论文检索

Zhanpeng Chen, Mingxiao Li, Ziyang Chen, Nan Du, Xiaolong Li, Yuexian Zou

Vision-language Models (VLMs) have shown remarkable capabilities in advancing general artificial intelligence, yet the irrational encoding of visual positions persists in inhibiting the models’ comprehensive perception performance across different levels of granularity. In this work, we propose Pyramid-descent Visual Position Encoding (PyPE), a novel approach designed to enhance the perception of visual tokens within VLMs. By assigning visual position indexes from the periphery to the center and expanding the central receptive field incrementally, PyPE addresses the limitations of traditional raster-scan methods and mitigates the long-term decay effects induced by Rotary Position Embedding (RoPE). Our method reduces the relative distance between interrelated visual elements and instruction tokens, promoting a more rational allocation of attention weights and allowing for a multi-granularity perception of visual elements and countering the over-reliance on anchor tokens. Extensive experimental evaluations demonstrate that PyPE consistently improves the general capabilities of VLMs across various sizes. Code is available at https://anonymous.4open.science/r/PyPE-34EE.

Vardaan Pahuja, Yadong Lu, Corby Rosset, Boyu Gou, Arindam Mitra, Spencer Whitehead, Yu Su, Ahmed Hassan Awadallah

Recent success in large multimodal models (LMMs) has sparked promising applications of agents capable of autonomously completing complex web tasks. While open-source LMM agents have made significant advances in offline evaluation benchmarks, their performance still falls substantially short of human-level capabilities in more realistic online settings. A key bottleneck is the lack of diverse and large-scale trajectory-level datasets across various domains, which are expensive to collect. In this paper, we address this challenge by developing a scalable recipe to synthesize the largest and most diverse trajectory-level dataset to date, containing over 94K successful multimodal web trajectories, spanning 49K unique URLs, 720K screenshots, and 33M web elements. In particular, we leverage extensive web exploration and refinement to obtain diverse task intents. The average cost is 28 cents per successful trajectory, making it affordable to a wide range of users in the community. Leveraging this dataset, we train Explorer, a multimodal web agent, and demonstrate strong performance on both offline and online web agent benchmarks such as Mind2Web-Live, Multimodal-Mind2Web, and MiniWob++. Additionally, our experiments highlight data scaling as a key driver for improving web agent capabilities. We hope this study makes state-of-the-art LMM-based agent research at a larger scale more accessible.

Peng Ding, Jun Kuang, ZongYu Wang, Xuezhi Cao, Xunliang Cai, Jiajun Chen, Shujian Huang

Large Language Models (LLMs) have shown impressive capabilities across various tasks but remain vulnerable to meticulously crafted jailbreak attacks. In this paper, we identify a critical safety gap: while LLMs are adept at detecting jailbreak prompts, they often produce unsafe responses when directly processing these inputs. Inspired by this insight, we propose SAGE(Self-Aware Guard Enhancement), a training-free defense strategy designed to align LLMs’ strong safety discrimination performance with their relatively weaker safety generation ability. SAGE consists of two core components: a Discriminative Analysis Module and a Discriminative Response Module, enhancing resilience against sophisticated jailbreak attempts through flexible safety discrimination instructions. Extensive experiments demonstrate SAGE’s effectiveness and robustness across various open-source and closed-source LLMs of different sizes and architectures, achieving an average 99% defense success rate against numerous complex and covert jailbreak methods while maintaining helpfulness on general benchmarks. We further conduct mechanistic interpretability analysis through hidden states and attention distributions, revealing the underlying mechanisms of this detection-generation discrepancy. Our work thus contributes to developing future LLMs with coherent safety awareness and generation behavior. Our code and datasets are publicly available at https://github.com/NJUNLP/SAGE.

