随机博弈框架下的多agent强化学习方法综述

多agent学习是在随机博弈的框架下,研究多个智能体间通过自学习掌握交互技巧的问题.单agent强化学习方法研究的成功,对策论本身牢固的数学基础以及在复杂任务环境中广阔的应用前景,使得多agent强化学习成为目前机器学习研究领域的一个重要课题.首先介绍了多agent系统随机博弈中基本概念的形式定义;然后介绍了随机博弈和重复博弈中学习算法的研究以及其他相关工作;最后结合近年来的发展,综述了多agent学习在电子商务、机器人以及军事等方面的应用研究,并介绍了仍存在的问题和未来的研究方向.

Survey of Multi-agent Reinforcement Learning in Markov Games

The research on multi-agent reinforcement learning is to deal with the problem of play skill between agents,just with the concept of stochastic game.All the things of the success of single agent reinforcement learning,the mathematics basis of the game theory,and the potential applications in complex task environment make the multi-agent learning an important topic in the field of machine learning.The formal definition of basic concepts in stochastic game is given first,and then the algorithms of learning in stochastic game and repeated game are introduced.Last,the research on the applications of multi-agent learning is summarized,and the problems remaining unsolved together with the future work are concluded.