Iterative Tree Search in General Game Playing with Incomplete Information

Armin Chitizadeh, Michael Thielscher

Abstract: General Game Playing (GGP) is concerned with the development of programs capable of effectively playing a game by just receiving its rules and without human intervention. The standard game representation language GDL has recently been extended so as to include games with incomplete information. The so-called Lifted HyperPlay technique, which is based on model sampling, provides a state-of-the-art solution to general game playing with incomplete information. However, this method is known not to model opponents properly, with the effect that it generates only pure strategies and is short-sighted when valuing information. In this paper, we overcome these limitations by adapting the classic idea of fictitious play to introduce an Iterative Tree Search algorithm for incomplete-information GGP. We demonstrate both theoretically and experimentally that our algorithm provides an improvement over existing solutions on several classes of games that have been discussed in the literature.