抄録
Recently, applications of multiagent systems are expected from the view point of the parallel and distributed processing of systems. Reinforcement learning attracts attention as an implementing method of multiagent systems. However, there is a problem that the more the number of agents to deal with increases, the slower the speed of learning becomes. To solve this problem, we propose a new reinforcement learning method that can learn quickly and reduce amount of memory. It tries to increase efficiency of the learning on a hunter game by paying attention to partial states of two agents among a large number of agents. In addition, the proposed method employs a switching algorithm and detects automatically a switching time by using a special index called golden cross.
