Computer go

Abstract

Generally we can divide a Go game into three stages: beginning game, middle game, and end game; and there is one sub problem that could appear in all three stages of the game – tsumego (Life and Death). Lift and Death problems concern about a group of stone can archive a life state, that are safe from being capture, by a specific sequence of moves or, in the counter part, to look for a sequence of play to kill opponent’s stone no matter how she/he defense it. All Go players do practice on the Life and Death problem as it comes up very often in game. Also there are books written by professional players where Life and Death questions are ranked into different levels. Those questions usually display an enclosed group of stone in a corner or on the side of a board, which exist one correct sequence of play to archive a firm life or a firm kill[Appendix 1]. For this kind of enclosed group which cannot escape and outside condition are ignored, we call it closed Life and Death. This project conducted two experiments on using ANN to recognize stones status enclosed in corner. The result is impressive and there maybe ways we can apply ANN not only to recognize the stone status but also to find the correct sequence of play. We believe that by creating a special partial connection between input layer and the first hidden layer can obtain more accurate result. With very similar shapes of life and dead patterns providing in our experiment, ANN showed its competent in distinguishing which is alive and which is dead. Also the performance of applying ANN to evaluate life status should have performance advantage over heuristic search.