Declaration of Originality

I Jason Carlton declare that this thesis is my own work and has not been submitted in any form for another degree or diploma at any university or other institute of tertiary education. Information derived from the published and unpublished work of others has been acknowledged in the text and a list of references is given in the bibliography.


-------------------

-------------------

Abstract:

Poker is ideal for testing automated reasoning under uncertainty as it an example of a game of incomplete information in which chance plays a role, and offers many properties not exhibited by other games such as chess, checkers and other highly studied games. Uncertainty is introduced through the physical randomness of shuffling, the imperfect information about an opponent's cards and the limited information available to construct psychological models of opponents and their playing behaviours. There are many benefits to be gained from designing and experimenting with poker playing programs. It is a game where competing agents must understand estimation, prediction, risk management, deception, and agent modeling.

This thesis describes work done on improving the knowledge representation, betting strategy, and opponent modeling of the Bayesian Poker Player (BPP), a poker-playing program developed at Monash University, which uses Bayesian networks to model the program's poker hand, the opponent's hand and the opponent's playing behaviour. An approach to model the dependence between opposing players hands is introduced as well as the various techniques used to refine and improve the way that the game is represented by the program. The opponent modeling aspect of the program is also of specific focus, with a number of beneficial improvements presented. Most significantly, a new betting strategy is described which makes use of decision networks for action selection. Experimental results show that a number of these enhancements represent a major advance in the strength of BPP and a discussion of the relative merits of each feature as well as a number of likely ways of improving play are provided.