论文检索

Xiaoyang Liu, Dawei Wang, Tian Li, Huizhi Liang, Gary Ushaw, Richard Davison

Reasoning capability is fundamental in enabling Large Language Models to perform complex multi-step inference. By sampling multiple reasoning paths and selecting the most frequent answer, Self Consistency (SC) remains highly effective but fails on challenging tasks where incorrect answers dominate the majority. Inspired by Metropolis Light Transport in physically-based rendering, where discovered high-contribution light paths guide subsequent sampling toward illumination sources, we propose Metropolis Self Consistency and its multi-LLM extension, Metropolis Cross Consistency, a probabilistic self- and cross-consistency verification framework for mathematical reasoning. Our approach employs an accept-reject mechanism to encourage high-quality reasoning paths, concentrating sampling in regions more likely to yield correct answers. Experiments on 9 LLMs across 4 challenging mathematical benchmarks demonstrate consistent improvements over SC. Even when combining models of vastly different capabilities, MCC maintains performance virtually matching the most capable model while significantly reducing computational cost compared to SC with the strongest model alone. While our implementation is training-free, adds minimal token overhead beyond SC, and requires no external reward model, our approach provides a flexible paradigm that can accommodate any scalar reward representing path correctness.

Hao Wang, Jindong Han, Wei Fan, Hao Liu

Climate research is pivotal for mitigating global environmental crises, yet the accelerating volume of multi-scale datasets and the complexity of analytical tools have created significant bottlenecks, constraining scientific discovery to fragmented and labor-intensive workflows. While the emergence Large Language Models (LLMs) offers a transformative paradigm to scale scientific expertise, existing explorations remain largely confined to simple Question-Answering (Q A) tasks. These approaches often oversimplify real-world challenges, neglecting the intricate physical constraints and the data-driven nature required in professional climate science. To bridge this gap, we introduce ClimAgent, a general-purpose autonomous framework designed to execute a wide spectrum of research tasks across diverse climate sub-fields. By integrating a unified tool-use environment with rigorous reasoning protocols, ClimAgent transcends simple retrieval to perform end-to-end modeling and analysis. To foster systematic evaluation, we propose ClimaBench, the first comprehensive benchmark for real-world climate discovery. It encompasses challenging problems spanning 5 distinct task categories derived from professional scenarios between 2000 and 2025. Experiments on ClimaBench demonstrate that ClimAgent significantly outperforms state-of-the-art baselines, achieving a 40.21% improvement over original LLM solutions in solution rigorousness and practicality. Our code are available at https://github.com/usail-hkust/ClimAgent.

Pengxiang Liu, Tao Ren, Wei Xiong, Tingrui Yang, Junjie Wang, Jun HU

Large Language Models (LLMs) have shown impressive reasoning capabilities in agents for complex interactive environments. However, these agents often suffer from hallucinations and lack grounding, leading to unreliable actions that conflict with real-world constraints. Existing approaches mitigate some issues through implicit imitation or sparse reinforcement learning but rely on fitting data distributions without explicitly understanding environmental constraints, often generating actions that are behaviorally distorted or environmentally impermissible. To address this, we introduce OntoGuard, an ontological framework designed to guard LLM agents by enforcing environmental and behavioral admissibility. These constraints are constructed by extracting knowledge from oracle demonstrations, supplemented with world knowledge inherent in LLMs and general knowledge bases. During inference, OntoGuard functions as an active interceptor, using a graph-based constraint-checking mechanism to reject invalid actions and prompt self-correction before acting. Experiments on both ScienceWorld and VirtualHome demonstrate OntoGuard’s advantage over state-of-the-art methods, validating its ability to enforce physical and behavioral constraints while preventing invalid actions.

