多重学習器を用いる強化学習

―報酬付与に遅れがある学習環境への適用―

抄録

In this report, we apply the proposed reinforcement learning with multiplex learning spaces to an environment with a delay time to get a reward and compare it with an ordinary reinforcement learning without considering the delay in experimental simulations. We adapt the proposed method by multiplexing some learning spaces corresponding to a different delay time each other. As the result of an experiment, the ordinary method couldn't get the best policy, but the proposed method could get it effectively.