Technique Analysis and Designing of Program with UCT Algorithm for NoGo
Rui Li,Yueqiu Wu,Andi Zhang,Chen Ma,Bo Chen,Shuliang Wang
DOI: https://doi.org/10.1109/ccdc.2013.6561055
2013-01-01
Abstract:As a typical example of dynamic search algorithm, the UCT algorithm was initially used on the computerized game of GO. This paper briefly introduces the Markov Decision process, the Multi-armed Bandit model, and the Upper-Confidence Bandit formula. It analyzes the source and structure of the UCT algorithm in theory, and proves that the UCT algorithm is suitable for the design of the program of NoGo. According to the characteristics of NoGo, in the paper we improved the algorithm in terms of move generation and data reuse. We also tried to establish an off-line knowledge database for research. With experimental data we have tested and evaluated the above methods. The above algorithm and technology have been successfully used in WTShadows-the NoGo game program, which enabled us to have won the champion in national competition.