论文检索

Zicheng Huang, Yajuan Tong, Xinhui Tu, Tingting He

Temporal knowledge graph question answering (TKGQA) addresses time-sensitive queries over temporal knowledge graphs, but existing approaches struggle with multi-hop reasoning and implicit temporal constraints. We introduce TempTool-R1, a novel tool-integrated reasoning framework that enables large language models to explicitly use temporal tools for precise reasoning. First, we design a unified temporal tool-based API capable of transforming implicit temporal cues into executable operations, establishing the structural foundation for tool interaction. In the second stage, supervised fine-tuning teaches the model to interweave chain-of-thought reasoning with think-then-tool usage, allowing it to call temporal tools during inference. Finally, we apply reinforcement learning with fine-grained, order-sensitive reward functions tailored for temporal tool use, further refining the model’s tool-use policy. Experiments on three challenging TKGQA benchmarks demonstrate that TempTool-R1 significantly outperforms existing methods. In particular, our approach excels on complex questions requiring multi-hop temporal reasoning, highlighting the effectiveness of temporal tool integration and reward optimization in improving TKGQA performance.

Jian Yang, Wei Zhang, Shuyue Guo, Yizhi LI, Linzheng Chai, Zhengmao Ye, Shukai Liu, Yuyang Song, Jiajun Wu, Che Liu 等

While large language models (LLMs) have mastered syntax-level code generation, complex algorithmic reasoning remains a challenge, typically addressed by scaling model depth and parameter count. Universal Transformers (UT) offer a compelling alternative by introducing a recurrent inductive bias that aligns with the recursive nature of programming logic. However, training looped architectures at scale has historically been hindered by severe instability and optimization difficulties associated with backpropagation through time (BPTT). We present LoopCoder (40B-A80B) pre-trained on 12T+ code and general tokens, along with LoopCoder-Thinking and LoopCoder-Instruct variants—the first large-scale looped transformer for code, achieving comparable performance to standard dense architectures with more parameters. Unlike prior approaches that restrict recurrence to small-scale tasks, we implement a comprehensive looped training protocol spanning both pre-training and post-training phases. We initiate the model via dense-to-loop transformation, folding a pre-trained dense checkpoint to initialize a recurrent block, followed by rigorous looped pre-training and specialized post-training for instruction following and reasoning. Our results establish a robust recipe for scaling coding intelligence via recurrent computation, proving that dense checkpoints serve as an optimal foundation for evolving into dynamic, looped reasoners.

Yuyin Zhou, Yongmei Liu

Recent advances in large language models (LLMs) have spurred interest in neuro-symbolic methods for logical reasoning based on auto-formalization, where LLMs first formalize problems into symbolic programs, for solvers to perform reasoning over.However, existing auto-formalization methods remain prone to both syntactic and semantic errors.Specifically, the absence of a program-level semantic verification mechanism leaves semantic errors largely unaddressed.In this paper, we propose a novel approach to semantic error correction via program preference retrieval-augmented generation (RAG).First, we conduct an in-depth analysis of semantic error patterns, and then automatically synthesize SemanticPref, a program preference dataset to model these patterns. Using the dataset as the knowledge base, we introduce PrefRAG, a general RAG framework for refinement in auto-formalization, which enables LLMs to detect and repair syntactic and semantic errors[<https://github.com/sysulic/PrefRAG>].Extensive evaluations across both in-distribution (ID) benchmarks (i.e., AR-LSAT and FOLIO) and out-of-distribution (OOD) datasets show that PrefRAG consistently outperforms strong baselines, achieving an average accuracy improvement of 2.39% on ID and 6.23% on OOD datasets.

