Applying determinized MCTS in Chinese Military Chess

Monte Carlo Tree Search (MCTS) algorithm has been proved to be very successful in many perfect information games such as Go and Amazon. This leads to a trend to apply MCTS in games with imperfect information. One popular method is called Determinized MCTS and its efficiency has been shown in many games. In this paper, we plan to apply determinized MCTS to Chinese Military Chess, which is a very popular game in China. We discuss how to generate initial belief state for AI agent according to some rules and domain knowledge of the game, and present an algorithm to update it online. We then apply this framework into determinized MCTS and show its efficiency in experiments.