought (CoT) based on problem difficulty. We first propose a Token Length Budget (TLB) metric to quantify difficulty, then leverage budget-aware reward shaping and budget preference optimization to implement DAST. DAST penalizes overlong responses for simple tasks while incentivizing sufficient reasoning for complex problems. Experiments on diverse datasets and model scales demonstrate that DAST effectively mitigates overthinking (reducing token usage by over 30\% on average) while preserving reasoning accuracy on complex problems. Our codes and models are available at https://github.com/AnonymousUser0520/AnonymousRepo01.
Authors: Shuo Jiang, Haonan Li, Ruochen Ren, Yanmin Zhou, Zhipeng Wang, Bin He
Abstract: Cutting-edge robot learning techniques including foundation models and imitation learning from humans all pose huge demands on large-scale and high-quality datasets which constitute one of the bottleneck in the general intelligent robot fields. This paper presents the Kaiwu multimodal dataset to address the missing real-world synchronized multimodal data problems in the sophisticated assembling scenario,especially with dynamics information and its fine-grained labelling. The dataset first provides an integration of human,environment and robot data collection framework with 20 subjects and 30 interaction objects resulting in totally 11,664 instances of integrated actions. For each of the demonstration,hand motions,operation pressures,sounds of the assembling process,multi-view videos, high-precision motion capture information,eye gaze with first-person videos,electromyography signals are all recorded. Fine-grained multi-level annotation based on absolute timestamp,and semantic segmentation labelling are performed. Kaiwu dataset aims to facilitate robot learning,dexterous manipulation,human intention investigation and human-robot collaboration research.
Authors: Hyungkyu Kang, Min-hwan Oh
Abstract: In this paper, we study offline preference-based reinforcement learning (PbRL), where learning is based on pre-collected preference feedback over pairs of trajectories. While offline PbRL has demonstrated remarkable empirical success, existing theoretical approaches face challenges in ensuring conservatism under uncertainty, requiring computationally intractable confidence set constructions. We address this limitation by proposing Adversarial Preference-based Policy Optimization (APPO), a computationally efficient algorithm for offline PbRL that guarantees sample complexity bounds without relying on explicit confidence sets. By framing PbRL as a two-player game between a policy and a model, our approach enforces conservatism in a tractable manner. Using standard assumptions on function approximation and bounded trajectory concentrability, we derive a sample complexity bound. To our knowledge, APPO is the first offline PbRL algorithm to offer both statistical efficiency and practical applicability. Experimental results on continuous control tasks demonstrate that APPO effectively learns from complex datasets, showing comparable performance with existing state-of-the-art methods.
Authors: Jillian Fisher, Ruth E. Appel, Chan Young Park, Yujin Potter, Liwei Jiang, Taylor Sorensen, Shangbin Feng, Yulia Tsvetkov, Margaret E. Roberts, Jennifer Pan, Dawn Song, Yejin Choi
Abstract: AI systems often exhibit political bias, influencing users' opinions and decisions. While political neutrality-defined as the absence of bias-is often seen as an ideal solution for fairness and safety, this position paper argues that true political neutrality is neither feasible nor universally desirable due to its subjective nature and the biases inherent in AI training data, algorithms, and user interactions. However, inspired by Joseph Raz's philosophical insight that "neutrality [...] can be a matter of degree" (Raz, 1986), we argue that striving for some neutrality remains essential for promoting balanced AI interactions and mitigating user manipulation. Therefore, we use the term "approximation" of political neutrality to shift the focus from unattainable absolutes to achievable, practical proxies. We propose eight techniques for approximating neutrality across three levels of conceptualizing AI, examining their trade-offs and implementation strategies. In addition, we explore two concrete applications of these approximations to illustrate their practicality. Finally, we assess our framework on current large language models (LLMs) at the output level, providing a demonstration of how it can be evaluated. This work seeks to advance nuanced discussions of political neutrality in AI and promote the development of responsible, aligned language models.
Authors: Angela Lopez-Cardona, Sebastian Idesis, Miguel Barreda-\'Angeles, Sergi Abadal, Ioannis Arapakis
Abstract: While Large Language Models (LLMs) have significantly advanced natural language processing, aligning them with human preferences remains an open challenge. Although current alignment methods rely primarily on explicit feedback, eye-tracking (ET) data offers insights into real-time cognitive processing during reading. In this paper, we present OASST-ETC, a novel eye-tracking corpus capturing reading patterns from 24 participants, while evaluating LLM-generated responses from the OASST1 dataset. Our analysis reveals distinct reading patterns between preferred and non-preferred responses, which we compare with synthetic eye-tracking data. Furthermore, we examine the correlation between human reading measures and attention patterns from various transformer-based models, discovering stronger correlations in preferred responses. This work introduces a unique resource for studying human cognitive processing in LLM evaluation and suggests promising directions for incorporating eye-tracking data into alignment methods. The dataset and analysis code are publicly available.
Authors: Hai-Long Sun, Zhun Sun, Houwen Peng, Han-Jia Ye
Abstract: Recent advancements in Large Language Models (LLMs) have demonstrated enhanced reasoning capabilities, evolving from Chain-of-Thought (CoT) prompting to advanced, product-oriented solutions like OpenAI o1. During our re-implementation of this model, we noticed that in multimodal tasks requiring visual input (e.g., geometry problems), Multimodal LLMs (MLLMs) struggle to maintain focus on the visual information, in other words, MLLMs suffer from a gradual decline in attention to visual information as reasoning progresses, causing text-over-relied outputs. To investigate this, we ablate image inputs during long-chain reasoning. Concretely, we truncate the reasoning process midway, then re-complete the reasoning process with the input image removed. We observe only a ~2% accuracy drop on MathVista's test-hard subset, revealing the model's textual outputs dominate the following reasoning process. Motivated by this, we propose Take-along Visual Conditioning (TVC), a strategy that shifts image input to critical reasoning stages and compresses redundant visual tokens via dynamic pruning. This methodology helps the model retain attention to the visual components throughout the reasoning. Our approach achieves state-of-the-art performance on average across five mathematical reasoning benchmarks (+3.4 points vs previous sota), demonstrating the effectiveness of TVC in enhancing multimodal reasoning systems.
Authors: Bram Rijsbosch, Gijs van Dijck, Konrad Kollnig
Abstract: AI-generated images have become so good in recent years that individuals often cannot distinguish them any more from "real" images. This development, combined with the rapid spread of AI-generated content online, creates a series of societal risks, particularly with the emergence of "deep fakes" that impersonate real individuals. Watermarking, a technique that involves embedding information within images and other content to indicate their AI-generated nature, has emerged as a primary mechanism to address the risks posed by AI-generated content. Indeed, watermarking and AI labelling measures are now becoming a legal requirement in many jurisdictions, including under the 2024 European Union AI Act. Despite the widespread use of AI image generation systems, the current status of the implementation of such measures remains largely unexamined. Moreover, the practical implications of the AI Act's watermarking and labelling requirements have not previously been studied. The present paper therefore both provides an empirical analysis of 50 widely used AI systems for image generation, embedded into a legal analysis of the AI Act. In our legal analysis, we identify four categories of generative AI image deployment scenarios relevant under the AI Act and outline how the legal obligations apply in each category. In our empirical analysis, we find that only a minority number of AI image generators currently implement adequate watermarking (38%) and deep fake labelling (8%) practices. In response, we suggest a range of avenues of how the implementation of these legally mandated techniques can be improved, and publicly share our tooling for the easy detection of watermarks in images.
Authors: Sohyun Lee, Nayeong Kim, Juwon Kang, Seong Joon Oh, Suha Kwak
Abstract: This paper studies continual test-time adaptation (CTTA), the task of adapting a model to constantly changing unseen domains in testing while preserving previously learned knowledge. Existing CTTA methods mostly focus on adaptation to the current test domain only, overlooking generalization to arbitrary test domains a model may face in the future. To tackle this limitation, we present a novel online test-time domain generalization framework for CTTA, dubbed TestDG. TestDG aims to learn features invariant to both current and previous test domains on the fly during testing, improving the potential for effective generalization to future domains. To this end, we propose a new model architecture and a test-time adaptation strategy dedicated to learning domain-invariant features, along with a new data structure and optimization algorithm for effectively managing information from previous test domains. TestDG achieved state of the art on four public CTTA benchmarks. Moreover, it showed superior generalization to unseen test domains.
Authors: Jiawei Lian, Jianhong Pan, Lefan Wang, Yi Wang, Shaohui Mei, Lap-Pui Chau
Abstract: Large language models (LLMs) are foundational explorations to artificial general intelligence, yet their alignment with human values via instruction tuning and preference learning achieves only superficial compliance. Here, we demonstrate that harmful knowledge embedded during pretraining persists as indelible "dark patterns" in LLMs' parametric memory, evading alignment safeguards and resurfacing under adversarial inducement at distributional shifts. In this study, we first theoretically analyze the intrinsic ethical vulnerability of aligned LLMs by proving that current alignment methods yield only local "safety regions" in the knowledge manifold. In contrast, pretrained knowledge remains globally connected to harmful concepts via high-likelihood adversarial trajectories. Building on this theoretical insight, we empirically validate our findings by employing semantic coherence inducement under distributional shifts--a method that systematically bypasses alignment constraints through optimized adversarial prompts. This combined theoretical and empirical approach achieves a 100% attack success rate across 19 out of 23 state-of-the-art aligned LLMs, including DeepSeek-R1 and LLaMA-3, revealing their universal vulnerabilities.
Authors: Raffi Khatchadourian, Tatiana Castro V\'elez, Mehdi Bagherzadeh, Nan Jia, Anita Raja
Abstract: Efficiency is essential to support ever-growing datasets, especially for Deep Learning (DL) systems. DL frameworks have traditionally embraced deferred execution-style DL code -- supporting symbolic, graph-based Deep Neural Network (DNN) computation. While scalable, such development is error-prone, non-intuitive, and difficult to debug. Consequently, more natural, imperative DL frameworks encouraging eager execution have emerged but at the expense of run-time performance. Though hybrid approaches aim for the "best of both worlds," using them effectively requires subtle considerations. Our key insight is that, while DL programs typically execute sequentially, hybridizing imperative DL code resembles parallelizing sequential code in traditional systems. Inspired by this, we present an automated refactoring approach that assists developers in determining which otherwise eagerly-executed imperative DL functions could be effectively and efficiently executed as graphs. The approach features novel static imperative tensor and side-effect analyses for Python. Due to its inherent dynamism, analyzing Python may be unsound; however, the conservative approach leverages a speculative (keyword-based) analysis for resolving difficult cases that informs developers of any assumptions made. The approach is: (i) implemented as a plug-in to the PyDev Eclipse IDE that integrates the WALA Ariadne analysis framework and (ii) evaluated on nineteen DL projects consisting of 132 KLOC. The results show that 326 of 766 candidate functions (42.56%) were refactorable, and an average relative speedup of 2.16 on performance tests was observed with negligible differences in model accuracy. The results indicate that the approach is useful in optimizing imperative DL code to its full potential.
