论文检索

Wenxuan Xu, Arvind Pillai, Subigya Nepal, Amanda C. Collins, Daniel M Mackin, Michael V. Heinz, Tess Z Griffin, Nicholas C. Jacobson, Andrew Campbell

Multimodal health sensing offers rich behavioral signals for assessing mental health, yet translating these numerical time-series measurements into natural language remains challenging. Current LLMs cannot natively ingest long-duration sensor streams, and paired sensor–text datasets are scarce. To address these challenges, we introduce LENS, a framework that aligns multimodal sensing data with language models to generate clinically grounded mental-health narratives. LENS first constructs a large-scale dataset by transforming Ecological Momentary Assessment (EMA) responses related to depression and anxiety symptoms into natural-language descriptions, yielding over 100,000 sensor–text QA pairs from 258 participants. To enable native time-series integration, we train a patch-level encoder that projects raw sensor signals directly into an LLM’s representation space. Our results show that LENS outperforms strong baselines on standard NLP metrics and task-specific measures of symptom-severity accuracy. A user study with 13 mental-health professionals further indicates that LENS-produced narratives are comprehensive and clinically meaningful. Ultimately, our approach advances LLMs as interfaces for health sensing, providing a scalable path toward models that can reason over raw behavioral signals and support downstream clinical decision-making.

Haoxiang Sun, Yingqian Min, Zhipeng Chen, Xin Zhao, Ji-Rong Wen

The rapid advancement of large reasoning models has saturated existing math benchmarks, underscoring the urgent need for more challenging evaluation frameworks. To address this, we introduce OlymMATH, a rigorously curated, Olympiad-level math benchmark comprising 350 problems, each with parallel English and Chinese versions. OlymMATH is the first benchmark to unify dual evaluation paradigms within a single suite: (1) natural language evaluation through OlymMATH-EASY and OlymMATH-HARD, comprising 200 computational problems with numerical answers for objective rule-based assessment, and (2) formal verification through OlymMATH-LEAN, offering 150 problems formalized in Lean 4 for rigorous process-level evaluation. All problems are manually sourced from printed publications to minimize data contamination, verified by experts, and span four core domains. Extensive experiments reveal the benchmark’s significant challenge, and our analysis also uncovers consistent performance gaps between languages and identifies cases where models employ heuristic "guessing" rather than rigorous reasoning. To further support community research, we release 582k+ reasoning trajectories, a visualization tool, and expert solutions at https://github.com/RUCAIBox/OlymMATH.

Tianci Liu, Ran Xu, Tony Yu, Ilgee Hong, Carl Yang, Tuo Zhao, Haoyu Wang

Reward modeling lies at the core of reinforcement learning from human feedback (RLHF), yet most existing reward models rely on scalar or pairwise judgments that fail to capture the multifaceted nature of human preferences. Recent studies have explored rubrics-as-rewards (RaR) that uses structured criteria to capture multiple dimensions of response quality. However, producing rubrics that are both reliable and scalable remains a key challenge. In this work, we introduce OpenRubrics, a diverse, large-scale collection of (prompt, rubric) pairs for training rubric-generation and rubric-based reward models. To elicit discriminative and comprehensive evaluation signals, we introduce Contrastive Rubric Generation (CRG), which derives both hard rules (explicit constraints) and principles (implicit qualities) by contrasting preferred and rejected responses. We further remove noisy rubrics via preserving preference–label consistency. Across multiple reward-modeling benchmarks, our rubric-based reward model, Rubric-RM, surpasses strong size-matched baselines by 8.4%. These gains transfer to policy models on instruction-following and biomedical benchmarks.The model weights and datasets are publicly available at https://huggingface.co/OpenRubrics.

Jiaju Chen, Yuxuan Lu, Xiaojie Wang, Huimin Zeng, Jing Huang, Jiri Gesi, Ying Xu, Bingsheng Yao, Dakuo Wang

Nearly all human work is collaborative; thus, the evaluation of real-world NLP applications often requires multiple dimensions that align with diverse human perspectives. As real human evaluator resources are often scarce and costly, the emerging "LLM-as-a-judge" paradigm sheds light on a promising approach to leverage LLM agents to believably simulate human evaluators. Yet, to date, existing LLM-as-a-judge approaches face two limitations: persona descriptions of agents are often arbitrarily designed, and the frameworks are not generalizable to other tasks. To address these challenges, we propose MAJ-EVAL, a Multi-Agent-as-Judge evaluation framework that can automatically construct multiple evaluator personas with distinct dimensions from relevant text documents (e.g., research papers), instantiate LLM agents with the personas, and engage in-group debates with multi-agents to generate multi-dimensional feedback. Our evaluation experiments in both the educational and medical domains demonstrate that MAJ-EVAL can generate evaluation results that better align with human experts’ ratings compared with conventional automated evaluation metrics and existing LLM-as-a-judge methods.

