Year: 2016
Type: dissertation
Abstract: This thesis uses a novel adaptation of Agent Case Embeddings (ACEs) and hierarchical clustering to perform an analysis on variations of Monte Carlo Tree Search (MCTS). The goal is to demonstrate similarities and differences in capabilities among some of the more popular variations of MCTS. Additionally, three new applications of the MCTS algorithm are proposed and analyzed: dynamic difficulty in games, level map generation, and real parameter optimization. For the dynamic difficulty applications, Elo ratings are used to distinguish the playing ability of different MCTS agents given unequal amounts of time to make play decisions. For level map generation and real optimization, a new MCTS-based algorithm called Multiple Pass MCTS is introduced and shown to have an impact on the quality of solutions. For real parameter optimization two new techniques are introduced in order to discretize the problems to enable the application of MCTS. These are compared with standard methods demonstrating the superiority of MCTS on some problems in a standard test set of optimization problems.
Author: Cameron McGuinness
Institution: University of Guelph
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2230649837