Authors: Lucia Celli, Giovanni Peccati
Abstract: Using entropic inequalities from information theory, we provide new bounds on the total variation and 2-Wasserstein distances between a conditionally Gaussian law and a Gaussian law with invertible covariance matrix. We apply our results to quantify the speed of convergence to Gaussian of a randomly initialized fully connected neural network and its derivatives - evaluated in a finite number of inputs - when the initialization is Gaussian and the sizes of the inner layers diverge to infinity. Our results require mild assumptions on the activation function, and allow one to recover optimal rates of convergence in a variety of distances, thus improving and extending the findings of Basteri and Trevisan (2023), Favaro et al. (2023), Trevisan (2024) and Apollonio et al. (2024). One of our main tools are the quantitative cumulant estimates established in Hanin (2024). As an illustration, we apply our results to bound the total variation distance between the Bayesian posterior law of the neural network and its derivatives, and the posterior law of the corresponding Gaussian limit: this yields quantitative versions of a posterior CLT by Hron et al. (2022), and extends several estimates by Trevisan (2024) to the total variation metric.
Authors: Junming Liu, Guosun Zeng, Ding Wang, Yanting Gao, Yufei Jin
Abstract: Multimodal data are often incomplete and exhibit Non-Independent and Identically Distributed (Non-IID) characteristics in real-world scenarios. These inherent limitations lead to both modality heterogeneity through partial modality absence and data heterogeneity from distribution divergence, creating fundamental challenges for effective federated learning (FL). To address these coupled challenges, we propose FedRecon, the first method targeting simultaneous missing modality reconstruction and Non-IID adaptation in multimodal FL. Our approach first employs a lightweight Multimodal Variational Autoencoder (MVAE) to reconstruct missing modalities while preserving cross-modal consistency. Distinct from conventional imputation methods, we achieve sample-level alignment through a novel distribution mapping mechanism that guarantees both data consistency and completeness. Additionally, we introduce a strategy employing global generator freezing to prevent catastrophic forgetting, which in turn mitigates Non-IID fluctuations. Extensive evaluations on multimodal datasets demonstrate FedRecon's superior performance in modality reconstruction under Non-IID conditions, surpassing state-of-the-art methods.
Authors: Hossein Ahmadi, Sajjad Emdadi Mahdimahalleh, Arman Farahat, Banafsheh Saffari
Abstract: The rapid growth of unlabeled time-series data in domains such as wireless communications, radar, biomedical engineering, and the Internet of Things (IoT) has driven advancements in unsupervised learning. This review synthesizes recent progress in applying autoencoders and vision transformers for unsupervised signal analysis, focusing on their architectures, applications, and emerging trends. We explore how these models enable feature extraction, anomaly detection, and classification across diverse signal types, including electrocardiograms, radar waveforms, and IoT sensor data. The review highlights the strengths of hybrid architectures and self-supervised learning, while identifying challenges in interpretability, scalability, and domain generalization. By bridging methodological innovations and practical applications, this work offers a roadmap for developing robust, adaptive models for signal intelligence.
Authors: Sonia Joseph, Praneet Suresh, Lorenz Hufe, Edward Stevinson, Robert Graham, Yash Vadi, Danilo Bzdok, Sebastian Lapuschkin, Lee Sharkey, Blake Aaron Richards
Abstract: Robust tooling and publicly available pre-trained models have helped drive recent advances in mechanistic interpretability for language models. However, similar progress in vision mechanistic interpretability has been hindered by the lack of accessible frameworks and pre-trained weights. We present Prisma (Access the codebase here: https://github.com/Prisma-Multimodal/ViT-Prisma), an open-source framework designed to accelerate vision mechanistic interpretability research, providing a unified toolkit for accessing 75+ vision and video transformers; support for sparse autoencoder (SAE), transcoder, and crosscoder training; a suite of 80+ pre-trained SAE weights; activation caching, circuit analysis tools, and visualization tools; and educational resources. Our analysis reveals surprising findings, including that effective vision SAEs can exhibit substantially lower sparsity patterns than language SAEs, and that in some instances, SAE reconstructions can decrease model loss. Prisma enables new research directions for understanding vision model internals while lowering barriers to entry in this emerging field.
Authors: Griffin Pitts, Viktoria Marcus, Sanaz Motamedi
Abstract: The use of chatbots equipped with artificial intelligence (AI) in educational settings has increased in recent years, showing potential to support teaching and learning. However, the adoption of these technologies has raised concerns about their impact on academic integrity, students' ability to problem-solve independently, and potential underlying biases. To better understand students' perspectives and experiences with these tools, a survey was conducted at a large public university in the United States. Through thematic analysis, 262 undergraduate students' responses regarding their perceived benefits and risks of AI chatbots in education were identified and categorized into themes. The results discuss several benefits identified by the students, with feedback and study support, instruction capabilities, and access to information being the most cited. Their primary concerns included risks to academic integrity, accuracy of information, loss of critical thinking skills, the potential development of overreliance, and ethical considerations such as data privacy, system bias, environmental impact, and preservation of human elements in education. While student perceptions align with previously discussed benefits and risks of AI in education, they show heightened concerns about distinguishing between human and AI generated work - particularly in cases where authentic work is flagged as AI-generated. To address students' concerns, institutions can establish clear policies regarding AI use and develop curriculum around AI literacy. With these in place, practitioners can effectively develop and implement educational systems that leverage AI's potential in areas such as immediate feedback and personalized learning support. This approach can enhance the quality of students' educational experiences while preserving the integrity of the learning process with AI.
Authors: Haoming Yang, Ke Ma, Xiaojun Jia, Yingfei Sun, Qianqian Xu, Qingming Huang
Abstract: Despite the remarkable performance of Large Language Models (LLMs), they remain vulnerable to jailbreak attacks, which can compromise their safety mechanisms. Existing studies often rely on brute-force optimization or manual design, failing to uncover potential risks in real-world scenarios. To address this, we propose a novel jailbreak attack framework, ICRT, inspired by heuristics and biases in human cognition. Leveraging the simplicity effect, we employ cognitive decomposition to reduce the complexity of malicious prompts. Simultaneously, relevance bias is utilized to reorganize prompts, enhancing semantic alignment and inducing harmful outputs effectively. Furthermore, we introduce a ranking-based harmfulness evaluation metric that surpasses the traditional binary success-or-failure paradigm by employing ranking aggregation methods such as Elo, HodgeRank, and Rank Centrality to comprehensively quantify the harmfulness of generated content. Experimental results show that our approach consistently bypasses mainstream LLMs' safety mechanisms and generates high-risk content, providing insights into jailbreak attack risks and contributing to stronger defense strategies.
Authors: Lang Feng, Weihao Tan, Zhiyi Lyu, Longtao Zheng, Haiyang Xu, Ming Yan, Fei Huang, Bo An
Abstract: Online fine-tuning vision-language model (VLM) agents with reinforcement learning (RL) has shown promise for equipping agents with multi-step, goal-oriented capabilities in dynamic environments. However, their open-ended textual action space and non-end-to-end nature of action generation present significant challenges to effective online exploration in RL, e.g., explosion of the exploration space. We propose a novel online fine-tuning method, Counterfactual Soft Reinforcement Learning (CoSo), better suited to the textual output space of VLM agents. Compared to prior methods that assign uniform uncertainty to all tokens, CoSo leverages counterfactual reasoning to dynamically assess the causal influence of individual tokens on post-processed actions. By prioritizing the exploration of action-critical tokens while reducing the impact of semantically redundant or low-impact tokens, CoSo enables a more targeted and efficient online rollout process. We provide theoretical analysis proving CoSo's convergence and policy improvement guarantees, and extensive empirical evaluations supporting CoSo's effectiveness. Our results across a diverse set of agent tasks, including Android device control, card gaming, and embodied AI, highlight its remarkable ability to enhance exploration efficiency and deliver consistent performance gains. The code is available at https://github.com/langfengQ/CoSo.
Authors: Hail Song, Wonsik Shin, Naeun Lee, Soomin Chung, Nojun Kwak, Woontack Woo
Abstract: Generating high-quality 3D models from 2D sketches is a challenging task due to the inherent ambiguity and sparsity of sketch data. In this paper, we present S3D, a novel framework that converts simple hand-drawn sketches into detailed 3D models. Our method utilizes a U-Net-based encoder-decoder architecture to convert sketches into face segmentation masks, which are then used to generate a 3D representation that can be rendered from novel views. To ensure robust consistency between the sketch domain and the 3D output, we introduce a novel style-alignment loss that aligns the U-Net bottleneck features with the initial encoder outputs of the 3D generation module, significantly enhancing reconstruction fidelity. To further enhance the network's robustness, we apply augmentation techniques to the sketch dataset. This streamlined framework demonstrates the effectiveness of S3D in generating high-quality 3D models from sketch inputs. The source code for this project is publicly available at https://github.com/hailsong/S3D.
Authors: Brian Choi, Shu Wang, Isabelle Choi, Kun Sun
Abstract: With the widespread deployment of deep neural network (DNN) models, dynamic watermarking techniques are being used to protect the intellectual property of model owners. However, recent studies have shown that existing watermarking schemes are vulnerable to watermark removal and ambiguity attacks. Besides, the vague criteria for determining watermark presence further increase the likelihood of such attacks. In this paper, we propose a secure DNN watermarking scheme named ChainMarks, which generates secure and robust watermarks by introducing a cryptographic chain into the trigger inputs and utilizes a two-phase Monte Carlo method for determining watermark presence. First, ChainMarks generates trigger inputs as a watermark dataset by repeatedly applying a hash function over a secret key, where the target labels associated with trigger inputs are generated from the digital signature of model owner. Then, the watermarked model is produced by training a DNN over both the original and watermark datasets. To verify watermarks, we compare the predicted labels of trigger inputs with the target labels and determine ownership with a more accurate decision threshold that considers the classification probability of specific models. Experimental results show that ChainMarks exhibits higher levels of robustness and security compared to state-of-the-art watermarking schemes. With a better marginal utility, ChainMarks provides a higher probability guarantee of watermark presence in DNN models with the same level of watermark accuracy.