Wenhan Han, Yifan Zhang, Zhixun Chen, Binbinliu, Mykola Pechenizkiy, Meng Fang, Yin Zheng

Multilingual large language models (LLMs) are advancing rapidly, with new models frequently claiming support for an increasing number of languages. However, existing evaluation datasets are limited and lack cross-lingual alignment, leaving assessments of multilingual capabilities fragmented in both language and skill coverage. To address this, we introduce MuBench, a benchmark covering 61 languages with 3.9M samples and evaluating a broad range of capabilities. We evaluate several state-of-the-art multilingual LLMs and find notable gaps between claimed and actual language coverage, particularly a persistent performance disparity between English and low-resource languages. Leveraging MuBench’s alignment, we propose Multilingual Consistency (MLC) as a complementary metric to accuracy for analyzing performance bottlenecks and guiding model improvement. MuBench provides flexible evaluation formats, including mixed-language testing. Experimental results show that increasing model size does not improve its ability to handle mixed-language contexts. We recruited human experts to evaluate translation quality and cultural sensitivity for 34k samples across 17 languages, and combined these assessments with an LLM-as-a-Judge approach to ensure overall data quality in low resource languages.

Fulong Fan, Peilin Liu, Liu FengZhe, Shuyan Yang, Gang Yan

Large language models perform well on many reasoning tasks, yet they often lack awareness of whether their current knowledge or reasoning state is complete. In non-interactive puzzle settings, the narrative is fixed and the underlying structure is hidden; once a model forms an early hypothesis under incomplete premises, it can propagate that error throughout the reasoning process, leading to unstable conclusions. To address this issue, we propose SABA, a reasoning framework that explicitly introduces self-awareness of missing premises before making the final decision. SABA formulates reasoning as a recursive process that alternates between structured state construction and obstacle resolution: it first applies Information Fusion to consolidate the narrative into a verifiable base state, and then uses Query-driven Structured Reasoning to identify and resolve missing or underspecified premises by turning them into queries and progressively completing the reasoning state through hypothesis construction and state refinement. Across multiple evaluation metrics, SABA achieves the best performance on all three difficulty splits of the non-interactive Detective Puzzle benchmark, and it also maintains leading results on multiple public benchmarks.

Guanyu Zheng, Wang Zhenyu, He Tingting, Xv Wang, Haochang Wang, Yaokai Huang, Tiejun Zhao

Lifelong knowledge editing aims to inject a stream of factual updates into large language models (LLMs) without retraining, yet existing memory-based editors often suffer from catastrophic forgetting as edits accumulate. We argue that a key factor is the coupled knowledge memory mechanism, where addressing (routing) and storage (writing via memory-module updates) are entangled. This entanglement makes it difficult to confine the effects of each edit to its intended scope, particularly in multi-domain and associated-fact editing streams, where updates either span diverse semantic domains or repeatedly modify related attributes of the same subject. Consequently, updating memory for one edit inadvertently alters the routing and stored representations of previously injected edits, leading to catastrophic forgetting as edits accumulate. We propose **DKME**, which decouples addressing from storage via two stages: decoupled semantic addressing learns a fact-aware manifold for scope-aware routing, and partitioned memory storage localizes edits to memory partitions identified by unsupervised clustering in the embedding space. Experiments on three benchmarks, including HalluEditBench, CKnowEdit, and WikiData_\text{counterfact}, demonstrate that DKME consistently achieves a more favorable trade-off between editing success and locality compared to baselines, while maintaining more stable performance as the edit scale increases.

Yunhao Feng, Yige Li, Yutao Wu, Yingshui Tan, Yanming Guo, Yifan Ding, Kun Zhai, Xingjun Ma, Yu-Gang Jiang

Large language model (LLM) agents execute tasks through multi-step workflows that combine planning, memory, and tool use. While this design enables autonomy, it also expands the attack surface for backdoor threats. Backdoor triggers injected into specific stages of an agent workflow can persist through multiple intermediate states and adversely influence downstream outputs. However, existing studies remain fragmented and typically analyze individual attack vectors in isolation, leaving the cross-stage interaction and propagation of backdoor triggers poorly understood from an agent-centric perspective.To fill this gap, we propose BackdoorAgent, a modular and stage-aware framework that provides a unified, agent-centric view of backdoor threats in LLM agents. BackdoorAgent structures the attack surface into three functional stages of agentic workflows, including planning attacks, memory attacks, and tool-use attacks, and instruments agent execution to enable systematic analysis of trigger activation and propagation across different stages.Building on this framework, we construct a standardized benchmark spanning four representative agent applications: Agent QA, Agent Code, Agent Web, and Agent Drive, covering both language-only and multimodal settings. Our empirical analysis shows that triggers implanted at a single stage can persist across multiple steps and propagate through intermediate states. For instance, when using a GPT-based backbone, we observe trigger persistence in 43.58% of planning attacks, 77.97% of memory attacks, and 60.28% of tool-stage attacks, highlighting the vulnerabilities of the agentic workflow itself to backdoor threats. Our code is available at https://github.com/Yunhao-Feng/BackdoorAgent.