Mengyang Li, Jingwen Wang, Pinlong Zhao

Direct Preference Optimization (DPO) has become a standard approach for aligning large language models with human preferences, yet existing methods treat all preference pairs uniformly during training. We identify two distinct sources of learning difficulty: Input Complexity (IC), capturing prompt understanding challenges, and Output Ambiguity (OA), measuring preference discrimination difficulty. Through systematic analysis, we demonstrate that these dimensions induce asymmetric learning dynamics, with IC-related competencies developing rapidly in early training while OA-related competencies emerge more gradually. Building on this observation, we propose DECOPO, a training framework that maintains separate, adaptive pacing schedules for each dimension. Experiments on UltraFeedback show that DECOPO achieves 42.3% length-controlled win rate on AlpacaEval 2.0 and 7.66 on MT-Bench, outperforming curriculum baselines by 2.1% and 0.21 points respectively, while matching full-data baseline performance with only 75% of training samples.

Zhenwen Liang, Ruosen Li, Yujun Zhou, Linfeng Song, Dian Yu, Xinya Du, Haitao Mi, Dong Yu

Assessing the quality of Large Language Model (LLM) outputs becomes especially challenging in high-branching settings, where a single prompt yields many plausible candidates. Existing verifiers typically operate on the surface text (e.g., reward models, LLM judges, majority voting) or on confidence proxies derived from token probabilities, both of which can be brittle: the former can be influenced by stylistic artifacts, while the latter is often miscalibrated. In this paper, we study a third source of information—the model’s hidden states—for binary correctness verification in tasks with a reliable success/failure signal (e.g., deterministic checkers or reference-grounded answers). We find that correct and incorrect solutions exhibit measurable geometric differences in their hidden-state trajectories. To isolate this signal with minimal modeling assumptions, we introduce Clue (Clustering and Experience-based Verification), a training-free, non-parametric verifier. Clue summarizes each reasoning trace by an activation delta—the difference between hidden states at the start and end of the explicit reasoning span—and predicts correctness by comparing this delta to two class centroids computed from labeled experience. Across math (AIME 24/25), scientific QA (GPQA), and a multi-domain benchmark (WebInstruct-verified), Clue improves selection and reranking, with particularly strong gains on smaller or less-calibrated models. For example, on AIME 24 with a 1.5B model, Clue raises accuracy from 56.7% (majority@64) to 70.0% (top-maj@16).

Jiaqi Shi, Xulong Zhang, Yuechan Li, Xiaoyang Qu, Jianzong Wang

High-resolution Multimodal Large Language Models (MLLMs) face prohibitive computational costs during inference due to the explosion of visual tokens. Existing acceleration strategies, such as token pruning or layer sparsity, suffer from severe "backbone dependency", performing well on Vicuna or Mistral architectures (e.g., LLaVA) but causing significant performance degradation when transferred to architectures like Qwen. To address this, we leverage truncated matrix entropy to uncover a universal three-stage inference lifecycle, decoupling visual redundancy into universal Intrinsic Visual Redundancy (IVR) and architecture-dependent Secondary Saturation Redundancy (SSR). Guided by this insight, we propose HalfV, a framework that first mitigates IVR via a unified pruning strategy and then adaptively handles SSR based on its specific manifestation. Experiments demonstrate that HalfV achieves superior efficiency-performance trade-offs across diverse backbones. Notably, on Qwen25-VL, it retains 96.8% performance at a 4.1\times FLOPs speedup, significantly outperforming state-of-the-art baselines. Our code is available at https://github.com/civilizwa/HalfV.

