基于大语言模型(LLM)的多智能体系统通常依赖预定义的协作结构,限制了其对不断变化的任务需求的适应性。尽管近期方法支持动态编排,但它们应用的轨迹级奖励为每个决策分配一个结果,往往难以识别单个结构变化的贡献。我们提出RoleFlow,一种自演化多角色智能体系统,它通过反馈驱动的原子编辑和按需角色合成来优化协作结构。为使结构自适应可学习,我们引入单编辑反事实策略优化(SECPO),该方法在相同链状态下将每个编辑与其无编辑对应项进行评估,并为编辑特定的标记分配针对性的 credit。在跨越四个领域的七个基准测试中,RoleFlow 取得了最优的平均性能,比最强的竞争对手高出 4.7 个点。它在所有三个多跳 QA 基准测试中领先 3.2-7.5 F1 点,在数学推理、网络交互和代码生成方面也表现出优势。进一步分析表明其具有跨基准可迁移性、训练数据效率和跨执行器泛化能力。
英文摘要:
Large language model (LLM) based multi-agent systems often rely on predefined collaboration structures, limiting their adaptability to evolving task requirements. Although recent approaches enable dynamic orchestration, trajectory-level rewards they apply assign one outcome to each decision, ofen struggling to identify the contributions of individual structural changes. We propose RoleFlow, a self-evolving multi-role agentic system that refines collaboration structures through feedback-driven atomic edits and on-demand role synthesis. To make structural adaptation learnable, we introduce Single-Edit Counterfactual Policy Optimization (SECPO), which evaluates each edit against its no-edit counterpart at the same chain state and assigns targeted credit to edit-specific tokens. Across seven benchmarks spanning four domains, RoleFlow achieves the optimal average performance, surpassing the strongest competitor by 4.7 points. It leads on all three multi-hop QA benchmarks by 3.2-7.5 F1 points and also shows supremacy on mathematical reasoning, web interaction, and code generation. Further analyses demonstrate cross-benchmark transferability, training-data efficiency, and cross-executor generalization.
From Agentification to Self-Evolving Agentic AI for Wireless Networks: Concepts, Approaches, and Future Research Directions
College of Computing and Data Science, Nanyang Technological University(计算与数据科学学院,南洋理工大学);College of Computer Science and Technology, Jilin University(计算机科学与技术学院,吉林大学);Department of Electrical and Computer Engineering, Western University(电气与计算机工程系,西方大学);Department of Electrical and Computer Engineering, Auburn University(电气与计算机工程系,阿伯茨罕大学);School of Electrical and Computer Engineering, University of Sydney(电气与计算机工程学院,悉尼大学)
EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences(深圳先进技术研究院,中国科学院);Tongyi Lab , Alibaba Group(通义实验室,阿里巴巴集团);Alibaba Group(阿里巴巴集团);SUAT(深圳大学)
TimeSage-EV: A Live Benchmark for Agentic Time Series Analysis in Evolving Environments
Eindhoven University of Technology(埃因霍温理工大学);University of Oxford(牛津大学);VulpiVox Intelligence(VulpiVox智能公司);Squirrel Ai Learning(Squirrel AI学习公司);Griffith University(格里菲斯大学)
HyEvo: Self-Evolving Hybrid Agentic Workflows for Efficient Reasoning
East China Normal University(华东师范大学);Beihang University(北京航空航天大学);Shanghai University of International Business and Economics(上海对外经贸大学);Fudan University(复旦大学)
OctoT2I: A Self-Evolving Agentic Text-to-Image Router
School of Electronic and Computer Engineering, Peking University(电子与计算机工程学院,北京大学);Guangdong Provincial Key Laboratory of Ultra High Definition Immersive Media Technology, Shenzhen Graduate School, Peking University(广东省超高清沉浸媒体技术重点实验室,北京大学深圳研究生院)
SAGE: A Self-Evolving Agentic Graph-Memory Engine for Structure-Aware Associative Memory
Institute for Artificial Intelligence, Peking University(北京大学人工智能研究院);School of Intelligence Science and Technology, Peking University(北京大学智能科学与技术学校);School of Computer Science and Technology, Beijing Institute of Technology(北京理工大学计算机科学与技术学校)
A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems
