Fei Lin, Tengchao Zhang, Ziyang Gong, Xiaotong Yu, Bohan Zhang, Yifan Zhou, Qihao Yang, Ji Dai 等 18 位作者
Self-Driving Chemical Laboratories (SDCLs) are moving chemical experimentation toward embodied automation, where machines must perceive laboratory scenes, reason about experimental states, and act under physical, procedural, and safety constraints. Although Multimodal Large Language Models (MLLMs) are increasingly considered as embodied brains for such systems, existing laboratory benchmarks mainly focus on safety, anomaly detection, or general scientific understanding, leaving their execution-oriented embodied capabilities underexplored. To fill this gap, we introduce BEAKER, a Benchmark for evaluating Embodied Actionability, Knowledge, Experimentation, and Reasoning in SDCLs. BEAKER contains 1,000 expert-curated question-answer samples across 11 image and video task types, covering key judgments required for chemical laboratory execution, including affordance grounding, trajectory planning, equipment, state, and safety recognition, apparatus setup, operation and step reasoning, and temporal procedure reasoning. All questions and annotations are manually created and reviewed by chemistry and embodied Artificial Intelligence (AI) experts, without AI-assisted generation or annotation. We evaluate 28 general-purpose, science-oriented, chemistry-specialized, and embodied MLLMs in an offline, single-turn setting. Results show that current models can follow coarse experimental procedures in videos, but still struggle to ground actions into operable regions and feasible trajectories, to recognize fine-grained laboratory apparatus, states, and safety norms, and to order steps in mid-length procedures. BEAKER thus reveals a clear gap between coarse procedural understanding and the precise, action-grounded perception required for laboratory execution, providing a diagnostic benchmark for future MLLMs in SDCLs.
Zuhao Yang, Kaichen Zhang, Sudong Wang, Keming Wu, Zhongyu Yang, Bo Li, Xiaojuan Qi, Shijian Lu 等 10 位作者
Training large multimodal models (LMMs) via reinforcement learning (RL) to natively invoke video-processing tools (e.g., cropping) has become a promising route to long-video understanding. However, most existing agentic Video-LMMs dispatch tool calls sequentially (i.e., one per turn). Sequential tool calling can propagate localization errors, accumulate irrelevant context, and require an additional main-model generation for each successive crop. We introduce ParaVT, a multi-agent RL framework for Parallel Video Tool calling, dispatching multiple time-window crops in a single turn, allowing evidence from multiple temporal windows to be examined concurrently. Yet applying standard Group Relative Policy Optimization (GRPO) to ParaVT reveals an obstacle we term the Tool Prior Paradox: inherited tool-use priors appear to support exploration while competing with the output format learned during supervised fine-tuning. Standard RL can also settle on tool skipping. A comparison with a model with weaker tool-use priors shows stable formatting but no tool calls during RL, consistent with a trade-off between format stability and tool exploration. We propose PARA-GRPO (Parseability-Anchored and Ratio-gAted GRPO), which augments standard RL with two complementary mechanisms: (i) a format reward that encourages tag closure and penalizes unclosed reasoning blocks, and (ii) a per-prompt randomization of the overview-frame budget to create a reward advantage for tool use when the overview provides insufficient evidence. Across six long-video understanding benchmarks, ParaVT improves over the Qwen3-VL baseline by +7.9% on average, with PARA-GRPO raising training-time format reward from 0.13 to a peak of 0.64. These findings highlight the need to account for inherited tool-use priors when designing RL methods for long-video understanding.
LLM agents are judged by what they deliver, and in professional work what gets delivered is a project, not a task. A project pays off only if the whole package holds, and one weak part can limit the value of the rest. Yet existing benchmarks mostly test agents on one task at a time, leaving a gap between the capability we measure and the outcome that makes agents useful. To address this gap, we introduce DeliverBench, which tests whether agents can carry a project from brief to final delivery. Its 11 projects, spanning financial due diligence, legal contract drafting, and drug safety reporting, are reconstructed with 13 domain experts from real materials and chain 150 subtasks through 2-15 rounds of new materials and client feedback. Because no reference outputs are injected, agents inherit their own errors from one task to the next. Expert rubrics then score every checkpoint as well as the final package. Across seven model-harness systems, even the best system scores only 65.0 out of 100 on final delivery and passes 50.0% of subtasks at a 90-point threshold. The failures are often failures of carrying work forward: in one audit project, all seven systems addressed the review findings, but only one preserved traceability. Together, these results suggest that task-level capability does not simply compose into delivery, which also depends on keeping the work coherent as it evolves.