George Ma, Anurag Koul, Qi Chen, Yawen Wu, Sachit Kuhar, Yu Yu, Aritra Sengupta, Varun Kumar, Murali Krishna Ramanathan

Large Language Models (LLMs) excel at code-related tasks but often struggle in realistic software repositories, where project-specific APIs and cross-file dependencies are crucial. Retrieval-augmented methods mitigate this by injecting repository context at inference time. Low inference time latency budget either affects retrieval quality or the added latency impacts user experience adversely. We address this limitation with SpecAgent, an agent that enhances both latency and code-generation quality by proactively exploring repository files during indexing and constructing speculative context that anticipates future edits in each file. This indexing-time asynchrony allows thorough context computation masking latency and the speculative nature of the context improves code-generation quality. Additionally, we identify the problem of future context leakage in existing benchmarks, which can inflate reported performance. To address this, we construct a synthetic, leakage-free benchmark that enables a more realistic evaluation of our agent against baselines. Experiments show that SpecAgent consistently achieves absolute gains of 9–11% (48–58% relative) compared to the best-performing baselines, while significantly reducing inference latency.

Akshith Reddy Putta, Jacob Devasier, Chengkai Li

Automated Fact-Checking has largely focused on verifying general knowledge against static corpora, overlooking high-stakes domains like law where truth is evolving and technically complex. We introduce CaseFacts, a benchmark for verifying colloquial legal claims against U.S. Supreme Court precedents. Unlike existing resources that map formal texts to formal texts, CaseFacts challenges systems to bridge the semantic gap between layperson assertions and technical jurisprudence while accounting for temporal validity. The dataset consists of 6,294 claims categorized as Supported, Refuted, or Overruled. We construct this benchmark using a multi-stage pipeline that leverages Large Language Models (LLMs) to synthesize claims from expert case summaries, employing a novel semantic similarity heuristic to efficiently identify and verify complex legal overrulings. Experiments with state-of-the-art LLMs reveal that the task remains challenging; notably, augmenting models with unrestricted web search degrades performance compared to closed-book baselines due to the retrieval of noisy, non-authoritative precedents. We release CaseFacts to spur research into legal fact verification systems.

Avinash Amballa, Yashas Malur Saidutta, Chi-Heng Lin, Vivek Kulkarni, Srinivas Chappidi

Large language models (LLMs) are increasingly being used to generate synthetic datasets for the evaluation and training of downstream models. However, prior work has noted that such generated data lacks diversity. In this paper, we propose VOYAGER, a novel principled approach to generate diverse datasets. Our approach is iterative and directly optimizes a mathematical quantity that optimizes the diversity of the dataset using the machinery of determinantal point processes. Furthermore, our approach is training-free, applicable to closed-source models, and scalable. In addition to providing theoretical justification for the working of our method, we also demonstrate through comprehensive experiments that VOYAGER improves diversity by \mathbf{1.5}-\mathbf{3} times compared to popular baseline approaches.

Ziyue Yan, Hongying Zan, Xinglin Lyu, Hongfei Xu

Paraphrasing uses different words, sentence structures, or expressions to convey similar semantics. It is an effective training data augmentation method to improve low-resource Natural Language Processing (NLP) tasks. Existing studies normally leverage parallel corpora to construct parabanks, regarding the Machine Translation (MT) results of source sentences as the paraphrases of the corresponding target sentences. As MT models are usually trained on the same parallel corpus, translation of the training set may suffer from overfitting, which leads to less diverse paraphrases. Training paraphrasers on the parabank generated via MT may also suffer from the information loss issue, as the parabank is derived from the parallel corpora, and the knowledge inside the parabank is a subset of that inside the parallel corpora. In this paper, we train bidirectional Multilingual Neural Machine Translation (MNMT) on the bi-directional bilingual parallel corpus, and use the MNMT model directly as a paraphrasing model by asking it to generate "translations" of the input language. As some source tokens also appear in the translation in the parallel corpus, we introduce "copy"/"not-copy" tags to indicate the existence/non-existence of source tokens in the target translation during training, and use the "not-copy" tag to encourage paraphrasing during inference. Manual and automatic evaluation results show that our ParaMNMT method can generate paraphrases of higher semantic consistency, literal fluency and sentential diversity compared to existing parabanks and LLMs. Our data augmentation experiments verify the effectiveness of ParaMNMT on improving low-resource NLP tasks.