Authors: Ryan Lagasse, Aidan Kierans, Avijit Ghosh, Shiri Dori-Hacohen
Abstract: We introduce a scaling law for fine-tuning large language models (LLMs) under fixed compute budgets that explicitly accounts for data composition. Conventional approaches measure training data solely by total tokens, yet the number of examples and their average token length -- what we term \emph{dataset volume} -- play a decisive role in model performance. Our formulation is tuned following established procedures. Experiments on the BRICC dataset \cite{salavati2024reducing} and subsets of the MMLU dataset \cite{hendrycks2021measuringmassivemultitasklanguage}, evaluated under multiple subsampling strategies, reveal that data composition significantly affects token efficiency. These results motivate refined scaling laws for practical LLM fine-tuning in resource-constrained settings.
Authors: Zara Siddique, Liam D. Turner, Luis Espinosa-Anke
Abstract: We introduce Dialz, a framework for advancing research on steering vectors for open-source LLMs, implemented in Python. Steering vectors allow users to modify activations at inference time to amplify or weaken a 'concept', e.g. honesty or positivity, providing a more powerful alternative to prompting or fine-tuning. Dialz supports a diverse set of tasks, including creating contrastive pair datasets, computing and applying steering vectors, and visualizations. Unlike existing libraries, Dialz emphasizes modularity and usability, enabling both rapid prototyping and in-depth analysis. We demonstrate how Dialz can be used to reduce harmful outputs such as stereotypes, while also providing insights into model behaviour across different layers. We release Dialz with full documentation, tutorials, and support for popular open-source models to encourage further research in safe and controllable language generation. Dialz enables faster research cycles and facilitates insights into model interpretability, paving the way for safer, more transparent, and more reliable AI systems.
Authors: Yuhao Li, Ling Luo, Uwe Aickelin
Abstract: Medical research, particularly in predicting patient outcomes, heavily relies on medical time series data extracted from Electronic Health Records (EHR), which provide extensive information on patient histories. Despite rigorous examination, labeling errors are inevitable and can significantly impede accurate predictions of patient outcome. To address this challenge, we propose an \textbf{A}ttention-based Learning Framework with Dynamic \textbf{C}alibration and Augmentation for \textbf{T}ime series Noisy \textbf{L}abel \textbf{L}earning (ACTLL). This framework leverages a two-component Beta mixture model to identify the certain and uncertain sets of instances based on the fitness distribution of each class, and it captures global temporal dynamics while dynamically calibrating labels from the uncertain set or augmenting confident instances from the certain set. Experimental results on large-scale EHR datasets eICU and MIMIC-IV-ED, and several benchmark datasets from the UCR and UEA repositories, demonstrate that our model ACTLL has achieved state-of-the-art performance, especially under high noise levels.
Authors: Reece O'Mahoney, Wanming Yu, Ioannis Havoutis
Abstract: Generative models have shown great promise as trajectory planners, given their affinity to modeling complex distributions and guidable inference process. Previous works have successfully applied these in the context of robotic manipulation but perform poorly when the required solution does not exist as a complete trajectory within the training set. We identify that this is a result of being unable to plan via stitching, and subsequently address the architectural and dataset choices needed to remedy this. On top of this, we propose a novel addition to the training and inference procedures to both stabilize and enhance these capabilities. We demonstrate the efficacy of our approach by generating plans with out of distribution boundary conditions and performing obstacle avoidance on the Franka Panda in simulation and on real hardware. In both of these tasks our method performs significantly better than the baselines and is able to avoid obstacles up to four times as large.
Authors: Kirill Vasilevski, Benjamin Rombaut, Gopi Krishnan Rajbahadur, Gustavo A. Oliva, Keheliya Gallaba, Filipe R. Cogo, Jiahuei Lin, Dayi Lin, Haoxiang Zhang, Bouyan Chen, Kishanthan Thangarajah, Ahmed E. Hassan, Zhen Ming Jiang
Abstract: Foundation Models (FMs) such as Large Language Models (LLMs) are reshaping the software industry by enabling FMware, systems that integrate these FMs as core components. In this KDD 2025 tutorial, we present a comprehensive exploration of FMware that combines a curated catalogue of challenges with real-world production concerns. We first discuss the state of research and practice in building FMware. We further examine the difficulties in selecting suitable models, aligning high-quality domain-specific data, engineering robust prompts, and orchestrating autonomous agents. We then address the complex journey from impressive demos to production-ready systems by outlining issues in system testing, optimization, deployment, and integration with legacy software. Drawing on our industrial experience and recent research in the area, we provide actionable insights and a technology roadmap for overcoming these challenges. Attendees will gain practical strategies to enable the creation of trustworthy FMware in the evolving technology landscape.
Authors: Lee Harris
Abstract: Classification is a common AI problem, and vector search is a typical solution. This transforms a given body of text into a numerical representation, known as an embedding, and modern improvements to vector search focus on optimising speed and predictive accuracy. This is often achieved through neural methods that aim to learn language semantics. However, our results suggest that these are not always the best solution. Our task was to classify rigidly-structured medical documents according to their content, and we found that using off-the-shelf semantic vector search produced slightly worse predictive accuracy than creating a bespoke lexical vector search model, and that it required significantly more time to execute. These findings suggest that traditional methods deserve to be contenders in the information retrieval toolkit, despite the prevalence and success of neural models.
Authors: Jianfeng Xu
Abstract: [Objective] This study focuses on addressing the current lack of a unified formal theoretical framework in machine learning, as well as the deficiencies in interpretability and ethical safety assurance. [Methods] A formal information model is first constructed, utilizing sets of well-formed formulas to explicitly define the ontological states and carrier mappings of typical components in machine learning. Learnable and processable predicates, along with learning and processing functions, are introduced to analyze the logical deduction and constraint rules of the causal chains within models. [Results] A meta-framework for machine learning theory (MLT-MF) is established. Based on this framework, universal definitions for model interpretability and ethical safety are proposed. Furthermore, three key theorems are proved: the equivalence of model interpretability and information recoverability, the assurance of ethical safety, and the estimation of generalization error. [Limitations] The current framework assumes ideal conditions with noiseless information-enabling mappings and primarily targets model learning and processing logic in static scenarios. It does not yet address information fusion and conflict resolution across ontological spaces in multimodal or multi-agent systems. [Conclusions] This work overcomes the limitations of fragmented research and provides a unified theoretical foundation for systematically addressing the critical challenges currently faced in machine learning.
Authors: Qiongqiong Wang, Hardik B. Sailor, Tianchi Liu, Ai Ti Aw
Abstract: Current speech-LLMs exhibit limited capability in contextual reasoning alongside paralinguistic understanding, primarily due to the lack of Question-Answer (QA) datasets that cover both aspects. We propose a novel framework for dataset generation from in-the-wild speech data, that integrates contextual reasoning with paralinguistic information. It consists of a pseudo paralinguistic label-based data condensation of in-the-wild speech and LLM-based Contextual Paralinguistic QA (CPQA) generation. The effectiveness is validated by a strong correlation in evaluations of the Qwen2-Audio-7B-Instruct model on a dataset created by our framework and human-generated CPQA dataset. The results also reveal the speech-LLM's limitations in handling empathetic reasoning tasks, highlighting the need for such datasets and more robust models. The proposed framework is first of its kind and has potential in training more robust speech-LLMs with paralinguistic reasoning capabilities.
Authors: Zijia Liu, Peixuan Han, Haofei Yu, Haoru Li, Jiaxuan You
Abstract: Large Language Models (LLMs) demonstrate impressive capabilities but lack robust temporal intelligence, struggling to integrate reasoning about the past with predictions and plausible generations of the future. Meanwhile, existing methods typically target isolated temporal skills, such as question answering about past events or basic forecasting, and exhibit poor generalization, particularly when dealing with events beyond their knowledge cutoff or requiring creative foresight. To address these limitations, we introduce \textit{Time-R1}, the first framework to endow a moderate-sized (3B-parameter) LLM with comprehensive temporal abilities: understanding, prediction, and creative generation. Our approach features a novel three-stage development path; the first two constitute a \textit{reinforcement learning (RL) curriculum} driven by a meticulously designed dynamic rule-based reward system. This framework progressively builds (1) foundational temporal understanding and logical event-time mappings from historical data, (2) future event prediction skills for events beyond its knowledge cutoff, and finally (3) enables remarkable generalization to creative future scenario generation without any fine-tuning. Strikingly, experiments demonstrate that Time-R1 outperforms models over 200 times larger, including the state-of-the-art 671B DeepSeek-R1, on highly challenging future event prediction and creative scenario generation benchmarks. This work provides strong evidence that thoughtfully engineered, progressive RL fine-tuning allows smaller, efficient models to achieve superior temporal performance, offering a practical and scalable path towards truly time-aware AI. To foster further research, we also release \textit{Time-Bench}, a large-scale multi-task temporal reasoning dataset derived from 10 years of news data, and our series of \textit{Time-R1} checkpoints.
Authors: Zhipei Xu, Xuanyu Zhang, Xing Zhou, Jian Zhang
Abstract: The rapid advancement of Artificial Intelligence Generated Content (AIGC) technologies, particularly in video generation, has led to unprecedented creative capabilities but also increased threats to information integrity, identity security, and public trust. Existing detection methods, while effective in general scenarios, lack robust solutions for human-centric videos, which pose greater risks due to their realism and potential for legal and ethical misuse. Moreover, current detection approaches often suffer from poor generalization, limited scalability, and reliance on labor-intensive supervised fine-tuning. To address these challenges, we propose AvatarShield, the first interpretable MLLM-based framework for detecting human-centric fake videos, enhanced via Group Relative Policy Optimization (GRPO). Through our carefully designed accuracy detection reward and temporal compensation reward, it effectively avoids the use of high-cost text annotation data, enabling precise temporal modeling and forgery detection. Meanwhile, we design a dual-encoder architecture, combining high-level semantic reasoning and low-level artifact amplification to guide MLLMs in effective forgery detection. We further collect FakeHumanVid, a large-scale human-centric video benchmark that includes synthesis methods guided by pose, audio, and text inputs, enabling rigorous evaluation of detection methods in real-world scenes. Extensive experiments show that AvatarShield significantly outperforms existing approaches in both in-domain and cross-domain detection, setting a new standard for human-centric video forensics.