Web services increasingly expose backend interfaces that let users delegate tasks to AI agents, which invoke service operations directly without navigating websites. This shift changes how existing Web vulnerabilities can be exploited. We investigate these changes through VELA, an executable benchmark covering 44 vulnerabilities across twelve Web application families. The benchmark preserves vulnerable backends and derives agent-facing interfaces from legitimate workflows. We evaluate three commercial agents across multiple interface configurations, using controlled comparisons to separate interface effects from agent behavior and independent oracles to verify attack outcomes. Our results show that the agent realizes every attack that direct interface invocation realizes in the reference configuration, so exploitability is set mainly by the interface: 91% of vulnerabilities remain exploitable under the reference design. Interface restrictions prevent some attacks by removing required operations or limiting input control, while retention remains high across vulnerability classes. No case-level loss in the reference configuration is attributed to agent refusal or input normalization, and in one case the agent completed an attack by supplying calls the attacker omitted. Across the tested configurations, interface design affects exploitability more than agent choice, moving retention from 39% to 95% against at most 4.5 points across agents, but some restrictive interfaces also limit legitimate workflows. Our study provides a systematic empirical account of how agent-mediated access reshapes existing Web vulnerabilities, grounding future Web architecture design in measured security outcomes and trade-offs between reducing attack exposure and preserving legitimate functionality in agent-facing services.
Unsafe deserialization turns untrusted data into attacker-controlled execution, and a single vulnerability can admit diverse exploit variants that evade defenses tailored to a known exploit. Existing defenses typically operate at a single enforcement point with limited visibility into the full attack. Combining controls across layers offers complementary protection, but coordinating them to preserve legitimate functionality and adapting them to new variants requires substantial expert effort. We propose JANUS, a red-blue adversarial framework for automated cross-layer defense synthesis. Starting from a known exploit, \sys iteratively refines deployment-specific policies in an isolated offline environment. A black-box attacker probes the current policy for exploit variants using only externally observable feedback. The defender summarizes each verified variant into its security-relevant effects, uses counterfactual replay to separate complementary coverage from redundant enforcement, and jointly selects and scopes controls across layers. Each candidate policy is validated against accumulated exploits and benign workflows. We evaluate JANUS on sixteen vulnerability deployments across six Java and PHP applications against gateway, application, system, and cross-layer defenses. JANUS reduces held-out attack success to 6.2%, compared with 18.8-55.0% for the baselines, fully mitigates 13 of 16 deployments, and preserves 98.4% of benign functionality tests at a 6.9% targeted P95 latency increase.
Learning dexterous manipulation from demonstrations is bottlenecked by data: the contact forces that determine whether a grasp succeeds are absent from every scalable source of human demonstrations. This paper builds on two observations. First, what survives the change from a human hand to a robot hand is the contact structure of a demonstration - which finger regions touch which object locations, and in what order - rather than its joint motion. Second, physical consistency need not be engineered per task: a single residual reinforcement learning (RL) policy, trained once across diverse demonstrations, can repair kinematic recordings into physically consistent, contact-annotated trajectories, and the same residual formulation restores dynamic feasibility after retargeting. These observations yield a three-stage pipeline that converts human motion-capture recordings into dexterous robot policies with no real-robot training data: physics refinement with a simulated MANO hand recovers contacts and forces, contact-anchored retargeting transfers the demonstrated contact structure through an objective independent of hand morphology, and residual policy learning adapts the result to robot actuation. The pipeline reconstructs 25,454 single-hand trajectories (success 7.3% -> 59.3%) and 25 dual-hand tasks (16.0% -> 62.4%) with one shared policy per setting, transfers one human dataset to four morphologically distinct robot hands (+62.4 pp), and executes four contact-rich bimanual tasks on physical hardware with zero real-robot training data.
Detailed image captioning is usually framed as an amount problem: a model should mention more visual facts while avoiding hallucination. However, autoregressive VLMs generate captions as ordered sequences, and later generations may be produced under weaker or noisier visual grounding. We therefore argue that long-form captioning should be studied not only by what facts are mentioned, but also by where they appear. We introduce a position-aware rubric representation that decomposes dense human references into atomic visual claims, verifies generated captions against those claims, and induces a quartile-wise composition of supported, hallucinated, and generic content. This lens reveals two qualitatively different failures in long captions: models tend to produce unsupported specificity in the middle, while the final portion often escapes into vague, non-evidential language rather than simply accumulating more hallucination. A same-reference human control does not exhibit these patterns; additional human audits examine the visual validity of the induced labels. Motivated by this diagnosis, we propose PAGER (Position-Aware Grounded Evidence Reinforcement), a GRPO training objective with separate signals for supported fine-grained middle evidence, evidence-bearing tail continuation, claim coverage, and content-aware length control. The result is a closed loop: the same ordered claim representation identifies where long captions fail and supplies the position-specific reward needed to repair those failures. Experiments on detailed-caption benchmarks show that PAGER improves position-wise grounding and pairwise caption preference while preserving competitive global caption quality.
