cs.AI（2025-09-28）

📊 共 26 篇论文 | 🔗 2 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (17 🔗1) 支柱二：RL算法与架构 (RL & Architecture) (9 🔗1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (17 篇)

#题目一句话要点标签🔗⭐
1 AnveshanaAI: A Multimodal Platform for Adaptive AI/ML Education through Automated Question Generation and Interactive Assessment AnveshanaAI：一个通过自动问题生成和交互式评估实现自适应AI/ML教育的多模态平台 large language model multimodal
2 HFuzzer: Testing Large Language Models for Package Hallucinations via Phrase-based Fuzzing HFuzzer：通过基于短语的模糊测试发现大语言模型中的包幻觉问题 large language model
3 Navigating the Labyrinth: Path-Sensitive Unit Test Generation with Large Language Models JUnitGenie：利用大语言模型实现路径敏感的单元测试生成，提升代码覆盖率。 large language model
4 MedLA: A Logic-Driven Multi-Agent Framework for Complex Medical Reasoning with Large Language Models MedLA：一种逻辑驱动的多智能体框架，用于大型语言模型进行复杂医学推理 large language model
5 Falcon: A Cross-Modal Evaluation Dataset for Comprehensive Safety Perception 提出Falcon数据集与FalconEye评估器，提升多模态大语言模型安全评估的全面性。 large language model multimodal
6 RADAR: A Risk-Aware Dynamic Multi-Agent Framework for LLM Safety Evaluation via Role-Specialized Collaboration RADAR：基于角色 специализирана协作的风险感知动态多智能体框架，用于LLM安全评估 large language model
7 GeoSQL-Eval: First Evaluation of LLMs on PostGIS-Based NL2GeoSQL Queries 提出GeoSQL-Eval，首次系统评估LLMs在PostGIS上的NL2GeoSQL查询能力。 large language model
8 PartnerMAS: An LLM Hierarchical Multi-Agent Framework for Business Partner Selection on High-Dimensional Features PartnerMAS：一种用于高维特征业务伙伴选择的LLM分层多智能体框架 large language model
9 Quant Fever, Reasoning Blackholes, Schrodinger's Compliance, and More: Probing GPT-OSS-20B 针对GPT-OSS-20B的安全评估揭示了多种对抗性攻击下的模型弱点 chain-of-thought
10 Multi-Value-Product Retrieval-Augmented Generation for Industrial Product Attribute Value Identification 提出MVP-RAG，解决工业产品属性值识别中的级联错误、OOD和泛化性问题。 large language model
11 From What to Why: A Multi-Agent System for Evidence-based Chemical Reaction Condition Reasoning 提出ChemMAS多智能体系统，解决化学反应条件推荐的可解释性问题 large language model
12 AdaPtis: Reducing Pipeline Bubbles with Adaptive Pipeline Parallelism on Heterogeneous Models AdaPtis：通过异构模型上的自适应流水线并行减少流水线气泡 large language model
13 Improving the Efficiency of LLM Agent Systems through Trajectory Reduction AgentDiet：通过轨迹缩减提升LLM Agent系统效率，降低计算成本 large language model
14 Uncovering Vulnerabilities of LLM-Assisted Cyber Threat Intelligence 揭示LLM辅助网络威胁情报的脆弱性以提升安全性 large language model
15 Benchmarking LLM-Assisted Blue Teaming via Standardized Threat Hunting CyberTeam：通过标准化威胁狩猎评估LLM辅助蓝队防御能力 large language model
16 Clean First, Align Later: Benchmarking Preference Data Cleaning for Reliable LLM Alignment PrefCleanBench：首个LLM对齐偏好数据清洗基准评测，提升对齐可靠性 large language model ✅
17 Beyond the Strongest LLM: Multi-Turn Multi-Agent Orchestration vs. Single LLMs on Benchmarks 多智能体协同超越最强LLM：在基准测试中胜过单一大语言模型 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (9 篇)

#题目一句话要点标签🔗⭐
18 Conditional Advantage Estimation for Reinforcement Learning in Large Reasoning Models 提出CANON：一种条件优势估计方法，提升大型推理模型在强化学习中的性能。 reinforcement learning large language model
19 SAC-Opt: Semantic Anchors for Iterative Correction in Optimization Modeling 提出SAC-Opt，通过语义锚点迭代修正优化建模中的逻辑错误。 SAC large language model
20 Formalization Driven LLM Prompt Jailbreaking via Reinforcement Learning 提出PASS框架，利用强化学习和形式化描述提升LLM提示越狱攻击的隐蔽性和有效性 reinforcement learning large language model
21 Taught Well Learned Ill: Towards Distillation-conditional Backdoor Attack 提出SCAR：一种蒸馏条件后门攻击方法，可注入隐蔽后门至教师模型。 distillation ✅
22 How LLMs Learn to Reason: A Complex Network Perspective 提出Annealed-RLVR算法，通过调控概念网络拓扑结构提升LLM推理能力 reinforcement learning large language model
23 Continual Learning to Generalize Forwarding Strategies for Diverse Mobile Wireless Networks 提出一种基于持续学习的通用转发策略，提升移动无线网络在多样场景下的性能。 reinforcement learning deep reinforcement learning DRL
24 Gradient Coupling: The Hidden Barrier to Generalization in Agentic Reinforcement Learning 提出梯度耦合理论，并通过解耦动作嵌入提升强化学习泛化性 reinforcement learning
25 EAPO: Enhancing Policy Optimization with On-Demand Expert Assistance EAPO：通过按需专家辅助增强策略优化，提升LLM推理能力 reinforcement learning large language model
26 Reasoning Scaffolding: Distilling the Flow of Thought from LLMs 提出推理支架(Reasoning Scaffolding)框架，提升小模型推理能力和逻辑一致性。 distillation large language model

⬅️ 返回 cs.AI 首页 · 🏠 返回主页