cs.CL（2026-04-27）

📊 共 29 篇论文 | 🔗 10 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (21 🔗7) 支柱二：RL算法与架构 (RL & Architecture) (7 🔗3) 支柱一：机器人控制 (Robot Control) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (21 篇)

#题目一句话要点标签🔗⭐
1 K-MetBench: A Multi-Dimensional Benchmark for Fine-Grained Evaluation of Expert Reasoning, Locality, and Multimodality in Meteorology K-MetBench：用于气象领域专家推理、局部性和多模态的细粒度评估基准 large language model multimodal ✅
2 AdapTime: Enabling Adaptive Temporal Reasoning in Large Language Models AdapTime：提出自适应时序推理方法，提升大语言模型处理时间信息的能力 large language model
3 The Chameleon's Limit: Investigating Persona Collapse and Homogenization in Large Language Models 揭示大语言模型人格崩塌现象，提出量化框架评估人口多样性 large language model
4 Generating Place-Based Compromises Between Two Points of View 提出基于同理心中立性的LLM提示方法，生成更易接受的观点折衷方案 large language model foundation model chain-of-thought
5 Zero-shot Large Language Models for Automatic Readability Assessment 提出基于零样本大语言模型的自动可读性评估方法，显著提升评估效果 large language model
6 A Multi-Dimensional Audit of Politically Aligned Large Language Models 提出多维度评估框架，用于审计政治倾向性大型语言模型的有效性、公平性、真实性和说服力。 large language model
7 Culture-Aware Machine Translation in Large Language Models: Benchmarking and Investigation 提出CanMT以解决大语言模型在文化翻译中的不足 large language model
8 Aligned Multi-View Scripts for Universal Chart-to-Code Generation 提出CharLuMA，利用多语言对齐脚本提升图表到代码生成效果 multimodal language conditioned ✅
9 MEG-RAG: Quantifying Multi-modal Evidence Grounding for Evidence Selection in RAG 提出MEG-RAG以解决多模态证据选择问题 large language model multimodal
10 OS-SPEAR: A Toolkit for the Safety, Performance,Efficiency, and Robustness Analysis of OS Agents OS-SPEAR：用于操作系统代理安全性、性能、效率和鲁棒性分析的工具包 large language model multimodal ✅
11 A Survey on Split Learning for LLM Fine-Tuning: Models, Systems, and Privacy Optimizations 针对LLM微调的分裂学习综述：模型、系统与隐私优化 large language model
12 Kwai Summary Attention Technical Report 提出Kwai Summary Attention (KSA)，通过可学习的摘要token压缩长文本上下文，降低序列建模成本。 large language model
13 Stabilizing Efficient Reasoning with Step-Level Advantage Selection 提出步级优势选择(SAS)以稳定高效推理，提升LLM在短上下文下的性能。 large language model
14 DepthKV: Layer-Dependent KV Cache Pruning for Long-Context LLM Inference DepthKV：针对长文本LLM推理的分层KV缓存剪枝方法 large language model
15 Skill Retrieval Augmentation for Agentic AI 提出技能检索增强（SRA）范式，解决Agentic AI中技能扩展瓶颈问题 large language model
16 The Pragmatic Persona: Discovering LLM Persona through Bridging Inference 提出基于桥接推理的LLM Persona发现框架，提升语义连贯性和Persona识别稳定性。 large language model ✅
17 Contextual Linear Activation Steering of Language Models 提出上下文线性激活调控(CLAS)，提升大语言模型在少量数据下的行为控制能力。 large language model
18 Can LLMs Act as Historians? Evaluating Historical Research Capabilities of LLMs via the Chinese Imperial Examination 提出ProHist-Bench基准，评估LLM在科举历史研究中的推理能力。 large language model ✅
19 Can You Make It Sound Like You? Post-Editing LLM-Generated Text for Personal Style 研究表明，用户可以通过编辑LLM生成文本，使其更贴近个人写作风格，但仍存在LLM痕迹。 large language model
20 Differentiable Faithfulness Alignment for Cross-Model Circuit Transfer 提出可微忠实度对齐(DFA)框架，用于跨模型传递神经回路信息。 zero-shot transfer ✅
21 PeeriScope: A Multi-Faceted Framework for Evaluating Peer Review Quality PeeriScope：一个多维度评估同行评审质量的综合框架 large language model ✅

🔬 支柱二：RL算法与架构 (RL & Architecture) (7 篇)

#题目一句话要点标签🔗⭐
22 MIPIC: Matryoshka Representation Learning via Self-Distilled Intra-Relational and Progressive Information Chaining MIPIC：通过自蒸馏关系对齐和渐进信息链实现Matryoshka表征学习 representation learning distillation
23 Rewarding the Scientific Process: Process-Level Reward Modeling for Agentic Data Analysis 提出DataPRM，用于提升Agentic数据分析中LLM的推理能力。 reinforcement learning large language model ✅
24 EPM-RL: Reinforcement Learning for On-Premise Product Mapping in E-Commerce 提出EPM-RL，利用强化学习解决电商环境下大规模、隐私敏感的产品匹配问题。 reinforcement learning
25 Structural Pruning of Large Vision Language Models: A Comprehensive Study on Pruning Dynamics, Recovery, and Data Efficiency 研究LVLM结构化剪枝，探索剪枝策略、恢复训练和数据效率的平衡。 distillation multimodal ✅
26 DPEPO: Diverse Parallel Exploration Policy Optimization for LLM-based Agents DPEPO：面向LLM智能体的多样化并行探索策略优化 reinforcement learning large language model ✅
27 Long-Context Aware Upcycling: A New Frontier for Hybrid LLM Scaling HyLo：通过混合架构和长上下文训练，将预训练Transformer LLM升级为长文本模型。 Mamba distillation
28 IRIS: Interleaved Reinforcement with Incremental Staged Curriculum for Cross-Lingual Mathematical Reasoning IRIS：结合交错强化学习与增量课程学习，提升跨语言数学推理能力 reinforcement learning curriculum learning

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
29 Benchmarking Source-Sensitive Reasoning in Turkish: Humans and LLMs under Evidential Trust Manipulation 评估大型语言模型在土耳其语中对证据来源可信度的推理能力 manipulation large language model

⬅️ 返回 cs.CL 首页 · 🏠 返回主页