World models (WMs) have demonstrated strong potential for end-to-end autonomous driving by learning predictive representations of future scene dynamics. However, generating future videos during inference introduces substantial computational overhead, leading many recent driving WMs to adopt a single front camera as input for efficient deployment. This design restricts spatial coverage in safety-critical maneuvers such as lane changes, merges, and turns. To address this limitation, we propose SV-WAM, a surround-view world-action model (WAM) that preserves full six-camera observations while maintaining efficient inference. SV-WAM leverages future-video prediction as dense training supervision for action learning within a shared generative model, rather than as an inference-time output. At the core of this design is an action-centered causal mask that prevents action tokens from attending to future-video tokens during joint action-video denoising. Consequently, the video branch can be discarded at deployment, enabling efficient action-only planning. Furthermore, we introduce a differentiable drivable-area compliance regularizer that penalizes vehicle-footprint corners approaching or crossing drivable boundaries, improving planning safety and boundary awareness. Extensive experiments on the closed-loop NAVSIMv2 benchmark and the open-loop nuScenes benchmark demonstrate that SV-WAM achieves state-of-the-art planning performance with low inference latency and competitive zero-shot transfer capability.
Penetration testing is essential for assessing and strengthening system security against real-world threats, yet traditional workflows remain highly manual, expertise-intensive, and difficult to scale. Although recent advances in Large Language Models (LLMs) offer promising opportunities for automation, existing applications rely on simplistic prompting without task decomposition or domain adaptation, resulting in unreliable black-box behavior and limited insight into model capabilities across penetration testing stages. To address this gap, we introduce PentestEval, the first comprehensive benchmark for evaluating LLMs across six decomposed penetration testing stages: Information Collection, Weakness Gathering and Filtering, Attack Decision-Making, Exploit Generation and Revision. PentestEval integrates expert-annotated ground truth with a fully automated evaluation pipeline across 346 tasks covering all stages in 12 realistic vulnerable scenarios. Our stage-level evaluation of 9 widely used LLMs reveals generally weak performance and distinct limitations across the stages of penetration-testing workflow. End-to-end pipelines reach only 31% success rate, and existing LLM-powered systems such as PentestGPT, PentestAgent, and VulnBot exhibit similar limitations, with the evaluated autonomous agents showing particularly low success rates under our benchmark settings. These findings highlight that reliable autonomous penetration testing requires stronger structured reasoning and fine-grained stage-level diagnosis. PentestEval provides the foundational benchmark needed for future research on fine-grained, stage-level evaluation, paving the way toward more reliable LLM-based automation.
Language model agents often struggle with complex tasks even when equipped with expert-written skills. Such skills encode declarative knowledge, whereas successful execution requires procedural knowledge tailored to the task at hand. Natural language skills may leave conditions and dependencies implicit, and tasks involving multiple skills further require agents to coordinate intermediate results and execution order. We introduce Think with Skills, a method that trains reasoning models to turn relevant skill descriptions into task-specific pseudocode plans before execution. The pseudocode explicitly represents operations, conditions, intermediate results, and dependencies, providing a structured bridge from skill descriptions to agent actions. We construct training data that pairs tasks and skills with pseudocode plans and execution trajectories, use supervised fine tuning to initialize this behavior, and further optimize end-to-end task performance with Group Relative Policy Optimization. On SkillsBench, Think with Skills raises task success from 35.0% to 62.0%. These results show that proceduralizing expert knowledge before execution substantially improves skill use.
Time series classification is an important analytical task across diverse domains. However, its practical application is often hindered by the scarcity of labeled data and the requirement for substantial computational resources. To address these challenges, this paper proposes EvoTSC, a novel genetic programming approach designed to automatically evolve lightweight feature learning models for time series classification. The core of EvoTSC is a carefully designed multi-layer program structure that strategically embeds diverse forms of prior expert knowledge into the evolutionary process, effectively guiding the search toward operations known to be highly effective for time series analysis. To mitigate the common overfitting problem in time series classification, a tailored Pareto tournament selection strategy is proposed to favor models that perform consistently well across varying training data subsets, promoting the discovery of highly generalizable models. Extensive experiments conducted on univariate time series classification datasets demonstrate that EvoTSC significantly outperforms eleven benchmark methods in most comparisons. Further analyses verify the contribution of each component and the resource efficiency of the evolved models.