Authors: Zijian Lin, Yang Zhang, Yougen Yuan, Yuming Yan, Jinjiang Liu, Zhiyong Wu, Pengfei Hu, Qun Yu
Abstract: Modern autoregressive speech synthesis models leveraging language models have demonstrated remarkable performance. However, the sequential nature of next token prediction in these models leads to significant latency, hindering their deployment in scenarios where inference speed is critical. In this work, we propose Speech Speculative Decoding (SSD), a novel framework for autoregressive speech synthesis acceleration. Specifically, our method employs a lightweight draft model to generate candidate token sequences, which are subsequently verified in parallel by the target model using the proposed SSD framework. Experimental results demonstrate that SSD achieves a significant speedup of 1.4x compared with conventional autoregressive decoding, while maintaining high fidelity and naturalness. Subjective evaluations further validate the effectiveness of SSD in preserving the perceptual quality of the target model while accelerating inference.
Authors: Mohamed S. Elmahdy, Marius Staring, Patrick J. H. de Koning, Samer Alabed, Mahan Salehi, Faisal Alandejani, Michael Sharkey, Ziad Aldabbagh, Andrew J. Swift, Rob J. van der Geest
Abstract: Accurate and efficient quantification of cardiac function is essential for the estimation of prognosis of cardiovascular diseases (CVDs). One of the most commonly used metrics for evaluating cardiac pumping performance is left ventricular ejection fraction (LVEF). However, LVEF can be affected by factors such as inter-observer variability and varying pre-load and after-load conditions, which can reduce its reproducibility. Additionally, cardiac dysfunction may not always manifest as alterations in LVEF, such as in heart failure and cardiotoxicity diseases. An alternative measure that can provide a relatively load-independent quantitative assessment of myocardial contractility is myocardial strain and strain rate. By using LVEF in combination with myocardial strain, it is possible to obtain a thorough description of cardiac function. Automated estimation of LVEF and other volumetric measures from cine-MRI sequences can be achieved through segmentation models, while strain calculation requires the estimation of tissue displacement between sequential frames, which can be accomplished using registration models. These tasks are often performed separately, potentially limiting the assessment of cardiac function. To address this issue, in this study we propose an end-to-end deep learning (DL) model that jointly estimates groupwise (GW) registration and segmentation for cardiac cine-MRI images. The proposed anatomically-guided Deep GW network was trained and validated on a large dataset of 4-chamber view cine-MRI image series of 374 subjects. A quantitative comparison with conventional GW registration using elastix and two DL-based methods showed that the proposed model improved performance and substantially reduced computation time.
Authors: Jiaxin Liu, Jia Wang, Saihui Hou, Min Ren, Huijia Wu, Long Ma, Renwang Pei, Zhaofeng He
Abstract: In recent years, the explosive advancement of deepfake technology has posed a critical and escalating threat to public security: diffusion-based digital human generation. Unlike traditional face manipulation methods, such models can generate highly realistic videos with consistency via multimodal control signals. Their flexibility and covertness pose severe challenges to existing detection strategies. To bridge this gap, we introduce DigiFakeAV, the new large-scale multimodal digital human forgery dataset based on diffusion models. Leveraging five of the latest digital human generation methods and a voice cloning method, we systematically construct a dataset comprising 60,000 videos (8.4 million frames), covering multiple nationalities, skin tones, genders, and real-world scenarios, significantly enhancing data diversity and realism. User studies demonstrate that the misrecognition rate by participants for DigiFakeAV reaches as high as 68%. Moreover, the substantial performance degradation of existing detection models on our dataset further highlights its challenges. To address this problem, we propose DigiShield, an effective detection baseline based on spatiotemporal and cross-modal fusion. By jointly modeling the 3D spatiotemporal features of videos and the semantic-acoustic features of audio, DigiShield achieves state-of-the-art (SOTA) performance on the DigiFakeAV and shows strong generalization on other datasets.
Authors: Liu Chang, Wang Dongbo, Liu liu, Zhao Zhixiao
Abstract: This study addresses the challenges in intelligent processing of Chinese ancient mathematical classics by constructing Guji_MATH, a benchmark for evaluating classical texts based on Suanjing Shishu. It systematically assesses the mathematical problem-solving capabilities of mainstream reasoning models under the unique linguistic constraints of classical Chinese. Through machine-assisted annotation and manual verification, 538 mathematical problems were extracted from 8 canonical texts, forming a structured dataset centered on the "Question-Answer-Solution" framework, supplemented by problem types and difficulty levels. Dual evaluation modes--closed-book (autonomous problem-solving) and open-book (reproducing classical solution methods)--were designed to evaluate the performance of six reasoning models on ancient Chinese mathematical problems. Results indicate that reasoning models can partially comprehend and solve these problems, yet their overall performance remains inferior to benchmarks on modern mathematical tasks. Enhancing models' classical Chinese comprehension and cultural knowledge should be prioritized for optimization. This study provides methodological support for mining mathematical knowledge from ancient texts and disseminating traditional culture, while offering new perspectives for evaluating cross-linguistic and cross-cultural capabilities of reasoning models.
Authors: Noah Amsel, David Persson, Christopher Musco, Robert M. Gower
Abstract: Computing the polar decomposition and the related matrix sign function, has been a well-studied problem in numerical analysis for decades. More recently, it has emerged as an important subroutine in deep learning, particularly within the Muon optimization framework. However, the requirements in this setting differ significantly from those of traditional numerical analysis. In deep learning, methods must be highly efficient and GPU-compatible, but high accuracy is often unnecessary. As a result, classical algorithms like Newton-Schulz (which suffers from slow initial convergence) and methods based on rational functions (which rely on QR decompositions or matrix inverses) are poorly suited to this context. In this work, we introduce Polar Express, a GPU-friendly algorithm for computing the polar decomposition. Like classical polynomial methods such as Newton-Schulz, our approach uses only matrix-matrix multiplications, making it GPU-compatible. Motivated by earlier work of Chen & Chow and Nakatsukasa & Freund, Polar Express adapts the polynomial update rule at each iteration by solving a minimax optimization problem, and we prove that it enjoys a strong worst-case optimality guarantee. This property ensures both rapid early convergence and fast asymptotic convergence. We also address finite-precision issues, making it stable in bfloat16 in practice. We apply Polar Express within the Muon optimization framework and show consistent improvements in validation loss on large-scale models such as GPT-2, outperforming recent alternatives across a range of learning rates.
Authors: Chaochen Gao, Xing Wu, Zijia Lin, Debing Zhang, Songlin Hu
Abstract: High-quality long-context instruction data is essential for aligning long-context large language models (LLMs). Despite the public release of models like Qwen and Llama, their long-context instruction data remains proprietary. Human annotation is costly and challenging, while template-based synthesis methods limit scale, diversity, and quality. We introduce LongMagpie, a self-synthesis framework that automatically generates large-scale long-context instruction data. Our key insight is that aligned long-context LLMs, when presented with a document followed by special tokens preceding a user turn, auto-regressively generate contextually relevant queries. By harvesting these document-query pairs and the model's responses, LongMagpie produces high-quality instructions without human effort. Experiments on HELMET, RULER, and Longbench v2 demonstrate that LongMagpie achieves leading performance on long-context tasks while maintaining competitive performance on short-context tasks, establishing it as a simple and effective approach for open, diverse, and scalable long-context instruction data synthesis.
Authors: Shaghayegh Hosseinpour, Sanchari Das
Abstract: Smishing, or SMS-based phishing, poses an increasing threat to mobile users by mimicking legitimate communications through culturally adapted, concise, and deceptive messages, which can result in the loss of sensitive data or financial resources. In such, we present a multi-channel smishing detection model that combines country-specific semantic tagging, structural pattern tagging, character-level stylistic cues, and contextual phrase embeddings. We curated and relabeled over 84,000 messages across five datasets, including 24,086 smishing samples. Our unified architecture achieves 97.89% accuracy, an F1 score of 0.963, and an AUC of 99.73%, outperforming single-stream models by capturing diverse linguistic and structural cues. This work demonstrates the effectiveness of multi-signal learning in robust and region-aware phishing.
Authors: Xiaojun Guo, Ang Li, Yifei Wang, Stefanie Jegelka, Yisen Wang
Abstract: Although Large Language Models (LLMs) have demonstrated remarkable progress, their proficiency in graph-related tasks remains notably limited, hindering the development of truly general-purpose models. Previous attempts, including pretraining graph foundation models or employing supervised fine-tuning, often face challenges such as the scarcity of large-scale, universally represented graph data. We introduce G1, a simple yet effective approach demonstrating that Reinforcement Learning (RL) on synthetic graph-theoretic tasks can significantly scale LLMs' graph reasoning abilities. To enable RL training, we curate Erd\~os, the largest graph reasoning dataset to date comprising 50 diverse graph-theoretic tasks of varying difficulty levels, 100k training data and 5k test data, all drived from real-world graphs. With RL on Erd\~os, G1 obtains substantial improvements in graph reasoning, where our finetuned 3B model even outperforms Qwen2.5-72B-Instruct (24x size). RL-trained models also show strong zero-shot generalization to unseen tasks, domains, and graph encoding schemes, including other graph-theoretic benchmarks as well as real-world node classification and link prediction tasks, without compromising general reasoning abilities. Our findings offer an efficient, scalable path for building strong graph reasoners by finetuning LLMs with RL on graph-theoretic tasks, which combines the strengths of pretrained LLM capabilities with abundant, automatically generated synthetic data, suggesting that LLMs possess graph understanding abilities that RL can elicit successfully. Our implementation is open-sourced at https://github.com/PKU-ML/G1, with models and datasets hosted on Hugging Face collections https://huggingface.co/collections/PKU-ML/g1-683d659e992794fc99618cf2 for broader accessibility.
URLs: https://github.com/PKU-ML/G1,, https://huggingface.co/collections/PKU-ML/g1-683d659e992794fc99618cf2
Authors: Minzhi Lin, Tianchi Xie, Mengchen Liu, Yilin Ye, Changjian Chen, Shixia Liu
Abstract: Understanding infographic charts with design-driven visual elements (e.g., pictograms, icons) requires both visual recognition and reasoning, posing challenges for multimodal large language models (MLLMs). However, existing visual-question answering benchmarks fall short in evaluating these capabilities of MLLMs due to the lack of paired plain charts and visual-element-based questions. To bridge this gap, we introduce InfoChartQA, a benchmark for evaluating MLLMs on infographic chart understanding. It includes 5,642 pairs of infographic and plain charts, each sharing the same underlying data but differing in visual presentations. We further design visual-element-based questions to capture their unique visual designs and communicative intent. Evaluation of 20 MLLMs reveals a substantial performance decline on infographic charts, particularly for visual-element-based questions related to metaphors. The paired infographic and plain charts enable fine-grained error analysis and ablation studies, which highlight new opportunities for advancing MLLMs in infographic chart understanding. We release InfoChartQA at https://github.com/CoolDawnAnt/InfoChartQA.
