cs.CL（2026-05-07）

📊 共 43 篇论文 | 🔗 3 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (29 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (11 🔗1) 支柱八：物理动画 (Physics-based Animation) (1) 支柱一：机器人控制 (Robot Control) (1) 支柱四：生成式动作 (Generative Motion) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (29 篇)

#题目一句话要点标签🔗⭐
1 Litespark Inference on Consumer CPUs: Custom SIMD Kernels for Ternary Neural Networks Litespark-Inference：面向消费级CPU的三元神经网络定制SIMD推理加速 large language model
2 The Cost of Context: Mitigating Textual Bias in Multimodal Retrieval-Augmented Generation 提出BAIR框架以解决多模态生成中的文本偏差问题 large language model multimodal
3 TableVista: Benchmarking Multimodal Table Reasoning under Visual and Structural Complexity 提出TableVista基准测试，揭示多模态大模型在复杂视觉与结构化表格推理中的性能瓶颈 foundation model multimodal
4 Decomposing the Basic Abilities of Large Language Models: Mitigating Cross-Task Interference in Multi-Task Instruct-Tuning 提出BADIT框架：通过基本能力分解与正交化LoRA专家缓解多任务指令微调中的跨任务干扰 large language model
5 Reflections and New Directions for Human-Centered Large Language Models 提出以人为中心的大语言模型（HCLLM）框架，实现全生命周期的价值对齐与责任部署 large language model
6 Towards Emotion Consistency Analysis of Large Language Models in Emotional Conversational Contexts 分析大语言模型在情感对话语境下的逻辑一致性与虚假信念易感性 large language model
7 Uncovering Entity Identity Confusion in Multimodal Knowledge Editing 揭示多模态知识编辑中的实体身份混淆问题，并提出基于I-E绑定约束的改进策略 multimodal
8 BioTool: A Comprehensive Tool-Calling Dataset for Enhancing Biomedical Capabilities of Large Language Models 提出BioTool数据集以增强大语言模型在生物医学领域的工具调用能力 large language model ✅
9 Negative Before Positive: Asymmetric Valence Processing in Large Language Models 揭示大语言模型中情感效价的非对称处理机制：基于激活修补与干预的深度分析 large language model
10 IntentGrasp: A Comprehensive Benchmark for Intent Understanding 提出IntentGrasp基准与意图微调（IFT）方法，显著提升大语言模型的意图理解能力 large language model
11 Cognitive Agent Compilation for Explicit Problem Solver Modeling 提出认知智能体编译（CAC）框架，通过显式建模实现教育场景下的可解释与可控问题求解 large language model
12 MELD: Multi-Task Equilibrated Learning Detector for AI-Generated Text 提出MELD多任务平衡学习检测器，通过辅助监督与对抗蒸馏提升AI生成文本检测的鲁棒性 large language model
13 Hallucination as an Anomaly: Dynamic Intervention via Probabilistic Circuits 提出PCNET，通过概率电路动态干预LLM幻觉问题，提升生成真实性。 large language model
14 One Turn Too Late: Response-Aware Defense Against Hidden Malicious Intent in Multi-Turn Dialogue 提出TurnGate防御框架，通过响应感知机制识别多轮对话中的隐蔽恶意意图 large language model ✅
15 SmellBench: Evaluating LLM Agents on Architectural Code Smell Repair 提出SmellBench评估框架，量化评估大模型智能体在架构级代码异味修复中的能力 large language model
16 Can LLMs Take Retrieved Information with a Grain of Salt? 提出基于交互设计的上下文确定性校准策略，显著提升LLM对检索信息置信度的判别与响应能力。 large language model
17 EMO: Pretraining Mixture of Experts for Emergent Modularity 提出EMO预训练框架，通过文档级约束实现混合专家模型（MoE）的涌现式模块化 large language model
18 Cited but Not Verified: Parsing and Evaluating Source Attribution in LLM Deep Research Agents 提出首个LLM深度研究代理引用评估框架，揭示了引用质量与事实准确性之间的严重脱节。 large language model
19 Algospeak, Hiding in the Open: The Trade-off Between Legible Meaning and Detection Avoidance 提出Algospeak评估框架，量化语言规避策略在内容可理解性与检测逃逸间的权衡。 large language model
20 Efficient Pre-Training with Token Superposition 提出Token叠加训练（TST）方法，通过两阶段训练显著提升大模型预训练效率 large language model
21 STALE: Can LLM Agents Know When Their Memories Are No Longer Valid? 提出STALE基准与CUPMem框架，解决LLM智能体在动态环境下的记忆失效与状态更新难题。 large language model
22 SEQUOR: A Multi-Turn Benchmark for Realistic Constraint Following 提出SEQUOR基准测试，揭示大模型在长多轮对话中遵循复杂约束的性能瓶颈 instruction following
23 Quantifying the Statistical Effect of Rubric Modifications on Human-Autorater Agreement 量化评估准则修改对人机评分一致性的统计影响，优化LLM作为裁判的评价效能 instruction following
24 UniPrefill: Universal Long-Context Prefill Acceleration via Block-wise Dynamic Sparsification 提出UniPrefill框架，通过块级动态稀疏化实现通用长上下文预填充加速 large language model
25 Navigating by Old Maps: The Pitfalls of Static Mechanistic Localization in LLM Post-Training 揭示大模型后训练中静态机制定位的局限性，提出电路演化分析框架以应对参数动态更新挑战 large language model
26 From Articles to Premises: Building PrimeFacts, an Extraction Methodology and Resource for Fact-Checking Evidence 提出PrimeFacts数据集与提取框架，通过LLM去语境化重写事实核查证据以提升自动验证性能。 large language model
27 Minimizing Modality Gap from the Input Side: Your Speech LLM Can Be a Prosody-Aware Text LLM 提出TextPro-SLM：通过输入端对齐策略缩小语音大模型模态鸿沟 large language model
28 Evaluation Awareness in Language Models Has Limited Effect on Behaviour 实证研究表明：大型推理模型中的“评估意识”对模型行为的影响极其有限 chain-of-thought
29 A Few Good Clauses: Comparing LLMs vs Domain-Trained Small Language Models on Structured Contract Extraction 提出领域专用小型语言模型Olava Extract，以低成本实现超越前沿大模型的合同结构化抽取能力。 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (11 篇)

