A Research on General Purpose Fast Monte Carlo Tree Search

2012 Fiscal Year Final Research Report

• Project/Area Number: 23700158
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Intelligent informatics
• Research Institution: Tokyo Institute of Technology (2012); The University of Tokyo (2011)
• Principal Investigator: YOSHIZOE Kazuki 東京工業大学, 大学院・情報理工学研究科, 東工大特別研究員 (80449115)
• Project Period (FY): 2011 – 2012
• Keywords: 探索 / 論理 / 推論アルゴリズム / モンテカルロ法 / 囲碁

Research Abstract

We have successfully parallelized the Monte-Carlo Tree Search algorithm which is known for its broad range of application. Our TDS-df-UCT algorithm achieved 3,200-fold speedup in the best case for synthesized game trees. Also, Accelerated UCT was invented to improve the speed of convergence for the cases existing algorithms fail to converge quickly. It was effective for several games. We have implemented a massively parallel Go program and a general purpose parallel search library on top of TDS-df-UCT.

Research Products

• [Journal Article] Accelerated UCT and Its Application to Two-Player Games (2012)
  • Author(s): Junichi Hashimoto, Akihiro Kishimoto, Kazuki Yoshizoe and Kokolo Ikeda
  • Journal Title: Advances in Computer Games 13 (ACG13) Volume: Vol.13 Pages: 1-12
  • DOI: DOI:10.1007/978-3-642-31866-5_1
  • Peer Reviewed
• [Presentation] Scalable Distributed Monte-Carlo Tree Search (2011)
  • Author(s): Kazuki Yoshizoe, Akihiro Kishimoto, Tomoyuki Kaneko, Haruhiro Yoshimoto and Yutaka Ishikawa
  • Organizer: The Fourth Annual Symposium on Combinatorial Search (SoCS 2011)
  • URL: http://www.aaai.org/ocs/index.php/SOCS/SOCS11/paper/view/4023
  • Place of Presentation: Barcelona, Spain
  • Year and Date: 2011-07-15
• [Book] コンピュータ囲碁-モンテカルロ法の理論と実践 (2012)
  • Author(s): 美添一樹(著), 山下宏(著)
  • Total Pages: 1-60
  • Publisher: 共立出版