Occasion Setting: A Neural Network Approach

Classical conditioning data show that a conditioned stimulus (CS) can act either as a simple CS -eliciting conditioned responses (CRs) by signaling the occurrence of an unconditioned stimulus (US) - or as an occasion setter - controlling the responses generated by another CS. In this article, the authors apply a simple extension of a network model of conditioning, originally presented by N. A. Schmajuk and J. J. DiCarlo (S-D; 1992), to the description of these 2 different CS functions. In the model, CS inputs are connected to the CR output both directly and indirectly through a hidden-unit layer that codes configurai stimuli. In this framework, a CS acts as (a) a simple stimulus through its direct connections with the output units and as (b) an occasion setter through its indirect configural connections via the hidden units. Computer simulations demonstrate that the network accounts for a large part of the data on occasion setting.