Proof Number Based Monte-Carlo Tree Search

Elliot DoeMark H. M. WinandsJakub KowalskiDennis J. N. J. SoemersDaniel G\'orskiCameron Browne

Mar 2023

摘要

This paper proposes a new game search algorithm, PN-MCTS, that combinesMonte-Carlo Tree Search (MCTS) and Proof-Number Search (PNS). These twoalgorithms have been successfully applied for decision making in a range ofdomains. We define three areas where the additional knowledge provided by theproof and disproof numbers gathered in MCTS trees might be used: final moveselection, solving subtrees, and the UCT formula. We test all possiblecombinations on different time settings, playing against vanilla UCT MCTS onseveral games: Lines of Action ($7$$\times$$7$ and $8$$\times$$8$), MiniShogi,Knightthrough, Awari, and Gomoku. Furthermore, we extend this new algorithm toproperly address games with draws, like Awari, by adding an additional layer ofPNS on top of the MCTS tree. The experiments show that PN-MCTS confidentlyoutperforms MCTS in 5 out of 6 game domains (all except Gomoku), achieving winrates up to 96.2% for Lines of Action.

展开全部

图表提取