Authors: Shenghai Yuan, Xianyi He, Yufan Deng, Yang Ye, Jinfa Huang, Bin Lin, Jiebo Luo, Li Yuan
Abstract: Subject-to-Video (S2V) generation aims to create videos that faithfully incorporate reference content, providing enhanced flexibility in the production of videos. To establish the infrastructure for S2V generation, we propose OpenS2V-Nexus, consisting of (i) OpenS2V-Eval, a fine-grained benchmark, and (ii) OpenS2V-5M, a million-scale dataset. In contrast to existing S2V benchmarks inherited from VBench that focus on global and coarse-grained assessment of generated videos, OpenS2V-Eval focuses on the model's ability to generate subject-consistent videos with natural subject appearance and identity fidelity. For these purposes, OpenS2V-Eval introduces 180 prompts from seven major categories of S2V, which incorporate both real and synthetic test data. Furthermore, to accurately align human preferences with S2V benchmarks, we propose three automatic metrics, NexusScore, NaturalScore and GmeScore, to separately quantify subject consistency, naturalness, and text relevance in generated videos. Building on this, we conduct a comprehensive evaluation of 18 representative S2V models, highlighting their strengths and weaknesses across different content. Moreover, we create the first open-source large-scale S2V generation dataset OpenS2V-5M, which consists of five million high-quality 720P subject-text-video triples. Specifically, we ensure subject-information diversity in our dataset by (1) segmenting subjects and building pairing information via cross-video associations and (2) prompting GPT-Image-1 on raw frames to synthesize multi-view representations. Through OpenS2V-Nexus, we deliver a robust infrastructure to accelerate future S2V generation research.
Authors: Mengru Wang, Ziwen Xu, Shengyu Mao, Shumin Deng, Zhaopeng Tu, Huajun Chen, Ningyu Zhang
Abstract: Precise control over language model generation is vital for ensuring both safety and reliability. Although prompt engineering and steering are commonly used to intervene in model behaviors, the vast number of parameters in models often results in highly intertwined internal representations. This interdependency can limit control precision and sometimes lead to unintended side effects. Recent research has explored the use of sparse autoencoders (SAE) to disentangle knowledge in high-dimensional spaces for steering. However, these applications have been limited to toy tasks owing to the nontrivial issue of locating atomic knowledge components. In this paper, we propose Steering Target Atoms (STA), a novel method that isolates and manipulates disentangled knowledge components to enhance safety. Comprehensive experiments demonstrate the effectiveness of our approach. Further analysis reveals that steering exhibits superior robustness and flexibility, particularly in adversarial scenarios. We also apply the steering strategy to the large reasoning model, confirming its effectiveness in precise reasoning control.
Authors: Zachary C. Brown, David Carlson
Abstract: The field of hypothesis generation promises to reduce costs in neuroscience by narrowing the range of interventional studies needed to study various phenomena. Existing machine learning methods can generate scientific hypotheses from complex datasets, but many approaches assume causal relationships are static over time, limiting their applicability to systems with dynamic, state-dependent behavior, such as the brain. While some techniques attempt dynamic causal discovery through factor models, they often restrict relationships to linear patterns or impose other simplifying assumptions. We propose a novel method that models dynamic graphs as a conditionally weighted superposition of static graphs, where each static graph can capture nonlinear relationships. This approach enables the detection of complex, time-varying interactions between variables beyond linear limitations. Our method improves f1-scores of predicted dynamic causal patterns by roughly 22-28% on average over baselines in some of our experiments, with some improvements reaching well over 60%. A case study on real brain data demonstrates our method's ability to uncover relationships linked to specific behavioral states, offering valuable insights into neural dynamics.
Authors: Qiuchen Wang, Ruixue Ding, Yu Zeng, Zehui Chen, Lin Chen, Shihang Wang, Pengjun Xie, Fei Huang, Feng Zhao
Abstract: Effectively retrieving, reasoning and understanding visually rich information remains a challenge for RAG methods. Traditional text-based methods cannot handle visual-related information. On the other hand, current vision-based RAG approaches are often limited by fixed pipelines and frequently struggle to reason effectively due to the insufficient activation of the fundamental capabilities of models. As RL has been proven to be beneficial for model reasoning, we introduce VRAG-RL, a novel RL framework tailored for complex reasoning across visually rich information. With this framework, VLMs interact with search engines, autonomously sampling single-turn or multi-turn reasoning trajectories with the help of visual perception tokens and undergoing continual optimization based on these samples. Our approach highlights key limitations of RL in RAG domains: (i) Prior Multi-modal RAG approaches tend to merely incorporate images into the context, leading to insufficient reasoning token allocation and neglecting visual-specific perception; and (ii) When models interact with search engines, their queries often fail to retrieve relevant information due to the inability to articulate requirements, thereby leading to suboptimal performance. To address these challenges, we define an action space tailored for visually rich inputs, with actions including cropping and scaling, allowing the model to gather information from a coarse-to-fine perspective. Furthermore, to bridge the gap between users' original inquiries and the retriever, we employ a simple yet effective reward that integrates query rewriting and retrieval performance with a model-based reward. Our VRAG-RL optimizes VLMs for RAG tasks using specially designed RL strategies, aligning the model with real-world applications. The code is available at https://github.com/Alibaba-NLP/VRAG.
Authors: Jintao Zhang, Zirui Liu, Mingyue Cheng, Shilong Zhang, Tingyue Pan, Qi Liu, Yanhu Xie
Abstract: Intraoperative hypotension (IOH) frequently occurs under general anesthesia and is strongly linked to adverse outcomes such as myocardial injury and increased mortality. Despite its significance, IOH prediction is hindered by event sparsity and the challenge of integrating static and dynamic data across diverse patients. In this paper, we propose \textbf{IOHFuseLM}, a multimodal language model framework. To accurately identify and differentiate sparse hypotensive events, we leverage a two-stage training strategy. The first stage involves domain adaptive pretraining on IOH physiological time series augmented through diffusion methods, thereby enhancing the model sensitivity to patterns associated with hypotension. Subsequently, task fine-tuning is performed on the original clinical dataset to further enhance the ability to distinguish normotensive from hypotensive states. To enable multimodal fusion for each patient, we align structured clinical descriptions with the corresponding physiological time series at the token level. Such alignment enables the model to capture individualized temporal patterns alongside their corresponding clinical semantics. In addition, we convert static patient attributes into structured text to enrich personalized information. Experimental evaluations on two intraoperative datasets demonstrate that IOHFuseLM outperforms established baselines in accurately identifying IOH events, highlighting its applicability in clinical decision support scenarios. Our code is publicly available to promote reproducibility at https://github.com/zjt-gpu/IOHFuseLM.
Authors: Alexander Gill, Abhilasha Ravichander, Ana Marasovi\'c
Abstract: Large language models (LLMs) are increasingly used for data generation. However, creating evaluation benchmarks raises the bar for this emerging paradigm. Benchmarks must target specific phenomena, penalize exploiting shortcuts, and be challenging. Through two case studies, we investigate whether LLMs can meet these demands by generating reasoning over-text benchmarks and comparing them to those created through careful crowdsourcing. Specifically, we evaluate both the validity and difficulty of LLM-generated versions of two high-quality reading comprehension datasets: CondaQA, which evaluates reasoning about negation, and DROP, which targets reasoning about quantities. We find that prompting LLMs can produce variants of these datasets that are often valid according to the annotation guidelines, at a fraction of the cost of the original crowdsourcing effort. However, we show that they are less challenging for LLMs than their human-authored counterparts. This finding sheds light on what may have been lost by generating evaluation data with LLMs, and calls for critically reassessing the immediate use of this increasingly prevalent approach to benchmark creation.
Authors: Mingzhe Du, Luu Anh Tuan, Yue Liu, Yuhao Qing, Dong Huang, Xinyi He, Qian Liu, Zejun Ma, See-kiong Ng
Abstract: Large Language Models (LLMs) generate functionally correct solutions but often fall short in code efficiency, a critical bottleneck for real-world deployment. In this paper, we introduce a novel test-time iterative optimization framework to address this, employing a closed-loop system where LLMs iteratively refine code based on empirical performance feedback from an execution sandbox. We explore three training strategies: Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Group Relative Policy Optimization (GRPO). Experiments on our Venus dataset and the APPS benchmark show that SFT and DPO rapidly saturate in efficiency gains. In contrast, GRPO, using reinforcement learning (RL) with execution feedback, continuously optimizes code performance, significantly boosting both pass@1 (from 47% to 62%) and the likelihood of outperforming human submissions in efficiency (from 31% to 45%). Our work demonstrates effective test-time code efficiency improvement and critically reveals the power of RL in teaching LLMs to truly self-improve code efficiency.
Authors: Shibbir Ahmed, Shahnewaz Karim Sakib, Anindya Bijoy Das
Abstract: This study presents a multimodal AI framework designed for precisely classifying medical diagnostic images. Utilizing publicly available datasets, the proposed system compares the strengths of convolutional neural networks (CNNs) and different large language models (LLMs). This in-depth comparative analysis highlights key differences in diagnostic performance, execution efficiency, and environmental impacts. Model evaluation was based on accuracy, F1-score, average execution time, average energy consumption, and estimated $CO_2$ emission. The findings indicate that although CNN-based models can outperform various multimodal techniques that incorporate both images and contextual information, applying additional filtering on top of LLMs can lead to substantial performance gains. These findings highlight the transformative potential of multimodal AI systems to enhance the reliability, efficiency, and scalability of medical diagnostics in clinical settings.
Authors: Peter David Fagan
Abstract: Neural network inference typically operates on raw input data, increasing the risk of exposure during preprocessing and inference. Moreover, neural architectures lack efficient built-in mechanisms for directly authenticating input data. This work introduces a novel encryption method for ensuring the security of neural inference. By constructing key-conditioned chaotic graph dynamical systems, we enable the encryption and decryption of real-valued tensors within the neural architecture. The proposed dynamical systems are particularly suited to encryption due to their sensitivity to initial conditions and their capacity to produce complex, key-dependent nonlinear transformations from compact rules. This work establishes a paradigm for securing neural inference and opens new avenues for research on the application of graph dynamical systems in neural network security.
