论文检索

Zihan Liu, Jianhui li, Zexin Wang, Fei Sun, Jingjing LI, Zheyuan Li, Ke Xiang, Hang Cui, Houhua Gong, Changhua Pei 等

Evidence-intensive analytical reports are expected to be fact-dense, quantitatively correct, and supported by figures. Yet one-shot long-form generation with large language models (LLMs) frequently produces fluent but under-supported drafts: core facts are missed, numbers drift, and key visuals are absent, making the report hard to trust. We propose EviReport, an evidence-tracked report-writing workflow that improves reliability by (i) organizing corpus evidence into compact, traceable units and retrieves query-relevant subgraphs into retrieval-ready packages (ii) leveraging a reasoning-focused LLM sketches a high-level plan for full coverage, then a chat-based LLM sharpens it into a detailed hierarchical outline with explicit scope and ordering (iii) rive generation with a facts-first iterative loop: extracting verifiable facts, composing strictly from those facts, then triggering gap-aware append queries to fill missing evidence To evaluate both correctness and completeness, we introduce EviReportBench, a benchmark instantiated on data-rich indicator reports that measures factual accuracy (claim verification), factual coverage (quiz-based evaluation), and visual evidence integration (image recall). Across 8 topics, experiments show that EviReport consistently outperforms strong baselines in factual coverage (2.16\times), factual accuracy (+8.9 points), and visual evidence integration (+34 points), approaching the quality of expert-written reports across multiple dimensions.

Jian Cui, Zhiyuan Ren, Desheng Weng, Yongqi Zhao, Gong Wenbin, Yu Lei, Zhenning Dong

This paper proposes ReaGeo, an end-to-end geocoding framework based on large language models, designed to overcome the limitations of traditional multi-stage approaches that rely on text or vector similarity retrieval over geographic databases, including workflow complexity, error propagation, and heavy dependence on structured geographic knowledge bases. The method converts geographic coordinates into geohash sequences, reformulating the coordinate prediction task as a text generation problem, and introduces a Chain-of-Thought mechanism to enhance the model’s reasoning over spatial relationships. Furthermore, reinforcement learning with a distance-deviation-based reward is applied to optimize the generation accuracy. Comprehensive experiments show that ReaGeo can accurately handle explicit address queries in single-point predictions and effectively resolve vague relative location queries. In addition, the model demonstrates strong predictive capability for non-point geometric regions, highlighting its versatility and generalization ability in geocoding tasks.

Chengyou Wang, Mingchen Shao, Jingbin Hu, Zeyu Zhu, Hongfei Xue, Bingshen Mu, Xin Xu, Xingyi Duan, Binbin Zhang, Zhu Pengcheng 等

Speech processing for low-resource dialects remains a fundamental challenge in developing inclusive and robust speech technologies. Despite its linguistic significance and large speaker population, the Wu dialect of Chinese has long been hindered by the lack of large-scale speech data, standardized evaluation benchmarks, and publicly available models. In this work, we present WenetSpeech-Wu, the first large-scale, multi-dimensionally annotated open-source speech corpus for the Wu dialect, comprising approximately 8,000 hours of diverse speech data. Building upon this dataset, we introduce WenetSpeech-Wu-Bench, the first standardized and publicly accessible benchmark for systematic evaluation of Wu dialect speech processing, covering automatic speech recognition (ASR), Wu-to-Mandarin translation, speaker attribute prediction, speech emotion recognition, text-to-speech (TTS) synthesis, and instruction-following TTS (instruct TTS). Furthermore, we release a suite of strong open-source models trained on WenetSpeech-Wu, establishing competitive performance across multiple tasks and empirically validating the effectiveness of the proposed dataset. Together, these contributions lay the foundation for a comprehensive Wu dialect speech processing ecosystem, and we open-source proposed datasets, benchmarks, and models to support future research on dialectal speech intelligence.

Xiangtao Meng, Yingkai Dong, Ning Yu, Li Wang, Zheng Li, Shanqing Guo

Text-to-image (T2I) generative models have achieved remarkable visual fidelity, yet remain vulnerable to generating unsafe content. Existing safety defenses typically intervene internally within the generative model, but suffer from severe concept entanglement, leading to degradation of benign generation quality—a trade-off we term the Safety Tax. To overcome this limitation, we advocate a paradigm shift from destructive internal editing to external safety rectification. Following this principle, we propose SafePatch, a structurally isolated safety module that performs external, interpretable rectification without modifying the base model. The core backbone of SafePatch is architecturally instantiated as a trainable clone of the base model’s encoder, allowing it to inherit rich semantic priors and maintain representation consistency. To enable interpretable safety rectification, we construct a strictly aligned counterfactual safety dataset (ACS) for differential supervision training. Across nudity and multi-category bench- marks and recent adversarial prompt attacks, SafePatch achieves robust unsafe suppression (7% unsafe on I2P) while preserving image quality and semantic alignment.