Zhongxiang Sun, Chenghao Yang, Haojie Ding, Boheng Zhang, Fan Yang, Qingsong Wen, Kun Wang
Embodied agents need physical intelligence to understand their surroundings through physical perception, physical knowledge retrieval, and physical reasoning. These capabilities must work together to interpret events and make quantitative judgments. Existing benchmarks advance physical understanding and general-purpose tool use, yet the coordination of these capabilities in physical scenarios remains insufficiently characterized. Inspired by Popper's Three Worlds, we introduce TriPhysBench, a benchmark of 410 questions from 200 videos that brings physical perception, physical knowledge retrieval, and physical reasoning into a shared evaluation setting. Its measurement-grounded and knowledge-grounded subtasks impose complementary demands on obtaining observations and acquiring external specifications, then applying them to quantitative problems. Among six representative vision-language models, the best achieves an answer accuracy of only 50.38%, revealing limited performance on tasks that require these capabilities to work together. We further introduce TriPhys, a framework that learns a unified policy for physical perception, knowledge retrieval, and quantitative reasoning from complete tool-integrated trajectories. The policy uses intermediate reasoning to guide retrieval and computation, then incorporates tool feedback into subsequent physical inference. On 110 questions from videos excluded from training, TriPhys improves answer and solution scores across three Qwen configurations, with a 17.7% relative gain in answer accuracy for TriPhys-27B over its Qwen3.6-27B backbone. This work provides a benchmark and a learning pathway toward embodied agents that move beyond recognizing events to understanding and reasoning about the physical processes behind them.
Changqing Shi, Anan Liu, Genbao Chen, Junchao Yang
The sparse path, comprising sparse operations and collective communication, is a critical component of large-scale recommendation models. However, these operations are often represented as opaque nodes in the training computation graph, hindering whole-graph optimization. We present SparseLift, a low-overhead compiler for the sparse path of large-scale recommendation training. It exposes sparse computation and communication to whole-graph analysis to expand the optimization space and improve training efficiency. By analyzing model structure and runtime information, SparseLift reconstructs opaque nodes as analyzable subgraphs, recovering the resource semantics of sparse computation and communication within the graph. Building on the recovered semantics, we design incremental cost models and a constraint-aware planner to automatically select and organize subgraph optimizations into an execution plan that is reused during subsequent steady-state training. We evaluate SparseLift on an industrial video-platform recommendation task with 100B sparse parameters and the public DLRM benchmark. SparseLift improves steady-state throughput for ordinary industrial training steps by 16.5\% over the native baseline. On DLRM, it achieves $1.60\times$ and $1.26\times$ model-execution speedups over DLRM-Baseline and TorchRec, respectively.
Yubin Kim, Viresh Pati, Jevon Twitty, Vinh Pham, Shihao Yang, Jiecheng Lu
Transformer architectures have established strong baselines in time series forecasting, yet they typically rely on positional encodings that assume uniform, index-based temporal progression. However, real-world systems, from shifting financial cycles to elastic biological rhythms, frequently exhibit "time-warped" dynamics where the effective flow of time decouples from the sampling index. In this work, we first formalize this misalignment and prove that rotary position embedding (RoPE) is mathematically incapable of representing non-affine temporal warping. To address this, we propose Symplectic Positional Embeddings (SyPE), a learnable encoding framework derived from Hamiltonian mechanics. SyPE strictly generalizes RoPE by extending the rotation group $\mathrm{SO}(2)$ to the symplectic group $\mathrm{Sp}(2,\mathbb{R})$, modulated by a novel input-dependent adaptive warp module. By allowing the attention mechanism to adaptively dilate or contract temporal coordinates end-to-end, our approach captures locally varying periodicities without requiring pre-defined warping functions. We implement this mechanism in StretchTime, a multivariate forecasting architecture that achieves state-of-the-art performance on standard benchmarks, demonstrating superior robustness on datasets exhibiting non-stationary temporal dynamics.
Zhichao Yang, Sepehr Janghorbani, Dongxu Zhang, Jun Han, Qian Qian, Andrew Ressler, Gregory D. Lyng, Sanjit Singh Batra 等 9 位作者
Reinforcement learning is a powerful approach for improving language models, but applying it to open-ended domains such as healthcare remains challenging because high-quality responses are often multidimensional, context-dependent, and difficult to capture with a single reward signal. Rubrics offer a natural form of supervision by providing a structured representation of the criteria that define a strong response. Existing rubric-based approaches, however, often rely on instance-specific criteria, creating two important limitations: (1) such rubrics are typically unavailable for unseen examples at test time; and (2) constructing new rubrics for every instance incurs substantial reward design costs. We introduce Health-SCORE, a scalable framework that captures recurring principles of response quality in the form of reusable meta-rubrics. Rather than relying on manually constructed instance-level criteria, Health-SCORE captures generalizable principles that can be applied across examples, tasks, and healthcare datasets. These meta-rubrics support two forms of model improvement: they can serve as structured, safety-aware reward signals for reinforcement learning during post-training, and they can be incorporated directly into prompts to guide model behavior at inference time. Across several open-ended healthcare benchmarks, models trained with Health-SCORE achieve performance competitive with models trained using human-authored instance-level rewards, while substantially reducing reward-design costs and improving generalizability.