Authors: Ziyin Zhang, Jiahao Xu, Zhiwei He, Tian Liang, Qiuzhi Liu, Yansi Li, Linfeng Song, Zhenwen Liang, Zhuosheng Zhang, Rui Wang, Zhaopeng Tu, Haitao Mi, Dong Yu
Abstract: Theorem proving serves as a major testbed for evaluating complex reasoning abilities in large language models (LLMs). However, traditional automated theorem proving (ATP) approaches rely heavily on formal proof systems that poorly align with LLMs' strength derived from informal, natural language knowledge acquired during pre-training. In this work, we propose DeepTheorem, a comprehensive informal theorem-proving framework exploiting natural language to enhance LLM mathematical reasoning. DeepTheorem includes a large-scale benchmark dataset consisting of 121K high-quality IMO-level informal theorems and proofs spanning diverse mathematical domains, rigorously annotated for correctness, difficulty, and topic categories, accompanied by systematically constructed verifiable theorem variants. We devise a novel reinforcement learning strategy (RL-Zero) explicitly tailored to informal theorem proving, leveraging the verified theorem variants to incentivize robust mathematical inference. Additionally, we propose comprehensive outcome and process evaluation metrics examining proof correctness and the quality of reasoning steps. Extensive experimental analyses demonstrate DeepTheorem significantly improves LLM theorem-proving performance compared to existing datasets and supervised fine-tuning protocols, achieving state-of-the-art accuracy and reasoning quality. Our findings highlight DeepTheorem's potential to fundamentally advance automated informal theorem proving and mathematical exploration.
Authors: Yuli Chen, Bo Cheng, Jiale Han, Yingying Zhang, Yingting Li, Shuhao Zhang
Abstract: Pruning has recently been widely adopted to reduce the parameter scale and improve the inference efficiency of Large Language Models (LLMs). Mainstream pruning techniques often rely on uniform layerwise pruning strategies, which can lead to severe performance degradation at high sparsity levels. Recognizing the varying contributions of different layers in LLMs, recent studies have shifted their focus toward non-uniform layerwise pruning. However, these approaches often rely on pre-defined values, which can result in suboptimal performance. To overcome these limitations, we propose a novel method called Dynamic Layerwise Pruning (DLP). This approach adaptively determines the relative importance of each layer by integrating model weights with input activation information, assigning pruning rates accordingly. Experimental results show that DLP effectively preserves model performance at high sparsity levels across multiple LLMs. Specifically, at 70% sparsity, DLP reduces the perplexity of LLaMA2-7B by 7.79 and improves the average accuracy by 2.7% compared to state-of-the-art methods. Moreover, DLP is compatible with various existing LLM compression techniques and can be seamlessly integrated into Parameter-Efficient Fine-Tuning (PEFT). We release the code at https://github.com/ironartisan/DLP to facilitate future research.
Authors: Haowei Yang, Haotian Lyu, Tianle Zhang, Dingzhou Wang, Yushang Zhao
Abstract: As e-commerce competition intensifies, balancing creative content with conversion effectiveness becomes critical. Leveraging LLMs' language generation capabilities, we propose a framework that integrates prompt engineering, multi-objective fine-tuning, and post-processing to generate marketing copy that is both engaging and conversion-driven. Our fine-tuning method combines sentiment adjustment, diversity enhancement, and CTA embedding. Through offline evaluations and online A/B tests across categories, our approach achieves a 12.5 % increase in CTR and an 8.3 % increase in CVR while maintaining content novelty. This provides a practical solution for automated copy generation and suggests paths for future multimodal, real-time personalization.
Authors: Zhaokun Wang, Jinyu Guo, Jingwen Pu, Lingfeng Chen, Hongli Pu, Jie Ou, Libo Qin, Wenhong Tian
Abstract: Current parameter-efficient fine-tuning methods for adapting pre-trained language models to downstream tasks are susceptible to interference from noisy data. Conventional noise-handling approaches either rely on laborious data pre-processing or employ model architecture modifications prone to error accumulation. In contrast to existing noise-process paradigms, we propose a noise-robust adaptation method via asymmetric LoRA poisoning experts (LoPE), a novel framework that enhances model robustness to noise only with generated noisy data. Drawing inspiration from the mixture-of-experts architecture, LoPE strategically integrates a dedicated poisoning expert in an asymmetric LoRA configuration. Through a two-stage paradigm, LoPE performs noise injection on the poisoning expert during fine-tuning to enhance its noise discrimination and processing ability. During inference, we selectively mask the dedicated poisoning expert to leverage purified knowledge acquired by normal experts for noise-robust output. Extensive experiments demonstrate that LoPE achieves strong performance and robustness purely through the low-cost noise injection, which completely eliminates the requirement of data cleaning.
Authors: Zefan Cai, Wen Xiao, Hanshi Sun, Cheng Luo, Yikai Zhang, Ke Wan, Yucheng Li, Yeyang Zhou, Li-Wen Chang, Jiuxiang Gu, Zhen Dong, Anima Anandkumar, Abedelkadir Asi, Junjie Hu
Abstract: Reasoning models have demonstrated impressive performance in self-reflection and chain-of-thought reasoning. However, they often produce excessively long outputs, leading to prohibitively large key-value (KV) caches during inference. While chain-of-thought inference significantly improves performance on complex reasoning tasks, it can also lead to reasoning failures when deployed with existing KV cache compression approaches. To address this, we propose Redundancy-aware KV Cache Compression for Reasoning models (R-KV), a novel method specifically targeting redundant tokens in reasoning models. Our method preserves nearly 100% of the full KV cache performance using only 10% of the KV cache, substantially outperforming existing KV cache baselines, which reach only 60% of the performance. Remarkably, R-KV even achieves 105% of full KV cache performance with 16% of the KV cache. This KV-cache reduction also leads to a 90% memory saving and a 6.6X throughput over standard chain-of-thought reasoning inference. Experimental results show that R-KV consistently outperforms existing KV cache compression baselines across two mathematical reasoning datasets.
Authors: Yichen Xie, Runsheng Xu, Tong He, Jyh-Jing Hwang, Katie Luo, Jingwei Ji, Hubert Lin, Letian Chen, Yiren Lu, Zhaoqi Leng, Dragomir Anguelov, Mingxing Tan
Abstract: The latest advancements in multi-modal large language models (MLLMs) have spurred a strong renewed interest in end-to-end motion planning approaches for autonomous driving. Many end-to-end approaches rely on human annotations to learn intermediate perception and prediction tasks, while purely self-supervised approaches--which directly learn from sensor inputs to generate planning trajectories without human annotations often underperform the state of the art. We observe a key gap in the input representation space: end-to-end approaches built on MLLMs are often pretrained with reasoning tasks in 2D image space rather than the native 3D space in which autonomous vehicles plan. To this end, we propose S4-Driver, a scalable self-supervised motion planning algorithm with spatio-temporal visual representation, based on the popular PaLI multimodal large language model. S4-Driver uses a novel sparse volume strategy to seamlessly transform the strong visual representation of MLLMs from perspective view to 3D space without the need to finetune the vision encoder. This representation aggregates multi-view and multi-frame visual inputs and enables better prediction of planning trajectories in 3D space. To validate our method, we run experiments on both nuScenes and Waymo Open Motion Dataset (with in-house camera data). Results show that S4-Driver performs favorably against existing supervised multi-task approaches while requiring no human annotations. It also demonstrates great scalability when pretrained on large volumes of unannotated driving logs.
Authors: Zheng Wang
Abstract: Fine-grained bird image classification (FBIC) is not only of great significance for ecological monitoring and species identification, but also holds broad research value in the fields of image recognition and fine-grained visual modeling. Compared with general image classification tasks, FBIC poses more formidable challenges: 1) the differences in species size and imaging distance result in the varying sizes of birds presented in the images; 2) complex natural habitats often introduce strong background interference; 3) and highly flexible poses such as flying, perching, or foraging result in substantial intra-class variability. These factors collectively make it difficult for traditional methods to stably extract discriminative features, thereby limiting the generalizability and interpretability of models in real-world applications. To address these challenges, this paper proposes a fine-grained bird classification framework based on strip-aware spatial perception, which aims to capture long-range spatial dependencies across entire rows or columns in bird images, thereby enhancing the model's robustness and interpretability. The proposed method incorporates two novel modules: extensional perception aggregator (EPA) and channel semantic weaving (CSW). Specifically, EPA integrates local texture details with global structural cues by aggregating information across horizontal and vertical spatial directions. CSW further refines the semantic representations by adaptively fusing long-range and short-range information along the channel dimension. Built upon a ResNet-50 backbone, the model enables jump-wise connection of extended structural features across the spatial domain. Experimental results on the CUB-200-2011 dataset demonstrate that our framework achieves significant performance improvements while maintaining architectural efficiency.
Authors: Celia Cintas, Miriam Rateike, Erik Miehling, Elizabeth Daly, Skyler Speakman
Abstract: We present a study on how and where personas -- defined by distinct sets of human characteristics, values, and beliefs -- are encoded in the representation space of large language models (LLMs). Using a range of dimension reduction and pattern recognition methods, we first identify the model layers that show the greatest divergence in encoding these representations. We then analyze the activations within a selected layer to examine how specific personas are encoded relative to others, including their shared and distinct embedding spaces. We find that, across multiple pre-trained decoder-only LLMs, the analyzed personas show large differences in representation space only within the final third of the decoder layers. We observe overlapping activations for specific ethical perspectives -- such as moral nihilism and utilitarianism -- suggesting a degree of polysemy. In contrast, political ideologies like conservatism and liberalism appear to be represented in more distinct regions. These findings help to improve our understanding of how LLMs internally represent information and can inform future efforts in refining the modulation of specific human traits in LLM outputs. Warning: This paper includes potentially offensive sample statements.
Authors: Andrei Chernov, Vitaliy Pozdnyakov, Ilya Makarov
Abstract: Recent studies in time series forecasting have explored formulating regression via classification task. By discretizing the continuous target space into bins and predicting over a fixed set of classes, these approaches benefit from stable training, robust uncertainty modeling, and compatibility with modern deep learning architectures. However, most existing methods rely on one-hot encoding that ignores the inherent ordinal structure of the underlying values. As a result, they fail to provide information about the relative distance between predicted and true values during training. In this paper, we propose to address this limitation by introducing binary cumulative encoding (BCE), that represents scalar targets into monotonic binary vectors. This encoding implicitly preserves order and magnitude information, allowing the model to learn distance-aware representations while still operating within a classification framework. We propose a convolutional neural network architecture specifically designed for BCE, incorporating residual and dilated convolutions to enable fast and expressive temporal modeling. Through extensive experiments on benchmark forecasting datasets, we show that our approach outperforms widely used methods in both point and probabilistic forecasting, while requiring fewer parameters and enabling faster training.
