cs.CL（2025-09-30）

📊 共 46 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) (38 🔗3) 支柱二：RL算法与架构 (RL & Architecture) (7 🔗2) 支柱一：机器人控制 (Robot Control) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (38 篇)

#题目一句话要点标签🔗⭐
1 Think Less, Label Better: Multi-Stage Domain-Grounded Synthetic Data Generation for Fine-Tuning Large Language Models in Telecommunications 提出基于领域知识图谱的多阶段合成数据生成方法，用于电信领域大语言模型微调。 large language model instruction following
2 CATCH: A Novel Data Synthesis Framework for High Therapy Fidelity and Memory-Driven Planning Chain of Thought in AI Counseling 提出CATCH框架以提升AI咨询的治疗忠实度和决策合理性 large language model chain-of-thought
3 BiasFreeBench: a Benchmark for Mitigating Bias in Large Language Model Responses BiasFreeBench：用于评估和缓解大语言模型偏见的综合基准 large language model
4 RoBiologyDataChoiceQA: A Romanian Dataset for improving Biology understanding of Large Language Models 提出RoBiologyDataChoiceQA，用于提升大语言模型在生物学理解方面的能力 large language model
5 TraceDet: Hallucination Detection from the Decoding Trace of Diffusion Large Language Models TraceDet：利用扩散大语言模型解码轨迹进行幻觉检测 large language model
6 Direct Token Optimization: A Self-contained Approach to Large Language Model Unlearning 提出直接Token优化（DTO）方法，实现大语言模型自包含式遗忘学习。 large language model
7 TAMA: Tool-Augmented Multimodal Agent for Procedural Activity Understanding 提出TAMA：工具增强的多模态Agent，用于程序性活动理解 multimodal
8 OceanGym: A Benchmark Environment for Underwater Embodied Agents OceanGym：水下具身智能体的综合基准环境，应对极端环境挑战。 embodied AI large language model ✅
9 Personalized Scientific Figure Caption Generation: An Empirical Study on Author-Specific Writing Style Transfer 研究个性化科学图表标题生成，探索作者风格迁移方法。 large language model multimodal
10 BatonVoice: An Operationalist Framework for Enhancing Controllable Speech Synthesis with Linguistic Intelligence from LLMs BatonVoice：利用LLM语言智能增强可控语音合成的运算主义框架 large language model instruction following
11 Latent Thinking Optimization: Your Latent Reasoning Language Model Secretly Encodes Reward Signals in Its Latent Thoughts 提出Latent Thinking Optimization，通过隐空间奖励建模提升LLM推理能力 large language model chain-of-thought
12 SafePassage: High-Fidelity Information Extraction with Black Box LLMs SafePassage：利用黑盒LLM实现高保真信息抽取，显著降低幻觉。 large language model
13 Catalog-Native LLM: Speaking Item-ID Dialect with Less Entanglement for Recommendation 提出IDIOMoE，通过Item-ID方言和MoE结构，增强LLM在推荐系统中的协同过滤能力。 large language model
14 TAU: A Benchmark for Cultural Sound Understanding Beyond Semantics TAU：一个用于文化声音理解的基准，超越语义层面 multimodal
15 Finetune Once: Decoupling General & Domain Learning with Dynamic Boosted Annealing 提出动态增强退火(DBA)方法，解耦通用和领域学习，高效微调LLM。 large language model
16 Uncertainty-Aware Answer Selection for Improved Reasoning in Multi-LLM Systems 提出基于校准对数似然的多LLM答案选择方法，提升推理性能。 large language model
17 Efficient Layer-wise LLM Fine-tuning for Revision Intention Prediction 提出IR-Tuning，一种高效的层级LLM微调框架，用于文本修订意图预测。 large language model
18 TUMIX: Multi-Agent Test-Time Scaling with Tool-Use Mixture TUMIX：基于工具使用混合的多Agent测试时扩展方法 large language model
19 Personalized Reasoning: Just-In-Time Personalization and Why LLMs Fail At It 提出PREFDISCO评估框架，揭示LLM在即时个性化推理上的局限性 large language model
20 Deconstructing Self-Bias in LLM-generated Translation Benchmarks 揭示LLM生成翻译评测基准中的自偏见问题，并提出缓解策略 large language model
21 PerQ: Efficient Evaluation of Multilingual Text Personalization Quality 提出PerQ：一种高效评估多语言文本个性化质量的指标 large language model
22 RoleConflictBench: A Benchmark of Role Conflict Scenarios for Evaluating LLMs' Contextual Sensitivity 提出RoleConflictBench基准，评估LLM在角色冲突场景下的情境敏感性 large language model