From extreme weather to agricultural consequences with agentic AI
Minghan Yu, Mengyu Dai, Yixin Wen, Haizhao Yang
AI 速览：提出WxCrop智能体，将可物理解释的天气指数与因果分析结合，预测历史与预报日期下作物损失。
Extreme weather is forecast days ahead, yet its consequences for agriculture remain hard to predict: a forecast says where a storm will strike, not which crops will suffer or how much is at risk, and the records that could tell span many sources. We present WxCrop, the first agent that judges agricultural hazards from physically interpretable weather indices examined by causal analysis, and the first that combines weather and agricultural information to predict crop loss reliably for past and forecast dates. We demonstrate it in Florida on hurricanes, freezes and dry heat, with a loss model calibrated on insurance records for each, and the same procedure extends to other hazards and areas. Three checks show reliability: the hurricane index detects recorded cyclone days with an AUC of 0.98, recorded losses in held-out years are reproduced to within 2%, and independent production records confirm the county rankings (rank correlation 0.63). Users ask the agent for each crop group's loss as a fraction of expected season value, every number sourced, so that governments, growers and scientists respond to agricultural disasters quickly.
AI 速览：PIVOT 将教师生成的纠正片段插入学生 rollout，并对这些片段施加 token 级 KL 监督，与 RLVR 目标联合训练，以缓解困难推理任务中稀疏奖励无学习信号的问题。
Reinforcement learning from verifiable rewards (RLVR) is limited by sparse outcome-level feedback on hard reasoning tasks: when all rollouts in a group fail, rewards provide no learning signal. Stronger teachers can provide additional guidance, but this raises a central question: how can such signals be incorporated into RLVR to support successful rollouts while remaining learnable for the student? We distinguish two complementary mechanisms: trajectory-level intervention, which improves access to reward-bearing trajectories, and token-level supervision, which provides dense targets to guide student learning. Our controlled experiments show that either mechanism alone can remain insufficient, motivating their combination. We therefore introduce PIVOT, which inserts teacher-generated corrections into student rollouts and applies token-level KL supervision to these segments alongside the RLVR objective. Beyond their combination, we systematically analyze where to intervene and how to supervise, yielding a recipe for effective guidance. Across hard logic games and mathematical reasoning benchmarks, PIVOT consistently outperforms existing methods, nearly doubling mean avg@8 on logic games from 19.5% for the best-performing baseline to 35.6%.
FARMS: A Verifiable Foundation for Automated Computational Science
Di Wu, Ling Liang, Haizhao Yang
AI 速览：作者提出 FARMS，将计算数学知识转化为可供 AI 解释与复用、并带有显式保证的“计算知识”基础。
Automated computational science could accelerate discovery by selecting, adapting and developing numerical methods. This prospect requires mathematical knowledge that AI systems can interpret and reuse with explicit guarantees. Here we present FARMS, a faithful and efficient foundation that transforms computational mathematics into AI-ready computing knowledge. FARMS connects formal statements to source context, algorithmic meaning and quantitative guarantees through source preparation, reusable representations and semantic review. Across 171 targets in randomized numerical linear algebra and numerical partial differential equations, 157 were accepted by the pipeline, with mean estimated costs of US \$0.85-2.29 per target across sources. Lean-checked implications from explicit source premises connect selected statements to computational skills, illustrated by a heat-flow solve and wave-response compression. FARMS therefore supports an affordable, reusable knowledge foundation for trustworthy automated computational science.
Qifan Wang, Qihang Fang, Boao Shi, Yushi Du, Pei Zhou, Siyan Dong, Yanchao Yang
Modeling human-object interaction (HOI) in real environments requires jointly representing whole-body motion, manipulated objects, and scene geometry, yet existing datasets trade off scalable capture against structured, scene-aligned supervision. We introduce SituHOI, a large-scale RGB-D dataset containing 18,429 annotated clips across 16 indoor rooms and approximately 86 hours of quality-screened activity. It pairs recordings from a single moving RGB-D camera with static scene scans, object geometry, language descriptions, and estimated human and object trajectories. Using the scans as persistent references, our pipeline combines camera localization and depth-based metric alignment with human mesh recovery, object tracking, and temporal refinement to register all geometric annotations in a shared floor-aligned frame. These data support FlexHOI-Bench, a goal-conditioned joint motion prediction benchmark with variable observation and prediction durations. Given observed histories, scene and object geometry, and a terminal object pose, models predict future human and object motion. We provide SharedHOI as a reference predictor and evaluate world-space accuracy and human-object coordination. On a MoCap subset, our pipeline reduces human G-MPJPE from 195.4 to 68.3 mm and object translation error from 43.0 to 28.7 mm. SharedHOI leads four adapted baselines on all three aggregate FlexHOI-Bench metrics. Data will be released upon acceptance. An anonymized code repository is available at https://anonymous.4open.science/r/iclr2027-anonymous-code-1DFB.
