cs.CL（2025-03-13）

📊 共 30 篇论文 | 🔗 5 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (24 🔗3) 支柱二：RL算法与架构 (RL & Architecture) (4 🔗2) 支柱一：机器人控制 (Robot Control) (2)

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

#题目一句话要点标签🔗⭐
1 Ensemble Learning for Large Language Models in Text and Code Generation: A Survey 综述：集成学习提升大语言模型在文本和代码生成中的性能 large language model multimodal
2 Compositional Subspace Representation Fine-tuning for Adaptive Large Language Models 提出CS-ReFT，通过组合子空间表示微调自适应大语言模型，解决多任务学习中的技能冲突问题。 large language model instruction following
3 Cognitive-Mental-LLM: Evaluating Reasoning in Large Language Models for Mental Health Prediction via Online Text 利用思维链LLM提升在线文本心理健康预测的推理能力 large language model chain-of-thought
4 DynaCode: A Dynamic Complexity-Aware Code Benchmark for Evaluating Large Language Models in Code Generation DynaCode：动态复杂性感知代码基准，用于评估代码生成大语言模型 large language model ✅
5 DarkBench: Benchmarking Dark Patterns in Large Language Models DarkBench：构建大型语言模型中暗黑模式的综合评测基准 large language model
6 Word-level Annotation of GDPR Transparency Compliance in Privacy Policies using Large Language Models 提出基于LLM的模块化流程，用于隐私政策中GDPR透明度合规性的词级别标注。 large language model
7 SCE: Scalable Consistency Ensembles Make Blackbox Large Language Model Generation More Reliable 提出可扩展一致性集成（SCE）框架，提升黑盒大语言模型生成可靠性 large language model
8 MMLU-ProX: A Multilingual Benchmark for Advanced Large Language Model Evaluation 提出MMLU-ProX多语言基准，用于全面评估大型语言模型的跨语言推理能力。 large language model
9 Source-primed Multi-turn Conversation Helps Large Language Models Translate Documents 提出基于源语言引导的多轮对话方法，提升大语言模型文档翻译质量 large language model
10 New Trends for Modern Machine Translation with Large Reasoning Models 利用大型推理模型，将机器翻译重构为动态推理任务，提升翻译质量。 multimodal chain-of-thought
11 "Well, Keep Thinking": Enhancing LLM Reasoning with Adaptive Injection Decoding 提出自适应注入解码，无需显式提示增强LLM推理能力 large language model chain-of-thought
12 Why Prompt Design Matters and Works: A Complexity Analysis of Prompt Search Space in LLMs 提出基于复杂性分析的Prompt设计理论框架，提升LLM推理能力 large language model chain-of-thought
13 Information Density Principle for MLLM Benchmarks 提出信息密度原则，用于评估和改进多模态大语言模型评测基准。 large language model multimodal ✅
14 Scalable Evaluation of Online Facilitation Strategies via Synthetic Simulation of Discussions 提出基于LLM的在线讨论模拟框架，用于大规模评估在线引导策略。 large language model
15 Gumiho: A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding Gumiho：一种混合架构，通过优先处理推测解码中的早期token来加速LLM推理。 large language model
16 OASST-ETC Dataset: Alignment Signals from Eye-tracking Analysis of LLM Responses OASST-ETC：利用眼动追踪分析LLM响应，提供对齐信号 large language model
17 NeurIPS 2023 LLM Efficiency Fine-tuning Competition NeurIPS 2023 LLM微调竞赛揭示基准数据集过度拟合问题，强调数据清洗的重要性。 large language model
18 G-Boost: Boosting Private SLMs with General LLMs G-Boost：利用通用LLM提升私有SLM性能的协同推理框架 large language model
19 Retrieval-Augmented Generation with Hierarchical Knowledge HiRAG：利用层级知识增强检索增强生成，提升领域任务性能 large language model
20 ZSMerge: Zero-Shot KV Cache Compression for Memory-Efficient Long-Context LLMs ZSMerge：面向长文本LLM的零样本KV缓存压缩，提升内存效率 large language model ✅
21 Who Relies More on World Knowledge and Bias for Syntactic Ambiguity Resolution: Humans or LLMs? 研究揭示LLM在歧义消解中过度依赖世界知识和偏见，缺乏人类的灵活性 large language model
22 Thinking Machines: A Survey of LLM based Reasoning Strategies 综述：基于LLM的推理策略研究，弥合语言能力与推理能力差距 large language model
23 Probing LLMs for Multilingual Discourse Generalization Through a Unified Label Set 提出统一标签集并探究LLM在跨语言篇章泛化能力，揭示中间层的重要性。 large language model
24 Adaptive Inner Speech-Text Alignment for LLM-based Speech Translation 提出自适应内部语音-文本对齐方法，提升基于LLM的语音翻译性能 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🔗⭐
25 Efficient Safety Alignment of Large Language Models via Preference Re-ranking and Representation-based Reward Modeling 提出基于偏好重排序和表征奖励建模的高效大语言模型安全对齐方法 reinforcement learning DPO direct preference optimization
26 RankPO: Preference Optimization for Job-Talent Matching 提出RankPO，通过偏好优化提升LLM在职位-人才匹配中的文本理解能力。 DPO direct preference optimization contrastive learning ✅
27 World Modeling Makes a Better Planner: Dual Preference Optimization for Embodied Task Planning 提出双重偏好优化D²PO，提升具身任务规划中LVLM的环境理解和规划能力 preference learning world model
28 Light-R1: Curriculum SFT, DPO and RL for Long COT from Scratch and Beyond Light-R1：一种基于公共数据的长文本推理模型训练方案 DPO ✅

🔬 支柱一：机器人控制 (Robot Control) (2 篇)

#题目一句话要点标签🔗⭐
29 PRISM: Preference Refinement via Implicit Scene Modeling for 3D Vision-Language Preference-Based Reinforcement Learning PRISM：通过隐式场景建模优化3D视觉-语言偏好强化学习 manipulation reinforcement learning chain-of-thought
30 HyperDAS: Towards Automating Mechanistic Interpretability with Hypernetworks HyperDAS：利用超网络自动实现机制可解释性 manipulation

⬅️ 返回 cs.CL 首页 · 🏠 返回主页