Duy Tung Doan, Quang Huy Phung, Dzung Nguyen, Khac-Hoai Nam Bui

Automated code generation remains a persistent challenge in software engineering, as conventional multi-agent frameworks are often constrained by static planning, isolated execution, high computational overhead, and limited adaptability to complex tasks. This paper introduces CollabCoder, a novel Plan-Code Co-Evolution framework that improves code generation through dynamic multi-agent collaboration. The core idea is to design a collaborative decision-making process between the plan module and the code module to decide which module should be executed for the debugging process. Extensive experiments on widely used benchmarks demonstrate that CollabCoder consistently improves code quality and robustness across tasks. Importantly, CollabCoder achieves performance comparable to or exceeding current state-of-the-art methods while reducing computational overhead, with efficiency gains becoming more pronounced as benchmark difficulty increases. On the more challenging LiveCodeBench and xCodeEval benchmarks, our approach improves performance by 11-20% over strong baselines while reducing the number of API calls by an average of 4-10 per execution.

Yikun Wang, Yibin Wang, Dianyi Wang, Zimian Peng, Qipeng Guo, Dacheng Tao, Jiaqi Wang

Recent progress in large language models (LLMs) has boosted mathematical reasoning, yet geometry remains challenging where auxiliary construction is often essential. Prior methods either underperform or depend on very large models (e.g., GPT-4o), making them costly. We argue that reinforcement learning with verifiable rewards (e.g., GRPO) can train smaller models to couple auxiliary construction with solid geometric reasoning. However, naively applying GRPO yields unconditional rewards, encouraging indiscriminate and sometimes harmful constructions. We propose Group Contrastive Policy Optimization (GCPO), an RL framework with two components: (1) Group Contrastive Masking, which assigns positive/negative construction rewards based on contextual utility, and (2) a Length Reward that encourages longer reasoning chains. On top of GCPO, we build GeometryZero, an affordable family of geometry reasoning models that selectively use auxiliary construction. Experiments on Geometry3K and MathVista show GeometryZero consistently outperforms RL baselines (e.g., GRPO, ToRL).

Yinan Wu, Jihang Jin, Xuhao Bao, Weiyan Zhang, Hanjing Yan, Tong Ruan, ChunMing Wang

Medical visual question answering (MedVQA) requires models to provide accurate answers given a medical image and a corresponding question. Recently, instruction tuning of general large vision–language models (LVLMs) has become a dominant paradigm for this task, enabling open-ended predictions and effective integration of multimodal information. However, existing methods synthesize instruction data from image–caption pairs that primarily focus on visual attributes, rather than knowledge-level QA generation. This situation limits the model’s ability to learn relevant medical knowledge during training, thereby restricting its performance on MedVQA. Hence, this paper proposes MedKInstruct, which incorporates a multimodal medical knowledge graph (MMKG) to assist LVLMs in synthesizing knowledge-intensive instruction data. Additionally, we design an MMKG path–based reward function to train a stronger MedVQA model through reinforcement learning. Experimental results on the public datasets Slake and VQA-RAD show that MedKInstruct outperforms previous methods by 4.16% and 4.50%. The source code is available at the following link: https://github.com/Sonder-hang/MedKinstruct

Jiayong Wan, Jiawei Chen, Zhaoxia Yin, Liu Shuyuan, Hang Su

Large Language Models (LLMs) are increasingly acting as autonomous agents, but their continuous interaction with the environment can lead to in-context reward hacking (ICRH), a phenomenon in which LLMs iteratively optimize their behavior to maximize proxy objectives, inadvertently producing harmful side effects. Existing defense methods are insufficient to address this risk, as ICRH arises not from adversarial inputs but from the model’s own over-optimization. To mitigate this issue, we propose LLM-based Constraint Optimization (LCO), a framework that effectively reduces ICRH without model fine-tuning. LCO consists of two modules: self-thought module, which guides the LLM to proactively deliberate and integrate potential safety constraints before execution; and guided evolutionary exploration module, which employs LLM-based crossover and mutation to constrain the model’s actions within a safe solution space while maintaining task performance. Experimental results demonstrate that LCO substantially alleviates ICRH in both output-refine and policy-refine scenarios. In particular, on the tweet engagement optimization task, LCO achieves a 39% reduction in the Toxicity Growth Rate (TGR) on GPT-4, while on the policy optimization benchmark, it reduces the ICRH Occurrence Rate by 15.23%, demonstrating safety improvement without sacrificing task performance.Our code is available at: https://github.com/Califoni/LCO_for_ICRH.

