Introduction and Evaluation of the Notion of Slump in Reinforcement Learning

Abstract

Reinforcement learning, which is inspired by behavioral psychology, is a sub-area of machine learning. It is concerned with how an agent takes actions in an environment to maximize the reward given by the environment. On the other hand, there is a condition of "slump" in the process of human skill learning in the real world, and it is reported that the occurrence and the resolution of the slump cause promotion of skills. In this paper, we propose to introduce the notion of slump in the reinforcement learning. By simulating the condition of the slump in the reinforcement learning, the efficiency of learning increases in some conditions. The formulation and the implementation of the slump in reinforcement learning are described, and the result of the evaluation of the performance is presented.