Experience Generation in Tic-Tac-Toe for General Game Learning

摘要

General Game Playing aims at developing game playing agents that are able to play a variety of games proficiently without game specific experience. This paper poses with a simplified experience-based learning approach on the basis of the Reinforcement Learning Algorithms. Through the selection of game states and induction of game experience, this approach reduces the needed experience in decision-making process, improving efficiency and making the AI player reach the specified goal such as victory, draw or defeat. In order to test the effectiveness of this approach, matches are played in several different games between AI player and a random player.