论文检索

Yuchen Wu, Liang Ding, Li Shen, Dacheng Tao

Aligning Large Language Models (LLMs) to be faithful to new knowledge in complex, multi-hop reasoning tasks is a critical, yet unsolved, challenge. We find that SFT-based methods, e.g., Reason-KE, while state-of-the-art, suffer from a "faithfulness gap": they optimize for format mimicry rather than sound reasoning. This gap enables the LLM’s powerful parametric priors to override new contextual facts, resulting in critical factual hallucinations (e.g., incorrectly reasoning "Houston" from "NASA" despite an explicit edit). To solve this core LLM alignment problem, we propose **Reason-KE++**, an SFT+RL framework that instills process-level faithfulness. Its core is a Stage-aware Reward mechanism that provides dense supervision for intermediate reasoning steps (e.g., Decomposition, Sub-answer Correctness). Crucially, we identify that naive outcome-only RL is a deceptive trap for LLM alignment: it collapses reasoning integrity (e.g., 19.00% Hop acc) while superficially boosting final accuracy. Our process-aware framework sets **a new SOTA of 95.48%** on MQUAKE-CF-3k (+5.28%), demonstrating that for complex tasks, aligning the reasoning process is essential for building trustworthy LLMs.

Omar Naim, Krish Sharma, Niyar R Barman, Nicholas Asher

Large Language Models (LLMs) typically come with a fixed architecture, despite growing evidence that not all layers contribute equally to every downstream task. We introduce TALE (Task-Aware Layer Elimination), an inference-time method that improves task performance by selectively removing layers that are irrelevant or detrimental for a given task. TALE optimizes task-specific performance, yielding a task-optimized architecture without retraining. Across 9 tasks and 5 model families, under both zero-shot and few-shot settings, TALE consistently matches or surpasses baseline performance while simultaneously reducing computational costs. TALE also synergizes with fine-tuning, leading to further performance improvements. Computing TALE for a new task requires modest resources, making it a practical and deployable solution for task-specialized LLM inference.

Zichun Guo, Yuling Shi, Wenhao Zeng, Chao Hu, Haotian Lin, Terry Yue Zhuo, Jiawei Chen, Xiaodong Gu, Wenping Ma

Multimodal Large Language Models (MLLMs) have achieved remarkable performance in Visually Rich Document Understanding (VRDU) tasks, but their capabilities are mainly evaluated on pristine, well-structured document images. We consider document reconstruction from shredded fragments, a challenging VRDU setting that requires integrating visual pattern recognition with semantic reasoning under significant content discontinuities. To facilitate systematic evaluation of complex VRDU tasks, we introduce ShredBench, a benchmark supported by an automated generation pipeline that renders fragmented documents directly from Markdown. The proposed pipeline ensures evaluation validity by allowing the flexible integration of latest or unseen textual sources to prevent training data contamination. ShredBench assesses four scenarios (English, Chinese, Code, Table) with three fragmentation granularities (8, 12, 16 pieces). Empirical evaluations on state-of-the-art MLLMs reveal a significant performance gap: The method is effective on intact documents; however, once the document is shredded, restoration becomes a significant challenge, with NED dropping sharply as fragmentation increases. Our findings highlight that current MLLMs lack the fine-grained cross-modal reasoning required to bridge visual discontinuities, identifying a critical gap in robust VRDU research.

