Supervised reinforcement learning in discrete environment domains
B Jensen, Daniel Ortíz-Arroyo, N Cruz-Corte, F Rodri · 2010
This paper describes a supervised reinforcement learning-based model for discrete environment domains. The model was tested within the domain of backgammon game. Our results show that a supervised actor-critic based learning model is capable of improving the initial performance and then eventually reach similar performance levels as those obtained by TD-Gammon, an artificial neural network player (ANN) trained by temporal differences.