Poker is an imperfect information game that requires
decision-making under conditions of uncertainty, much like
many real-world applications. Strong poker players
have to skillfully deal with multiple opponents, risk management,
opponent modeling, deception and unreliable information. These
features make poker an interesting area for Artificial
Intelligence research. This thesis describes work done on
improving the knowledge representation, betting strategy, and
opponent modeling of *Loki*, a poker-playing program at
the University of Alberta. First, a randomized betting strategy that
returns a *probability triple* is introduced. A probability triple is
a probabilistic representation of betting decisions that
indicates the likelihood of each betting action occurring in a given
situation. Second, real-time simulations are used to compute the
expected values of betting decisions. These simulations use
*selective sampling* to maximize the information obtained with each
simulation trial. Experimental results show that each of these enhancements
represents a major advance in the strength of Loki.

- Contents
- List of Figures
- List of Tables
- Introduction
- Poker
- Loki-1
- Probability triples
- Selective sampling simulation
- Other examples of selective sampling simulation
- Conclusions and future work
- Bibliography
- Table of Abbreviations
- About this document ...