Inverse game theory: learning the nature of a game through play

Abstract

ABSTRACT Real world adversarial dynamics such as those encountered in Computer and Network security require modelswhich allow for both imperfect and incomplete information. Recently game theoretic models and speci callysignaling games have been at the forefront of interest for modeling these scenarios. We propose a modi cationof signaling games, a type of Bayesian game, which we believe can serve as a model for these scenarios. Byincorporating real world data into the model, these games could allow interested parties to learn the true natureof the game that they are already playing - though without the rulebook.Keywords: Bayesian games, imperfect information, incomplete information, signaling games, game theory 1. INTRODUCTION Inverse game theory involves the learning of the nature of a game. In certain scenarios the payo s, identities,and/or actions available to the players may be unavailable. Our previous work on the same topic involved theuse of a multiplayer betting game to discover the utility functions of opponents through play, and exploiting thatknowledge to improve the player's winnings.