Zhenyu He, Qingping Yang, Wei Shen, Xiaojian Zhong, Kechi Zhang, Chenxin An, Wenlei Shi, Tianle Cai, Di He, Jiaze Chen 等

Automated software engineering, particularly resolving real-world issues on benchmarks like SWE-bench, remains a significant challenge for Large Language Models (LLMs). To address this, we introduce SWE-Swiss, a two-phase training recipe that systematically develops these capabilities. Our approach first decomposes issue resolution into three core skills: Localization, Repair, and Unit Test Generation. In the first phase, we perform multi-task Supervised Fine-Tuning (SFT) on three new, meticulously curated datasets to build a versatile foundation. The second phase applies targeted Reinforcement Learning (RL), using direct feedback from test execution to boost the critical skill of code repair. The resulting model, SWE-Swiss-32B, establishes a new state-of-the-art for open-source models in its size class, achieving a 60.2% score on the SWE-bench Verified benchmark and placing it in the same top-tier performance bracket as much larger models. Finally, we show that despite its specialized training, SWE-Swiss-32B demonstrates strong generalization to other common LLM benchmarks. To accelerate research in the community, we are open-sourcing the models and our complete training datasets.

Huangwei Chen, Wu Li, Junhao Jia, Yining Chen, Xiaotao Pang, Ya-Long Chen, Li Gonghui, Haishuai Wang, Jiajun Bu, Lei Wu

The initial outpatient consultation is critical for clinical decision-making, yet it is often conducted by a single physician under time pressure, making it prone to cognitive biases and incomplete evidence capture. Although the Multi-Disciplinary Team (MDT) reduces these risks, they are costly and difficult to scale to real-time intake. We propose Aegle, a synchronous virtual MDT framework that brings MDT-level reasoning to outpatient consultations via a graph-based multi-agent architecture. Aegle formalizes the consultation state using a structured SOAP representation, separating evidence collection from diagnostic reasoning to improve traceability and bias control. An orchestrator dynamically activates specialist agents, which perform decoupled parallel reasoning and are subsequently integrated by an aggregator into a coherent clinical note. Experiments on ClinicalBench and a real-world RAPID-IPN dataset across 24 departments and 53 metrics show that Aegle consistently outperforms state-of-the-art proprietary and open-source models in documentation quality and consultation capability.

Jiachen Qian, Zhaolu Kang

The rapid proliferation of Multimodal Large Language Models (MLLMs) has ushered in the era of the “Agentic Economy,” where Mobile Agents autonomously execute high-stakes financial transactions. While these agents demonstrate impressive operational capabilities, their adversarial robustness remains a glaring blind spot. In this paper, we identify a systemic vulnerability termed Visual Dominance Hallucination (VDH), where imperceptible adversarial visual cues can act as a “super-stimulus,” overriding textual price evidence in our evaluated screenshot-based price-constrained settings and forcing the agent into irrational economic decisions. We propose PriceBlind, a stealthy, white-box adversarial attack framework for controlled screenshot-based evaluation. Unlike prior works that rely on conspicuous artifacts like pop-ups, PriceBlind exploits the modality gap in CLIP-based encoders via a novel Semantic-Decoupling Loss. Rather than literally making a luxury item “look cheap,” this regularizer weakens the consistency between high-price text and visual value cues by aligning the image embedding with a low-cost/value-associated anchor region while preserving pixel-level fidelity. On our main E-ShopBench benchmark with clear price constraints, screenshot-based white-box evaluation yields ASRs around 80% on the evaluated agents. Under the evaluated single-turn coordinate-selection protocol in a simplified layout-aware setting, our Ensemble-DI-FGSM strategy also yields non-trivial black-box transfer, with ASR roughly 35–41% across GPT-4o, Gemini-1.5-Pro, and Claude-3.5-Sonnet. In the same screenshot-based setting, standard robust encoders reduce ASR only partially, while a Verify-then-Act stack with robust encoders lowers ASR to below 10% at some clean-accuracy cost.