#题目一句话要点标签🔗⭐
30 UniSD: Towards a Unified Self-Distillation Framework for Large Language Models 提出UniSD框架以解决自蒸馏在大语言模型中的挑战 contrastive learning distillation feature matching
31 StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction StraTA：通过策略轨迹抽象激励Agentic强化学习，提升长程决策能力 reinforcement learning large language model
32 MANTRA: Synthesizing SMT-Validated Compliance Benchmarks for Tool-Using LLM Agents MANTRA：合成SMT验证的合规性基准，用于工具型LLM Agent world model world models large language model
33 Continuous Latent Diffusion Language Model 提出Cola DLM：一种连续潜在扩散语言模型，用于高效灵活的文本生成。 representation learning large language model
34 When2Speak: A Dataset for Temporal Participation and Turn-Taking in Multi-Party Conversations for Large Language Models 提出When2Speak数据集与四阶段生成流水线，解决大语言模型在多方对话中的介入时机决策问题。 reinforcement learning reward shaping large language model
35 Milestone-Guided Policy Learning for Long-Horizon Language Agents 提出里程碑引导的策略学习框架BEACON，解决长程语言智能体训练中的信用分配与样本效率难题。 reinforcement learning policy learning reward shaping ✅
36 Estimating the Black-box LLM Uncertainty with Distribution-Aligned Adversarial Distillation 提出分布对齐对抗蒸馏（DisAAD）框架，实现黑盒大模型的高效不确定性量化 distillation large language model
37 Beyond Negative Rollouts: Positive-Only Policy Optimization with Implicit Negative Gradients 提出POPO框架：通过仅正样本策略优化实现大语言模型推理能力的提升 reinforcement learning PPO large language model
38 Rethinking RL for LLM Reasoning: It's Sparse Policy Selection, Not Capability Learning 提出ReasonMaxxer方法：通过稀疏策略选择替代强化学习以提升大模型推理能力 reinforcement learning large language model
39 A$^2$TGPO: Agentic Turn-Group Policy Optimization with Adaptive Turn-level Clipping 提出A$^2$TGPO算法，通过自适应轮次裁剪优化智能体大模型的强化学习过程奖励分配。 reinforcement learning large language model
40 MemReranker: Reasoning-Aware Reranking for Agent Memory Retrieval 提出MemReranker重排序模型，通过多阶段知识蒸馏增强智能体记忆检索的推理能力 contrastive learning distillation

🔬 支柱八：物理动画 (Physics-based Animation) (1 篇)

#题目一句话要点标签🔗⭐
41 MIST: Multimodal Interactive Speech-based Tool-calling Conversational Assistants for Smart Homes 提出MIST多模态交互式语音工具调用数据集，以解决智能家居场景下复杂时空约束与动态状态追踪难题。 spatiotemporal large language model multimodal

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
42 Lightweight Stylistic Consistency Profiling: Robust Detection of LLM-Generated Textual Content for Multimedia Moderation 提出LiSCP轻量级风格一致性分析方法，实现多媒体内容中LLM生成文本的鲁棒检测 manipulation large language model multimodal

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
43 Bridging Passive and Active: Enhancing Conversation Starter Recommendation via Active Expression Modeling 提出PA-Bridge框架，通过主动表达建模打破对话推荐中的反馈循环与回声室效应 penetration large language model

⬅️ 返回 cs.CL 首页 · 🏠 返回主页