论文检索

Fengran Mo, Yifan Gao, Sha Li, Hansi Zeng, Xin Liu, Zhaoxuan Tan, Xian Li, Jianshu Chen, Dakuo Wang, Meng Jiang

Large Language Models (LLMs) have become a popular interface for human–AI interaction, supporting information seeking and task assistance through natural, multi-turn dialogue. To respond to users within multi-turn dialogues, the context-dependent user intent evolves across interactions, requiring contextual interpretation, query reformulation, and dynamic coordination between retrieval and generation. Existing studies usually follow static “rewrite, retrieve, and generate” pipelines, which optimize different procedures separately and overlook the mixed-initiative action optimization simultaneously. Although the recent developments in deep search agents demonstrate the effectiveness in jointly optimizing retrieval and generation via reasoning, these approaches focus on single-turn scenarios, which might lack the ability to handle multi-turn interactions. We introduce a conversational agent that interleaves search and reasoning across turns, enabling exploratory and adaptive behaviors learned through reinforcement learning (RL) training with tailored rewards towards evolving user goals. The experimental results across four widely used conversational benchmarks demonstrate the effectiveness of our methods by surpassing several existing strong baselines.

Jiacheng Shi, Hongfei Du, Xinyuan Song, Y. Alicia Hong, Yanfu Zhang, Ashley Gao

Neural speech codecs provide discrete representations for speech language models, but emotional cues are often degraded during quantization. Existing codecs mainly optimize acoustic reconstruction, leaving emotion expressiveness insufficiently modeled at the representation level. We propose an emotion-guided neural speech codec that explicitly preserves emotional information while maintaining semantic fidelity and prosodic naturalness. Our framework combines emotion–semantic guided latent modulation, relation-preserving emotional–semantic distillation, and emotion-weighted semantic alignment to retain emotionally salient cues under compression. Extensive evaluations across speech reconstruction, emotion recognition, and downstream text to speech generation demonstrate improved emotion consistency and perceptual quality without sacrificing content accuracy.

Chenyang Yang, Yike Shi, Qianou Ma, Michael Xieyang Liu, Christian Kaestner, Tongshuang Wu

Prompt underspecification is a common challenge when interacting with LLMs. In this paper, we present an in-depth analysis of this problem, showing that while LLMs can often infer unspecified requirements by default (41.1%), such behavior is fragile: Under-specified prompts are 2x as likely to regress across model or prompt changes, sometimes with accuracy drops exceeding 20%. This instability makes it difficult to reliably build LLM applications. Moreover, simply specifying all requirements does not consistently help, as models have limited instruction-following ability and requirements can conflict. Standard prompt optimizers likewise provide little benefit. To address these issues, we propose requirements-aware prompt optimization mechanisms that improve performance by 4.8% on average over baselines. We further advocate for a systematic process of proactive requirements discovery, evaluation, and monitoring to better manage prompt underspecification in practice.

Haoqin Sun, Jinghua Zhao, Xuechen Wang, Shiwan Zhao, Jiaming Zhou, Hui Wang, Xi Yang, Yequan Wang, Yonghua Lin

The advancement of Multimodal Emotion Recognition (MER) in Chinese is significantly hindered by the scarcity of high-quality, spontaneous dialogue datasets compared to their English counterparts. In this work, we introduce EmotionTalk, the first interactive Chinese multimodal dataset designed to capture the nuance of authentic emotional interplay. Collected from 19 professional actors, the dataset spans 23.6 hours of dyadic conversations across diverse scenarios. A key contribution of EmotionTalk is its multi-grained annotation system, which integrates standard categorical and dimensional labels with fine-grained emotional speaking style captions, enabling research into interpretable emotion analysis. We establish comprehensive benchmarks for emotion recognition and captioning tasks, verifying the dataset’s effectiveness and the necessity of multimodal fusion. EmotionTalk serves as a critical resource for bridging the gap in non-English affective computing and is publicly released for the research community.

