Static strategies and inference for the game of Phantom Go

Playing the game with partially observable information is a very challenging issue in AI field as its high complexity. Phantom game is a kind of such games, which is usually with large state space. One of them, Phantom Go, is the variant game of computer Go with imperfect information. It is a great challenge and attractive topic in AI for its uncertainty of the hidden information and the complexity from computer Go. In the recent years, the research of IS-MCTS (Information Set Monte-Carlo Search) has boosted the development of Phantom games. Determinization is the very crucial processing in IS-MCTS, which reveals the imperfect information and provides perfect board configuration to the Monte-Carlo tree. As a result, advanced methods that make use of the knowledge by rational players to predict the opponent's information is highly required. This paper proposes two static strategies and an inference model to demonstrate how to use professional knowledge to improve the search quality. These methods are universal and will greatly improve the playing strength of the Phantom Go program.