23 LD-MoLE: Learnable Dynamic Routing for Mixture of LoRA Experts 提出LD-MoLE，通过可学习动态路由实现LoRA专家混合，提升LLM微调性能。 large language model
24 Submodular Context Partitioning and Compression for In-Context Learning 提出Sub-CP框架，利用子模目标进行上下文分块和压缩，提升ICL性能。 large language model
25 Evaluation Sheet for Deep Research: A Use Case for Academic Survey Writing 提出评估框架，用于评估大语言模型在学术综述写作中的表现 large language model
26 Training Matryoshka Mixture-of-Experts for Elastic Inference-Time Expert Utilization 提出Matryoshka MoE，实现MoE模型在推理时专家利用的弹性调整 large language model
27 CreAgentive: An Agent Workflow Driven Multi-Category Creative Generation Engine 提出CreAgentive以解决多类别创作生成的四大问题 large language model
28 Automatic Fact-checking in English and Telugu 构建英-泰双语数据集，探索大型语言模型在英语和泰卢固语事实核查中的有效性。 large language model
29 Fast-dLLM v2: Efficient Block-Diffusion LLM Fast-dLLM v2：高效块扩散语言模型，加速并行文本生成。 large language model
30 VietBinoculars: A Zero-Shot Approach for Detecting Vietnamese LLM-Generated Text VietBinoculars：一种零样本越南语LLM生成文本检测方法 large language model
31 Explaining novel senses using definition generation with open language models 利用开源语言模型生成定义解释词义演变，性能超越闭源模型。 large language model
32 IMProofBench: Benchmarking AI on Research-Level Mathematical Proof Generation IMProofBench：用于评估AI在研究级数学证明生成能力的新基准 large language model
33 Limited Preference Data? Learning Better Reward Model with Latent Space Synthesis LENS：通过潜在空间合成提升有限偏好数据下的奖励模型学习 large language model ✅
34 RE$^2$: Improving Chinese Grammatical Error Correction via Retrieving Appropriate Examples with Explanation RE$^2$：通过检索带解释的合适示例来改进中文语法纠错 large language model
35 Understanding the Mixture-of-Experts with Nadaraya-Watson Kernel 提出KERN以替代Softmax解决MoE模型中的路由问题 large language model
36 ReFACT: A Benchmark for Scientific Confabulation Detection with Positional Error Annotations ReFACT：提出一个科学知识捏造检测基准，包含位置错误标注。 large language model ✅
37 Atomic Thinking of LLMs: Decoupling and Exploring Mathematical Reasoning Abilities 提出数学原子能力解耦方法，探索大语言模型数学推理能力的本质 large language model
38 The Media Bias Detector: A Framework for Annotating and Analyzing the News at Scale 提出Media Bias Detector框架，用于大规模标注和分析新闻媒体的偏见 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🔗⭐
39 TruthRL: Incentivizing Truthful LLMs via Reinforcement Learning TruthRL：通过强化学习激励大型语言模型生成更真实可靠的答案 reinforcement learning large language model
40 Efficient and Transferable Agentic Knowledge Graph RAG via Reinforcement Learning 提出基于强化学习的高效可迁移Agentic知识图谱RAG框架KG-R1 reinforcement learning large language model ✅
41 Mem-α: Learning Memory Construction via Reinforcement Learning 提出Mem-α，通过强化学习训练LLM Agent有效管理复杂记忆系统，解决长程信息理解问题。 reinforcement learning large language model
42 MENLO: From Preferences to Proficiency -- Evaluating and Modeling Native-like Quality Across 47 Languages MENLO：提出多语言原生质量评估框架，提升LLM在47种语言上的表现。 reinforcement learning reward shaping large language model
43 One-Token Rollout: Guiding Supervised Fine-Tuning of LLMs with Policy Gradient 提出One-Token Rollout算法，利用策略梯度指导LLM的监督微调，提升泛化能力。 reinforcement learning policy learning large language model
44 Judging with Confidence: Calibrating Autoraters to Preference Distributions 提出概率化自动评分器以解决主观偏差问题 reinforcement learning large language model
45 dParallel: Learnable Parallel Decoding for dLLMs dParallel：面向扩散语言大模型的并行可学习解码方法 distillation large language model ✅

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
46 Unspoken Hints: Accuracy Without Acknowledgement in LLM Reasoning 提出系统性研究以揭示LLM推理中的提示影响 manipulation RLHF large language model

⬅️ 返回 cs.CL 首页 · 🏠 返回主页