Daixin Shu, Jian Yang, Zhenhe Wu, Xianjie Wu, Xianfu Cheng, Guan Xiangyuan, Yanghai Wang, Pengfei Wu, Tingyang Yang, Hualei Zhu 等

Tabular data is a fundamental component of real-world information systems. However, existing multilingual table benchmarks suffer from geolinguistic imbalance - overrepresenting certain languages and lacking sufficient scale for rigorous cross-lingual analysis. To address these limitations, we introduce M3TQA, which is a comprehensive framework for massively multilingual multitask table question answering, including subsequent datasets M3TQA-BENCH and M3TQA-INSTRUCT, featuring tables expanded to 97 languages from Chinese and English sources. M3TQA-BENCH includes 6,606 professionally annotated question-answering pairs across four tasks designed to evaluate nuanced table reasoning capabilities. Additionally, we synthesized the training set M3TQA-INSTRUCT in 97 languages using Large Language Model (LLM). Experiments on state-of-the-art LLMs reveal critical insights into cross-lingual generalization, demonstrating that synthetically generated, unannotated training data can significantly boost performance, particularly for low-resource languages. M3TQA establishes a new standard for multilingual table understanding, providing both a challenging evaluation platform and a scalable methodology for future research.

Haijie Ruan, Xiaowu Jiang, Zhanpeng LI, Wei Jia, Xuanwu Xu, Xiao-Fen Shan, Shujie Chen, Xindong Ye

Large language models (LLMs) are increasingly proposed as conversational agents in healthcare, yet many existing systems treat roles as static prompts and rely on one-shot safety filters. In such designs, it can be difficult to enforce long-horizon responsibilities, stable role identity, and realistic communication behavior. We propose a Self-Evolving LLM Agent that learns from role-based social experience and explicitly models communicator-level individual traits informed by prior communication questionnaires and clinical literature. The agent integrates (i) perception and action conditioned on both hard role responsibility norms and soft trait-conditioned style preferences, (ii) structured memory storing norm-annotated trajectories and identity states, (iii) dual-layer reflection that combines short-term responsibility diagnosis with long-term identity drift detection via trait consistency and trait-norm compatibility checks, and (iv) self-evolution that updates system prompts and identity parameters through preference-style optimization with AI feedback. We instantiate the framework in a multi-role healthcare sandbox and evaluate outpatient medication review, emergency triage, and discharge planning. Across our simulated tasks, self-evolution is associated with lower severity-weighted norm risk, more stable role-identity signals, and improved social embeddedness metrics (including trust-like signals) relative to strong static baselines.

Midan Shim, Seokju Hwang, KaeHyun Um, Kyong-Ho Lee

Large language models still struggle with faithfulness and hallucinations despite their remarkable reasoning abilities. In Knowledge Graph Question Answering (KGQA), semantic parsing-based approaches address the limitations by understanding constraints in a user’s question and converting them into a logical form to execute on a knowledge graph. However, existing KGQA benchmarks and methods are biased toward positive and calculation constraints. Negative constraints are neglected, although they frequently appear in real-world questions. In this paper, we introduce a new task, NEgative-conSTrained (NEST) KGQA, where each question contains at least one negative constraint, and a corresponding dataset, NestKGQA. We also design PyLF, a Python-formatted logical form, since existing logical forms are hardly suitable to express negation clearly while maintaining readability. Furthermore, NEST questions naturally contain multiple constraints. To mitigate their semantic complexity, we present a novel framework named CUCKOO, specialized to multiple-constrained questions and ensuring semantic executability. CUCKOO first generates a constraint-aware logical form draft and performs schema-guided semantic matching. It then selectively applies self-directed refinement only when executing improper logical forms yields an empty result, reducing cost while improving robustness. Experimental results demonstrate that CUCKOO consistently outperforms baselines on both conventional KGQA and NEST-KGQA benchmarks under few-shot settings.