Yuxing Chen, Guoqing Luo, Zijun Wu, Lili Mou

Large Language Models (LLMs) exhibit social biases, which can lead to harmful stereotypes and unfair outcomes. We propose Multi-Persona Thinking (MPT), a simple inference-time framework that reduces social bias by encouraging reasoning from multiple perspectives. MPT guides the model to consider contrasting social identities, such as male and female, together with a neutral viewpoint. These viewpoints then interact through an iterative reasoning process to identify and correct biased judgments. This design transforms the potential weakness of persona assignment into a mechanism to mitigate bias. We evaluate MPT on two widely used bias benchmarks with both open-source and closed-source models. Our results show that MPT achieves a lower bias than the existing prompting-based methods while maintaining the core reasoning ability.

Shixin Jiang, Zhihao Zhu, Jiafeng Liang, Yang Wu, Ming Liu, Bing Qin

Tool-integrated reasoning (TIR) enables large language models (LLMs) to invoke external tools for tasks beyond their internal capacity but often suffers from tool overuse.Existing approaches leverage imitation learning or reward shaping to improve efficiency, yet mainly target single-tool scenarios and ignore the varying invocation costs across tools in multi-tool reasoning (MTIR). To address these gaps, we propose EMTIR-GRPO, a simple yet effective RL algorithm for cost-aware MTIR. Built upon GRPO, we introduce a composite reward considering format completeness, answer correctness, and tool efficiency.By incorporating a cost-aware coefficient with group optimal cost estimation, EMTIR-GRPO explicitly models heterogeneous tool costs and encourages more cost-effective tool-use strategies. Experiments on MTIR-QA and MTIR-TC demonstrate significant efficiency gains (e.g., \Delta+10.9 on Tool-Star-7B and \Delta+3.6 on ReCall-7B) while maintaining or even improving accuracy (e.g., 55.4 vs. 52.0 on Tool-Star-7B). Additional budget-constrained and tool-free evaluations further validate its effectiveness in maximizing cost-efficiency and reducing cognitive offloading.

Chengtao Gan, Zhiqiang Liu, Long Jin, Yushan Zhu, Lei Liang, Wen Zhang

Real-world scenarios involve massive heterogeneous structured data (e.g., tables, knowledge graphs), making effective reasoning over such diverse data increasingly important. Unified structured data question answering has emerged as a prominent research trend, aiming to answer natural language questions across different structured data types within a single framework. However, existing unified methods share a common limitation: they rely on a set of predefined functions, which restricts their ability to perform complex reasoning beyond these predefined operations. To overcome this fundamental limitation, we propose \textbf{CRAFTQA}, a novel adaptive code-driven framework comprising two core modules, CodeSTEP and CRAFT. The \textbf{CodeSTEP} module is a paradigm that generates a complete executable Python code sequence, which contains step-by-step code-based reasoning operations based on the question.The \textbf{CRAFT} module dynamically generates custom code functions for operations beyond the predefined function set, and seamlessly integrates with CodeSTEP to significantly enhance flexibility in handling complex reasoning. Comprehensive experiments on multiple structured datasets demonstrate that CRAFTQA achieves remarkable improvements in complex reasoning scenarios compared to existing unified methods.

Ruijun Huang, Zhiqiao Kang, Yuxuan Zhu, Junxiong Li, Jiahao Zhao, Minghuan Tan, Feng Jiang, Min Yang

The accurate extraction of scientific measurements from literature is a critical yet challenging task in AI4Science, enabling large-scale analysis and integration of quantitative research findings. However, Large Language Models (LLMs) frequently exhibit severe hallucinations, which significantly undermine the reliability of automated scientific document understanding systems. To address this problem, we propose MeasHalu, a novel framework for mitigating scientific measurement hallucinations through enhanced reasoning and targeted optimization. We first present a fine-grained taxonomy of measurement-specific hallucinations, categorizing errors across quantities, units, modifiers, and relations. Our approach incorporates a two-stage reasoning-aware fine-tuning strategy using augmented scientific data and process-based supervision. Furthermore, we introduce a progressive reward curriculum designed to penalize specific hallucination types, significantly improving extraction faithfulness. Experimental results demonstrate that MeasHalu substantially reduces hallucination rates and improves overall accuracy on the MeasEval benchmark. This work provides a targeted solution to a key bottleneck in automated scientific knowledge extraction, facilitating more trustworthy and scalable machine-assisted scientific literature analysis.

