Constiution of chu maps by using EDA-RL

The Channel Theory have been used to represent information flows qualitatively. Chu maps are often utilized for showing classifications in the Channel Theory. This framework is very useful and Chu maps have been used for interface design. This framework has a difficulty such that the domain knowledge represented in this framework is static. That is, there is no way to modify existing domain knowledge. In this paper, we discuss a systematic way to describe domain knowledge by an evolutionary approach on the context of reinforcement learning problems. The domain knowledge is represented by feature functions in EDA-RL, Estimation of Distribution Algorithms for solving Reinforcement Learning, proposed by us. In addition, model search procedure is introduced. Therefore, the changes of the structure of probabilistic models correspond to the modification of the domain knowledge.