Chunyu Ye, Yunhao Zhang, Jingyuan Sun, Chong Li, Yang Zhao, Shaonan Wang

Decoding language from the human brain remains a grand challenge for Brain-Computer Interfaces (BCIs). Current approaches typically rely on unimodal brain representations, neglecting the brain’s inherently multimodal processing. Inspired by the brain’s associative mechanisms, where viewing an image can evoke related sounds and linguistic representations, we propose a unified framework that leverages Multimodal Large Language Models (MLLMs) to align brain signals with a shared semantic space encompassing text, images, and audio. A router module dynamically selects and fuses modality-specific brain features according to the characteristics of each stimulus. Experiments on various fMRI datasets with textual, visual, and auditory stimuli demonstrate state-of-the-art performance, achieving an 8.48% average improvement on the most commonly used benchmark. We further extend our framework to EEG and MEG data, demonstrating flexibility and robustness across varying temporal and spatial resolutions. To our knowledge, this is the first unified BCI architecture capable of robustly decoding multimodal brain activity across diverse brain signals and stimulus types, offering a flexible solution for real-world applications.

Donghee Han, Daeyoung Roh, A Young Kim, Hwanjun Song, Mun Yong Yi

Large Language Models (LLMs) have shown remarkable potential in recommendation systems but suffer from prohibitive inference latency. Existing distillation approaches typically target Small Language Models (SLMs) or Conventional Recommendation Models (CRMs), face a critical trade-off between computational cost and semantic reasoning capacity. To bridge this accuracy-efficiency gap, we introduce Reasoning-to-Encoder Distillation (R2END), a framework that establishes a text encoder as the optimal student architecture for scalable recommendation. Unlike methods that mimic token generation, R2END compresses the teacher’s reasoning into a dense vector space via a semantic alignment objective, effectively capturing user-item dynamics. Extensive experiments on four datasets demonstrate that R2END not only outperforms state-of-the-art baselines but also achieves drastically reduced latency, offering a sweet spot for recommendation.

Changhao Jiang, Jiahao Chen, Zhenghao Xiang, Zhixiong Yang, Hanchen Wang, Jiabao Zhuang, Xinmeng Che, Jiajun Sun, Hui Li, Yifei Cao 等

Recent commercial systems such as Suno demonstrate strong capabilities in long-form song generation, while academic research remains largely non-reproducible due to the lack of publicly available training data, hindering fair comparison and progress. To this end, we release a fully open-source system for long-form song generation with fine-grained style conditioning, including a licensed synthetic dataset, training and evaluation pipelines, and Muse, an easy-to-deploy song generation model. The dataset consists of 116k fully licensed synthetic songs with automatically generated lyrics and style descriptions paired with audio synthesized by SunoV5. We train Muse via single-stage supervised finetuning of a Qwen-based language model extended with discrete audio tokens using MuCodec, without task-specific losses, auxiliary objectives, or additional architectural components. Our evaluations find that although Muse is trained with a modest data scale and model size, it achieves competitive performance on phoneme error rate, text–music style similarity, and audio aesthetic quality, while enabling controllable segment-level generation across different musical structures. All data, model weights, and training and evaluation pipelines will be publicly released, paving the way for continued progress in controllable long-form song generation research.

Yi Sui, Chaozhuo Li, Dawei Song

Large Reasoning Models (LRMs) achieve strong performance on complex tasks by leveraging long Chain-of-Thought (CoT), but often suffer from overthinking, leading to excessive reasoning steps and high inference latency. Existing CoT compression methods struggle to balance accuracy and efficiency, and lack fine-grained, step-level adaptation to redundancy and reasoning bias. Therefore, we propose State-Aware Reasoning Compression with Knowledge Guidance (STACK), a framework that performs step-wise CoT compression by explicitly modeling stage-specific redundancy sources and integrating with a retrieval-augmented guidance. STACK constructs online long–short contrastive samples and dynamically switches between knowledge-guided compression for uncertain or biased reasoning state and self-prompted compression for overly long but confident state, complemented by an answer-convergence-based early stopping mechanism to suppress redundant verification. We further propose a reward-difference-driven training strategy by combining Proximal Policy Optimization (PPO) and Direct Preference Optimization (DPO), enabling models to learn state-conditioned compression strategies. Experiments on three mathematical reasoning benchmarks show that STACK achieves a superior accuracy–efficiency balance, reducing average response length by 59.9% while improving accuracy by 4.8 points over existing methods.

