Blackjack as a test bed for learning strategies in neural networks

Blackjack or twenty-one is a card game where the player attempts to beat the dealer, by obtaining a sum of card values that is equal to or less than 21 so that his total is higher than the dealer´s. The probabilistic nature of the game makes it an interesting test bed problem for learning algorithms, though the problem of learning a good playing strategy is not obvious. Learning with teacher systems are not very useful since the target outputs for a given stage of the game are not known. Instead, the learning system has to explore different actions and develop a certain strategy by selectively retaining the actions that maximize the player´s performance. The paper explores the use of blackjack as a test bed for learning strategies in neural networks, and specifically with reinforcement learning techniques. Furthermore, performance comparisons with previous related approaches are also reported