Jianpeng Hu, Yanzeng Li, Jialun Zhong, Lei Zou, Wenfa Qi

The Retrieval-augmented generation (RAG) system based on Large language model (LLM) has made significant progress. It can effectively reduce factuality hallucinations, but faithfulness hallucinations still exist. Previous methods for detecting faithfulness hallucinations either neglect to capture the models’ internal reasoning processes or handle those features coarsely, making it difficult for discriminators to learn. This paper proposes a semantic-level internal reasoning graph-based method for detecting faithfulness hallucination. Specifically, we first extend the layer-wise relevance propagation algorithm from the token level to the semantic level, constructing an internal reasoning graph based on attribution vectors. This provides a more faithful semantic-level representation of dependency. Furthermore, we design a general framework based on a small pre-trained language model to utilize the dependencies in LLM’s reasoning for training and hallucination detection, which can dynamically adjust the pass rate of correct samples through a threshold. Experimental results demonstrate that our method achieves better overall performance compared to state-of-the-art baselines on RAGTruth and Dolly-15k. Implementation available here: https://anonymous.4open.science/r/SIRG-1022.

Pengfei He, Shaowei Wang, Tse-Hsun Chen

Retrieval-Augmented Generation (RAG) enhances code generation by incorporating retrieved code examples into prompts, but the resulting long-context inputs impose substantial memory and computational overhead. Existing prompt compression techniques are largely designed for natural language and fail to account for the structural and semantic properties of code, while also lacking fine-grained control over compression ratios. We propose CodePromptZip, a code-aware prompt compression framework for RAG that enables precise length control while preserving critical information. Motivated by type-aware ablation studies, CodePromptZip leverages static analysis to rank code tokens by information gain and applies a dynamic compression strategy to retain the most informative tokens under a given budget. For incomplete or unparsable code snippets, CodePromptZip employs a language-model-based compressor trained on analyzable samples and augmented with a copy mechanism to preserve key tokens. Extensive experiments on three code-related tasks demonstrate that CodePromptZip consistently outperforms entropy-based and distillation-based baselines, achieving improvements of 23.4%, 28.7%, and 8.7%, respectively, while providing accurate control over compression ratios.

Fuwen Luo, Zihao Wan, Ziyue Wang, Yaluo Liu, Pau Tong Lin Xu, Xuanjia Qiao, Xiaolong Wang, Peng Li, Yang Liu

Hieroglyphs, as logographic writing systems, encode rich semantic and cultural information within their internal structural composition. Yet, current advanced Large Language Models (LLMs) and Multimodal LLMs (MLLMs) usually remain structurally blind to this information. LLMs process characters as textual tokens, while MLLMs additionally view them as raw pixel grids. Both fall short to model the underlying logic of character strokes. Furthermore, existing structural analysis methods are often script-specific and labor-intensive. In this paper, we propose Hieroglyphic Stroke Analyzer (HieroSA), a novel and generalizable framework that enables MLLMs to automatically derive stroke-level structures from character bitmaps without handcrafted data. It transforms modern logographic and ancient hieroglyphs character images into explicit, interpretable line-segment representations in a normalized coordinate space, allowing for cross-lingual generalization. Extensive experiments demonstrate that HieroSA effectively captures character-internal structures and semantics, bypassing the need for language-specific priors. Experimental results highlight the potential of our work as a graphematics analysis tool for a deeper understanding of hieroglyphic scripts.

Suhyun Lee, Palakorn Achananuparp, Neemesh Yadav, Ee-Peng Lim, Yang Deng

Large language models (LLMs) are increasingly explored as scalable tools for mental health counseling, yet evaluating their safety remains challenging due to the interactional and context-dependent nature of clinical harm. Existing evaluation frameworks predominantly assess isolated responses using coarse-grained taxonomies or static datasets, limiting their ability to diagnose how harms emerge and accumulate over multi-turn counseling interactions. In this work, we introduce R-MHSafe, a role-aware mental health safety taxonomy that characterizes clinically significant harm in terms of the interactional roles an AI counselor adopts, including perpetrator, instigator, facilitator, or enabler, combined with clinically grounded harm categories. Then, we propose MHSafeEval, a closed-loop, agent-based evaluation framework that formulates safety assessment as trajectory-level discovery of harm through adversarial multi-turn interactions, guided by role-aware modeling. Using R-MHSafe and MHSafeEval, we conduct a large-scale evaluation across state-of-the-art LLMs. Our results reveal substantial role-dependent and cumulative safety failures that are systematically missed by existing static benchmarks, and show that our framework significantly improves failure-mode coverage and diagnostic granularity.