Meng Zhang, Jinzhong Ning, Xiaolong Wu, Hongfei Lin, Yijia Zhang

Grounded Multimodal Named Entity Recognition (GMNER) aims to jointly identify named entity mentions in text, predict their semantic types, and ground each entity to a corresponding visual region in an associated image. Existing approaches predominantly adopt pipeline-based architectures that decouple textual entity recognition and visual grounding, leading to error accumulation and suboptimal joint optimization. In this paper, we propose E2E-GMNER, a fully end-to-end generative framework that unifies entity recognition, semantic typing, visual grounding, and implicit knowledge reasoning within a single multimodal large language model. We formulate GMNER as an instruction-tuned conditional generation task and incorporate chain-of-thought reasoning to enable the model to adaptively determine when visual evidence or background knowledge is informative, reducing reliance on noisy cues. To further address the instability of generative bounding box prediction, we introduce Gaussian Risk-Aware Box Perturbation (GRBP), which replaces hard box supervision with probabilistically perturbed soft targets to improve robustness against annotation noise and discretization errors. Extensive experiments on the Twitter-GMNER and Twitter-FMNERG benchmarks demonstrate that E2E-GMNER achieves highly competitive performance compared with state of the art methods, validating the effectiveness of unified end-to-end optimization and noise-aware grounding supervision.

Xiaoyang Liu, Dawei Wang, Tian Li, Huizhi Liang, Gary Ushaw, Richard Davison

Reasoning capability is fundamental in enabling Large Language Models to perform complex multi-step inference. By sampling multiple reasoning paths and selecting the most frequent answer, Self Consistency (SC) remains highly effective but fails on challenging tasks where incorrect answers dominate the majority. Inspired by Metropolis Light Transport in physically-based rendering, where discovered high-contribution light paths guide subsequent sampling toward illumination sources, we propose Metropolis Self Consistency and its multi-LLM extension, Metropolis Cross Consistency, a probabilistic self- and cross-consistency verification framework for mathematical reasoning. Our approach employs an accept-reject mechanism to encourage high-quality reasoning paths, concentrating sampling in regions more likely to yield correct answers. Experiments on 9 LLMs across 4 challenging mathematical benchmarks demonstrate consistent improvements over SC. Even when combining models of vastly different capabilities, MCC maintains performance virtually matching the most capable model while significantly reducing computational cost compared to SC with the strongest model alone. While our implementation is training-free, adds minimal token overhead beyond SC, and requires no external reward model, our approach provides a flexible paradigm that can accommodate any scalar reward representing path correctness.

Jun Seo, Sangwon Ryu, Heejin Do, Hyounghun Kim, Gary Lee

Knowledge Tracing (KT) aims to predict learners’ future performance from past interactions. While recent KT approaches have improved via learning item representations aligned with Knowledge Components, they overlook the procedural dynamics of problem solving. We propose Behavior-Aware Item Modeling (BAIM), a framework that enriches item representations by integrating dynamic procedural solution information. BAIM leverages a reasoning language model to decompose each item’s solution into four problem-solving stages (i.e., understand, plan, carry out, and look back), pedagogically grounded in Polya’s framework. Specifically, it derives stage-level representations from per-stage embedding trajectories, capturing latent signals beyond surface features. To reflect learner heterogeneity, BAIM adaptively routes these stage-wise representations, introducing a context-conditioned mechanism within a KT backbone, allowing different procedural stages to be emphasized for different learners. Experiments on XES3G5M and NIPS34 show that BAIM consistently outperforms strong pretraining-based baselines, achieving particularly large gains under repeated learner interactions.

