Can opponent models aid poker player evolution?

We investigate the impact of Bayesian opponent modeling upon the evolution of a player for a simplified poker game. Through the evolution of artificial neural networks using NEAT we create and compare players both utilizing and ignoring Bayesian opponent beliefs. We test the effectiveness of this model against various collections of dynamic and partially randomized opponents and find that using a Bayesian opponent model enhances our AI players even when dealing with a previously unseen collection of players. We further utilize the inherent recurrency of our evolved players in order to recognize the opponent models of multiple players. Through ablative studies upon the inputs of the network, we show that utilization of an opponent model as an evolutionary aid yields significantly stronger players in this case.