Lishui Fan, Mouxiang Chen, Zhongxin Liu

Large language models (LLMs) have achieved impressive performance in code generation. Despite the remarkable success, we observed that LLMs often misunderstand or overlook some problem-specific undertrained keywords during code generation, compromising the accuracy of the generated code. After explicitly explaining these undertrained keywords using well-trained terms in the prompt, LLMs are more likely to generate correct code implementation. Inspired by this observation, we propose a novel technique named SEK(Self-Explained Keywords), which empowers an LLM for better code generation by extracting and explaining the key terms in the problem description with the LLM itself. Comprehensive experiments across four benchmarks, i.e., HumanEval(+), MBPP(+), APPS and BigCodeBench, with five representative LLMs, show that SEK can significantly improve LLMs in code generation, yielding substantial and consistent gains. For instance, SEK improves the Pass@1 of DeepSeek-Coder-V2-Instruct from 85.4% to 93.3% on the Humaneval benchmark. Further analysis confirms that SEK enables the LLMs to shift their attention from low-frequency keywords to their corresponding explanations.

Yilei Tu, Andrew Xue, Freda Shi

While multilingual large language models generally perform adequately, and sometimes even rival English performance on high-resource languages (HRLs), they often significantly underperform on low-resource languages (LRLs). Among several prompting strategies aiming at bridging the gap, multilingual in-context learning (ICL) has been particularly effective when demonstration in target languages is unavailable. However, there lacks a systematic understanding when and why it works well.In this work, we systematically analyze multilingual ICL, using demonstrations in HRLs to enhance cross-lingual transfer. We show that demonstrations in mixed HRLs consistently outperform English-only ones across the board, particularly for tasks written in LRLs. Surprisingly, our ablation study show that the presence of irrelevant non-English sentences in the prompt yields measurable gains, suggesting the effectiveness of multilingual exposure itself. Our results highlight the potential of strategically leveraging multilingual resources to bridge the performance gap for underrepresented languages.

Kening Zheng, Junkai Chen, Yibo Yan, Xin Zou, Huiyu Zhou, Xuming Hu

Hallucination issues continue to affect multimodal large language models (MLLMs), with existing research mainly addressing object-level or attribute-level hallucinations, neglecting the more complex relation hallucinations that require advanced reasoning. Current benchmarks for relation hallucinations lack detailed evaluation and effective mitigation, and their datasets often suffer from biases due to systematic annotation processes. To address these challenges, we introduce Reefknot, a comprehensive benchmark targeting relation hallucinations, comprising over 20,000 real-world samples. We provide a systematic definition of relation hallucinations, integrating perceptive and cognitive perspectives, and construct a relation-based corpus using the Visual Genome scene graph dataset. Our comparative evaluation reveals significant limitations in current MLLMs’ ability to handle relation hallucinations. Additionally, we propose a novel confidence-based mitigation strategy, which reduces the hallucination rate by an average of 9.75% across three datasets, including Reefknot. Our work offers valuable insights for achieving trustworthy multimodal intelligence. The dataset and code are released at https://github.com/JackChen-seu/Reefknot.

Zengyi Gao, Yukun Cao, Hairu Wang, Ao Ke, Yuan Feng, S Kevin Zhou, Xike Xie

To mitigate the hallucination and knowledge deficiency in large language models (LLMs), Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) has shown promising potential by utilizing KGs as an external resource to enhance LLM reasoning.However, existing KG-RAG approaches struggle with a trade-off between flexibility and retrieval quality. Modular methods prioritize flexibility by avoiding the use of KG-fine-tuned models during retrieval, leading to fixed retrieval strategies and suboptimal retrieval quality. Conversely, coupled methods embed KG information within models to improve retrieval quality but at the expense of flexibility.In this paper, we propose a novel flexible modular KG-RAG framework, termed FRAG, which synergizes the advantages of both approaches. FRAG estimates the hop range of reasoning paths based solely on the query and classifies it as either simple or complex.To match the complexity of the query, tailored pipelines are applied to ensure efficient and accurate reasoning path retrieval, thus fostering the final reasoning process. By using the query text instead of the KG to infer the structural information of reasoning paths and employing adaptable retrieval strategies, FRAG improves retrieval quality while maintaining flexibility. Moreover, FRAG does not require extra LLM fine-tuning or calls, significantly boosting efficiency and conserving resources. Extensive experiments show that FRAG achieves state-of-the-art performance with high efficiency and low resource consumption. The code for our method is publicly available at https://github.com/gzy02/FRAG.