Authors: Yuchong Li, Xiaojun Zeng, Chihua Fang, Jian Yang, Fucang Jia, Lei Zhang
Abstract: Hepato-pancreato-biliary (HPB) disorders represent a global public health challenge due to their high morbidity and mortality. Although large language models (LLMs) have shown promising performance in general medical question-answering tasks, the current evaluation benchmarks are mostly derived from standardized examinations or manually designed questions, lacking HPB coverage and clinical cases. To address these issues, we systematically eatablish an HPB disease evaluation benchmark comprising 3,535 closed-ended multiple-choice questions and 337 open-ended real diagnosis cases, which encompasses all the 33 main categories and 465 subcategories of HPB diseases defined in the International Statistical Classification of Diseases, 10th Revision (ICD-10). The multiple-choice questions are curated from public datasets and synthesized data, and the clinical cases are collected from prestigious medical journals, case-sharing platforms, and collaborating hospitals. By evalauting commercial and open-source general and medical LLMs on our established benchmark, namely ClinBench-HBP, we find that while commercial LLMs perform competently on medical exam questions, they exhibit substantial performance degradation on HPB diagnosis tasks, especially on complex, inpatient clinical cases. Those medical LLMs also show limited generalizability to HPB diseases. Our results reveal the critical limitations of current LLMs in the domain of HPB diseases, underscoring the imperative need for future medical LLMs to handle real, complex clinical diagnostics rather than simple medical exam questions. The benchmark will be released at https://clinbench-hpb.github.io.
Authors: Ahmed Elhady, Eneko Agirre, Mikel Artetxe
Abstract: Continued pretraining (CPT) is a popular approach to adapt existing large language models (LLMs) to new languages. When doing so, it is common practice to include a portion of English data in the mixture, but its role has not been carefully studied to date. In this work, we show that including English does not impact validation perplexity, yet it is critical for the emergence of downstream capabilities in the target language. We introduce a language-agnostic benchmark for in-context learning (ICL), which reveals catastrophic forgetting early on CPT when English is not included. This in turn damages the ability of the model to generalize to downstream prompts in the target language as measured by perplexity, even if it does not manifest in terms of accuracy until later in training, and can be tied to a big shift in the model parameters. Based on these insights, we introduce curriculum learning and exponential moving average (EMA) of weights as effective alternatives to mitigate the need for English. All in all, our work sheds light into the dynamics by which emergent abilities arise when doing CPT for language adaptation, and can serve as a foundation to design more effective methods in the future.
Authors: Jun Wu, Yirong Xiong, Jiangtao Wen, Yuxing Han
Abstract: Despite rapid advancements in the research and deployment of large language models (LLMs), the statistical distribution of model parameters, as well as their influence on initialization, training dynamics, and downstream efficiency, has received surprisingly little attention. A recent work introduced BackSlash, a training-time compression algorithm. It first demonstrated that pre-trained LLM parameters follow generalized Gaussian distributions (GGDs) better. By optimizing GG priors during training, BackSlash can reduce parameters by up to 90\% with minimal performance loss. Building on this foundational insight, we propose a unified, end-to-end framework for LLM optimization based on the GG model. Our contributions are threefold: (1) GG-based initialization scheme that aligns with the statistical structure of trained models, resulting in faster convergence and improved accuracy; (2) DeepShape, a post-training regularization method that reshapes weight distributions to match a GG profile, improving compressibility with minimized degradation in performance; and (3) RF8, a compact and hardware-efficient 8-bit floating-point format designed for GG-distributed-initialized BackSlash training, enabling low-cost inference without compromising accuracy. Experiments across diverse model architectures show that our framework consistently yields smaller and faster models that match or outperform standard training baselines. By grounding LLM development in principled statistical modeling, this work forges a new path toward efficient, scalable, and hardware-aware AI systems. The code is available on our project page: https://huggingface.co/spaces/shifeng3711/gg_prior.
Authors: Yang Zheng, Mengqi Huang, Nan Chen, Zhendong Mao
Abstract: Text-guided 3D editing aims to precisely edit semantically relevant local 3D regions, which has significant potential for various practical applications ranging from 3D games to film production. Existing methods typically follow a view-indiscriminate paradigm: editing 2D views indiscriminately and projecting them back into 3D space. However, they overlook the different cross-view interdependencies, resulting in inconsistent multi-view editing. In this study, we argue that ideal consistent 3D editing can be achieved through a \textit{progressive-views paradigm}, which propagates editing semantics from the editing-salient view to other editing-sparse views. Specifically, we propose \textit{Pro3D-Editor}, a novel framework, which mainly includes Primary-view Sampler, Key-view Render, and Full-view Refiner. Primary-view Sampler dynamically samples and edits the most editing-salient view as the primary view. Key-view Render accurately propagates editing semantics from the primary view to other key views through its Mixture-of-View-Experts Low-Rank Adaption (MoVE-LoRA). Full-view Refiner edits and refines the 3D object based on the edited multi-views. Extensive experiments demonstrate that our method outperforms existing methods in editing accuracy and spatial consistency.
Authors: Yuxi Sun, Aoqi Zuo, Wei Gao, Jing Ma
Abstract: Large Language Models (LLMs) often exhibit knowledge disparities across languages. Encouraging LLMs to \textit{abstain} when faced with knowledge gaps is a promising strategy to reduce hallucinations in multilingual settings. Current abstention strategies for multilingual scenarios primarily rely on generating feedback in various languages using LLMs and performing self-reflection. However, these methods can be adversely impacted by inaccuracies and biases in the generated feedback. To address this, from a causal perspective, we introduce \textit{CausalAbstain}, a method that helps LLMs determine whether to utilize multiple generated feedback responses and how to identify the most useful ones. Extensive experiments demonstrate that \textit{CausalAbstain} effectively selects helpful feedback and enhances abstention decisions with interpretability in both native language (\textsc{Casual-native}) and multilingual (\textsc{Causal-multi}) settings, outperforming strong baselines on two benchmark datasets covering encyclopedic and commonsense knowledge QA tasks. Our code and data are open-sourced at https://github.com/peachch/CausalAbstain.
Authors: Haesung Pyun, Yoonah Park, Yohan Jo
Abstract: In dialogue state tracking (DST), in-context learning comprises a retriever that selects labeled dialogues as in-context examples and a DST model that uses these examples to infer the dialogue state of the query dialogue. Existing methods for constructing training data for retrievers suffer from three key limitations: (1) the synergistic effect of examples is not considered, (2) the linguistic characteristics of the query are not sufficiently factored in, and (3) scoring is not directly optimized for DST performance. Consequently, the retriever can fail to retrieve examples that would substantially improve DST performance. To address these issues, we present CombiSearch, a method that scores effective in-context examples based on their combinatorial impact on DST performance. Our evaluation on MultiWOZ shows that retrievers trained with CombiSearch surpass state-of-the-art models, achieving a 20x gain in data efficiency and generalizing well to the SGD dataset. Moreover, CombiSearch attains a 12% absolute improvement in the upper bound DST performance over traditional approaches when no retrieval errors are assumed. This significantly increases the headroom for practical DST performance while demonstrating that existing methods rely on suboptimal data for retriever training.
Authors: Femi Bello, Anubrata Das, Fanzhi Zeng, Fangcong Yin, Liu Leqi
Abstract: It has been hypothesized that neural networks with similar architectures trained on similar data learn shared representations relevant to the learning task. We build on this idea by extending the conceptual framework where representations learned across models trained on the same data can be expressed as linear combinations of a \emph{universal} set of basis features. These basis features underlie the learning task itself and remain consistent across models, regardless of scale. From this framework, we propose the \textbf{Linear Representation Transferability (LRT)} Hypothesis -- that there exists an affine transformation between the representation spaces of different models. To test this hypothesis, we learn affine mappings between the hidden states of models of different sizes and evaluate whether steering vectors -- directions in hidden state space associated with specific model behaviors -- retain their semantic effect when transferred from small to large language models using the learned mappings. We find strong empirical evidence that such affine mappings can preserve steering behaviors. These findings suggest that representations learned by small models can be used to guide the behavior of large models, and that the LRT hypothesis may be a promising direction on understanding representation alignment across model scales.
Authors: Li Zhang, Morgan Gray, Jaromir Savelka, Kevin D. Ashley
Abstract: Large Language Models (LLMs) demonstrate potential in complex legal tasks like argument generation, yet their reliability remains a concern. Building upon pilot work assessing LLM generation of 3-ply legal arguments using human evaluation, this paper introduces an automated pipeline to evaluate LLM performance on this task, specifically focusing on faithfulness (absence of hallucination), factor utilization, and appropriate abstention. We define hallucination as the generation of factors not present in the input case materials and abstention as the model's ability to refrain from generating arguments when instructed and no factual basis exists. Our automated method employs an external LLM to extract factors from generated arguments and compares them against the ground-truth factors provided in the input case triples (current case and two precedent cases). We evaluated eight distinct LLMs on three tests of increasing difficulty: 1) generating a standard 3-ply argument, 2) generating an argument with swapped precedent roles, and 3) recognizing the impossibility of argument generation due to lack of shared factors and abstaining. Our findings indicate that while current LLMs achieve high accuracy (over 90%) in avoiding hallucination on viable argument generation tests (Tests 1 & 2), they often fail to utilize the full set of relevant factors present in the cases. Critically, on the abstention test (Test 3), most models failed to follow instructions to stop, instead generating spurious arguments despite the lack of common factors. This automated pipeline provides a scalable method for assessing these crucial LLM behaviors, highlighting the need for improvements in factor utilization and robust abstention capabilities before reliable deployment in legal settings. Link: https://lizhang-aiandlaw.github.io/An-Automated-Pipeline-for-Evaluating-LLM-Generated-3-ply-Case-Based-Legal-Arguments/
Authors: Keyuan Cheng, Zijian Kan, Zhixian He, Zhuoran Zhang, Muhammad Asif Ali, Ke Xu, Lijie Hu, Di Wang
Abstract: Knowledge Editing, which efficiently modifies the knowledge in large language models, has gathered great attention. Current benchmarks primarily use multi-hop question answering to assess and analyze newly injected or updated knowledge. However, we argue that these benchmarks fail to effectively evaluate how well the updated models apply this knowledge in real-life scenarios, particularly when questions require complex reasoning, involving one-to-many relationships or multi-step logical intersections. To fill in this gap, we introduce a new benchmark, COMPKE: Complex Question Answering under Knowledge Editing, which includes 11,924 complex questions that reflect real-life situations. We conduct an extensive evaluation of four knowledge editing methods on COMPKE, revealing that their effectiveness varies notably across different models. For instance, MeLLo attains an accuracy of 39.47 on GPT-4O-MINI, but this drops sharply to 3.83 on QWEN2.5-3B. We further investigate the underlying causes of these disparities from both methodological and model-specific perspectives. The datasets are available at https://github.com/kzjkzj666/CompKE.