University of Glasgow(格拉斯哥大学);University of Sheffield(谢菲尔德大学);Mohamed bin Zayed University of Artificial Intelligence(穆罕默德·本·扎耶德人工智能大学);National University of Singapore(新加坡国家大学);University of Cambridge(剑桥大学);University College London(伦敦大学学院);University of Aberdeen(阿伯丁大学);Leiden University(莱顿大学)
MobEvolve: An Agentic Self-Evolving Heuristic System for Interpretable Human Mobility Generation
The Hong Kong Polytechnic University(香港理工大学);McGill University(麦吉尔大学);MIT(麻省理工学院);Tsinghua University(清华大学);Autorité régionale de transport métropolitain(大都会交通地区管理局);Université de Montréal(蒙特利尔大学);Mila – Quebec AI Institute(魁北克人工智能研究所)
SSE-Bio: A Structured Self-Evolving Agent with Agentic Retrieval Policy for Multi-Hop Biomedical Reasoning
School of Computing Science, University of Glasgow(格拉斯哥大学计算科学学院);Brigham and Women’s Hospital, Harvard Medical School, Harvard University(哈佛大学医学院附属布莱根妇女医院);Language Technology Lab, University of Cambridge(剑桥大学语言技术实验室);School of Natural and Computing Science, University of Aberdeen(阿伯丁大学自然与计算科学学院);School of Cancer Sciences, University of Glasgow(格拉斯哥大学癌症科学学院);Cancer Research UK Scotland Institute(英国癌症研究苏格兰研究所)
Adapting to Evolving Requirements: Agentic AI for Retail Supply Chain Operations
School of Business and Management, Hong Kong University of Science and Technology(香港科技大学工商管理学院);School of Management, University of Science and Technology of China(中国科学技术大学管理学院);Institute of Operations Research and Analytics, National University of Singapore(新加坡国立大学运筹学与分析研究所);NUS Business School, National University of Singapore(新加坡国立大学商学院)
PhotoCraft: Agentic Reasoning with Hierarchical Self-Evolving Memory for Deep Image Search
Pattern Recognition Center, WeChat AI, Tencent Inc.(微信AI模式识别中心,腾讯公司);Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所);Zhongguancun Academy(中关村学院);Nanyang Technological University(南洋理工大学);Shanghai Jiao Tong University(上海交通大学)
CHIME: Credit-Aware Hierarchical Memory Evolution for Long-Horizon Agentic Planning
Institute of Artificial Intelligence, Xiamen University(厦门大学人工智能学院);Key Laboratory of Digital Protection and Intelligent Processing of Intangible Cultural Heritage of Fujian and Taiwan (Xiamen University), Ministry of Culture and Tourism(文化和旅游部闽台非物质文化遗产数字化保护与智能处理重点实验室(厦门大学));Zhejiang University(浙江大学);Alibaba Group(阿里巴巴集团)
University of British Columbia(不列颠哥伦比亚大学);Vector Institute(向量研究所);Kolors Team, Kuaishou Technology(快手团队);Carnegie Mellon University(卡内基梅隆大学);University of Waterloo(滑铁卢大学);Etude AI;Tsinghua University(清华大学);Georgia Institute of Technology(佐治亚理工学院)
Agentic AI for Bilevel Long-Term Optimization of Policy-Driven Physical Layer Systems
Key Laboratory of EMW Information (MoE), College of Future Information Technology, Fudan University(复旦大学未来信息技术学院电磁波信息科学教育部重点实验室);James Watt School of Engineering, University of Glasgow(格拉斯哥大学詹姆斯·瓦特工程学院);School of Engineering, Edith Cowan University(埃迪斯科文大学工程学院);Information Systems Technology and Design Pillar, Singapore University of Technology and Design(新加坡科技设计大学信息系统技术与设计系)
VideoEvolve: Co-Evolving Memory and Retrieval for Long Video Understanding
Alibaba Group(阿里巴巴集团);University of Science and Technology of China(中国科学技术大学);Shanghai Jiao Tong University(上海交通大学);University of Chinese Academy of Sciences(中国科学院大学);Beihang University(北京航空航天大学)
Towards Agentic Self-Learning LLMs in Search Environment
The Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences(认知与决策智能复杂系统实验室,自动化研究所,中国科学院);School of Artificial Intelligence, University of Chinese Academy of Sciences(人工智能学院,中国科学院大学);Xiaohongshu Inc(小红书公司);Meituan Inc(美团公司);Shanghai Artificial Intelligence Laboratory(上海人工智能实验室);Tsinghua University(清华大学)