Guofeng Quan, Wenfeng Feng, Chuzhan Hao, Guochao Jiang, Yuewei Zhang, Hao Henry Wang

Speculative decoding accelerates inference in large language models (LLMs) by generating draft tokens for target model verification. Current approaches for obtaining draft tokens rely on lightweight draft models or additional model structures to generate draft tokens and retrieve context from databases. Due to the draft model’s small size and limited training data, model-based speculative decoding frequently becomes less effective in out-of-domain scenarios. Additionally, the time cost of the drafting phase results in a low upper limit on acceptance length during the verification step, limiting overall efficiency. This paper proposes RASD (Retrieval-Augmented Speculative Decoding), which adopts retrieval methods to enhance model-based speculative decoding. We introduce tree pruning and tree fusion to achieve this. Specifically, we develop a pruning method based on the draft model’s probability distribution to construct the optimal retrieval tree. Second, we employ the longest prefix matching algorithm to merge the tree generated by the draft model with the retrieval tree, resulting in a unified tree for verification. Experimental results demonstrate that RASD achieves state-of-the-art inference acceleration across tasks such as DocQA, Summary, Code, and In-Domain QA. Moreover, RASD exhibits strong scalability, seamlessly integrating with various speculative decoding approaches, including both generation-based and retrieval-based methods.

Jiakai Tang, Shiqi Shen, ZhipengWang ZhipengWang, Gong Zhi, Xueyang Feng, Zexu Sun, Haoran Tan, Xu Chen

Dialogue assistants have become ubiquitous in modern applications, fundamentally reshaping human daily communication patterns and information access behaviors. In real-world conversational interactions, however, user queries are often volatile, ambiguous, and diverse, making it difficult accurately and efficiently grasp the user’s underlying intentions. To address this challenge, we propose a simple yet effective deliberative agent framework that leverages human thought process to build high-level domain knowledge. To further achieve efficient knowledge accumulation and retrieval, we design a tree-structured knowledge base to store refined experience and data. Moreover, we construct a new benchmark, User-Intent-Understanding (UIU), which covers multi-domain, multi-tone, and sequential multi-turn personalized user queries. Extensive experiments demonstrate the effectiveness of our proposed method across multi-step evaluations.

Weihong Du, Wenrui Liao, Binyu Yan, Hongru Liang, Anthony G Cohn, Wenqiang Lei

Large language model (LLM) based agents have shown great potential in following human instructions and automatically completing various tasks. To complete a task, the agent needs to decompose it into easily executed steps by planning. Existing studies mainly conduct the planning by inferring what steps should be executed next starting from the agent’s initial state. However, this forward reasoning paradigm doesn’t work well for complex tasks. We propose to study this issue in Minecraft, a virtual environment that simulates complex tasks based on real-world scenarios. We believe that the failure of forward reasoning is caused by the big perception gap between the agent’s initial state and task goal. To this end, we leverage backward reasoning and make the planning starting from the terminal state, which can directly achieve the task goal in one step. Specifically, we design a backward reasoning based agent (BAR). It is equipped with a recursive goal decomposition module, a state consistency maintaining module and a stage memory module to make robust, consistent, and efficient planning starting from the terminal state. Experimental results demonstrate the superiority of BAR over existing methods and the effectiveness of proposed modules.

