活性化拡散モデルに基づく強化学習エージェントの転移学習手法

抄録

This paper proposes a policy transfer method of a reinforcement learning agent for suitable learning in unknown or dynamic environments based on a spreading activation model in the cognitive psychology.The reinforcement learning agent saves policies learned in various environments and learns flexibly by partially using suitable policy according to the environment. In the proposed method, an undirected graph is created between policies, and the network is constructed by them. The agent updates the activate value that policy has according to the environment while repeating processes of recall, activation, spreading, attenuation and learns based on the network. Agent uses this network in transfer learning. Experimental simulations comparing the proposed method with several existing methods are conducted to confirm the usefulness of the proposed method. Simulation results show that the reinforcement learning agent achieves task by selecting the optimal one from policies with the proposed method.