Liyong Wang, Junliang Xing, Tianyu Hu, Jianfei Jiang, Jihuai Zhao, Huimin Ma

Audio-Visual Speech Recognition enhances speech recognition robustness in noisy conditions by leveraging visual cues. However, current Multimodal LLMs suffer from a fundamental temporal gap. This gap is characterized by limited fine-grained temporal modeling in vision encoders and progressive temporal semantic degradation throughout the deep layers of LLM decoders. To bridge this gap, we propose a novel framework that deeply stacks temporal tokens across both the encoding and decoding stages. Specifically, we enhance the vision encoder with a temporal-aware attention module and temporal rotary positional embeddings to precisely capture the sequential evolution and dynamics of lip movements. Furthermore, we stack hierarchical temporal tokens that incorporate temporally enriched features into multiple layers of the LLM decoder in a bottom-up manner. Extensive experiments on the LRS2 and LRS3 benchmarks demonstrate that our approach achieves high efficiency and firm performance, outperforming existing supervised, self-supervised, and LLM-based methods by 6.1% on LRS2 and 7.8% on LRS3.

Youmi Ma, Naoaki Okazaki

Advances in mechanistic interpretability have identified special attention heads, known as retrieval heads, that are responsible for retrieving information from the context. However, the role of these retrieval heads in improving model performance remains unexplored. This work investigates whether retrieval heads can be leveraged to enhance the long-context capabilities of LLMs. Specifically, we propose RetMask, a method that generates training signals by contrasting normal model outputs with those from an ablated variant in which the retrieval heads are masked. This mechanism-based approach achieves substantial improvements: +2.28 points on HELMET at 128K for Llama-3.1, with +70% gains on generation with citation and +32% on passage re-ranking, while preserving performance on general tasks. Experiments across three model families demonstrate that RetMask consistently improves long-context performance, with gains correlating with the sparsity of the retrieval score distribution: models with sparser distributions, where retrieval capabilities are concentrated in a small set of heads, respond more strongly, while those with less sparse distributions show more modest gains. These results validate the functional role of retrieval heads and show that mechanistic insights can be transformed into performance enhancements.

Nobin Sarwar, Shubhashis Roy Dipta, Zheyuan Liu, Vaidehi Patil

With the growing adoption of VLMs, DMs, LLMs, and AFMs, these multimodal foundation models can inadvertently encode sensitive, copyrighted, biased, or unsafe cross-modal associations that originate from their training data. Retraining after deletion requests or policy updates is often impractical, and targeted forgetting remains difficult because knowledge is distributed across shared representations. Multimodal unlearning addresses this challenge by enabling selective removal across modalities while retaining overall utility. This survey offers a unified, system-oriented view of multimodal unlearning across vision, language, audio, and video, grounded in recent advances, emerging applications, and open problems. Our taxonomy enables systematic comparison across model architectures and modalities, clarifying trade-offs among deletion strength, retention, efficiency, reversibility, and robustness. This survey highlights open problems and practical considerations to support future research and deployment of multimodal unlearning.

Shiqi Zhang, Xinbei Ma, Yunqing Xu, Zouying Cao, Pengrui Lu, Haobo Yuan, Tiancheng Shen, Zhuosheng Zhang, Hai Zhao, Ming-Hsuan Yang

Large Language Models (LLMs) exhibit strong reasoning abilities for planning long-horizon, real-world tasks, yet existing agent benchmarks focus on task completion while neglecting time efficiency in parallel and asynchronous operations. To address this, we present ParaCook, a benchmark for time-efficient collaborative planning. Inspired by the Overcooked game, ParaCook provides an environment for various challenging interaction planning of multi-agent systems that are instantiated as cooking tasks, with a simplified action space to isolate the core challenge of strategic parallel planning. Through a comprehensive evaluation of state-of-the-art LLMs, we find that current approaches achieve suboptimal plans, which struggle with parallel actions or coordination. Our analysis also reveals LLMs’ potential on abstract tasks where they can focus on high-level parallel optimization. ParaCook provides a scalable evaluation framework with adjustable complexity, establishing a foundation for developing and assessing time efficiency-aware multi-agent planning.