Ruiqiao Bai, Xue Han, Shuo Lei, Junlan Feng, Yanyan Luo, Chao Deng

Applying Large Language Models (LLM) to solve math problems is one of the hottest research topics at present. Traditional Chain-of-Thought-based methods typically generate the reasoning path in a chain structure, leading to unnecessary interference caused by non-zero self-attention among weakly related reasoning steps. Such a setting also differs from humans’ typical graph-structured reasoning habit (with an inter-step relationship graph in mind). To solve the problem, this paper proposes a novel decoding method for Transformer-based LLM, named Self-attention-based Graph-of-Thought (SaGoT). SaGoT constructs a thought graph simultaneously as an LLM inference (based on a newly defined inter-step self-attention indicator), and generates reasoning steps with a novel graph-structured self-attention mechanism. It is a significant contribution for SaGoT to enable an LLM’s graph-like reasoning ability by modifying its inner working operations, compared to SOTA prompting methods that are ex-post, rely on huge LLMs and redundant reasoning step generation to form a graph (inefficient & non-human-like). In addition, SaGoT is a training-free technique that can be seamlessly incorporated into pre-trained Transformer-based LLMs. Our experimental results have shown that SaGoT could significantly enhance mathematical reasoning accuracy without the reliance on huge computationally over-expensive LLMs. It also avoids SOTA methods’ performance degradation issues when the LLM is too small to comprehend complex prompts. Moreover, SaGoT integrates intrinsic interpretability into the LLM’s reasoning procedure, intuitively assisting humans in understanding how an LLM views the relationships among its reasoning steps, and why the LLM succeeds or fails.

Yijiong Yu, Huiqiang Jiang, Xufang Luo, Qianhui Wu, Chin-Yew Lin, Dongsheng Li, Yuqing Yang, Yongfeng Huang, Lili Qiu

Long-context language models (LCLMs) can process long context, but still exhibit position bias, also known as “lost in the middle”, which indicates placing key information in the middle of the context will significantly affect performance. To mitigating this, we first explore the micro-level manifestations of position bias, concluding that attention weights are a micro-level expression of position bias. Then we identify that, in addition to position embeddings, positional information in hidden states also contributes to position bias, and it manifests itself in specific channels of hidden states, called positional hidden states. Based on these, we propose a method to mitigate position bias by scaling positional hidden states. Experiments on NaturalQuestions Multi-document QA, KV retrieval and LongBench, using various models including RoPE models, context window-extended models, and Alibi models, demonstrate the effectiveness and generalizability of our approach. Our method can improve performance by up to 15.2% in “lost in the middle” benchmark by modifying just one channel of hidden states. Our code is available at https://aka.ms/PositionalHidden.

Jingbao Luo, Ming Liu, Ran Liu, Yongpan Sheng, Xin Hu, Gang Li, Peng Wu

With the popularity of large language models and their high-quality text generation capabilities, researchers are using them as auxiliary tools for text summary writing. Although summaries generated by these large language models are smooth and capture key information sufficiently, the quality of their output depends on the prompt, and the generated text is somewhat procedural to a certain extent. We construct LecSumm to verify whether language models truly capture human writing preferences, in which we recruit 200 college students to write summaries for lecture notes on ten different machine-learning topics and analyze writing preferences in real-world human summaries through the dimensions of length, content depth, tone & style, and summary format. We define the method of capturing human writing preferences by language models as finetuning pre-trained models with data and designing prompts to optimize the output of large language models. The results of translating the analyzed human writing preferences into prompts and conducting experiments show that both models still fail to capture human writing preferences effectively. Our LecSumm dataset brings new challenges to finetuned and prompt-based large language models on the task of human-centered text summarization.

Yiwei Fu, Yuxing Zhang, Chunchun Chen, Jianwen Ma, Quan Yuan, Rong-Cheng Tu, Xinli Huang, Wei Ye, Xiao Luo, Minghua Deng

This paper studies the problem of text-attributed graph clustering, which aims to cluster each node into different groups using both textual attributes and structural information. Although graph neural networks (GNNs) have been proposed to solve this problem, their performance is usually limited when uncertain nodes are near the cluster boundaries due to label scarcity. In this paper, we introduce a new perspective of leveraging large language models (LLMs) to enhance text-attributed graph clustering and develop a novel approach named Multi-agent Collaboration with Ranking Guidance (MARK). The core of our MARK is to generate reliable guidance using the collaboration of three LLM-based agents as ranking-based supervision signals. In particular, we first conduct the coarse graph clustering, and utilize a concept agent to induce the semantics of each cluster. Then, we infer the robustness under perturbations to identify uncertain nodes and use a generation agent to produce synthetic text that closely aligns with their topology. An inference agent is adopted to provide the ranking semantics for each uncertain node in comparison to its synthetic counterpart. The consistent feedback between uncertain and synthetic texts is identified as reliable guidance for fine-tuning the clustering model within a ranking-based supervision objective. Experimental results on various benchmark datasets validate the effectiveness of the proposed MARK compared with competing baselines.

