A Proposal of Reinforcement Learning with Fuzzy Environment Evaluation Rules and Its Application to Chess

Abstract

The machine learning method is proposed to learn techniques of specialists. A machine has to learn techniques by trial and error when there are not enough training data. Reinforcement learning is a powerful machine learning system, which is able to learn without giving training data to a learning unit. But it is impossible for the reinfocement learning to support large environments because the number of if-then rules, which explain a relationship between one environment and one action, is a huge combination. We propose a new reinforcement learning with fuzzy environment evaluation. The fuzzy environment evaluation rule shows a relationship between one environment and one evaluation. This machine learning system is made up from a fuzzy evaluation, an environment simulator and MinMax search. The learning unit renews the evaluation in every action. The fuzzy evaluation of inexperienced environments is reasoned by fuzzy rules. The fuzzy evaluation, the environment simulator and Min Max search present the best policy in a huge environment. We dealt with chess as an example of target environment. Then we will show the excellent results of chess against the GNU chess. The fuzzy evaluation of inexperienced environments is reasoned by fuzzy rule sets, machine has.