Discounted regret measures the performance of an online learner by assigning exponentially decaying weights to past losses, reflecting the greater importance of recent losses in non-stationary online learning. While previous work obtains uniform discounted regret bounds over a continuous range of discount factors using a specialized meta-algorithm, its results are limited to convex functions. In this paper, we demonstrate that strongly adaptive regret bounds can directly imply uniform discounted regret guarantees for convex, strongly convex, and exp-concave loss functions. Our key observation is that discounted regret against a fixed comparator admits an exact decomposition into a convex combination of interval regrets. For convex functions, this reduction yields an $O(\sqrt{\log T/(1-λ)})$ discounted regret bound for all discount factors $λ\in(0,1)$, matching the previous rate over a broader range of discount factors. For strongly convex and exp-concave functions, we provide regret bounds of $O\left(\log T\left(1+\log\frac{1}{1-λ}\right)\right)$.
Leo Betthauser, Yingqi Liu, Uday Shah, Jishang Wei, Sainyam Galhotra, Liang Gou, Vedant Dharnidharka, Hao Yang
Many log-analysis tasks involve choosing among known options, such as detecting anomalous sessions or identifying attack techniques, and benefit from calibrated probabilities. We present Lev-3B, an open 3B-parameter decision model built on Granite-4.1-3B-Base with a shared pointer readout, and study its training recipe. The recipe has two key parts. First, training follows a curriculum: the model learns bounded natural-language decisions, using no log data, before it sees any logs. Second, the log stage mixes the target tasks with general log data spanning 35 auxiliary tasks unrelated to either evaluation task. We evaluate one shared checkpoint on sources seen in training and on sources excluded from training, calibration and model selection. The two-stage recipe transfers substantially better than training on log data alone. Adding general log data improves performance on excluded sources. At a fraction of their serving cost, Lev-3B outperforms prompted frontier models on anomaly detection, including on an excluded logging system, but trails them on attack-technique classification. Its probabilities are well calibrated on familiar sources, but on the excluded logging system it confidently misses anomalies. We release the checkpoint, recipe, data manifests, and prediction-level artifacts.
In this paper, we investigate decentralized online convex optimization with compressed communication, where $n$ learners over a network jointly minimize a sequence of global loss functions based only on their local losses and compressed messages from their neighbors. The best known regret bounds for this problem are $O(ω^{-1/2}nρ^{-1}\sqrt{T\ln n})$ and $O(ω^{-1}nρ^{-2}\ln n\ln T)$ for convex and strongly convex loss functions, respectively, where $ρ\in(0,1)$ is the spectral gap of the communication matrix and $ω\in(0,1]$ is the compression quality factor. Compared with the corresponding lower bounds of $Ω(ω^{-1/2}nρ^{-1/4}\sqrt{T})$ and $Ω(ω^{-1}nρ^{-1/2}\ln T)$, these upper bounds exhibit polynomial gaps in their dependence on $ρ^{-1}$. To close these gaps, we first develop Ascent-gossip, which combines an online compression strategy with a Chebyshev recursion and attains a faster consensus rate than existing compressed gossip methods. Furthermore, by adopting Ascent-gossip, we propose an online algorithm that achieves optimal regret bounds of $\widetilde O(ω^{-1/2}nρ^{-1/4}\sqrt{T})$ and $\widetilde O(ω^{-1}nρ^{-1/2}\ln T)$ for convex and strongly convex loss functions, respectively. Up to logarithmic factors in $n$ and $ρ^{-1}$, these results match the lower bounds in terms of $T$, $n$, $ρ$, and $ω$.
