A comparison of some neural network models of classical conditioning

Classical conditioning is a form of temporal learning that may be useful in intelligent control. Three neural network models of classical conditioning are compared: the Sutton-Barto model, the Klopf model, and the Grossberg-Schmajuk model. All are based on Hebbian learning, but they differ in how events are remembered. Although these models can learn to associate two events occurring at different times, they also show behaviors that may not be satisfactory in an intelligent control system. It is concluded that these models of classical conditioning fall far short of being good learners of temporal associations.<
>