Suchen Liu, Yuanfeng Song, Jun Gao, Xing Chen

The growth of complex data fuels demand for automated insight discovery. While LLMs and agent technologies have advanced data analysis, existing methods struggle with maintaining contextual coherence, limited coverage due to single-path exploration, and rigid planning that fails to adapt to dynamic data discovery. We propose DataSeer, a collaborative multi-agent framework for automated insight discovery. Our first contribution is a Manager-Centric Collaborative Framework, where the Manager ensures cross-episode contextual coherence through a dual-layer memory system with compression, consolidation, and retrieval, alongside dynamic prompt editing, coordinating the overall process between the Planner and Executor. Second, we optimize the planning and execution components: the Planner employs multi-role discussion for adaptive sub-goal generation and plan refinement; the Executor is endowed with tactical autonomy for exploratory execution and incorporates real-time multi-dimensional self-assessment to guarantee insight quality. Third, we design Multi-Branch Reasoning that executes multiple discovery trajectories and synthesizes outcomes through LLM-based aggregation, improving coverage and reducing single-path stochasticity. Experiments on InsightBench and InsightEval show that DataSeer outperforms baselines, achieving improvements of 18.7% and 12.1% in insight-level scores, and 11.6% and 10.3% in summary-level scores, respectively.

Qingjia Huang, Jingyu Zhang, Jianguo Wu, Yakai Li, Weijuan Zhang, Yankai Rong, Junyi Yao, Shengzhi Zhang, Xiaoqi Jia

The assessment of jailbreak attacks against large language models currently suffers from inconsistent evaluation criteria and methods, leading to unreliable estimates of attack success rates. We propose JailMeter, an evidence-based evaluation framework designed to more faithfully measure jailbreak effectiveness. Inspired by the Information Bottleneck theory, JailMeter applies dual-feedback optimization to filter jailbreak noise from model responses while preserving content relevant to the original malicious question. This process produces concise evidence for a rigorous assessment under which an attack is validated only when the response captures the malicious intent and delivers a complete answer, thereby signaling a substantive bypass of model safety alignment. We evaluate JailMeter on JailMeter-Eva, a challenging benchmark containing 330 human-labeled, non-rejected jailbreak instances. JailMeter achieves an accuracy of 97.27%, substantially outperforming existing evaluation methods. To support large-scale evaluation, we further distill JailMeter into a small language model, JailMeterSLM, which maintains comparable reliability with significantly reduced computational costs. Code and dataset are available at https://github.com/Magi2B0y/JailMeter.

Wenhan Liu, Xinyu Ma, Yutao Zhu, Yuchen Li, Daiting Shi, Dawei Yin, Zhicheng Dou

Agentic search has recently emerged as a powerful paradigm, where an agent interleaves multi-step reasoning with on-demand retrieval to solve complex questions. Despite its success, how to design a retriever for agentic search remains largely underexplored. Existing search agents typically rely on similarity-based retrievers, while similar passages are not always useful for final answer generation. In this paper, we propose a novel retriever training framework tailored for agentic search. Unlike retrievers designed for single-turn retrieval-augmented generation (RAG) that only rely on local passage utility, we propose to use both local query-passage relevance and global answer correctness to measure passage utility in a multi-turn agentic search. We further introduce an iterative training strategy, where the search agent and the retriever are optimized bidirectionally and iteratively. Different from RAG retrievers that are only trained once with fixed questions, our retriever is continuously improved using evolving and higher-quality queries from the agent. Extensive experiments on seven single-hop and multi-hop QA benchmarks demonstrate that our retriever, termed Agentic-R, consistently outperforms strong baselines across different search agents.

