Research On Open-Region UCT Algorithms Combined With Ko

UCT (Upper Confidence Bound Applied to Trees) algorithm has a great significance, that it has improved computer go program level to a large extent. They may be achieve professional level in small board, such as 9×9. But in 19×19 board, the playing level is modest. This is mainly due to multiple local problems occurring at the same time. Computer go search algorithm can not achieve a good performance.One way to solve this issue is to use combination of global search and local search algorithm. So we will need a search area in every local search. If it is a closed region, local search will become very easy. So we introduce a unique method to determine local search area. And from these results, we find some problems of Ko.The main works are as follows:1) We discuss some characteristics of Computer Go, encountered trouble in the filed of Computer Go and simply introduce several classical algorithms.2) We propose a method to determine local search area. It is achieved by controlling power value. Finally, we through the contrast experiments to illustrate the effectiveness of our method.3) From open-region UCT algorithm experiment, we found that the performance of open-region UCT algorithm is very poor when they deal with Ko. Therefore, we add the identification and treatment of Ko in UCT algorithm. Finally, through some experiments, we find that the performance of UCT algorithm has been advanced.