A new associative conditioning element for consummatory classical conditioning


Rogers, Brendan L.

Available versions


This thesis documents the development, operation, and performance of a new nonlinear analog system designed to exhibit a wide range of (mainly) consummatory classical conditioning behaviour. This new system, referred to hereafter as the Associative Conditioning Element (ACE), may be categorised as a functioning Artificial Neural Network (ANN) element of unusually large mechanistic and behavioural complexity. However, depending upon the perspective of the reader, ACE may also be considered as a neuronal model of classical conditioning, or as a psychological theory of animal learning and memory. The performance specifications for ACE consist entirely of empirical results from behavioural animal experiments, primarily focussing upon the Nictitating Membrane (third eyelid) Response (NMR) of the rabbit. However, the widespread applicability of classical conditioning means that ACE'S behaviour also generalises to many other response systems and animal species, and where appropriate NMR results are not available, those from other response systems are utilised. ACE is assessed by comparing the results of computer simulations of its operation with corresponding empirical results. ACE produces behaviour attributable to a single biological neuron, or small group of neurons. Despite the widespread ANN view that neural systems consist of densely interconnected networks of very simple elements, the demonstration of basic classical conditioning behaviour by presynaptic facilitation in the marine mollusc Aplysia, and the sparsely interconnected organisation of relatively humble creatures in general, suggests that individual neurons may indeed have considerable information processing capability. ACE begins to bridge the gap between the excessive simplicity of standard ANN elements, and the considerable complexity of biological neurons which is only beginning to be understood and appreciated. An exploration of the functional relationship between memory and learning has yielded a new nonlinear subsystem of interacting CS-specific (or synaptic) memory types, collectively referred to here as the Neural Multiprocess Memory Model (NMMM). The NMMM is progressively developed from the standard ANN adaptive synaptic weighting, initially producing spontaneous regression and recovery behaviour, then U-shaped memory retention, and finally comprehensive adaptive associability behaviour. The adaptive associability mechanism supports both negatively accelerated and sigmoidal acquisition curves, latent inhibition, learned irrelevance, the Partial Reinforcement Effect (PRE), and accelerated learning following alternating acquisition/extinction training sessions. A new nonlinear Conditioned Stimulus Trace Circuit (CSTC) has been developed with sufficient complexity to accurately model the shape of the mean NMR topography, and the way in which it changes subject to variations in CS duration and amplitude. This CS 'trace' is available for generation of an appropriately timed Conditioned Response (CR), and to selectively gate the effect of experience subsequent to CS presentation upon the associative strength of the CS. The latter role enables the acquisition of a CR which is timed to peak approximately at the time reinforcement is expected, the production of anticipatory CRs, and the implementation of trace conditioning. An associative Short Term Memory (STM) is made available by the NMMM which supports a modified intratrial version of the Rescorla-Wagner model, in which expectation of reinforcement and actual reinforcement experienced are able to be compared to determine change in associative strength, despite the asynchronous nature of CS and US presentation, and their different temporal profiles. This allows ACE to exhibit behaviour such as stimulus amplitude effects, acquisition of conditioned excitation and inhibition, extinction, overshadowing, compound conditioning effects (e.g., blocking and unblocking), and discriminative stimulus effects. And finally, the learning rules controlling changes in associative STM also support the nonassociative phenomena of secondary extinction, and reinstatement. As a result, ACE is able to capture a sense of the prevailing temporal context, and so, for example, support the acquisition of a conditioned inhibitor by successive discrimination, without relying upon conditioned excitation to background stimuli.

Publication year


Thesis supervisor

John G. Wallace

Publication type

Thesis (PhD)


Copyright © 1991 Brendan L. Rogers.

Thesis note

Submitted in fulfillment of the requirements for the degree of Doctor of Philosophy, Swinburne Institute of Technology, 1991.