cs.CL（2026-03-19）

📊 共 28 篇论文 | 🔗 4 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (22 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (6 🔗2)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (22 篇)

#题目一句话要点标签🔗⭐
1 Entropy trajectory shape predicts LLM reasoning reliability: A diagnostic study of uncertainty dynamics in chain-of-thought 基于熵轨迹形状预测LLM推理可靠性，诊断思维链中的不确定性动态 chain-of-thought
2 UGID: Unified Graph Isomorphism for Debiasing Large Language Models 提出UGID框架，通过统一图同构性来消除大语言模型中的偏见。 large language model
3 What Really Controls Temporal Reasoning in Large Language Models: Tokenisation or Representation of Time? 提出MultiTempBench多语言时间推理基准，揭示LLM时间推理能力受分词质量和时间表示影响。 large language model ✅
4 Evaluating Counterfactual Strategic Reasoning in Large Language Models 评估大语言模型在反事实情境下的策略推理能力 large language model
5 Implicit Grading Bias in Large Language Models: How Writing Style Affects Automated Assessment Across Math, Programming, and Essay Tasks 揭示大语言模型评分中的隐性偏见：写作风格如何影响数学、编程和论文任务的自动评估 large language model
6 GAIN: A Benchmark for Goal-Aligned Decision-Making of Large Language Models under Imperfect Norms GAIN：用于评估大语言模型在不完善规范下目标对齐决策的基准测试。 large language model
7 Multimodal Task Interference: A Benchmark and Analysis of History-Target Mismatch in Multimodal LLMs 提出多模态任务干扰基准，分析多模态LLM中历史-目标不匹配问题 multimodal
8 TARo: Token-level Adaptive Routing for LLM Test-time Alignment 提出TARo：一种Token级自适应路由方法，用于LLM测试时对齐，提升推理能力。 large language model instruction following
9 EntropyCache: Decoded Token Entropy Guided KV Caching for Diffusion Language Models EntropyCache：利用解码Token熵引导扩散语言模型的KV缓存，实现高效推理。 large language model chain-of-thought ✅
10 Cross-Modal Rationale Transfer for Explainable Humanitarian Classification on Social Media 提出跨模态推理迁移方法，用于社交媒体人道主义分类的可解释性研究。 multimodal
11 Hypothesis-Conditioned Query Rewriting for Decision-Useful Retrieval 提出假设条件查询重写（HCQR）框架，提升RAG在决策型检索任务中的性能。 large language model
12 When Names Change Verdicts: Intervention Consistency Reveals Systematic Bias in LLM Decision-Making 提出ICE-Guard框架以检测LLM决策中的系统性偏见 large language model
13 UT-ACA: Uncertainty-Triggered Adaptive Context Allocation for Long-Context Inference 提出UT-ACA框架，通过不确定性触发的自适应上下文分配解决长文本推理中的挑战。 large language model
14 Parallelograms Strike Back: LLMs Generate Better Analogies than People 大型语言模型在类比生成任务中优于人类，更符合平行四边形模型 large language model
15 Evaluating LLM-Generated Lessons from the Language Learning Students' Perspective: A Short Case Study on Duolingo 评估LLM生成的语言学习课程：Duolingo案例研究，强调个性化与专业领域适应性。 large language model
16 WASD: Locating Critical Neurons as Sufficient Conditions for Explaining and Controlling LLM Behavior WASD：通过定位关键神经元作为充分条件来解释和控制LLM行为 large language model
17 How Auditory Knowledge in LLM Backbones Shapes Audio Language Models: A Holistic Evaluation 探究LLM先验听觉知识对音频语言模型的影响：一个全面评估 large language model
18 Automatic detection of Gen-AI texts: A comparative framework of neural models 提出并比较多种神经模型，用于自动检测生成式AI文本，解决学术、编辑和社会领域难题。 large language model
19 Cross-Lingual LLM-Judge Transfer via Evaluation Decomposition 提出基于评估分解的跨语言LLM评判迁移方法，解决多语言评估难题。 large language model
20 Inducing Sustained Creativity and Diversity in Large Language Models 提出一种解码方案，用于激发大语言模型持续的创造性和多样性，解决探索性搜索中结果同质化问题。 large language model
21 Scalable Prompt Routing via Fine-Grained Latent Task Discovery 提出基于细粒度潜在任务发现的可扩展Prompt路由方法 large language model
22 Is Evaluation Awareness Just Format Sensitivity? Limitations of Probe-Based Evidence under Controlled Prompt Structure 研究表明，基于探针的评估意识证据可能仅反映提示结构，而非模型真正的评估能力。 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (6 篇)

#题目一句话要点标签🔗⭐
23 VEPO: Variable Entropy Policy Optimization for Low-Resource Language Foundation Models 提出VEPO，通过强化学习优化低资源语言大模型的翻译质量与token化效率。 reinforcement learning large language model foundation model
24 MoRI: Learning Motivation-Grounded Reasoning for Scientific Ideation in Large Language Models MoRI：提出一种基于动机推理的科学构思大语言模型框架，提升技术深度和科学依据。 reinforcement learning large language model ✅
25 Adaptive Decoding via Test-Time Policy Learning for Self-Improving Generation 提出基于强化学习的自适应解码策略，提升LLM在不同任务上的生成质量 reinforcement learning policy learning large language model
26 Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation Nemotron-Cascade 2：利用级联强化学习和多域策略蒸馏训练30B MoE大模型 distillation IMoS
27 PowerFlow: Unlocking the Dual Nature of LLMs via Principled Distribution Matching PowerFlow：通过原则性分布匹配解锁大语言模型的双重能力 reinforcement learning large language model
28 Cooperation and Exploitation in LLM Policy Synthesis for Sequential Social Dilemmas 提出基于LLM策略合成框架，解决序贯社会困境中的合作与利用问题 reinforcement learning large language model ✅

⬅️ 返回 cs.CL 首页 · 🏠 返回主页