Artificial Intelligence-Creating World Class Game Playing Agent
The intended project is creation of a world class poker-playing agent using the Counterfactual Regret Minimization (CFRM) algorithm pioneered by the University Of Alberta's Computer Poker Research Group. The poker variation is "One-on-one No Limit Texas Hold'em".
The project will be based primarily on the following papers:
1. "Robust Strategies and Counter-Strategies: Building a Champion Level Computer Poker Player"- Michael Bradley Johanson, 2007
2. "Regret Minimization in Games with Incomplete Information"
These papers will be used as a blueprint for implementation of the CFRM to the game of No Limit Texas Hold'em. The winning bidder will read and familiarize themselves with these papers, as well as other hand picked important academic work, and then implement the process described in these publications.
Desired features for the agent include :
• Ability to display statistics (for example, its own probability distribution over
actions for a specific information set)
• Ability to determine its exploitability in its own abstraction
• Ability to work with different bucketing techniques
• Ability to work with various abstractions
• Distributed implementation
• Academic background
• Experience in Game Theory
• Strong mathematical and analytical skills
• Fluency in Probabilities Theory is a big plus
• Knowledge or experience with poker is a plus
The information in the provided papers includes:
• Detailed description of the CFRM technique
• General description of the data representation for the poker game
• General description of the abstractions used for poker agents
• General description of bucketing techniques
• Sample pseudocode for tree traversal for the algorithmic implementation of
the CFRM technique (calculation of utilities)
Information required but not provided in the papers
The papers do not include the following required information :
• Precise “One-on-one No Limit Texas Hold’em” rules
• Algorithm for the calculation of hands’ winning probabilities (this includes
preflop 2-card hands up to 7-card hands after the river)
• Hand strength chart
The missing information can be obtained without difficulty. An extensive amount of information has been published on using the CFRM in Texas Hold'em games and I can provide direction in finding necessary missing pieces. Although the exact algorithm is not provided, it can be replicated by following the description in the papers.
I have included the the second of the two mentioned papers above for reference and have other recommended resources if necessary. If there are any questions remaining regarding the scope of the project or my goals please do not hesitate to ask and I will provide more clarity. This project provides an opportunity to work on cutting edge applications of Artificial Intelligence techniques along with interesting optimization challenges. New ground will be broken on this project in both the realms of artificial intelligence and poker research. Similar additional projects will be available should this first project be a success.
Thank you for your consideration.