cs.CL（2026-05-12）

📊 共 45 篇论文 | 🔗 2 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (26 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (14) 支柱一：机器人控制 (Robot Control) (4) 支柱七：动作重定向 (Motion Retargeting) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (26 篇)

#题目一句话要点标签🔗⭐
1 Checkup2Action: A Multimodal Clinical Check-up Report Dataset for Patient-Oriented Action Card Generation 提出Checkup2Action数据集，用于评估多模态临床报告生成患者导向行动卡片的能力 large language model multimodal
2 Human-Grounded Multimodal Benchmark with 900K-Scale Aggregated Student Response Distributions from Japan's National Assessment of Academic Ability 提出Gakucho基准，用于评估多模态大语言模型在真实日本K-12学业评估中的表现。 large language model multimodal ✅
3 Scalable Token-Level Hallucination Detection in Large Language Models 提出TokenHD，实现大规模语言模型中token级别幻觉检测，无需步骤分割。 large language model
4 From Token to Token Pair: Efficient Prompt Compression for Large Language Models in Clinical Prediction 提出MedTPE方法，用于临床预测中LLM的EHR序列高效压缩。 large language model
5 Question Difficulty Estimation for Large Language Models via Answer Plausibility Scoring 提出Q-DAPS方法，通过答案合理性评分估计大语言模型问题难度 large language model
6 Pretraining Exposure Explains Popularity Judgments in Large Language Models 通过预训练数据暴露度解释大语言模型中的流行度判断 large language model
7 Correcting Selection Bias in Sparse User Feedback for Large Language Model Quality Estimation: A Multi-Agent Hierarchical Bayesian Approach 提出多代理层次贝叶斯方法以纠正稀疏用户反馈中的选择偏差 large language model
8 Large Language Models for Causal Relations Extraction in Social Media: A Validation Framework for Disaster Intelligence 提出基于专家知识的评估框架，验证大型语言模型在灾害情报中提取因果关系的能力。 large language model
9 Reconstruction of Personally Identifiable Information from Supervised Finetuned Models 提出COVA算法，用于从监督微调模型中重建个人身份信息（PII）。 large language model instruction following
10 Towards Visually-Guided Movie Subtitle Translation for Indic Languages 提出视觉引导的电影字幕翻译方法，提升印地语等低资源语言的翻译质量 multimodal visual grounding
11 Task-Adaptive Embedding Refinement via Test-time LLM Guidance 提出基于测试时LLM指导的任务自适应嵌入精炼方法，提升零样本检索和分类性能。 instruction following ✅
12 ORBIT: Preserving Foundational Language Capabilities in GenRetrieval via Origin-Regulated Merging ORBIT：通过源头调控合并，在生成式检索中保留基础语言能力 large language model
13 Stories in Space: In-Context Learning Trajectories in Conceptual Belief Space 提出概念信念空间，以几何视角理解LLM上下文学习中的信念动态变化 large language model
14 Training-Inference Consistent Segmented Execution for Long-Context LLMs 提出训练-推理一致的分段执行框架，提升长文本LLM的效率和可扩展性 large language model
15 Three Regimes of Context-Parametric Conflict: A Predictive Framework and Empirical Validation 提出上下文参数冲突三阶段框架，预测并验证大语言模型知识更新行为 large language model
16 Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting 提出基于协方差感知的GRPO方法，通过高斯核优势重加权稳定大语言模型推理能力。 large language model
17 The Algorithmic Caricature: Auditing LLM-Generated Political Discourse Across Crisis Events 提出算法漫画方法，通过对比真实与生成政治言论，评估LLM在危机事件中生成内容的人口真实性。 large language model
18 Mitigating Context-Memory Conflicts in LLMs through Dynamic Cognitive Reconciliation Decoding 提出DCRD动态认知协调解码，缓解大语言模型中的上下文-记忆冲突 large language model
19 Geometric Factual Recall in Transformers 揭示Transformer中几何事实记忆机制，突破参数线性增长瓶颈 chain-of-thought
20 Latent Causal Void: Explicit Missing-Context Reconstruction for Misinformation Detection 提出隐性因果空洞(LCV)，通过显式重建缺失上下文来提升信息检测性能。 large language model
21 Metaphor Is Not All Attention Needs 诗意越狱并非仅依赖注意力机制，而是源于风格不规则性对LLM处理方式的改变 large language model
22 Do Language Models Encode Knowledge of Linguistic Constraint Violations? 提出稀疏自编码器以检测语言模型中的语法约束违规特征 large language model
23 Safety-Oriented Evaluation of Language Understanding Systems for Air Traffic Control 提出安全导向评估框架以解决空中交通管制语言理解系统的可靠性问题 large language model
24 Robust LLM Unlearning Against Relearning Attacks: The Minor Components in Representations Matter 提出MCU方法，通过优化表征中的次要成分，增强LLM抗重学习攻击的卸载能力 large language model
25 StoicLLM: Preference Optimization for Philosophical Alignment in Small Language Models StoicLLM：小模型哲学对齐的偏好优化方法 large language model
26 Freeze Deep, Train Shallow: Interpretable Layer Allocation for Continued Pre-Training 提出LayerTracer以解决大语言模型继续预训练中的层分配问题 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (14 篇)