Haonan Shangguan, Xiaocui Yang, Shi Feng, Daling Wang, Yifei Zhang, Feiliang Ren, Ge Yu

Current approaches for Multimodal Sentiment Analysis (MSA) primarily leverage the knowledge and reasoning capabilities of parameter-heavy (Multimodal) LLMs for classification, overlooking autonomous multimodal sentiment reasoning generation in resource-constrained environments.In this paper, we focus on the Resource-Limited Joint Multimodal Sentiment Reasoning and Classification task, JMSRC, which simultaneously performs multimodal sentiment reasoning chain generation and sentiment classification only with a lightweight model.We propose a Multimodal Chain-of-Thought Reasoning Distillation model, MulCoT-RD, designed for JMSRC that employs a "Teacher-Assistant-Student" distillation paradigm to address deployment constraints in resource-limited environments.We first leverage a high-performance Multimodal Large Language Model (MLLM) to generate the initial reasoning dataset and train a medium-sized assistant model with a multi-task learning mechanism. A lightweight student model is jointly trained to perform efficient multimodal sentiment reasoning generation and classification.Extensive experiments on four datasets demonstrate that MulCoT-RD with only 3B parameters and achieves strong performance on JMSRC, while exhibiting robust generalization and enhanced interpretability.

Junkun Qiu, Min Huang, Qinghai Miao

While LLM-based agents have emerged as a focal point for automating data science tasks, they continue to grapple with inefficient context management, "silent failures" (where code executes correctly but fails the task objectives), and error propagation inherent in sequential generation. In this paper, we propose Tree-Notebook, an agentic framework designed to mimic the iterative cognitive process of human data scientists. At its core, Tree-Notebook conceptualizes Jupyter Notebook cells as nodes within a tree structure, facilitating organized and efficient context retrieval. We formalize the task-solving process as a Partially Observable Markov Decision Process (POMDP) over a dynamic tree, utilizing an entropy-based information gain function for path evaluation to enhance adaptability in real-world environments. Furthermore, we introduce the "Data Shadow" system, which resolves silent failures by performing real-time tracking of data distributions, provenance, and semantic constraints. Experimental results demonstrate that Tree-Notebook achieves state-of-the-art (SOTA) performance on both InfiAgent-DABench and DSBench. To further evaluate robustness, we introduce an augmented version of InfiAgent-DABench to simulate complex environments, where Tree-Notebook consistently maintains its SOTA standing. Code is available at: https://github.com/QJK-BUAA/Tree-Notebook

Xinping Zhao, Shouzheng Huang, Yan Zhong, Xinshuo Hu, Meishan Zhang, Baotian Hu, Min Zhang

Retrieval-Augmented Generation (RAG) effectively improves the accuracy of Large Language Models (LLMs). However, retrieval noises significantly undermine the quality of LLMs’ generation, necessitating the development of denoising mechanisms. Previous works extract evidence straightforwardly without deep thinking, which may risk filtering out key clues and struggle with generalization. To this end, we propose EviOmni, which learns to extract rational evidence via reasoning first and then extracting. Specifically, EviOmni integrates evidence reasoning and evidence extraction into one unified trajectory, followed by knowledge token masking to avoid information leakage, optimized via on-policy reinforcement learning with verifiable rewards in terms of answer, length, and format. Extensive experiments on five benchmark datasets show the superiority of EviOmni, which provides compact and high-quality evidence, enhances the accuracy of downstream tasks, and supports both traditional and agentic RAG systems.