Authors: Sa Zhu, Huashan Chen, Wanqian Zhang, Jinchao Zhang, Zexian Yang, Xiaoshuai Hao, Bo Li
Abstract: Given a text query, partially relevant video retrieval (PRVR) aims to retrieve untrimmed videos containing relevant moments, wherein event modeling is crucial for partitioning the video into smaller temporal events that partially correspond to the text. Previous methods typically segment videos into a fixed number of equal-length clips, resulting in ambiguous event boundaries. Additionally, they rely on mean pooling to compute event representations, inevitably introducing undesired misalignment. To address these, we propose an Uneven Event Modeling (UEM) framework for PRVR. We first introduce the Progressive-Grouped Video Segmentation (PGVS) module, to iteratively formulate events in light of both temporal dependencies and semantic similarity between consecutive frames, enabling clear event boundaries. Furthermore, we also propose the Context-Aware Event Refinement (CAER) module to refine the event representation conditioned the text's cross-attention. This enables event representations to focus on the most relevant frames for a given text, facilitating more precise text-video alignment. Extensive experiments demonstrate that our method achieves state-of-the-art performance on two PRVR benchmarks. Code is available at https://github.com/Sasa77777779/UEM.git.
Authors: Yongdong chi, Hanqing Wang, Zonghan Yang, Jian Yang, Xiao Yan, Yun Chen, Guanhua Chen
Abstract: Text-to-SQL transforms the user queries from natural language to executable SQL programs, enabling non-experts to interact with complex databases. Existing prompt-based methods craft meticulous text guidelines and examples to facilitate SQL generation, but their accuracy is hindered by the large semantic gap between the texts and the low-resource SQL programs. In this work, we propose Pi-SQL, which incorporates the high-resource Python program as a pivot to bridge between the natural language query and SQL program. In particular, Pi-SQL first generates Python programs that provide fine-grained step-by-step guidelines in their code blocks or comments, and then produces an SQL program following the guidance of each Python program. The final SQL program matches the reference Python program's query results and, through selection from candidates generated by different strategies, achieves superior execution speed, with a reward-based valid efficiency score up to 4.55 higher than the best-performing baseline. Extensive experiments demonstrate the effectiveness of Pi-SQL, which improves the execution accuracy of the best-performing baseline by up to 3.20.
Authors: SeungBum Ha, Saerom Park, Sung Whan Yoon
Abstract: Machine unlearning (MU) aims to expunge a designated forget set from a trained model without costly retraining, yet the existing techniques overlook two critical blind spots: "over-unlearning" that deteriorates retained data near the forget set, and post-hoc "relearning" attacks that aim to resurrect the forgotten knowledge. We first derive the over-unlearning metric OU@{\epsilon}, which represents the collateral damage to the nearby region of the forget set, where the over-unlearning mainly appears. Next, we expose an unforeseen relearning threat on MU, i.e., the Prototypical Relearning Attack, which exploits the per-class prototype of the forget class with just a few samples, and easily restores the pre-unlearning performance. To counter both blind spots, we introduce Spotter, a plug-and-play objective that combines (i) a masked knowledge-distillation penalty on the nearby region of forget set to suppress OU@{\epsilon}, and (ii) an intra-class dispersion loss that scatters forget-class embeddings, neutralizing prototypical relearning attacks. On CIFAR-10, as one of validations, Spotter reduces OU@{\epsilon}by below the 0.05X of the baseline, drives forget accuracy to 0%, preserves accuracy of the retain set within 1% of difference with the original, and denies the prototype-attack by keeping the forget set accuracy within <1%, without accessing retained data. It confirms that Spotter is a practical remedy of the unlearning's blind spots.
Authors: Guobin Zhu, Rui Zhou, Wenkang Ji, Shiyu Zhao
Abstract: Although Multi-Agent Reinforcement Learning (MARL) is effective for complex multi-robot tasks, it suffers from low sample efficiency and requires iterative manual reward tuning. Large Language Models (LLMs) have shown promise in single-robot settings, but their application in multi-robot systems remains largely unexplored. This paper introduces a novel LLM-Aided MARL (LAMARL) approach, which integrates MARL with LLMs, significantly enhancing sample efficiency without requiring manual design. LAMARL consists of two modules: the first module leverages LLMs to fully automate the generation of prior policy and reward functions. The second module is MARL, which uses the generated functions to guide robot policy training effectively. On a shape assembly benchmark, both simulation and real-world experiments demonstrate the unique advantages of LAMARL. Ablation studies show that the prior policy improves sample efficiency by an average of 185.9% and enhances task completion, while structured prompts based on Chain-of-Thought (CoT) and basic APIs improve LLM output success rates by 28.5%-67.5%. Videos and code are available at https://windylab.github.io/LAMARL/
Authors: Lu\'is Cruz, Jo\~ao Paulo Fernandes, Maja H. Kirkeby, Silverio Mart\'inez-Fern\'andez, June Sallou, Hina Anwar, Enrique Barba Roque, Justus Bogner, Joel Casta\~no, Fernando Castor, Aadil Chasmawala, Sim\~ao Cunha, Daniel Feitosa, Alexandra Gonz\'alez, Andreas Jedlitschka, Patricia Lago, Henry Muccini, Ana Oprescu, Pooja Rani, Jo\~ao Saraiva, Federica Sarro, Raghavendra Selvan, Karthik Vaidhyanathan, Roberto Verdecchia, Ivan P. Yamshchikov
Abstract: The environmental impact of Artificial Intelligence (AI)-enabled systems is increasing rapidly, and software engineering plays a critical role in developing sustainable solutions. The "Greening AI with Software Engineering" CECAM-Lorentz workshop (no. 1358, 2025) funded by the Centre Europ\'een de Calcul Atomique et Mol\'eculaire and the Lorentz Center, provided an interdisciplinary forum for 29 participants, from practitioners to academics, to share knowledge, ideas, practices, and current results dedicated to advancing green software and AI research. The workshop was held February 3-7, 2025, in Lausanne, Switzerland. Through keynotes, flash talks, and collaborative discussions, participants identified and prioritized key challenges for the field. These included energy assessment and standardization, benchmarking practices, sustainability-aware architectures, runtime adaptation, empirical methodologies, and education. This report presents a research agenda emerging from the workshop, outlining open research directions and practical recommendations to guide the development of environmentally sustainable AI-enabled systems rooted in software engineering principles.
Authors: Yile Liu, Ziwei Ma, Xiu Jiang, Jinglu Hu, Jing Chang, Liang Li
Abstract: With the rapid adoption of large language models (LLMs) in natural language processing, the ability to follow instructions has emerged as a key metric for evaluating their practical utility. However, existing evaluation methods often focus on single-language scenarios, overlooking the challenges and differences present in multilingual and cross-lingual contexts. To address this gap, we introduce MaXIFE: a comprehensive evaluation benchmark designed to assess instruction-following capabilities across 23 different languages with 1667 verifiable instruction tasks. MaXIFE integrates both Rule-Based Evaluation and Model-Based Evaluation, ensuring a balance of efficiency and accuracy. We applied MaXIFE to evaluate several leading commercial LLMs, establishing baseline results for future comparisons. By providing a standardized tool for multilingual instruction-following evaluation, MaXIFE aims to advance research and development in natural language processing.
Authors: Genta Indra Winata, David Anugraha, Emmy Liu, Alham Fikri Aji, Shou-Yi Hung, Aditya Parashar, Patrick Amadeus Irawan, Ruochen Zhang, Zheng-Xin Yong, Jan Christian Blaise Cruz, Niklas Muennighoff, Seungone Kim, Hanyang Zhao, Sudipta Kar, Kezia Erina Suryoraharjo, M. Farid Adilazuarda, En-Shiun Annie Lee, Ayu Purwarianti, Derry Tanti Wijaya, Monojit Choudhury
Abstract: High-quality datasets are fundamental to training and evaluating machine learning models, yet their creation-especially with accurate human annotations-remains a significant challenge. Many dataset paper submissions lack originality, diversity, or rigorous quality control, and these shortcomings are often overlooked during peer review. Submissions also frequently omit essential details about dataset construction and properties. While existing tools such as datasheets aim to promote transparency, they are largely descriptive and do not provide standardized, measurable methods for evaluating data quality. Similarly, metadata requirements at conferences promote accountability but are inconsistently enforced. To address these limitations, this position paper advocates for the integration of systematic, rubric-based evaluation metrics into the dataset review process-particularly as submission volumes continue to grow. We also explore scalable, cost-effective methods for synthetic data generation, including dedicated tools and LLM-as-a-judge approaches, to support more efficient evaluation. As a call to action, we introduce DataRubrics, a structured framework for assessing the quality of both human- and model-generated datasets. Leveraging recent advances in LLM-based evaluation, DataRubrics offers a reproducible, scalable, and actionable solution for dataset quality assessment, enabling both authors and reviewers to uphold higher standards in data-centric research. We also release code to support reproducibility of LLM-based evaluations at https://github.com/datarubrics/datarubrics.
Authors: Minghao Liu, Zhitao He, Zhiyuan Fan, Qingyun Wang, Yi R. Fung
Abstract: Text-guided image editing has seen rapid progress in natural image domains, but its adaptation to medical imaging remains limited and lacks standardized evaluation. Clinically, such editing holds promise for simulating surgical outcomes, creating personalized teaching materials, and enhancing patient communication. To bridge this gap, we introduce MedEBench, a comprehensive benchmark for evaluating text-guided medical image editing. It consists of 1,182 clinically sourced image-prompt triplets spanning 70 tasks across 13 anatomical regions. MedEBench offers three key contributions: (1) a clinically relevant evaluation framework covering Editing Accuracy, Contextual Preservation, and Visual Quality, supported by detailed descriptions of expected change and ROI (Region of Interest) masks; (2) a systematic comparison of seven state-of-the-art models, revealing common failure patterns; and (3) a failure analysis protocol based on attention grounding, using IoU between attention maps and ROIs to identify mislocalization. MedEBench provides a solid foundation for developing and evaluating reliable, clinically meaningful medical image editing systems. Project website: https://mliuby.github.io/MedEBench_Website/