cs.CL（2025-03-17）

📊 共 26 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) (24 🔗4) 支柱二：RL算法与架构 (RL & Architecture) (2)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (24 篇)

#题目一句话要点标签🔗⭐
1 HiDe-LLaVA: Hierarchical Decoupling for Continual Instruction Tuning of Multimodal Large Language Model HiDe-LLaVA：提出分层解耦方法，用于多模态大语言模型持续指令调优 large language model multimodal ✅
2 Levels of Analysis for Large Language Models 借鉴认知科学分析框架，提升大型语言模型的可理解性 large language model
3 LLM-Match: An Open-Sourced Patient Matching Model Based on Large Language Models and Retrieval-Augmented Generation LLM-Match：基于大语言模型和RAG的开源患者匹配模型，优于现有方案。 large language model
4 TablePilot: Recommending Human-Preferred Tabular Data Analysis with Large Language Models TablePilot：利用大语言模型推荐更符合人类偏好的表格数据分析方案 large language model
5 Code-Driven Inductive Synthesis: Enhancing Reasoning Abilities of Large Language Models with Sequences 提出CodeSeq数据集，提升大语言模型在归纳推理中的代码生成能力 large language model
6 A Multi-Stage Framework with Taxonomy-Guided Reasoning for Occupation Classification Using Large Language Models 提出基于分类体系引导推理的多阶段框架，利用大语言模型进行职业分类 large language model
7 HICD: Hallucination-Inducing via Attention Dispersion for Contrastive Decoding to Mitigate Hallucinations in Large Language Models 提出HICD，通过注意力分散诱导幻觉，用于对比解码以缓解大语言模型中的幻觉问题 large language model
8 Can Language Models Follow Multiple Turns of Entangled Instructions? MultiTurnInstruct：系统评估LLM在多轮纠缠指令下的执行能力，揭示模型在记忆、推理和冲突解决间的权衡。 large language model instruction following ✅
9 ThinkPatterns-21k: A Systematic Study on the Impact of Thinking Patterns in LLMs 系统性研究思维模式对大语言模型影响，提出ThinkPatterns-21k数据集。 large language model instruction following
10 Mitigating KV Cache Competition to Enhance User Experience in LLM Inference CacheOPT通过缓解KV缓存竞争，提升LLM推理用户体验 large language model
11 AccelGen: Heterogeneous SLO-Guaranteed High-Throughput LLM Inference Serving for Diverse Applications AccelGen：面向多样化应用，提供异构SLO保障的高吞吐量LLM推理服务系统 large language model
12 CoDet-M4: Detecting Machine-Generated Code in Multi-Lingual, Multi-Generator and Multi-Domain Settings CoDet-M4：提出一种多语言、多生成器、多领域环境下的机器生成代码检测框架 large language model
13 Pensez: Less Data, Better Reasoning -- Rethinking French LLM Pensez：通过少量高质量数据，提升法语LLM的推理能力 large language model
14 MetaScale: Test-Time Scaling with Evolving Meta-Thoughts MetaScale：通过演进的元思考实现大语言模型测试时自适应缩放 large language model
15 DLPO: Towards a Robust, Efficient, and Generalizable Prompt Optimization Framework from a Deep-Learning Perspective DLPO：从深度学习视角构建鲁棒、高效、泛化的提示优化框架 large language model ✅
16 KVShare: An LLM Service System with Efficient and Effective Multi-Tenant KV Cache Reuse KVShare：通过高效多租户KV缓存复用提升LLM服务系统性能 large language model
17 Valid Text-to-SQL Generation with Unification-based DeepStochLog 提出基于Unification的DeepStochLog框架，保证Text-to-SQL生成结果的有效性。 large language model ✅
18 Computation Mechanism Behind LLM Position Generalization 揭示LLM位置泛化能力背后的计算机制，发现注意力logits的解耦现象 large language model
19 A Survey on Transformer Context Extension: Approaches and Evaluation Transformer长文本扩展综述：方法与评估 large language model
20 Improving Complex Reasoning with Dynamic Prompt Corruption: A soft prompt Optimization Approach 提出动态Prompt腐蚀(DPC)方法，提升大语言模型在复杂推理任务中的Prompt Tuning效果。 large language model
21 REPA: Russian Error Types Annotation for Evaluating Text Generation and Judgment Capabilities 提出REPA数据集，评估LLM在俄语环境下的文本生成和判断能力，揭示俄语LLM Judge的性能差距。 large language model
22 ClusComp: A Simple Paradigm for Model Compression and Efficient Finetuning ClusComp：一种用于模型压缩和高效微调的简单范式 large language model
23 A Framework to Assess Multilingual Vulnerabilities of LLMs 提出多语言LLM脆弱性评估框架，揭示低资源语言中的安全风险。 large language model
24 nvBench 2.0: Resolving Ambiguity in Text-to-Visualization through Stepwise Reasoning 提出nvBench 2.0基准与Step-Text2Vis模型，解决文本到可视化任务中歧义查询的难题。 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (2 篇)

#题目一句话要点标签🔗⭐
25 Enhancing LLM Reasoning with Iterative DPO: A Comprehensive Empirical Investigation 提出迭代DPO框架，以低计算成本显著提升LLM推理能力 reinforcement learning DPO direct preference optimization
26 TNCSE: Tensor's Norm Constraints for Unsupervised Contrastive Learning of Sentence Embeddings 提出TNCSE，通过约束张量范数进行无监督对比学习，提升句子嵌入表示 contrastive learning

⬅️ 返回 cs.CL 首页 · 🏠 返回主页