Feiyu Duan, Xuanjing Huang, Zhongyu Wei

The rapid advancement of large language models (LLMs) has accelerated progress toward universal AI assistants. However, existing benchmarks for personalized assistants remain misaligned with real-world user-assistant interactions, failing to capture the complexity of external contexts and users’ cognitive states. To bridge this gap, we propose LifeSim, a user simulator that models user cognition through the Belief-Desire-Intention (BDI) model within physical environments for coherent life trajectories generation, and simulates intention-driven user interactive behaviors. Based on LifeSim, we introduce LifeSim-Eval, a comprehensive benchmark for multi-scenario, long-horizon personalized assistance. LifeSim-Eval covers 8 life domains and 1,200 diverse scenarios, and adopts a multi-turn interactive method to assess models’ abilities to complete explicit and implicit intentions, recover user profiles, and produce high-quality responses. Under both single-scenario and long-horizon settings, our experiments reveal that current LLMs face significant limitations in handling implicit intention and long-term user preference modeling.

Yibo Zhang, Kaiwen Luo, Liang Lin, Shilinlu Yan, Jin Wang, Yaoqi Guo, Yitian Chen, Yalan Qin, Zhenhong Zhou, Kun Wang 等

While Audio Large Models (ALLMs) have achieved remarkable proficiency, their robustness remains brittle in real-world deployment. Existing evaluations largely rely on synthetic Gaussian noise or simplistic single-source interference, failing to capture the intricate, multi-layered acoustic dynamics—or "Acoustic Ecology"—that characterize authentic physical environments. To bridge this ecological gap, we introduce RSA-Bench, a comprehensive robustness benchmark designed to stress-test ALLMs through high-fidelity auditory scene simulations. Unlike traditional methods, we construct evaluation samples by naturally superimposing diverse environmental soundscapes—spanning Pasture, Extreme Weather, Classroom, and Outdoors—onto clean speech signals across a spectrum of interference intensities. By evaluating models on six core tasks ranging from fundamental perception to complex reasoning, our study unveils three macro-level insights: (I) The Perception-Cognition Gap: Models maintain relative resilience in low-level recognition but suffer a functional collapse in high-order reasoning tasks under stress; (II) Scenario Sensitivity: "Vocal-like" interference (e.g., children playing) proves significantly more destructive than mechanical noise, challenging the model’s auditory attention mechanisms; and (III) The Denoising Paradox: Standard speech enhancement often exacerbates performance degradation, as ALLMs prove highly sensitive to the semantic distortions introduced by denoising artifacts.

Manjie Xu, Isabella Yin, Xinyi Tu, Chi Zhang, Yixin Zhu

LLMs struggle with Semantic Inertia: the inability to inhibit pre-trained priors (e.g., “Lava is Dangerous”) when dynamic, in-context rules contradict them. We probe this phenomenon using , where physical laws are mutable text rules, enabling precise evaluation of models’ ability to override learned priors when rules change. We quantatively observe that larger models can exhibit inverse scaling: they perform worse than smaller models when natural language reasoning requires suppressing pre-trained associations (e.g., accepting “Lava is Safe”). Our analysis attributes this to natural language encoding, which entangles descriptive semantics and logical rules, leading to persistent hallucinations of familiar physics despite explicit contradictory rules. Here we show that representing dynamics as executable code, rather than descriptive text, reverses this trend and enables effective prior inhibition. We introduce LCV, which fine-tunes models on counterfactual pairs and identifies states with contradictory rules, thereby forcing attention to logical constraints rather than visual semantics. This training-time approach outperforms expensive inference-time search methods in both efficiency and accuracy. Our results demonstrate that representation fundamentally determines whether scaling improves or impairs contextual reasoning. This challenges the assumption that larger models are universally better, with implications for domains that require dynamic overriding of learned priors.