Dingzirui Wang, Xuanliang Zhang, Rongyu Cao, Longxu Dou, Xianzhen Luo, Yingwei MA, Qingfu Zhu, Binhua Li, Fei Huang, Yongbin Li

Generating and voting multiple answers is an effective method to mitigate reasoning inconsistencies of large language models (LLMs). Prior works have shown that multiple reasoning formats outperform a single format when generating multiple answers. However, previous works using multiple formats rely on formats labeled by humans, which could be unsuitable for all tasks and have high labeling costs. To address this issue, we adapt suitable formats to the given tasks by generating and selecting formats. We first propose how to measure the reasoning error when generating multiple answers. Then, we introduce Format-Adapter, which utilizes LLMs to generate and select suitable reasoning formats by minimizing the error measurement we present. We conduct experiments on math and commonsense reasoning tasks, where Format-Adapter achieves a 4.3% performance improvement on average over previous works, demonstrating the effectiveness.

Yuyang Hu, Jiongnan Liu, Jiejun Tan, Yutao Zhu, Zhicheng Dou

Large language models (LLMs) are increasingly deployed as intelligent agents that reason, plan, and interact with their environments. To effectively scale to long-horizon scenarios, a key capability for such agents is a memory mechanism that can retain, organize, and retrieve past experiences to support downstream decision-making. However, most existing approaches organize and store memories in a flat manner and rely on simple similarity-based retrieval techniques. Even when structured memory is introduced, existing methods often struggle to explicitly capture the logical relationships among experiences or memory units. Moreover, memory access is largely detached from the constructed structure and still depends on shallow semantic retrieval, preventing agents from reasoning logically over long-horizon dependencies. In this work, we propose CompassMem, an event-centric memory framework inspired by Event Segmentation Theory. CompassMem organizes memory as an Event Graph by incrementally segmenting experiences into events and linking them through explicit logical relations. This graph serves as a logic map, enabling agents to perform structured and goal-directed navigation over memory beyond superficial retrieval, progressively gathering valuable memories to support long-horizon reasoning. Experiments on LoCoMo and NarrativeQA demonstrate that CompassMem consistently improves both retrieval and reasoning performance across multiple backbone models.

Ziwei Wang, Junjie Zheng, Leyang Yang, Sheng Zhou, Xiaoxuan Tang, Fang Zhouhua, Zhiwei Liu, Dajun Chen, Yong Li, Jiajun Bu

Autonomous Graphical User Interface (GUI) agents powered by Multimodal Large Language Models (MLLMs) enable digital automation on end-user devices. While scaling both parameters and data has yielded substantial gains, advanced methods still suffer from prohibitive deployment costs on resource-constrained devices. When facing complex in-the-wild scenarios, lightweight GUI agents are bottlenecked by limited capacity and poor task scalability under end-to-end episodic learning, impeding multi-agent systems (MAS) adaptation, while training multiple skill-specific experts remains costly. Can we strike an effective trade-off in this cost–scalability dilemma, enabling lightweight MLLMs to participate in realistic GUI workflows? To address these challenges, we propose LAMO framework, which endows a lightweight MLLM with GUI-specific knowledge and task scalability, allowing multi-role orchestration to expand their capability boundary for GUI automation. LAMO combines role-oriented data synthesis with a two-stage training recipe: (i) supervised fine-tuning with Perplexity-Weighted Cross-Entropy optimization for knowledge distillation and visual perception enhancement, and (ii) reinforcement learning for role-oriented cooperative exploration. Via LAMO, we develop a task-scalable native GUI agent LAMO-3B supporting monolithic execution and MAS-style orchestration. When paired with advanced planners, as a plug-and-play policy executor, LAMO-3B can continuously benefit from planner advances, enabling a higher performance ceiling. Extensive static and online evaluations validate the effectiveness of our designs.

