Title :
Monte Carlo Go Using Previous Simulation Results
Author :
Toyoda, Takuma ; Kotani, Yoshiyuki
Author_Institution :
Grad. Sch. of Eng., Tokyo Univ. of Agric. & Technol., Koganei, Japan
Abstract :
The researches on Go using Monte Carlo method are treated as hot topics in these years. In particular, Monte Carlo Tree Search algorithm such as UCT made great contributions to the development of computer Go. When Monte Carlo method was used for Go, the previous simulation results were not usually stored. In this paper, we suggest a new idea of using previous simulation results (PSR heuristic) and try to improve the method. Experimental results showed that this heuristic is very effective in blitz and on larger Go board.
Keywords :
Monte Carlo methods; computer games; tree searching; Monte Carlo Go; Monte Carlo tree search algorithm; UCT; previous simulation results; Monte Carlo; UCT; artificial intelligence; computer Go; heuristic;
Conference_Titel :
Technologies and Applications of Artificial Intelligence (TAAI), 2010 International Conference on
Conference_Location :
Hsinchu City
Print_ISBN :
978-1-4244-8668-7
Electronic_ISBN :
978-0-7695-4253-9
DOI :
10.1109/TAAI.2010.39