Jiang Li, Zehua Duo, Tian Lan, Feilong Bao, Guanglai Gao, Xiangdong Su

Temporal knowledge graph embedding (TKGE) aims to model the temporal evolution of relational facts. However, existing approaches predominantly rely on discrete timestamp lookup tables and high-dimensional embedding spaces, which lack explicit structural constraints for continuous-time dynamics. As a result, temporal patterns are often captured through capacity scaling rather than principled dynamic modeling, leading to limited parameter efficiency and scalability.To address these limitations, we propose , a physics-inspired framework that embeds temporal dynamics into a symplectic phase space. Our model introduces a structure-preserving Hamiltonian evolution mechanism based on a pairwise-decoupled Hamiltonian generator and its Cayley transform, ensuring that temporal transformations adhere to the symplectic group \mathrm{Sp}(2d) and preserve phase-space volume with linear computational complexity. In addition, we design a Time-Aware Parameter Modulation mechanism that integrates continuous Rotary Time Embeddings via Feature-wise Linear Modulation, enabling smooth temporal evolution while capturing event-driven variations. Theoretical analysis establishes the geometric validity of the proposed framework. Extensive experiments on standard TKGE benchmarks demonstrate that achieves competitive performance with substantially lower embedding dimensions. Furthermore, empirical results show that the proposed continuous Hamiltonian evolution facilitates generalization to unseen timestamps by learning transferable temporal dynamics from the underlying geometric structure.

Zihan Chen, Yiming Zhang, Hengguang Zhou, Zenghui Ding, Yining Sun, Cho-Jui Hsieh

Current benchmarks are inadequate for evaluating progress in reinforcement learning (RL) for large language models (LLMs). Despite recent benchmark gains reported for RL, we find that training on these benchmarks’ training sets achieves nearly the same performance as training directly on the test sets, suggesting that the benchmarks cannot reliably separate further progress. To study this phenomenon, we introduce a diagnostic suite and the Oracle Performance Gap (OPG) metric that quantifies the performance difference between training on the train split versus the test split of a benchmark. We further analyze this phenomenon with stress tests and find that, despite strong benchmark scores, existing RL methods struggle to generalize across distribution shifts, varying levels of difficulty, and counterfactual scenarios: shortcomings that current benchmarks fail to reveal. We conclude that current benchmarks are insufficient for evaluating generalization and propose three core principles for designing more faithful benchmarks: sufficient difficulty, balanced evaluation, and distributional robustness.

Cheng Liu, Xiaolei Liu, Xingyu Li, Bangzhou Xin, Kangyi Ding

Existing jailbreak defense paradigms primarily rely on static detection of prompts, outputs, or internal states, often neglecting the dynamic evolution of risk during decoding. This oversight leaves risk signals embedded in decoding trajectories underutilized, constituting a critical blind spot in current defense systems. In this work, we empirically demonstrate that hidden states in critical layers during the decoding phase carry stronger and more stable risk signals than input jailbreak prompts. Specifically, the hidden representations of tokens generated during jailbreak attempts progressively approach high-risk regions in the latent space. Based on this observation, we propose TrajGuard, a training-free, decoding-time defense framework. TrajGuard aggregates hidden-state trajectories via a sliding window to quantify risk in real time, triggering a lightweight semantic adjudication only when risk within a local window persistently exceeds a threshold. This mechanism enables the immediate interruption or constraint of subsequent decoding. Extensive experiments across 12 jailbreak attacks and various open-source LLMs show that TrajGuard achieves an average defense rate of 95%. Furthermore, it reduces detection latency to 5.2 ms/token while maintaining a false positive rate below 1.5%. These results confirm that hidden-state trajectories during decoding can effectively support real-time jailbreak detection, highlighting a promising direction for defenses without model modification.

Sanjeda Akter, Ibne Farabi Shihab, Anuj Sharma

Proxy optimization, where AI systems exploit evaluator weaknesses rather than improve intended objectives, threatens both reinforcement learning (reward hacking) and LLM alignment (evaluator gaming). We introduce the Evaluator Stress Test (EST), an invariance-based framework that detects proxy gaming by separating exploitable sensitivity (e.g., format, physics bugs) from content-driven improvements using controlled perturbations with semantic validity audits. We validate EST across both domains. In RL, across 15 environments and 5 algorithms (2,156 expert-annotated episodes), EST achieves 78.4% precision and 81.7% recall. In LLM alignment, across 4 tasks, 2 model scales, 2 training methods, and 2 judges (1,200 human-annotated instances), EST achieves 74.2% precision and 78.6% recall with early warning signals preceding quality decline. Cross-domain analysis reveals that proxy-true correlation tracking transfers directly between domains, while perturbation design requires domain adaptation. Closed-loop mitigation improves human win-rate by 8.3 points (LLM) and reduces hacking by 54.6% (RL). We release benchmarks for both domains: 2,156 RL episodes and 1,200 LLM instances.

