多智能体博弈中的分布式学习:原理与算法
Distributed Learning for Multi-agent Games:Theory and Algorithms
谭少林 1谷海波 2刘克新2
作者信息
- 1. 中关村实验室,北京 100094
- 2. 中关村实验室,北京 100094;北京航空航天大学自动化科学与电气工程学院,北京 100191
- 折叠
摘要
自主智能决策是未来无人系统发展的核心技术,而博弈学习是实现自主智能决策的关键方法之一.围绕多智能体博弈中分布式学习领域,系统地介绍其基本问题、研究背景及意义;针对连续动作空间博弈与离散动作空间博弈两种典型博弈类型,综述多智能体博弈分布式学习算法的构建及收敛性研究进展;给出博弈学习领域尚待突破的挑战性问题.
Abstract
Autonomous intelligent decision-making is a core technique of future unmanned system development,and game-theoretic learning is one of the key methods to realize autonomous intelligent decision-making.The rapid development field of the distributed learning for multi-agent games is centered on,a systematic introduction of its basic problems,research background and significance is performed.Then,regarding to two typical classes of games,including continuous action space games and discrete action space games,the recent construction and convergence research progresses of the distributed game-theoretic learning algorithms are overviewed.Finally,several challenging problems to be broken through in the future game-theoretic learning field are pointed out.
关键词
博弈学习/分布式算法/智能决策/纳什均衡/多智能体系统/集群智能Key words
game-theoretic learning/distributed algorithms/intelligent decision-making/Nash equilibrium/multi-agent systems/swarm intelligence引用本文复制引用
基金项目
国家自然科学基金(T2322023)
国家自然科学基金(62103015)
国家自然科学基金(92067204)
湖南省自然科学基金(2022JJ20018)
空间智能控制技术重点实验室项目(HTKJ2022KL502013)
出版年
2024
NETL
NSTL
万方数据