Hengyi Feng, Zeang Sheng, Meiyi Qiang, Yang Li, Wentao Zhang

Despite the remarkable success of multimodal large language models (MLLMs) in generative tasks, we observe that they exhibit a counterintuitive deficiency in the zero-shot multimodal retrieval task. In this work, we investigate the underlying mechanisms that hinder MLLMs from being effective retrievers. With the help of sparse autoencoders (SAEs), we decompose MLLM output representations into interpretable semantic concepts to probe their intrinsic behavior. Our analysis reveals that the representation space of MLLMs is overwhelmingly dominated by textual semantics; and the visual semantics essential for multimodal retrieval only constitute a small portion. We find that this imbalance is compounded by the heavy focus of MLLMs on bridging image-text modalities, which facilitates generation but homogenizes embeddings and finally diminishes the discriminative power required for multimodal retrieval. We further discover that the specific feature components that contribute most to the similarity computations of MLLMs are actually distractors that greatly reduce retrieval performance. Building on these insights, we propose , a test-time adaptation approach that applies a whitening transformation to adjust the geometry of MLLM representation spaces. Empirical results show that this simple intervention consistently improves zero-shot multimodal retrieval performance across diverse MLLMs without fine-tuning efforts.

Yuwei Guo, Zihan Zhao, Xiaowei Liu, Xiangning Yu, Qun Ma, Deyu Zhou, Xiao Xue

Multi-agent simulation based on LLMs has increasingly emerged as a new paradigm for exploring complex social phenomena and validating theoretical hypotheses. However, traditional experimental design in the social sciences relies heavily on interdisciplinary expert knowledge, involving cumbersome procedures and high technical barriers. While LLM-driven agents demonstrate broad prospects for designing experiments, their limitations regarding reliability and scientific rigor continue to significantly hinder their in-depth application in social science research. To address these challenges, this paper proposes FSTS, an automated framework for multi-agent experiment design based on script generation. Drawing on the concept of the "Decision Theater," the framework deconstructs experimental design into three core phases: Script Composition, Script Finalization, and Actor Generation. Tests across multiple scenarios indicate that the agents generated by this framework can enact the script within the "experimental theater," reproducing results consistent with real-world situations. The proposal of FSTS not only effectively lowers the barrier for social science experimental design but also provides scientifically grounded decision support for policy-making.

Changhai Zhou, Shiyang Zhang, Yuhua Zhou, Jun Gao, Qian Qiao, Shichao Weng, Weizhong Zhang, Cheng Jin

Deploying and fine-tuning Large Language Models (LLMs) on resource-constrained edge devices requires navigating a strict trade-off between memory footprint and task performance. Existing quantization-aware fine-tuning methods typically decouple weight precision and adapter capacity, overlooking that a layer’s ability to adapt is constrained by the information preserved in its frozen weights. Layers that are highly sensitive to quantization—whether due to representational specialization or accumulated error propagation—can become bottlenecks that adapter rank alone cannot recover. To address this issue, we introduce QR-Adaptor, a unified framework that jointly optimizes per-layer quantization bit-width and LoRA rank. We formulate resource allocation as a multi-objective discrete search guided by empirical layer-wise sensitivity, and implement it with a three-stage pipeline comprising KL-based sensitivity profiling, evolutionary exploration, and Bayesian refinement. Extensive experiments across LLaMA and Qwen models, including modern instruction tuning on OpenOrca and comparisons with strong PEFT baselines such as QDoRA, show that QR-Adaptor establishes a strong Pareto frontier: under a strict 4-bit memory budget, it matches or approaches 16-bit baselines while using substantially less memory.

Lei Wang, Min Huang, Eduard Dragut

Qualitative analysis is essential for studying complex social and behavioral phenomena, yet existing large language model (LLM) approaches face key limitations. Fully automated pipelines often compromise methodological rigor, while fully manual coding remains costly and labor-intensive. Although recent work emphasizes human–AI collaboration, existing multi-agent systems focus primarily on theme-level outputs, provide limited human oversight, and overlook fine-grained, data-level coding quality.We introduce CentaurTA, an iterative, self-improving human–agent framework for scalable thematic analysis. CentaurTA places humans in the loop to oversee and guide analysis, using expert feedback as a persistent learning signal to drive prompt-level refinement. By combining structured human feedback with rubric-based evaluation, CentaurTA provides fine-grained supervision for both open coding and theme construction while preserving methodological rigor. Experiments across multiple datasets, baselines, and LLM families show that CentaurTA improves coding alignment and transparency, highlighting the central role of human feedback in reliable qualitative analysis. Our code and data are available at https://github.com/Tom-Owl/CentaurTA.