[4B3-GS-1] Fundamental AI, theory (2)

[4B3-GS-1-04] Automatic adjustment of decay factor in Tree Reuse MCTS

〇Kota Yokogawa¹, Koichi Moriyama¹, Atsuko Mutoh¹, Tohgoroh Matsui², Nobuhiro Inuzuka¹ (1. Nagoya Institute of Technology, 2. Chubu University)

Keywords:Monte-Calro Tree Search

MCTS is a best-first search algorithm that gradually expands a search tree based on the results of random searches. One of its extensions is called "Tree Reuse". It stops repetitive searches by leaving a subtree from a child node that will be the root node next. However, it causes a problem in a non-deterministic environment where same choices give different results. It is often alleviated by attenuating the information the subtree has, but it is difficult to know in advance the attenuation level, called a decay factor, appropriate in the environment. Therefore, this work proposes a method automatically adjusting the decay factor using information obtained from the environment. Experiments using multiple game environments show that the proposal is better than the conventional Tree Reuse MCTS in the finish rate of the games.