Aaron Mueller, Andrew Lee, Shruti Joshi, Ekdeep Singh Lubana, Dhanya Sridhar, Patrik Reizinger

A goal of interpretability is to recover disentangled representations of latent concepts (features) from the activations of neural networks. The quality of features is typically evaluated in isolation, and under implicit independence assumptions that may not hold in practice. Thus, it is unclear to what extent common featurization methods such as sparse autoencoders (SAEs) and probes disentangle one concept from another. We propose a multi-concept evaluation setting using concepts such as sentiment, domain, voice, and tense. We evaluate how well featurizers produce disentangled representations of each concept, observing that features are typically sensitive to only one concept, but also that concepts are distributed across many features. Then, we steer these features, measuring whether each concept is independently manipulable, and whether features interact. Even in idealized settings, steering a feature often affects many concepts, despite a near absence of interaction effects. These results suggest that correlational metrics are insufficient to establish steering selectivity, and that demonstrating that two features operate in separate spaces is insufficient to claim that they will be selective for one concept. These results underscore the importance of multi-concept evaluations in interpretability research.

Shira Wein, Anna Serbina, Jiyuan Ji, Nathan Wolf, Jason DeGraaff, Prajakta Kini, Maria Leonor Pacheco

Translationese refers to the statistical patterns that distinguish translated texts from original texts, which are often subtle and imperceptible to human readers. When translated texts appear in either training or testing data, these patterns can negatively affect model performance or warp model evaluation. We approach the task of discerning whether a text was originally written in English or translated into English by fine-tuning contemporary foundation models at distinct item lengths and achieve state-of-the-art performance (94% Macro F1). Given that these linguistic cues are subtle and often imperceptible to humans, we analyze the features which enable our model’s high performance. Employing a suite of interpretability-based techniques, we find that: (1) our high accuracy is enabled by a collection of linguistic features, a number of which correspond with linguistic theories of translationese, and (2) pretrained neural models are adept at picking up these features without any fine-tuning.

Muyang Li, Runze Wu, Xiangyu Zhao, Bo Han, Daoyi Dong, Tongliang Liu

The post-training stage of Large Language Models (LLMs) typically involves Supervised Fine-Tuning (SFT) followed by preference alignment to ensure LLM to generate safe, helpful, and instruction-aligned content. The SFT model critically serves as both the initialization and reference model for subsequent preference alignment. However, an essential yet often neglected question is the optimal selection of the SFT checkpoint for this role. We show that checkpoint selection substantially affects final performance, and that the common practice of choosing the minimum validation-loss checkpoint often fails, due to a fundamental conflict between SFT’s focus on imitation and alignment’s goal of response discriminability. To this end, we propose RewardRank, a simple, effective, training-free metrics for estimating initial implicit alignment between reference model and preference objective. Empirical evidence suggests that, using our selected model as reference can gain up to 67.6% relative increase on length-controlled win rate on the popular Zephyr recipe comparing to baselines.

Chuhan Wang, Xintong Li, Jennifer Yuntong Zhang, Junda Wu, Chengkai Huang, Lina Yao, Julian McAuley, Jingbo Shang

Multimodal large language models often struggle with faithful reasoning in complex visual scenes, where intricate entities and relations require precise visual grounding at each step. This reasoning unfaithfulness frequently manifests as hallucinated entities, mis-grounded relations, skipped steps, and over-specified reasoning. Existing preference-based approaches, typically relying on textual perturbations or answer-conditioned rationales, fail to address this challenge as they allow models to exploit language priors to bypass visual grounding. To address this, we propose SceneAlign, a framework that leverages scene graphs as structured visual information to perform controllable structural interventions. By identifying reasoning-critical nodes and perturbing them through four targeted strategies that mimic typical grounding failures, SceneAlign constructs hard negative rationales that remain linguistically plausible but are grounded in inaccurate visual facts. These contrastive pairs are used in Direct Preference Optimization to steer models toward fine-grained, structure-faithful reasoning. Across seven visual reasoning benchmarks, SceneAlign consistently improves answer accuracy and reasoning faithfulness, highlighting the effectiveness of grounding-aware alignment for multimodal reasoning.

Myeong Seok Oh, Dong-Yun Kim, Hanseok Oh, Chaean Kang, Joeun Kang, Xiaonan Wang, Hyunjung Park, Young Cheol Jung, Hansaem Kim

Current text anonymization evaluation relies on span-based metrics that fail to capture what an adversary could actually infer, and assumes a single data subject, ignoring multi-subject scenarios. To address these limitations, we present SPIA (Subject-level PII Inference Assessment), the first benchmark that shifts the unit of evaluation from text spans to individuals, comprising 675 documents across legal and online domains with novel subject-level protection metrics. Extensive experiments show that even when over 90% of PII spans are masked, subject-level inference protection drops as low as 33%, leaving the majority of personal information recoverable through contextual inference. Furthermore, target-subject-focused anonymization leaves non-target subjects substantially more exposed than the target subject. We show that subject-level inference-based evaluation is essential for ensuring safe text anonymization in real-world settings.