Yanjun Zhao, Tianxin Wei, Jiaru Zou, Xuying Ning, Yuanchen Bei, Lingjie Chen, Simmi Rana, Wendy H. Yang, Hanghang Tong, Jingrui He

Understanding scientific papers requires more than answering isolated questions or summarizing content. It involves an integrated reasoning process that grounds textual and visual information, interprets experimental evidence, synthesizes information across sources, and critically evaluates scientific claims. However, existing benchmarks typically assess these abilities in isolation, making it difficult to evaluate scientific paper understanding as a unified set of interacting cognitive abilities. In this work, we introduce PaperMind , a benchmark designed to evaluate integrated and agent-oriented scientific reasoning over research papers. PaperMind is constructed from real scientific papers across seven domains, including agriculture, biology, chemistry, computer science, medicine, physics, and economics. It comprises four complementary task families that collectively operationalize distinct cognitive facets of scientific paper reasoning, including multimodal grounding, experimental interpretation, cross-source evidence reasoning, and critical assessment. By analyzing model behavior across multiple tasks, PaperMind enables a diagnostic evaluation of integrated scientific reasoning behaviors that are difficult to assess through isolated task evaluations. Extensive experiments on both open-source and closed-source multimodal LLMs reveal consistent performance gaps across tasks, highlighting persistent challenges in integrated scientific reasoning and critique. Our benchmark and dataset are available at https://github.com/Yanjun-Zhao/PaperMind.

Hayden Helm, Tianyi Chen, Harvey McGuinness, Paige Lee, Brandon Duderstadt, Carey Priebe

In this paper we provide evidence that our virtual model of U.S. congresspersons based on a collection of language models moves towards satisfying the definition of a digital twin. In particular, we introduce and provide high-level descriptions of a daily-updated dataset that contains every Tweet from every U.S. congressperson during their respective terms. We demonstrate that a modern language model equipped with congressperson-specific subsets of this data producing Tweets that are largely indistinguishable from actual Tweets posted by their physical counterparts. We illustrate how generated Tweets can be used to predict roll-call vote behaviors and to quantify the likelihood of congresspersons crossing party lines, thereby assisting stakeholders in allocating resources and potentially impacting real-world legislative dynamics. We conclude with a discussion of the limitations and important extensions of our analysis.

Yuxuan Dong, Xinyu Zhang, Lingling Zhang, Han Lai, Pengyu Li, Bifan Wei, Yaqiang Wu, Jun Liu

Despite the remarkable progress of Large Language Models (LLMs) in abstract reasoning tasks, they continue to struggle with physics problem solving due to difficulties in decoding implicit constraints and maintaining physical consistency. To address these challenges, Process Reward Models (PRMs) have emerged as a promising approach to verify intermediate reasoning steps. Existing PRMs attempt to mitigate reasoning errors but typically rely on scalar scoring, which lacks the explanatory power necessary to diagnose complex physical misconceptions. In this work, we introduce PhysPRM, a Generative PRM that treats evaluation as a generative task to produce fine-grained diagnoses comprising critiques, final judgments, and specific error types. To facilitate this, we develop an automated data synthesis pipeline to construct PhysPRM30K, a comprehensive training dataset, and PhysProcessBench, a rigorously human-verified benchmark. By employing a two-stage training paradigm that integrates Supervised Fine-Tuning with Group Relative Policy Optimization, PhysPRM significantly enhances the physics reasoning capabilities of various LLMs. Extensive experiments demonstrate that PhysPRM improves performance across seven benchmarks in both Best-of-N and critique refinement strategies.

