cs.CL（2026-04-14）

📊 共 31 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) (26 🔗4) 支柱二：RL算法与架构 (RL & Architecture) (4) 支柱一：机器人控制 (Robot Control) (1)

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

#题目一句话要点标签🔗⭐
1 Learning Chain Of Thoughts Prompts for Predicting Entities, Relations, and even Literals on Knowledge Graphs 提出RALP以解决知识图谱中链式推理问题 large language model chain-of-thought ✅
2 Mining Large Language Models for Low-Resource Language Data: Comparing Elicitation Strategies for Hausa and Fongbe 利用提示工程从大语言模型中挖掘低资源语言数据 large language model
3 PolicyLLM: Towards Excellent Comprehension of Public Policy for Large Language Models 提出PolicyBench和PolicyMoE，提升大语言模型在公共政策理解与应用能力 large language model
4 Generating Effective CoT Traces for Mitigating Causal Hallucination 提出一种CoT轨迹生成方法，缓解小模型在事件因果识别中的因果幻觉问题 large language model chain-of-thought
5 Token-Level Policy Optimization: Linking Group-Level Rewards to Token-Level Aggregation via Sequence-Level Likelihood 提出TEPO，通过序列似然和KL散度约束优化LLM的token级策略，提升数学推理能力。 large language model chain-of-thought
6 TimeMark: A Trustworthy Time Watermarking Framework for Exact Generation-Time Recovery from AIGC 提出TimeMark：一种可信的时间水印框架，用于从AIGC中精确恢复生成时间 large language model TAMP
7 SpecBound: Adaptive Bounded Self-Speculation with Layer-wise Confidence Calibration SpecBound：通过分层置信度校准的自适应有界自推测解码，加速LLM自回归推理。 large language model
8 One Token Away from Collapse: The Fragility of Instruction-Tuned Helpfulness 揭示指令微调大语言模型对微小词汇约束的脆弱性，并分析其内在原因。 large language model
9 MetFuse: Figurative Fusion between Metonymy and Metaphor 提出MetFuse数据集，研究隐喻和转喻的融合，并提升隐喻和转喻识别效果。 large language model ✅
10 The role of System 1 and System 2 semantic memory structure in human and LLM biases 通过语义记忆网络结构分析人类和LLM偏见差异，揭示认知机制 large language model
11 NaviRAG: Towards Active Knowledge Navigation for Retrieval-Augmented Generation NaviRAG：面向检索增强生成的主动知识导航框架 large language model
12 Adaptive Test-Time Scaling for Zero-Shot Respiratory Audio Classification 提出TRIAGE框架，自适应调整呼吸音频零样本分类的计算量 large language model
13 Transforming External Knowledge into Triplets for Enhanced Retrieval in RAG of LLMs 提出Tri-RAG，通过结构化三元组知识表示增强LLM的RAG检索效率。 large language model
14 Latent-Condensed Transformer for Efficient Long Context Modeling 提出Latent-Condensed Transformer，高效处理长文本建模中的KV缓存和计算复杂度问题 large language model
15 Agentic Insight Generation in VSM Simulations 提出解耦式Agent架构，提升VSM仿真中洞察生成的准确性和鲁棒性 large language model
16 KoCo: Conditioning Language Model Pre-training on Knowledge Coordinates 提出KoCo方法，通过知识坐标引导语言模型预训练，提升下游任务性能并加速收敛。 large language model
17 From Myopic Selection to Long-Horizon Awareness: Sequential LLM Routing for Multi-Turn Dialogue DialRouter：面向多轮对话的序列LLM路由方法，提升长时交互性能 large language model
18 Masked by Consensus: Disentangling Privileged Knowledge in LLM Correctness 通过共识掩盖：解耦LLM正确性中的特权知识 large language model
19 Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors 通过零空间约束将激活引导编译到权重中，实现隐蔽后门攻击 large language model
20 CompliBench: Benchmarking LLM Judges for Compliance Violation Detection in Dialogue Systems CompliBench：用于评估LLM在对话系统中违规检测能力的新基准 large language model
21 ContextLens: Modeling Imperfect Privacy and Safety Context for Legal Compliance ContextLens：建模不完善的隐私和安全上下文以实现法律合规 large language model
22 Towards Robust Real-World Spreadsheet Understanding with Multi-Agent Multi-Format Reasoning 提出SpreadsheetAgent，通过多模态多Agent推理实现鲁棒的真实世界电子表格理解。 large language model ✅
23 CodeSpecBench: Benchmarking LLMs for Executable Behavioral Specification Generation CodeSpecBench：用于评估LLM生成可执行行为规范能力的基准测试 large language model ✅
24 Thought-Retriever: Don't Just Retrieve Raw Data, Retrieve Thoughts for Memory-Augmented Agentic Systems Thought-Retriever：通过检索思维而非原始数据，增强记忆增强型Agentic系统 large language model
25 Beyond Majority Voting: Efficient Best-Of-N with Radial Consensus Score 提出基于径向共识评分的高效Best-of-N方法，提升LLM答案选择的可靠性。 large language model
26 AgenticAI-DialogGen: Topic-Guided Conversation Generation for Fine-Tuning and Evaluating Short- and Long-Term Memories of LLMs AgenticAI-DialogGen：用于微调和评估LLM记忆的Topic引导对话生成框架 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (4 篇)

#题目一句话要点标签🔗⭐
27 Enhance-then-Balance Modality Collaboration for Robust Multimodal Sentiment Analysis 提出EBMC框架，通过增强弱模态和平衡模态协作，提升多模态情感分析的鲁棒性。 distillation multimodal
28 Teaching LLMs Human-Like Editing of Inappropriate Argumentation via Reinforcement Learning 提出基于强化学习的LLM训练方法，实现更符合人类习惯的不当论证编辑 reinforcement learning large language model
29 Meet Dynamic Individual Preferences: Resolving Conflicting Human Value with Paired Fine-Tuning 提出Preference-Paired Fine-Tuning框架，解决LLM对动态个体偏好对齐难题 DPO large language model
30 ReasonXL: Shifting LLM Reasoning Language Without Sacrificing Performance ReasonXL：通过迁移LLM推理语言，在不牺牲性能的前提下实现多语言推理 reinforcement learning large language model

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
31 Decoding by Perturbation: Mitigating MLLM Hallucinations via Dynamic Textual Perturbation 提出DeP框架，通过动态文本扰动缓解多模态大语言模型中的幻觉问题 manipulation large language model multimodal

⬅️ 返回 cs.CL 首页 · 🏠 返回主页