Exact and Monte-Carlo algorithms for combinatorial games
Year: 2014
Type: dissertation
Abstract: This thesis concerns combinatorial games and algorithms that can be used to play them.Basic definitions and results about combinatorial games are covered, and an implementation of the minimax algorithm with alpha-beta pruning is presented.Following this, we give a description and implementation of the common UCT (Upper Confidence bounds applied to Trees) variant of MCTS (Monte-Carlo tree search).Then, a framework for testing the behavior of UCT as first player, at various numbers of iterations (namely 2,7, ... 27), versus minimax as second player, is described.Finally, we present the results obtained by applying this framework to the 2.2 million smallest non-trivial positional games having winning sets of size either 2 or 3.It is seen that on almost all different classifications of the games studied, UCT converges quickly to near-perfect play.
Author: Anders Leino
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W318077175