JunJian Wang, Lidan Zhao, Xi Sheryl Zhang

Large Language Models (LLMs) exhibit impressive reasoning capabilities but often suffer from Embodied Semantic Hallucinations—generating plans that are semantically fluent but physically unsafe due to a lack of grounded common sense. Existing safety alignment methods, such as RLHF or naive safety prompting, typically fall into a Safety-Utility Trade-off, resulting in severe over-rejection of benign household instructions. To address this, we propose MADRA (Multi-Agent Debate for Risk Awareness), a training-free cognitive architecture that mimics System-2 deliberation. MADRA introduces a meta-cognitive Critical Agent that evaluates peer debates using a structured argumentation framework derived from the Toulmin Model, effectively mitigating the "herd mentality" in multi-agent systems. We also introduce SafeAware-VH, a benchmark featuring adversarial safe instructions designed to probe agents’ sensitivity to physical risks. Extensive experiments demonstrate that MADRA breaks the Pareto frontier, achieving over 90% rejection of unsafe tasks while maintaining high utility, significantly outperforming standard Chain-of-Thought and single-agent reflection baselines.

Longteng Guo, Xuanxu Lin, Dongze Hao, Tongtian Yue, Pengkang Huo, Jiatong Ma, Yuchen Liu, Jing Liu

Scientific reasoning is a key aspect of human intelligence, requiring the integration of multimodal inputs, domain expertise, and multi-step inference across various subjects. Existing benchmarks for multimodal large language models (MLLMs) often fail to capture the complexity and traceability of reasoning processes necessary for rigorous evaluation. To fill this gap, we introduce SciVQR, a multimodal benchmark covering 54 subfields in mathematics, physics, chemistry, geography, astronomy, and biology. SciVQR includes domain-specific visuals, such as equations, charts, and diagrams, and challenges models to combine visual comprehension with reasoning. The tasks range from basic factual recall to complex, multi-step inferences, with 46% including expert-authored solutions. SciVQR not only evaluates final answers but also examines the reasoning process, providing insights into how models reach their conclusions. Our evaluation of leading MLLMs, including both proprietary and open-source models, reveals significant limitations in handling complex multimodal reasoning tasks, underscoring the need for improved multi-step reasoning and better integration of interdisciplinary knowledge in advancing MLLMs toward true scientific intelligence. The dataset and evaluation code are publicly available at https://github.com/CASIA-IVA-Lab/SciVQR.

Mafizur Rahman, Lijun Qian

Analog in-memory computing (AIMC) offers substantial efficiency gains for transformer inference but introduces hardware-induced noise that can distort attention behavior. Prior studies primarily focus on AIMC evaluations for vision tasks and CNN-based models. They largely overlook how hardware-induced noise perturbs internal attention dynamics in NLP models. In this work, we present the first fine-grained analysis of analog vulnerability in pretrained transformers, examining projection submodules, attention heads, and layer-wise dynamics across multiple NLP tasks. Results show that query (Q), key (K), and value (V) projections are the most sensitive components, while feed-forward layers remain comparatively robust. Also, analog noise yields depth-dependent degradation in higher layers, leading to scattered attention and disrupted token routing. This pre-deployment analysis mitigates potential resource misuse before physical deployment and offers practical guidance for designing noise-resilient analog NLP transformers.

Qiguang Chen, Chengyu Luan, Jiajun Wu, Qiming Yu, Yi Yang, Yizhuo Li, Jingqi Tong, Xiachong Feng, Libo Qin, Wanxiang Che

Large vision-language models (LVLMs) have made substantial advances in reasoning tasks at the Olympiad level. Nevertheless, current Olympiad-level multimodal reasoning benchmarks for these models often emphasize single-image analysis and fail to exploit contextual information across multiple images. We present OMIBench, a benchmark designed to evaluate Olympiad-level reasoning when the required evidence is distributed over multiple images. It contains problems from biology, chemistry, mathematics, and physics Olympiads, together with manually annotated rationales and evaluation protocols for both exact and semantic answer matching. Across extensive experiments on OMIBench, we observe meaningful performance gaps in existing models. Even the strongest LVLMs, such as Gemini-3-Pro, attain only about 50% on the benchmark. These results position OMIBench as a focused resources for studying and improving multi-image reasoning in LVLMs.