Tian Lan, Xiangdong Su, Xu Liu, Ruirui Wang, Ke Chang, Jiang Li, Guanglai Gao

As large language models (LLMs) are increasingly applied to various NLP tasks, their inherent biases are gradually disclosed. Therefore, measuring biases in LLMs is crucial to mitigate its ethical risks. However, most existing bias evaluation datasets are focus on English andNorth American culture, and their bias categories are not fully applicable to other cultures. The datasets grounded in the Chinese language and culture are scarce. More importantly, these datasets usually only support single evaluation task and cannot evaluate the bias from multiple aspects in LLMs. To address these issues, we present a Multi-task Chinese Bias Evaluation Benchmark (McBE) that includes 4,077 bias evaluation instances, covering 12 single bias categories, 82 subcategories and introducing 5 evaluation tasks, providing extensive category coverage, content diversity, and measuring comprehensiveness. Additionally, we evaluate several popular LLMs from different series and with parameter sizes. In general, all these LLMs demonstrated varying degrees of bias. We conduct an in-depth analysis of results, offering novel insights into bias in LLMs.

Jianghao Chen, Zhenlin Wei, Zhenjiang Ren, Ziyong Li, Jiajun Zhang

Recent progress in o1-like models has significantly enhanced the reasoning abilities of Large Language Models (LLMs), empowering them to tackle increasingly complex tasks through reflection capabilities, such as making assumptions, backtracking, and self-refinement. However, effectively evaluating such reflection capabilities remains challenging due to the lack of appropriate benchmarks. To bridge this gap, we introduce LR^2Bench, a novel benchmark designed to evaluate the Long-chain Reflective Reasoning capabilities of LLMs. LR^2Bench comprises 850 samples across six Constraint Satisfaction Problems (CSPs) where reflective reasoning is crucial for deriving solutions that meet all given constraints. Each type of task focuses on distinct constraint patterns, such as knowledge-based, logical, and spatial constraints, providing a comprehensive evaluation of diverse problem-solving scenarios. We conduct extensive evaluation on both conventional models and o1-like models. Our experimental results reveal that even the most advanced reasoning-specific models, such as DeepSeek-R1 and OpenAI o1-preview, struggle with tasks in LR^2Bench, achieving an average Exact Match score of only 20.0% and 23.6%, respectively. These findings underscore the significant room for improvement in the reflective reasoning capabilities of current LLMs.

Wenhao Liu, Siyu An, Junru Lu, Muling Wu, Tianlong Li, Xiaohua Wang, Changze Lv, Xiaoqing Zheng, Di Yin, Xing Sun 等

Role-Playing Agents (RPAs) have shown remarkable performance in various applications, yet they often struggle to recognize and appropriately respond to hard queries that conflict with their role-play knowledge. To investigate RPAs’ performance when faced with different types of conflicting requests, we develop an evaluation benchmark that includes contextual knowledge conflicting requests, parametric knowledge conflicting requests, and non-conflicting requests to assess RPAs’ ability to identify conflicts and refuse to answer appropriately without over-refusing. Through extensive evaluation, we find that most RPAs behave significant performance gaps toward different conflict requests. To elucidate the reasons, we conduct an in-depth representation-level analysis of RPAs under various conflict scenarios. Our findings reveal the existence of rejection regions and direct response regions within the model’s forwarding representation, and thus influence the RPA’s final response behavior. Therefore, we introduce a lightweight representation editing approach that conveniently shifts conflicting requests to the rejection region, thereby enhancing the model’s refusal accuracy. The extensive experiments validate the effectiveness of our editing method, improving RPAs’ refusal ability of conflicting requests while maintaining their general role-playing capabilities.