Hao Yang, Zhuo Ma, Yang Liu, Yilong Yang, Guancheng Wang, JianFeng Ma
Large vision-language models (LVLMs) have emerged as a powerful paradigm for multimodal intelligence, but their growing deployment also expands the attack surface of prompt injection. Despite this growing concern, existing attacks still suffer from a critical limitation: the injected prompt for one modality only steers the model's interpretation of that singular input. Alternatively, these attacks remain multimodal but fail to achieve cross-modal prompt perturbation. To bridge this gap, we introduce a novel cross-modal prompt injection attack CrossMPI, which can steer the model's interpretation of both textual and visual inputs via image-only prompt injection. Our design is underpinned by the following key breakthroughs. First, we turn the focus of the injected prompt perturbation optimization from the visual embedding space (typically with only $10^5$ parameters) to the model hidden state space (for multimodal information integration and with $10^7$ parameters). Then, two strategies are adopted to mitigate the optimization challenges posed by the larger parameter space. To constrain the optimized model parameter space, we introduce a layer selection strategy that identifies the layers most critical to multimodal integration. Interestingly, deviating from the past experience, our analysis reveals that the optimal layers for LVLM prompt perturbation reside in the middle of the model rather than the last. To constrain the image perturbation space, we propose a new distance-decremental perturbation budget assignment strategy that allocates budgets decrementally as the pixel distance to semantic-critical regions increases. Extensive experiments across multiple LVLMs and datasets show that our method significantly outperforms baseline approaches.
Modern large language models (LLMs) often exhibit overconfident errors, assigning high confidence to incorrect predictions and thereby undermining reliability in real-world use. Existing calibration methods typically rely on post-hoc adjustment or retraining, operating at the output level without addressing the underlying causes of miscalibration. We propose CoCHI, (Class-oriented Causal Head Intervention), a framework that improves calibration by identifying and intervening on internal components responsible for distinct confidence behaviours. By analysing predictions through correctness-confidence patterns, CoCHI enables targeted intervention on model internals, reducing overconfident errors while preserving correct predictions. Our approach provides a mechanistic perspective on calibration, linking reliability to identifiable structures within the model rather than treating it as an output-level issue. Experiments across multiple LLMs show consistent improvements in calibration metrics, including Expected Calibration Error, Negative Log-Likelihood, and Brier Score, while maintaining accuracy on five multiple-choice question answering benchmarks. These results suggest that effective calibration requires not only adjusting outputs, but also addressing the underlying causal mechanisms within the model.
Zhichao Shi UCAS, Cehao Yang The Hong Kong University of Science and Technology, Hao Zhou, Xiaojun Wu The Hong Kong University of Science and Technology and IDEA Institution, Huajie Li International Digital Economy Academy, International Digital Economy Academy, Xuhui Jiang International Digital Economy Academy, International Digital Economy Academy, Chengjin Xu International Digital Economy Academy, Yuanzhuo Wang Chinese Academy of Sciences 等 9 位作者
Synthetic data has emerged as a crucial solution to the data scarcity bottleneck in large language models (LLMs), particularly for specialized domains and low-resource languages. However, the broader adoption of existing synthetic data tools is severely hindered by convoluted workflows, fragmented data standards, and limited scalability across modalities.To address these limitations, we develop DataArc-SynData-Toolkit, an open-source framework featuring: (1) a configuration-driven, end-to-end pipeline equipped with an intuitive visual interface and simplified CLI for exceptional usability; (2) a unified, quality-controllable synthesis paradigm that standardizes multi-source data generation to ensure high reusability; and (3) a highly modular architecture designed for seamless multimodal, multilingual, and multi-task adaptation.We apply the toolkit in multiple application scenarios. Experimental results demonstrate that our toolkit achieves an optimal balance between generation efficiency and data quality. By offering an end-to-end and visually interactive pipeline, DataArc-SynData-Toolkit significantly lowers the technical barrier to synthetic data generation and subsequent model training, accelerating its practical deployment in real-world applications.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) · 2026 · 正式发表
Xiaowei Yuan, Ziyang Huang, Zhao Yang Meituan, Yequan Wang Peking University and Beijing Academy of Artificial Intelligence, Jun Zhao Institute of automation, Chinese academy of science, Kang Liu Institute of automation, Chinese academy of science, Chinese Academy of Sciences
Retrieval-Augmented Generation (RAG) has become a standard paradigm for grounding Large Language Models (LLMs) with external knowledge. However, RAG performance often degrades substantially when faced with noisy, outdated, or conflicting retrieved information. In this work, we empirically demonstrate that Prior-Guided Reasoning—a strategy that explicitly elicits the model’s parametric knowledge as prior information to guide reasoning on retrieved documents—effectively mitigates the impact of external conflicts. Building on this, we propose BrPr (Bernoulli-gated reinforcement learning for Prior-Guided reasoning), a framework that achieves robust performance across varying degrees of external inconsistency. Furthermore, by employing a Bernoulli-gated dropout mechanism during training, BrPr distills the prior-driven reasoning capability into the model parameters, enabling efficient latent reasoning without explicit prior generation. The experimental results demonstrate that BrPr consistently exhibits superior robustness to external conflicts and noise.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Tengchao Yang, Sichen Guo, Mengzhao Jia, Jiaming Su, Yuanyang Liu, Zhihan Zhang Amazon, Meng Jiang University of Notre Dame