Mingzi Zuo, Lei Zhang, Hailiang Sun, Shengzhi Huo, Changyu Dong, Xin Wang, Bo Wang, Hao Liu

The widespread dissemination of toxic content on online platforms poses a critical threat to user experience. Toxicity detection in speech receives significantly less research attention than its text counterpart. Most existing methods rely on high-resource languages and employ a cascaded pipeline combining automatic speech recognition (ASR) and text classifiers. These designs limit robustness in low-resource languages and discard important acoustic cues. To address the lack of datasets, we construct PolySpeechTox, the first toxicity-annotated speech dataset spanning 53 languages and accent varieties, with a focus on low-resource languages and multiple accents. Based on PolySpeechTox, we conduct the first systematic study of toxic speech detection under low-resource, multilingual, and multi-accent conditions. We propose SoftPrompt-TSD, a prompt-based adaptation framework that leverages a frozen audio language model to perform end-to-end toxicity detection without ASR. The decomposed soft-prompt design balances global task alignment, cross-lingual generalization, and language-specific or accent-specific calibration. On PolySpeechTox, SoftPrompt-TSD achieves a micro-averaged ROC-AUC of 98.07%, mitigating the severe failures observed in baseline methods for several languages. In three generalization experiments, SoftPrompt-TSD demonstrates superior generalization capability and maintains robust performance against distribution shifts.

Yifei Zhang, Xu Yang, Xiao Yang, Bowen Xian, Qizheng Li, Shikai Fang, Jingyuan Li, Jian Wang, Minrui Xu, Yuge Zhang 等

LLM-based agents for machine learning engineering (MLE) predominantly rely on tree search, a form of gradient-free optimization that uses scalar validation scores to rank candidates. As LLM reasoning capabilities improve, exhaustive enumeration becomes increasingly inefficient compared to directed updates, analogous to how accurate gradients enable efficient descent over random search. We introduce Gome, an MLE agent that operationalizes gradient-based optimization. Gome maps structured diagnostic reasoning to gradient computation, success memory to momentum, and multi-trace execution to distributed optimization. Under a closed-world protocol that isolates architectural effects from external knowledge, Gome achieves a state-of-the-art 35.1% any-medal rate on MLE-Bench with a restricted 12-hour budget on a single V100 GPU. Scaling experiments across 10 models reveal a critical crossover: with weaker models, tree search retains advantages by compensating for unreliable reasoning through exhaustive exploration; as reasoning capability strengthens, gradient-based optimization progressively outperforms, with the gap widening at frontier-tier models. Given the rapid advancement of reasoning-oriented LLMs, this positions gradient-based optimization as an increasingly favorable paradigm. We release our codebase and GPT-5 traces at: https://github.com/microsoft/RD-Agent.

Zhihan Zhou, Daqian Shi, Lida Shi, Rui Song, Peiqiang Qiu, Xiaolei Diao, Hao Xu

Research on ancient Chinese language is of great significance for tracing Chinese history and civilization. In the field of large language models, studies on the pre-Qin excavated documents such as Oracle Bone Inscriptions, Bronze Inscriptions, and Bamboo Book of Chu remain insufficient. This is because these ancient characters have a low level of digitization, training corpora are extremely scarce, and they typically contain complex and rich semantic information. Therefore, we propose an ancient character semantic-aware embedding for large language models. This embedding integrates both the glyph and lexicality of ancient characters and maps them to the modern Chinese semantic space. We also design a two-stage method for lightweight and parameter-efficient training of the embedding. Finally, we conduct extensive experiments on excavated documents from the pre-Qin period, and the results demonstrate the effectiveness of our approach.

Zhen Bi, Zhenlin Hu, Xueshu Chen, Mingyang Chen, Cheng Deng, Yida Xue, Zhen Wang, Qing Shen, Ningyu Zhang, Jungang Lou

The reasoning capabilities of Large Language Models (LLMs) are increasingly attributed to training data quality rather than mere parameter scaling. However, existing data-centric paradigms often equate quality with factuality or diversity and ignore the internal logical complexity of training samples. In this work, we propose that natural language harbors Structured Logical Knowledge manifested through entailment relationships and logical topologies. To quantify this, we introduce Structured Logical Knowledge Density (SLKD), a novel metric that measures logical information content by decomposing natural language into executable predicates and logical primitives. Our analysis reveals a significant logical disparity in current datasets where sparse logical signals predominate. Consequently, we propose a density-aware re-cognizing optimization strategy that prioritizes high-density logical samples to align training with the model’s reasoning boundary. Extensive experiments demonstrate that our approach enhances reasoning performance and generalization without increasing total data volume. These results, further validated within a reinforcement learning framework, suggest that elevating logical density is more critical than expanding data scale for realizing the full cognitive potential of LLMs. The anonymized code is available in the Appendix C.

