Conditioned stimulus duration in classical trace conditioning: test of a real-time neural network model.

In classical trace conditioning, the interstimulus interval (ISI) is equal to the conditioned stimulus (CS) duration plus the trace interval (TI), the interval between CS offset and unconditioned stimulus (US) onset. The Sutton-Barto-Desmond neural-network model of classical conditioning predicts that, with a sufficiently long TI, conditioning will be faster with a CS of relatively long duration than with one of shorter duration. This prediction is illustrated with simulations and tested with the rabbit nictitating membrane response. Animals were trained with a tone CS of 350- or 700-ms duration. The TI was fixed at 300 ms, so that the ISI for the two durations was 650 or 1000 ms, respectively. Another factor in the experimental design was tone intensity (63 or 83 dB). Consistent with the model's prediction, conditioning was faster with the longer ISI, but only with the louder tone. The results have implications for computational models of classical conditioning.