cs.CL（2026-01-30）

📊 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) (27 🔗4) 支柱二：RL算法与架构 (RL & Architecture) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (27 篇)

#题目一句话要点标签🔗⭐
1 ReGuLaR: Variational Latent Reasoning Guided by Rendered Chain-of-Thought 提出ReGuLaR，利用渲染的思维链指导变分隐空间推理，提升计算效率和推理效果。 large language model chain-of-thought ✅
2 Towards Resiliency in Large Language Model Serving with KevlarFlow KevlarFlow：面向大规模语言模型服务，提升硬件故障下的系统韧性 large language model
3 Character as a Latent Variable in Large Language Models: A Mechanistic Account of Emergent Misalignment and Conditional Safety Failures 揭示大语言模型中角色扮演诱导的潜在风险，强调行为倾向而非孤立错误 large language model
4 FNF: Functional Network Fingerprint for Large Language Models 提出功能网络指纹FNF，用于检测大型语言模型的知识产权侵权。 large language model ✅
5 DIFFA-2: A Practical Diffusion Large Language Model for General Audio Understanding DIFFA-2：一种实用的扩散大语言模型，用于通用音频理解 large language model ✅
6 Large Language Model Agents Are Not Always Faithful Self-Evolvers 揭示LLM Agent自我进化中经验依赖的非忠实性问题 large language model
7 Residual Context Diffusion Language Models 提出残差上下文扩散（RCD）模块，提升扩散语言模型（dLLM）的推理精度和效率。 large language model instruction following
8 Towards the Holographic Characteristic of LLMs for Efficient Short-text Generation 针对短文本生成，论文揭示LLM全息特性并提出高效插件HOLO large language model chain-of-thought
9 InstructDiff: Domain-Adaptive Data Selection via Differential Entropy for Efficient LLM Fine-Tuning InstructDiff：通过差分熵进行领域自适应数据选择，高效微调大语言模型 large language model instruction following
10 MM-THEBench: Do Reasoning MLLMs Think Reasonably? 提出MM-THEBench，评估推理多模态大模型中间CoT中的幻觉问题 large language model multimodal
11 LLMs Explain't: A Post-Mortem on Semantic Interpretability in Transformer Models 探讨LLMs的语义可解释性问题及其局限性 large language model
12 DART-ing Through the Drift: Dynamic Tracing of Knowledge Neurons for Adaptive Inference-Time Pruning DART：通过动态追踪知识神经元实现自适应推理时剪枝 large language model ✅
13 Rethinking LLM-as-a-Judge: Representation-as-a-Judge with Small Language Models via Semantic Capacity Asymmetry 提出基于表征的INSPECTOR框架，利用小模型实现高效、可靠、可解释的LLM评判。 large language model
14 SpanNorm: Reconciling Training Stability and Performance in Deep Transformers SpanNorm：平衡深度Transformer训练稳定性和性能的新型归一化方法 large language model
15 Are LLM Evaluators Really Narcissists? Sanity Checking Self-Preference Evaluations 提出评估器质量基线，消除LLM自偏好评估中的噪声，提升评估可靠性 large language model
16 Bias Beyond Borders: Political Ideology Evaluation and Steering in Multilingual LLMs 提出跨语言对齐引导(CLAS)框架，用于缓解多语言LLM中的政治偏见 large language model
17 UPA: Unsupervised Prompt Agent via Tree-Based Search and Selection 提出UPA：一种基于树搜索与选择的无监督Prompt Agent，用于自动Prompt优化。 large language model
18 Deep Search with Hierarchical Meta-Cognitive Monitoring Inspired by Cognitive Neuroscience 提出DS-MCM框架，通过分层元认知监控提升深度搜索Agent的性能与鲁棒性 large language model
19 MiTa: A Hierarchical Multi-Agent Collaboration Framework with Memory-integrated and Task Allocation MiTa：一种分层多智能体协作框架，集成记忆与任务分配，提升复杂任务效率。 large language model
20 A Unified View of Attention and Residual Sinks: Outlier-Driven Rescaling is Essential for Transformer Training 揭示Transformer训练中Outlier驱动的重缩放机制，提升模型性能与量化鲁棒性 large language model
21 Leveraging LLMs For Turkish Skill Extraction 利用大型语言模型进行土耳其语技能提取，填补低资源语言技能提取空白。 large language model
22 When Meanings Meet: Investigating the Emergence and Quality of Shared Concept Spaces during Multilingual Language Model Training 研究多语言模型训练中共享概念空间的涌现与质量，揭示跨语言对齐的训练动态。 large language model
23 Sparse or Dense? A Mechanistic Estimation of Computation Density in Transformer-based LLMs 提出一种基于机制可解释性的方法，用于量化Transformer LLM中的计算密度。 large language model
24 AR-BENCH: Benchmarking Legal Reasoning with Judgment Error Detection, Classification and Correction AR-BENCH：提出法律判决错误检测、分类与纠正的评测基准 large language model
25 Models Know Models Best: Evaluation via Model-Preferred Formats 提出基于模型偏好格式的动态评估方法，提升大语言模型zero-shot能力。 large language model
26 Layer-wise Swapping for Generalizable Multilingual Safety 提出层级交换方法，提升低资源语言大模型安全性 large language model
27 $ρ$-$\texttt{EOS}$: Training-free Bidirectional Variable-Length Control for Masked Diffusion LLMs 提出$ρ$-$ exttt{EOS}$，实现Masked扩散LLM的免训练双向变长控制。 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (1 篇)

#题目一句话要点标签🔗⭐
28 Autonomous Chain-of-Thought Distillation for Graph-Based Fraud Detection 提出FraudCoT，通过自主CoT蒸馏和LLM-GNN协同训练提升图文属性图上的欺诈检测。 distillation chain-of-thought

⬅️ 返回 cs.CL 首页 · 🏠 返回主页