Effective math tutoring requires not only solving problems but also diagnosing students’ difficulties and guiding them step by step. While multimodal large language models (MLLMs) show promise, existing benchmarks largely overlook these tutoring skills. We introduce MMTutorBench, the first benchmark for AI math tutoring, consisting of 770 problems built around pedagogically significant key-steps. Each problem is paired with problem-specific rubrics that enable fine-grained evaluation across six dimensions, and structured into three tasks—Insight Discovery, Operation Formulation, and Operation Execution. We evaluate 12 leading MLLMs and find clear performance gaps between proprietary and open-source systems, substantial room compared to human tutors, and consistent trends across input variants: OCR pipelines degrade tutoring quality, few-shot prompting yields limited gains, and our rubric-based LLM-as-a-Judge proves highly reliable. These results highlight both the difficulty and diagnostic value of MMTutorBench for advancing AI tutoring.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Siran Liu, Guoxia Wang, Sa Wang, Jinle Zeng Baidu, Haoyang Xie, Siyu Lou Beijing University of Post and Telecommunications, Jiabin Yang, Dianhai Yu 等 10 位作者
The quadratic complexity of attention mechanisms poses a critical bottleneck for large language models processing long contexts. While dynamic sparse attention methods offer input-adaptive efficiency, they face fundamental trade-offs: requiring preprocessing, lacking global evaluation, violating query independence, or incurring high computational overhead. We present RRAttention, a novel dynamic sparse attention method that simultaneously achieves all desirable properties through a head round-robin (RR) sampling strategy. By rotating query sampling positions across attention heads within each stride, RRAttention maintains query independence while enabling efficient global pattern discovery with stride-level aggregation. Our method reduces complexity from O(L^2) to O(L^2/S^2) and employs adaptive Top-\tau selection for optimal sparsity. Extensive experiments on natural language understanding (HELMET) and multimodal video comprehension (Video-MME) demonstrate that RRAttention recovers over 99% of full attention performance while computing only half of the attention blocks, achieving 2.4\times speedup at 128K context length and outperforming existing dynamic sparse attention methods. The code is available at https://github.com/PaddlePaddle/PaddleFleet (see ‘Research/RRAttention‘).
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Fei Zhang, Zhe Zhao, HaiBin Wen, Tianshuo Wei, Zaixi Zhang Princeton University, Chao Yang Shanghai Jiaotong University, Ye Wei City University of Hong Kong
The rapid advance of automatic survey generation (ASG) has created a critical evaluation challenge. Existing evaluation methods suffer from both cognitive dimensional simplification and methodological unreliability, primarily due to the over-reliance on the ”LLM-as-a-Judge” approach. To bridge this gap, we establish Bloom-Eval, a six-tiered benchmark based on Bloom’s Taxonomy that reliably evaluates ASG systems by prioritizing deterministic algorithms and introducing our GRADE approach for abstract abilities. Furthermore, we construct a large-scale, cross-disciplinary dataset of over 3,000 high-quality papers. Our empirical study on this benchmark reveals that while leading ASG systems are proficient format organizers, they remain unqualified knowledge integrators. This work aims to redefine ASG evaluation standards, shifting the research focus from the formal mimicry of surface structure to the cognitive deepening of intellectual content. Our method provides the ASG field with a systematic, reproducible, and theoretically grounded benchmark to guide future research.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Zihan Chen Singapore University of Technology and Design, Howard Hao Yang Zhejiang University, Tony Quek Singapore University of Technology and Design, Kai Fong Ernest Chong Singapore University of Technology and Design
Federated fine-tuning of large language models (LLMs) provides a privacy-preserving approach to deploying pervasive generative AI services, yet the substantial memory overhead of first-order (FO) gradient computation presents significant practical challenges. While zeroth-order (ZO) optimization methods offer memory-efficient alternatives, they remain susceptible to performance degradation brought by data heterogeneity. Specifically, direct ZO-for-FO substitution is incompatible with existing strategies tailored for cross-client discrepancies. In response, we propose a new federated LLM fine-tuning framework, with a holistic revamped design of the entire ZO gradient processing pipeline. Crucially, with our proposed global adaptive optimization and local personalized perturbation, we present a unified solution for incorporating ZO gradients in federated learning, from local personalized perturbation sampling and ZO gradient transmission, to global ZO gradient reconstruction and aggregation with adaptive momentum, thereby directly addressing the challenges of inefficiencies and cross-client discrepancies. Our convergence analysis and experiment results demonstrate the superiority of our proposed framework over diverse heterogeneous data settings, both in terms of generalization and efficiency.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表