Hwan Chang, Hwanhee Lee

Large language models (LLMs) risk retaining unauthorized or sensitive information from their training data, which raises privacy concerns. LLM unlearning seeks to mitigate these risks by selectively removing specified data while maintaining overall model performance. However, most existing work focuses on methods to achieve effective forgetting and does not provide a detailed analysis of the retain set, the portion of training data that is not targeted for removal. In this paper, we investigate the effects of unlearning on various subsets of the retain set through a case study on entity unlearning. We introduce the \textit{Syntactically Similar Neighbor Set}, a group of queries that share similar syntactic structures with the data targeted for removal, and show that this subset suffers the greatest performance drop during unlearning. Moreover, when used for regularization, this set not only preserves performance on syntactically similar queries but also delivers comparable or improved results across other data subsets. Our results highlight that syntactic similarity is a critical factor, potentially more so than domain or entity relationships, in achieving effective and practical LLM unlearning.

Yu Li, Qizhi Pei, Mengyuan Sun, Honglin Lin, Chenlin Ming, Xin Gao, Jiang Wu, Conghui He, Lijun Wu

Large language models (LLMs) have demonstrated remarkable capabilities, especially the recent advancements in reasoning, such as o1 and o3, pushing the boundaries of AI. Despite these impressive achievements in mathematics and coding, the reasoning abilities of LLMs in domains requiring cryptographic expertise remain underexplored. In this paper, we introduce CipherBank, a comprehensive benchmark designed to evaluate the reasoning capabilities of LLMs in cryptographic decryption tasks. CipherBank comprises 2,358 meticulously crafted problems, covering 262 unique plaintexts across 5 domains and 14 subdomains, with a focus on privacy-sensitive and real-world scenarios that necessitate encryption. From a cryptographic perspective, CipherBank incorporates 3 major categories of encryption methods, spanning 9 distinct algorithms, ranging from classical ciphers to custom cryptographic techniques. We evaluate state-of-the-art LLMs on CipherBank, e.g., GPT-4o, DeepSeek-V3, and cutting-edge reasoning-focused models such as o1 and DeepSeek-R1. Our results reveal significant gaps in reasoning abilities not only between general-purpose chat LLMs and reasoning-focused LLMs but also in the performance of current reasoning-focused models when applied to classical cryptographic decryption tasks, highlighting the challenges these models face in understanding and manipulating encrypted data. Through detailed analysis and error investigations, we provide several key observations that shed light on the limitations and potential improvement areas for LLMs in cryptographic reasoning.These findings underscore the need for continuous advancements in LLM reasoning capabilities.

Shuai Niu, Jing Ma, Hongzhan Lin, Liang Bai, Zhihua Wang, Wei Bi, Yida Xu, Guo Li, Xian Yang

Large language models (LLMs) have shown remarkable performance in vision-language tasks, but their application in the medical field remains underexplored, particularly for integrating structured time series data with unstructured clinical notes. In clinical practice, dynamic time series data, such as lab test results, capture critical temporal patterns, while clinical notes provide rich semantic context. Merging these modalities is challenging due to the inherent differences between continuous signals and discrete text. To bridge this gap, we introduce ProMedTS, a novel self-supervised multimodal framework that employs prompt-guided learning to unify these heterogeneous data types. Our approach leverages lightweight anomaly detection to generate anomaly captions that serve as prompts, guiding the encoding of raw time series data into informative prompt embeddings. These prompt embeddings are aligned with textual representations in a shared latent space, preserving fine-grained temporal nuances alongside semantic insights. Furthermore, our framework incorporates tailored self-supervised objectives to enhance both intra- and inter-modal alignment. We evaluate ProMedTS on disease diagnosis tasks using real-world datasets, and the results demonstrate that our method consistently outperforms state-of-the-art approaches.