Jinghan Yu, Junhao Xiao, Chenyu Zhu, Jiaming Li, Jia Li, HanMing Deng, Xirui Wang, Guoli Jia, Jianjun Li, Xiang Bai 等

Existing text-guided image editing methods primarily rely on end-to-end pixel-level inpainting paradigm. Despite its success in simple scenarios, this paradigm still significantly struggles with compositional editing tasks that require precise local control and complex multi-object spatial reasoning. This paradigm is severely limited by 1) the implicit coupling of planning and execution, 2) the lack of object-level control granularity, and 3) the reliance on unstructured, pixel-centric modeling. To address these limitations, we propose I2E, a novel "Decompose-then-Action” paradigm that revisits image editing as an actionable interaction process within a structured environment. I2E utilizes a Decomposer to transform unstructured images into discrete, manipulable object layers and then introduces a physics-aware Vision-Language-Action Agent to parse complex instructions into a series of atomic actions via Chain-of-Thought reasoning. Further, we also construct I2E-Bench, a benchmark designed for multi-instance spatial reasoning and high-precision editing. Experimental results on I2E-Bench and multiple public benchmarks demonstrate that I2E significantly outperforms state-of-the-art methods in handling complex compositional instructions, maintaining physical plausibility, and ensuring multi-turn editing stability.

Yuyang Liu, Jingya Wang, Liuzhenghao Lv, Yonghong Tian

Large language models (LLMs) have demonstrated significant reasoning capabilities in scientific discovery but struggle to bridge the gap to physical execution in wet-labs. In these irreversible environments, probabilistic hallucinations are not merely incorrect; they can cause equipment damage or experimental failure. To address this, we propose BioProAgent, a neuro-symbolic framework that anchors probabilistic planning in a deterministic Finite State Machine (FSM). We introduce a State-Augmented Planning mechanism that enforces a rigorous Design-Verify-Rectify workflow, ensuring hardware compliance before execution. Furthermore, we address the context bottleneck inherent in complex device schemas by Semantic Symbol Grounding, reducing token consumption by \sim6\times through symbolic abstraction. In the extended BioProBench benchmark, BioProAgent achieves 95.6% physical compliance (compared to 21.0% for ReAct), demonstrating that neuro-symbolic constraints are essential for reliable autonomy in irreversible physical environments. Code at https://github.com/YuyangSunshine/bioproagent and Website at https://yuyangsunshine.github.io/BioPro-Project/ .

Jihang Jin, Ronghao Chen, Hao Zhang, Ziyan Liu, Huacan Wang, Qi Ye, Jingping Liu

Visual scale recognition is a fundamental aspect for humans to perceive physical quantities in the real world, and it is crucial for enabling human-like intelligence in multimodal large language models (MLLMs). However, existing benchmarks typically focus on a single type of quantity (e.g., time) or a specific format (e.g., dials), lacking a comprehensive evaluation of scale recognition capabilities. To address these problems, we propose ScaleBench, a visual scale recognition benchmark built using images from COCO, Open Images, and Flickr, designed to comprehensively evaluate the scale recognition capabilities of MLLMs. To ensure high data quality, we develop detailed annotation guidelines and procedures, resulting in a total of 6,574 annotated samples. Based on this benchmark, we evaluate multiple closed-source and open-source MLLMs. Experimental results reveal that the best-performing model achieves only 42.60% accuracy, far lower than the 97.40% of humans. Furthermore, we conduct in-depth experimental analyses and provide future research directions. Our benchmark and implementation codes are available at https://github.com/Sonder-hang/ScaleBench.