Knowledge reuse of learning agent based on factor information of behavioral rules

Abstract

In this study, we attempt to extract knowledge by collecting results from multiple environments using an autonomous learning agent. A common factor of the environment is extracted by applying non-negative matrix factorization to the set of learning results of the reinforcement learning agent. In transfer learning of knowledge management of agents, as the number of experienced tasks increases, the knowledge database becomes larger and the cost of knowledge selection increases. By the proposed approach, an agent that can adapt to multiple environments can be developed without increasing cost of knowledge selection.