Year: 2012
Type: dissertation
Abstract: The aim of General Game Playing (GGP) is to create intelligent agents that automatically learn how to play many different games at an expert level without any human intervention. One of the main challenges such agents face is to automatically learn knowledge-based heuristics in real-time, whether for evaluating game positions or for search guidance. \nIn this thesis we approach this challenge with Monte-Carlo Tree Search (MCTS), which in recent years has become a popular and effective search method in games. For competitive play such an approach requires an effective search-control mechanism for guiding the simulation playouts. In here we describe our GGP agent, CADIAPLAYER, and introduce several schemes for automatically learning search guidance based on both statistical and reinforcement learning techniques. Providing GGP agents with the knowledge relevant to the game at hand in real time is, however, a challenging task. This thesis furthermore proposes two extensions for MCTS in the context of GGP, aimed at improving the effectiveness of the simulations in real time based on in-game statistical feedback. Also we present a way to extend MCTS solvers to handle simultaneous move games. Finally, we study how various game-tree properties affect MCTS performance.
Source: Opin vísindi (Opin vísindi)
Author: Hilmar Finnsson
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W1584247264