An investigation into hybrid artificial intelligence techniques applied to the game of Go

Churchill, JJ, 2006. An investigation into hybrid artificial intelligence techniques applied to the game of Go. PhD, Nottingham Trent University.

Preview

Text 10183426.pdf - Published version Download (29MB) | Preview

Abstract

The use of Artificial Intelligence (AI) techniques has allowed many advances in computing and increasingly permeates modern computing applications within industry, commerce and entertainment. This thesis describes an investigation into hybrid AI techniques which focuses on applying these techniques to the game of Go, a two-player game of zero chance, and looks into how to combine AI techniques in a complimentary and effective manner. The hybrid approach was driven by the complexity of Go, which whilst being simple to learn and having very few rules, remains an incredibly difficult game to master. This applies even more so for a computer since the brute force approach which worked so well for computer Chess fails to produce even moderate Go program. Therefore a further level of intelligence is required; the program must be smarter, more efficient and contain within it many abstract concepts that human beings so readily absorb.

The thesis is roughly split into two investigations involving the fields of Artificial Neural Networks (ANNs) and Genetic Algorithms (GAs). The former has shown some promise in the field of computer Go and naturally lends itself to be included as a component in a larger AI system. Different training methods and algorithms are compared with reference to Go and several network architectures of varying levels of complexity are tested. The ANNs developed are also used within a Minimax framework to show the improvement over a state of the art game tree search algorithm.

Later chapters discuss a GA-oriented approach to Go which resulted from attempts to create a flexible humanistic playing algorithm based on game tree search. This was initiated after the limitations of a standard brute force approach, even with the addition of ANN encapsulated knowledge, were fully realised. Additionally the ANNs developed earlier in the thesis are used to incorporate expert knowledge within the GA.

Item Type:	Thesis
Creators:	Churchill, J.J.
Date:	2006
ISBN:	9781369316278
Identifiers:	Number Type PQ10183426 Other
Divisions:	Schools > School of Science and Technology
Record created by:	Linda Sullivan
Date Added:	25 Sep 2020 14:46
Last Modified:	07 Sep 2023 09:15
URI:	https://irep.ntu.ac.uk/id/eprint/40957

Actions (login required)

Edit View

Statistics

Views

Downloads