Opponent Modelling in Poker
Finnegan Southey (University of Alberta, Canada)
NICTA SML SEMINARDATE: 2005-11-15
TIME: 09:00:00 - 10:00:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Poker is a domain that captures many of the aspects of interesting real-world problems: stochasticity, partial observability, limited data, high variance, and real-time performance constraints. Uncertainty in poker stems from two key sources, the shuffled deck and an adversary whose strategy is unknown. One approach to playing poker is to find a pessimistic game theoretic solution (i.e., a Nash equilibrium), but human players have idiosyncratic weaknesses that can be exploited if some model or counter-strategy can be learned by observing their play. However, games against humans last for at most a few hundred hands, so learning must be very fast to be useful.
I will present recent work on applying Bayesian methods to opponent modelling in poker to exploit opponent weaknesses, with an emphasis on methods that can perform in real-time. The overall approach addresses two key subproblems: (i) inferring a posterior over opponent strategies given a prior distribution and observations of their play, and (ii) playing an appropriate response to that distribution. Results are shown on both a small, research version of the game, and on a large, real-world version as well.
BIO:
Finnegan Southey is a post-doctoral fellow at the University of Alberta, Department of Computing Science and is part of the Alberta Ingenuity Centre for Machine Learning. He obtained his doctorate from the University of Waterloo in 2004. He works in several areas, including statistical machine learning, satisfiability, constrained optimization, and planning. Much of his recent work has been applied to games, including commercial video games and poker.