Zizhen Wang, Bo Feng, Zhengfeng Lai, Shiyu Li, Yang Lu, Meng Cao, Ping Huang, Xiaoming Simon Wang

Evaluating video captioning remains a critical challenge for Visual Large Language Models (VLLMs). Existing metrics primarily rely on matching generated text against ground-truth references. This paradigm suffers from the “one-to-many” nature of video description, where high-quality captions are often penalized for lexical mismatches or valid shifts in visual focus. Furthermore, such assessments are typically one-dimensional, failing to provide a fine-grained analysis of caption quality. To address this, we redefine caption quality through the lens of information fidelity: A caption must maximize the coverage of salient visual information while ensuring strict factuality. We introduce CapQuiz, a novel reference-free benchmark that assesses captions based on their utility in answering human-verified, fine-grained, multiple-choice questions derived from the video. CapQuiz features a hierarchical taxonomy of 10 question types (spanning Descriptive and Inferential categories) across 24 diverse video domains. Extensive experiments demonstrate that CapQuiz correlates significantly better with human judgments than existing metrics and offers interpretable insights into model performance. We will release the benchmark to facilitate reproducible research.

Mingfei Lu, Yi Zhang, Mengjia Wu, Yue Feng

Legal consultation question answering (Legal CQA) presents unique challenges compared to traditional legal QA tasks, including the scarcity of high-quality training data, complex task composition, and strong contextual dependencies. To address these, we construct JurisCQAD, a large-scale dataset of over 43,000 real-world Chinese legal queries annotated with expert-validated positive and negative responses, and design a structured task decomposition that converts each query into a legal element graph integrating entities, events, intents, and legal issues. We further propose JurisMA, a modular multi-agent framework supporting dynamic routing, statutory grounding, and stylistic optimization. Combined with the element graph, the framework enables strong context-aware reasoning, effectively capturing dependencies across legal facts, norms, and procedural logic. Trained on JurisCQAD and evaluated on a refined LawBench, our system significantly outperforms both general-purpose and legal-domain LLMs across multiple lexical and semantic metrics, demonstrating the benefits of interpretable decomposition and modular collaboration in Legal CQA.

Shangbin Feng, Wenxuan Ding, Alisa Liu, Zifeng Wang, Weijia Shi, Yike Wang, Shannon Zejiang Shen, Xiaochuang Han, Hunter Lang, Chen-Yu Lee 等

This position paper argues that in many realistic (i.e., complex, contextualized, subjective) scenarios, one LLM is not enough to produce a reliable output. We challenge the status quo of relying solely on a single general-purpose LLM and argue for multi-LLM collaboration to better represent the extensive diversity of data, skills, and people. We first posit that a single LLM underrepresents real-world data distributions, heterogeneous skills, and pluralistic populations, and that such representation gaps cannot be trivially patched by further training a single LLM. We then organize existing multi-LLM collaboration methods into a hierarchy, based on the level of access and information exchange, ranging from API-level, text-level, logit-level, to weight-level collaboration. Based on these methods, we highlight how multi-LLM collaboration addresses challenges that a single LLM struggles with, such as reliability, democratization, and pluralism. Finally, we identify the limitations of existing multi-LLM methods and motivate future work. We envision multi-LLM collaboration as an essential path toward compositional intelligence and collaborative AI development.

Jinbiao Wei, Yilun Zhao, Kangqi Ni, Arman Cohan

End-to-end GUI agents for real desktop environments require large amounts of high-quality interaction data, yet collecting human demonstrations is expensive and existing synthetic pipelines often suffer from limited task diversity or noisy, goal-drifting trajectories. We present a trajectory expansion framework Anchor that bootstraps scalable desktop supervision from a small set of verified seed demonstrations. Starting from each seed, we identify branch points that correspond to meaningful state changes and propose new, state-grounded task variants conditioned on the current GUI context. An executing agent then follows the proposed instructions to generate new trajectories, while a verifier enforces task completion via state-aware checks and trajectory-level consistency. To improve supervision quality, we further apply task-conditioned step-level filtering to remove ungrounded actions and denoise post-branch segments to maintain coherent intent. Experiments on standard desktop benchmarks, OSWorld and WindowsAgentArena, show that models fine-tuned on our expanded corpus achieve consistent improvements over zero-shot agents and representative synthesis baselines, and generalize across applications and operating systems.