Multi-agent reinforcement learning with adaptive mimetism
Takuro Yamaguchi, M. Miura, M. Yachida · 2002
To learn the group behavior of a multi-agent system, it is important to selectively share the learning results in order to speed up learning without homogenizing the agents' behaviors. This paper describes a new method designed to permit multiple agents in an environment to learn cooperatively. The advantage of our method is to dynamically switch the learning mode between mimetism and reinforcement learning according to the situation. Mimetism seeks stability in group behavior, while individual reinforcement learning seeks the best solution. Accordingly, selective mimetism that allows the agents to partially share learning results works to prevent homogenization among the agents.