Yassine Bouziane, Youness Moukafih, Mounir Ghogho

Large language models (LLMs) have recently achieved impressive results on multiple-choice question answering (MCQA), with retrieval-augmented generation (RAG) emerging as an effective strategy for improving the performance of smaller models. However, existing RAG formulations face persistent challenges: retrieving too many passages often introduces noise, and even when relevant content is retrieved, models may still struggle with partially relevant or conflicting information. Moreover, while LLMs perform strongly on English benchmarks, their accuracy declines substantially on Arabic multi-task evaluations, revealing ongoing issues in cross-lingual transfer and domain adaptation. In this paper, we propose a novel approach, using Arabic as a representative case study, that jointly models the relevance of both the question and its candidate answers when selecting contextual passages. The method employs a lightweight reranker trained with a hybrid regression–triplet loss objective to identify passages that provide discriminative and reliable evidence. Extensive experiments across multiple model sizes and humanities domains show that our approach consistently outperforms both standard RAG baselines and reranker baselines, delivering two- to threefold improvements while remaining competitive with considerably larger models.

Zeqi Chen, Zhaoyang Chu, Yi Gui, Feng Guo, Yao Wan, Chuan Shi

Large Language Models (LLMs) have demonstrated remarkable performance in code intelligence tasks such as code generation, summarization, and translation. However, their reliance on linearized token sequences makes them brittle to long-range program dependencies and superficial lexical shifts such as identifier renaming. Existing structure-aware approaches typically treat structure as serialized text prompts or auxiliary training objectives, which often inflate context length or rely on internalized structural priors, failing to provide explicit guidance during inference. To address these limitations, we propose CGBridge, a novel plug-and-play method that enhances LLMs with Code Graph information through an external, trainable Bridge module. It aligns Code Property Graph structure with code semantics and compresses them into compact soft-prefixes, decoupling structural reasoning from textual generation without updating the backbone. Experiments across multiple code LLM backbones and scales show consistent gains over both text-only adaptation and graph-augmented baselines. Furthermore, CGBridge remains robust under identifier renaming and enables over 4× faster inference than LoRA-tuned models, demonstrating both effectiveness and efficiency in structure-aware code understanding.

Kaiyue Sun, Rongyao Fang, Chengqi Duan, Xian Liu, Aoxue Li, Xihui Liu

Text-to-image (T2I) generative models have achieved remarkable progress, demonstrating exceptional capability in synthesizing high-quality images from textual prompts. While existing research and benchmarks have extensively evaluated the ability of T2I models to follow the literal meaning of prompts, their ability to reason over prompts with domain knowledge to uncover implicit meaning and contextual nuances remains underexplored. To bridge this gap, we introduce T2I-ReasonBench, a novel benchmark designed to explore the knowledge-driven reasoning capabilities of T2I models.T2I-ReasonBench comprises 800 meticulously designed prompts organized into four dimensions: (1) Idiom Interpretation, (2) Textual Image Design, (3) Entity Reasoning, and (4) Scientific Reasoning. These dimensions challenge models to integrate domain knowledge, infer implicit meaning, and resolve contextual ambiguities. To quantify the performance, we introduce a two-stage evaluation framework: a large language model (LLM) generates prompt-specific question-criterion pairs that evaluate if the image includes the essential elements resulting from correct reasoning; a multimodal LLM (MLLM) then scores the generated image against these criteria. Our comprehensive study across 16 state-of-the-art diffusion and unified multimodal models (UMMs) reveal two primary bottlenecks. First, many models lack the foundational reasoning ability to fully comprehend complex prompts. Second, even models with stronger reasoning modules exhibit a persistent gap between their internal understanding and the final generated image. This highlights an urgent need for the next generation of T2I systems to not only improve their reasoning capability but also to enhance integration between reasoning and synthesis.

Atharva Naik, Prakam, Yash Mathur, Darsh Agrawal, Manav Nitin Kapadnis, Yuwei An, Clayton Marr, Carolyn Rose, David R. Mortensen

While many benchmarks evaluate the reasoning abilities of Large Language Models (LLMs), few isolate reasoning as a capability independent of domain knowledge. We introduce a new benchmark for inductive reasoning inspired by Sound Law Induction (SLI) in historical linguistics and formulated in a simple multi-step Programming by Example (PBE) framework. The task requires inducing a cascade of string rewrite programs that transform inputs into target outputs. We present PBEBench, a fully automated evaluation approach that generates such problems with controllable difficulty and ordering constraints, enabling scalable and contamination-resistant evaluation of sequential inductive reasoning. Using this approach, we construct three datasets that show a large gap between models that leverage test-time compute or long chain-of-thought reasoning and those that do not. Although recent models such as GPT-5 and gpt-oss-120b show promise, solve rates remain below 5% on hard PBEBench instances with long program cascades, even under computationally expensive scaling strategies. Finally, we show that PBEBench scores are more predictive of performance on real SLI than are other inductive reasoning benchmarks. We will release code and data to support further research.

