cs.CL（2026-02-11）

📊 共 24 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) (16 🔗1) 支柱二：RL算法与架构 (RL & Architecture) (8 🔗1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (16 篇)

#题目一句话要点标签🔗⭐
1 Conversational Behavior Modeling Foundation Model With Multi-Level Perception 提出基于多层感知的对话行为建模基础模型，用于构建自然全双工交互系统。 foundation model chain-of-thought
2 The CLEF-2026 FinMMEval Lab: Multilingual and Multimodal Evaluation of Financial AI Systems CLEF 2026 FinMMEval：首个金融AI系统多语言多模态评测框架 large language model multimodal
3 SteuerLLM: Local specialized large language model for German tax law analysis 提出SteuerLLM，一个针对德国税法分析的本地专业大语言模型。 large language model
4 Can Large Language Models Make Everyone Happy? 提出MisAlign-Profile基准，用于评估大语言模型在安全、价值和文化维度上的对齐权衡。 large language model
5 Canvas-of-Thought: Grounding Reasoning via Mutable Structured States 提出Canvas-CoT，通过可变结构化状态提升多模态大语言模型的推理能力 large language model multimodal chain-of-thought
6 Data Repetition Beats Data Scaling in Long-CoT Supervised Fine-Tuning 长CoT监督微调中，数据重复优于数据扩增 large language model chain-of-thought
7 Beyond Confidence: The Rhythms of Reasoning in Generative Models 提出Token Constraint Bound以解决LLM预测稳定性问题 large language model
8 When to Memorize and When to Stop: Gated Recurrent Memory for Long-Context Reasoning 提出GRU-Mem，通过门控循环记忆网络解决长文本推理中效率和稳定性问题 large language model
9 C-MOP: Integrating Momentum and Boundary-Aware Clustering for Enhanced Prompt Evolution C-MOP：融合动量与边界感知聚类，提升Prompt进化效果 large language model ✅
10 Training-Induced Bias Toward LLM-Generated Content in Dense Retrieval 揭示稠密检索中训练诱导的LLM生成内容偏见，并驳斥了困惑度的解释力 large language model
11 UMEM: Unified Memory Extraction and Management Framework for Generalizable Memory UMEM：统一记忆提取与管理框架，提升LLM Agent记忆泛化能力 large language model
12 ISD-Agent-Bench: A Comprehensive Benchmark for Evaluating LLM-based Instructional Design Agents 提出ISD-Agent-Bench以解决LLM代理评估标准化问题 large language model
13 On the Robustness of Knowledge Editing for Detoxification 提出面向鲁棒性的知识编辑解毒框架，评估大语言模型有害行为抑制的可靠性 large language model
14 TestExplora: Benchmarking LLMs for Proactive Bug Discovery via Repository-Level Test Generation TestExplora：通过仓库级测试生成，评估LLM在主动缺陷发现中的能力。 large language model
15 LATA: A Tool for LLM-Assisted Translation Annotation LATA：一种用于LLM辅助翻译标注的工具，提升跨语言对齐精度。 large language model
16 The Landscape of Prompt Injection Threats in LLM Agents: From Taxonomy to Analysis AgentPI：针对LLM Agent中Prompt注入威胁的全面分析与基准测试 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (8 篇)

#题目一句话要点标签🔗⭐
17 How Do Decoder-Only LLMs Perceive Users? Rethinking Attention Masking for User Representation Learning 提出梯度引导软掩码，提升Decoder-only LLM用户表征学习效果 representation learning contrastive learning large language model ✅
18 Safety Recovery in Reasoning Models Is Only a Few Early Steering Steps Away SafeThink：通过早期引导步骤实现推理模型中的安全性恢复 reinforcement learning multimodal chain-of-thought
19 DataChef: Cooking Up Optimal Data Recipes for LLM Adaptation via Reinforcement Learning DataChef：通过强化学习自动生成LLM适配的最佳数据配方 reinforcement learning large language model
20 Reinforced Curriculum Pre-Alignment for Domain-Adaptive VLMs 提出RCPA：强化课程预对齐方法，提升领域自适应视觉-语言模型性能 reinforcement learning large language model multimodal
21 Neuro-Symbolic Synergy for Interactive World Modeling 提出神经符号协同框架NeSyS，提升交互式世界建模的表达性和鲁棒性 world model large language model
22 Deep Learning-based Method for Expressing Knowledge Boundary of Black-Box LLM 提出LSCL，一种基于深度学习的黑盒LLM知识边界表达方法 distillation large language model
23 Online Causal Kalman Filtering for Stable and Effective Policy Optimization 提出在线因果卡尔曼滤波策略优化算法，解决LLM强化学习中不稳定的重要性采样问题 reinforcement learning large language model
24 Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters Step 3.5 Flash：以11B活跃参数实现前沿水平的智能体能力，兼顾推理与效率 reinforcement learning IMoS

⬅️ 返回 cs.CL 首页 · 🏠 返回主页