70% stars - 3077 reviews

  • Game theory and AI: a unified approach to poker games

    About The File:
    File Format | PDF
    File Size | 1.1 MB
    Pages | 102
    Language | English
    Category | Poker
    Description:  This thesis focuses on decision making in partially observable card games and, in particular, poker games. An attempt is made to outline both the game theoretic, as an agent-centric approach to such games, analyzing differences and similarities, as well as strong and weaker points and finally proposing a view to make a tradeoff between these. The game theoretic approach for this type of games would specify a Nash- equilibrium, i.e., a pair of policies that are a best response to each other. Although a policy found in this way guarantees a minimum payoff, it is conservative in the sense that it is unable to exploit any weaknesses the opponent might have. This motivates an agent-centric perspective, in which we propose modeling a simple poker game as a Partial Observable Markov Decision Process (POMDP) for a player who is playing against a fixed opponent whose policy is known (e.g. by repeated play). The resulting deterministic policy is a best response against the fixed opponent policy. Such a best-response policy does exploit weaknesses in the opponent's policy, thus yielding the maximum payoff attainable. In order for the results obtained for such a simplified poker game to be of significance for real-life poker games, various methods for dealing with large (PO)MDPs are treated. These could be used to tackle larger games using the best-response approach.
    Download | Free
    Share the Book
    ​