#题目一句话要点标签🔗⭐
27 OmniThoughtVis: A Scalable Distillation Pipeline for Deployable Multimodal Reasoning Models 提出OmniThoughtVis，用于可部署多模态推理模型的可扩展蒸馏流水线。 distillation large language model multimodal
28 Combining On-Policy Optimization and Distillation for Long-Context Reasoning in Large Language Models 提出dGRPO方法，结合On-Policy优化与知识蒸馏，提升大语言模型长文本推理能力。 reinforcement learning distillation large language model
29 A Study on Hidden Layer Distillation for Large Language Model Pre-Training 研究大型语言模型预训练中的隐层蒸馏方法，探索其性能瓶颈与潜在价值 distillation large language model
30 Qwen-Scope: Turning Sparse Features into Development Tools for Large Language Models Qwen-Scope：将稀疏特征转化为大语言模型开发工具 reinforcement learning large language model
31 Enhancing Multilingual Counterfactual Generation through Alignment-as-Preference Optimization 提出Macro框架，通过偏好对齐优化提升多语言反事实生成效果 DPO direct preference optimization large language model
32 TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection TextSeal：一种用于溯源和蒸馏保护的局部化LLM水印方案 distillation large language model
33 SkillGraph: Skill-Augmented Reinforcement Learning for Agents via Evolving Skill Graphs 提出SkillGraph，通过演化技能图增强智能体强化学习能力 reinforcement learning large language model
34 On Predicting the Post-training Potential of Pre-trained LLMs 提出RuDE框架，通过判别而非生成方式预测预训练LLM的后训练潜力，提升模型选择效率。 reinforcement learning large language model foundation model
35 Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation 揭示On-Policy蒸馏效率之谜：提出EffOPD加速训练并保持性能。 distillation large language model
36 TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching TokenRatio：通过比例匹配实现原则性的Token级别偏好优化 DPO direct preference optimization instruction following
37 SAGE: Scalable Automated Robustness Augmentation for LLM Knowledge Evaluation SAGE：一种可扩展的自动化鲁棒性增强框架，用于评估LLM知识 reinforcement learning large language model
38 Learning Agentic Policy from Action Guidance 提出ActGuide-RL，利用行动指导提升LLM智能体策略探索能力，降低对SFT数据的依赖。 reinforcement learning large language model
39 StepCodeReasoner: Aligning Code Reasoning with Stepwise Execution Traces via Reinforcement Learning StepCodeReasoner：通过强化学习对齐代码推理与逐步执行轨迹 reinforcement learning
40 Agent-BRACE: Decoupling Beliefs from Actions in Long-Horizon Tasks via Verbalized State Uncertainty Agent-BRACE：通过语言化的状态不确定性解耦信念与动作，解决长时程任务中的挑战。 reinforcement learning large language model

🔬 支柱一：机器人控制 (Robot Control) (4 篇)

#题目一句话要点标签🔗⭐
41 When Emotion Becomes Trigger: Emotion-style dynamic Backdoor Attack Parasitising Large Language Models 提出Paraesthesia，一种基于情感风格的动态后门攻击方法，寄生于大型语言模型。 manipulation large language model instruction following
42 YFPO: A Preliminary Study of Yoked Feature Preference Optimization with Neuron-Guided Rewards for Mathematical Reasoning YFPO：一种神经元引导奖励的Yoked特征偏好优化方法，用于提升数学推理能力。 manipulation preference learning large language model
43 PreScam: A Benchmark for Predicting Scam Progression from Early Conversations PreScam：一个基于早期对话预测诈骗进展的基准数据集 manipulation
44 Is Child-Directed Language Optimized for Word Learning? A Computational Study of Verb Meaning Acquisition 通过计算模型研究儿童语言对动词学习的优化效果 manipulation

🔬 支柱七：动作重定向 (Motion Retargeting) (1 篇)

#题目一句话要点标签🔗⭐
45 Predicting Disagreement with Human Raters in LLM-as-a-Judge Difficulty Assessment without Using Generation-Time Probability Signals 提出一种无需生成概率信号的LLM难度评估不一致性预测方法 geometric consistency large language model

⬅️ 返回 cs.CL 首页 · 🏠 返回主页