Abstract
We aim to accelerate learning processes in reinforcement learning by transfer learning. Its concept is that knowledge to solve similar tasks accelerates a learning process of a target task. We have proposed that the basic transfer method based on forbidden rule set that is a set of rules which cause to immediately failure of a target task. However, the basic method works poorly for the "Different purpose task," which has same state transition probability and different purpose. In this article, we discuss the effective method to transfer for the different purpose task. In detail, we propose the additional information that some explorations which based on knowledge in the database for the effective transfer. We perform simple experiments to show the effectiveness of the proposed method.