Zhiyi Duan, Xiangren Wang, Jiangshan Guan, Bing Jia, Qianli Xing

Teacher sentiment analysis is pivotal for understanding instructional dynamics, yet it remains challenging because classroom expressions are professionally regulated performances rather than spontaneous outbursts. However, existing approaches typically treat sentiment as a static, monolithic label, failing to capture this structured heterogeneity. To effectively model this complexity, we decompose teacher sentiment into three granularities: coarse-level performativity, medium-level intra-class heterogeneity, and fine-level cross-modal complementarity. Guided by this perspective, we propose CF-TSA, a coarse-to-fine multimodal framework. Specifically, we employ CLS-guided cross-modal attention to recover effective signals from regulated displays (coarse-level), thresholded substyle discovery to identify latent pedagogical styles (medium-level), and substyle-aware contrastive learning to align dynamic multimodal cue compositions (fine-level). Experiments on T-MED and CMU-MOSEI demonstrate that CF-TSA consistently outperforms state-of-the-art baselines, validating the effectiveness of the coarse-to-fine perspective and the hierarchical modeling.

Chen Lifan, Wei Ding, Jingwen Yang, Xiuze Zhou, Jingan Chen, Fan Lin

Financial report generation is a complex task that requires gathering and reasoning over multi-source information. Recent advances in Large Language Models have made them a promising solution for automating this process. However, the reasoning paths in traditional Chain-of-Thought paradigms are inherently constrained by predefined, static computational topologies, rendering them ill-equipped to handle the dynamic uncertainties of real-world financial environments. To tackle this challenge, we propose Cogito, a cognitively grounded agentic framework for professional financial report generation. At its core, Cogito is driven by Dynamic Graph of Thoughts, a novel reasoning mechanism that models the agent’s reasoning process as an evolving topology for adaptive exploration.We further introduce a Social Collaboration Mechanism to facilitate coordinated agent interaction. Finally, Cogito is instantiated as a multi-agent system, where four specialized agents collaboratively execute the end-to-end report generation task. Extensive experiments on enterprise- and industry-level financial report generation benchmarks demonstrate the superiority of Cogito in data quality, analytical validity, and presentation quality.

Zhixiong Zhuang, Maria-Irina Nicolae, Hui-Po Wang, Mario Fritz

The integration of large language models (LLMs) into a wide range of applications has highlighted the critical role of well-crafted system prompts, which require extensive testing and domain expertise. These prompts enhance task performance but may also encode sensitive information and filtering criteria, posing security risks if exposed. Recent research shows that system prompts are vulnerable to extraction attacks, while existing defenses are either easily bypassed or require constant updates to address new threats. In this work, we introduce ProxyPrompt, a novel defense mechanism that prevents prompt leakage by replacing the original prompt with a proxy. This proxy maintains the original task’s utility while obfuscating the extracted prompt, ensuring attackers cannot reproduce the task or access sensitive information. Comprehensive evaluations on 264 LLM and system prompt pairs show that ProxyPrompt protects 94.70% of prompts from extraction attacks, outperforming the next-best defense, which only achieves 42.80%.

Yuhao Wu, Yushi Bai, Zhiqiang Hu, Juanzi Li, Roy Ka-Wei Lee

Long-form text generation remains a significant challenge for large language models (LLMs), particularly in maintaining coherence, ensuring logical consistency, and preserving text quality as sequence length increases. To address these limitations, we propose \textit{SuperWriter}-Agent, an agent-based framework designed to enhance the quality and consistency of long-form text generation. \textit{SuperWriter}-Agent introduces explicit structured thinking-through planning and refinement stages—into the generation pipeline, guiding the model to follow a more deliberate and cognitively grounded process akin to that of a professional writer. Based on this framework, we construct a supervised fine-tuning dataset to train a 7B \textit{SuperWriter}-LM. We further develop a hierarchical Direct Preference Optimization (DPO) procedure that uses Monte Carlo Tree Search (MCTS) to propagate final quality assessments and optimize each generation step accordingly. Empirical results across diverse benchmarks demonstrate that \textit{SuperWriter}-LM achieves state-of-the-art performance, surpassing even larger-scale baseline models in both automatic evaluation and human evaluation. Furthermore, comprehensive ablation studies demonstrate the effectiveness of hierarchical DPO and underscore the value of incorporating structured thinking steps to improve the quality of long-form text generation.

