An Adaptation System in Unknown Environments Using a Mixture Probability Model and Clustering Distributions

Abstract

Adaptation to dynamic environments is required in an agent system using Reinforcement Learning (RL). A mixture model of Bayesian network was introduced into the learning system for quickly adapting to such environments. This increases the computational complexity for training the parameters of the system. Therefore, reducing such complexities is necessary when there are limitations in the processing resources. In this paper, we introduce a mixture probability into RL for allowing an agent to adjust to environmental changes. We also introduce a new clustering method that enables one to select fewer elements of the mixture probability in order to reduce the computational complexity and simultaneously maintain the system’s performance. Computer simulations are presented to investigate the effectiveness of our proposed method.