Application of Neuroevolution in Blackjack

Abstract

Blackjack is one of the few casino games with an extremely low house edge. In the past, many brute force simulations have been done to derive basic strategy. Classical brute force methods are tedious, time consuming, and often require hundreds of millions of games played to achieve results. In this project, I use reinforcement learning, specifically neuroevolution (NE), which is an attempt to simulate biological evolution, to see if an artificial neural net (ANN) can evolve to learn basic strategy and achieve the theoretical maxima provided by a basic strategy simulation. Two main simulations are run in this project, one using basic strategy charts and the other using the evolved ANN. These are then compared to see how effective the ANN was in learning strategy as well as how quickly it was able to learn.