Edward Y Chang

Do frontier LLMs reason causally, or do they pattern-match, yielding under pressure and hedging under uncertainty? We frame causal judgment as evaluation along three axes, Utility, Safety, and Wise Refusal, across Pearl’s Ladder. We introduce Recursive Causal Audit (RCA), a process-integrity evaluator whose Judge has no access to gold labels: it checks whether a model’s answer is entailed by itsown derivation, internally consistent, and not dominated by user hints under pressure. RCA unifies persona and pressure: prompt tone is the intervention that regulates pressure-induced drift. For fine diagnostic resolution we use CAUSALT3, with explicit trap families and standardized pressure protocols. CAUSALT3 reveals a Skepticism Trap (Claude Haiku rejects 60% of valid L1 links) and a Scaling Paradox (GPT-5.2 underperforms GPT-4-Turbo by 55 points on L3, driven by paralysis rather than hallucination). Under RCA, operating points shift toward the high-Utility, high-Safety quadrant without retraining, consistent with much of the observed failure arising from how answers are rendered under pressure rather than from missing causal knowledge.

Ruiyang Huang, Chenxi Wang, Tinghe Zhang, Fengrui Liu, Jiayan Sun, Haocheng Wang, Yifan Wu

While LLM-based Multi-Agent Systems (MAS) demonstrate remarkable problem-solving capabilities, their interconnectivity acts as a conduit for the rapid spread of malicious injections. Addressing the limitations of static defenses, we present TopoSHIELD, a framework that reshapes the flow of malice via risk-aware topological evolution. Our approach utilizes a spatio-temporal graph neural network to monitor interaction dynamics, calculating node risk entropy (NRE) and edge attack conductivity (EAC) to pinpoint vulnerabilities. Guided by these metrics, TopoSHIELD executes precise structural interventions, pruning high-risk edges and isolating compromised communities to block attack diffusion. Empirically, TopoSHIELD reduces toxicity by 58% on GPT-4o while preserving high utility (>90% success rate), outperforming existing baselines in both suppression efficiency and scalability.

Jiale Chang, Ying Li, Siliang Tang, Yueting Zhuang

Multimodal large language models excel at vision-language tasks but remain prone to hallucinations that undermine their reliability. Existing approaches predominantly treat hallucinations as classification errors, overlooking the heterogeneous behaviors of attention heads and their dynamic influences during inference. We revisit MLLM reasoning from an energy perspective and identify that hallucinations stem from imbalances between visual potential and language prior potential: when visual information is ambiguous or language priors dominate, attention heads tend to be driven by linguistic statistical patterns, generating content inconsistent with visual evidence. We propose AHEAD, a framework that quantifies the energetic properties of each attention head during object generation through two potential networks—the Visual Grounding Potential Network and the Language Prior Potential Network—and dynamically adjusts their contributions at inference time. Specifically, we amplify attention heads with strong visual grounding capacity while suppressing those overly reliant on language priors. Experiments across multiple benchmarks demonstrate that AHEAD significantly reduces hallucination rates without fine-tuning the base MLLM while maintaining generation quality.

Zhichen Liu, Kaitong Qin, Linhan He, Yang Xu

As Large Language Models asymptotically approach human-level fluency in natural language generation, solely relying on surface-level semantic artifacts for detecting LLM-generated texts has become increasingly precarious. Existing detectors often falter when facing three critical challenges: adversarial perturbations, cross-domain shifts, and the rapid temporal evolution of the foundation model. To address these issues, we propose , a novel framework that reformulates text detection as a signal processing task within the time-frequency domain. Unlike previous methods that analyze static token probability distributions, models the generated output as a probability signal, upon which a differentiable Continuous Wavelet Transform is applied to convert them into learnable spectral representations. This process reveals the intrinsic “spectral fingerprints” in machine-generated texts–patterns that remain invisible in time domain. Comprehensive evaluations on three well-curated datasets (RAID, EvoBench, and Domain-Shift) show that our method achieves a new state-of-the-art. It not only achieves superior accuracy but also exhibits remarkable robustness against sophisticated attacks, generalization across out-of-distribution topics and unseen evolving LLMs. Our results validate the efficacy of spectral analysis as a promising paradigm for LLM-generated texts detection.