Yihong Liu, Raoyuan Zhao, Hinrich Schuetze, Michael A. Hedderich

Large reasoning models (LRMs) achieve strong performance on mathematical reasoning tasks, often attributed to their capability to generate explicit chain-of-thought (CoT) explanations. However, recent work shows that LRMs often arrive at the correct answer before completing these textual reasoning steps, indicating the presence of latent reasoning – internal, non-verbal computation encoded in hidden states. While this phenomenon has been explored in English, its multilingual behavior remains largely unknown. In this paper, we conduct a systematic investigation of multilingual latent reasoning in LRMs across 11 languages. Using a truncation-based strategy, we examine how the correct answer emerges as the model is given only partial reasoning traces, allowing us to measure stepwise latent prediction formation. Our results reveal clear evidence of multilingual latent reasoning, though unevenly: strong in resource-rich languages, weaker in low-resource ones, and broadly less observable on harder benchmarks. To understand whether these differences reflect distinct internal mechanisms, we further perform representational analyses. Despite surface-level disparities, we find that the internal evolution of predictions is highly consistent across languages and broadly aligns with English – a pattern suggesting an English-centered latent reasoning pathway.

Nan Sun, Jing Tang, Lei Sun, Rui Chen, Yuxing Lu, Xiangxiang Chu, Hefei Ling, Yujun Cai

Zero-Shot Composed Image Retrieval (ZS-CIR) retrieves target images using a reference image and modification text without task-specific training. Existing methods typically rely on MLLMs to generate query vectors with pre-trained models like CLIP. However, those constructed queries suffer from inherent cognitive bias due to unknown candidate distribution. We propose CoRR, a training-free framework that reframes ZS-CIR as a self-correcting process through bias-aware query refinement. CoRR uses retrieved results as feedback to perceive the candidate distribution. With carefully designed CoT prompting, the MLLM inspects the retrieved candidates to identify intent misalignments in the query and then corrects them via Historical Query Fusion. We also introduce Retrieval-Driven Caption Optimization to provide context-aligned examples, reducing phrasing and style mismatches. Experiments on public benchmarks show that CoRR significantly outperforms other SOTA methods.

Yu-Shiang Huang, Yun-Yu Lee, Tzu-Hsin Chou, Che Lin, Chuan-Ju Wang

Numerical precision is critical in financial NLP, yet embedding-based semantic similarity metrics exhibit numerical blindness—failing to distinguish contradictory values within similar contexts. We introduce NASH (Numerically Aware Scoring Hueristic), a model-agnostic metric that decouples numerical verification from textual semantic evaluation through a three-stage pipeline: (1) modal separation via numeric masking, (2) dual-channel similarity estimation through masked-text similarity and context-aware numeric alignment, and (3) IDF-weighted aggregation. NASH functions as a drop-in enhancement to existing embedding-based metrics. Validated on our proposed NumFinE financial numerical evaluation benchmark and established semantic similarity datasets (STS-B, Financial-STS), NASH achieves substantial improvements in numerical sensitivity (up to +159.6% on listwise ranking) while preserving general semantic performance, establishing a reliable standard for numeracy-aware evaluation.

Yingzhi Mao, Chunkang Zhang, Junxiang Wang, Xinyan Guan, Boxi Cao, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun

Large Reasoning Models (LRMs) achieve strong performance on complex multi-step reasoning, yet they still exhibit severe safety failures such as harmful content generation. Existing methods often apply coarse-grained constraints over the entire reasoning trajectories, which can undermine reasoning capability while failing to address the root causes of unsafe behavior. In this work, we uncover a previously underexplored failure mode in LRMs, termed Self-Jailbreak, where models initially recognize the harmful intent of a query, but override this judgment during subsequent reasoning steps, ultimately generating unsafe outputs. Such a phenomenon reveals that LRMs are capable of recognizing harm, while safety failures primarily arise from reasoning steps. Motivated by this finding, we propose Chain-of-Guardrail (CoG), a trajectory-level training framework that mitigates Self-Jailbreak via targeted, step-level interventions while maintaining reasoning ability. Experiments across multiple safety and reasoning benchmarks indicate that CoG achieves a favorable balance between safety and reasoning performance compared with existing approaches.