Off-campus UMass Amherst users: To download dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.
Non-UMass Amherst users, please click the view more button below to purchase a copy of this dissertation from Proquest.
(Some titles may also be available free of charge in our Open Access Dissertation Collection, so please check there first.)
Computational and behavioral investigations of real-time models of classical conditioning
Abstract
Real-time models of classical conditioning were evaluated using simulation and behavioral techniques. Performance criteria included ability to account for aspects of the conditioned response (CR) during conditioning of the rabbit nictitating membrane response (NMR), such as development of CR topography and the dependency of conditioning on the interstimulus interval (ISI) elapsing between onsets of conditioned and unconditioned stimuli (CSs and USs). The models investigated included the Sutton-Barto-Desmond (SBD) model (Moore, Desmond, Berthier, Blazis, Sutton, and Barto, 1986), which predicts CR topography, and the Temporal Differences (TD) model (Sutton and Barto, 1987), which does not. Simulation studies addressed the impact of modifying the TD model to describe topography. The resultant model, called the TD$\sb{RTS}$ model, replicates the learning phenomena described by the TD model but is highly parameter-sensitive. Simulation studies showed the performance of the TD, TD$\sb{RTS}$, and SBD models to be similar in most respects. However, model predictions differ for simultaneous and backwards conditioning protocols involving single and multiple CSs. Behavioral verification of new predictions of the models was carried out using rabbit NMR conditioning. Experiment 1 showed that with fixed-trace-interval conditioning, conditioning increases with CS duration, a result consistent with the SBD but not the TD model. Experiment 1 also showed a novel CS intensity by ISI interaction that none of the models explain. Experiment 2 replicated Experiment 1 and analysed response topography to illustrate problems caused by the CS representation assumed by the models. Experiments 3 and 4 investigated compound conditioning using one CS followed by simultaneous presentation of a second CS and US. Experiments 3 and 4 showed that conditioning to the compound occurred more readily than in controls when the first CS was less salient than the second. Experiment 3 presented weak evidence that the simultaneous CS was inhibitory. Of the models explored, the TD model most fully accounts for the results of Experiments 3 and 4. These computational and behavioral experiments suggested that the models might be better served by alternative CS representations, and highlighted the need for a resolution of issues in simultaneous conditioning. These points and future directions are discussed.
Subject Area
Experimental psychology|Psychobiology|Computer science|Psychology
Recommended Citation
Blazis, Diana Evelyn-Jennings, "Computational and behavioral investigations of real-time models of classical conditioning" (1990). Doctoral Dissertations Available from Proquest. AAI9035381.
https://scholarworks.umass.edu/dissertations/AAI9035381