Simulating animal conditioning : investigating Halperin's neuro-connector model

• Main Author: Hallam, B.
• Published: University of Edinburgh 2001
• Subjects: 006.3
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.651976

This dissertation describes simulation work which starts an investigation into Halperin's [1990] claim that her Neuro-Connector model is 'plausible as a general model of vertebrate learning'. One way of testing a claim such as this is to examine the model's ability to reproduce learning phenomena which are <i>not</i> part of the model's development history. Choosing a phenomenon which is <i>not</i> one of those that the model was specifically built to reproduce tests the model's ability to generalise. This gives creedance to the claim that the model represents or explains some actual underlying mechanism. This implementation was made to produce results similar to those of standard animal experiments for the following classical conditioning phenomena: acquisition, extinction, reacquisition of an extinguished response, the effect of interruptions, changing the inter-stimulus interval, the effect of closely-massed trials, pre-exposure to the CS, second-order conditioning, and postponed conditioning. However, in some cases implausible assumptions or starting conditions are necessary in order to achieve these results. The implementation could not produce results mirroring those of animal for the effects of partial reinforcement, variations in US duration, pre-exposure to the US, or varying the trace interval. It can produce backward conditioning under some circumstances; backward conditioning is a source of debate amongst animal researchers. Other classical conditioning phenomena were not investigated. The simulation only implements the main features of the model. An implementation of the full model should be able to reproduce sensory pre-conditioning and probably US duration effects despite these not being possible with the implementation used. Preliminary investigation of instrumental conditioning showed that reward training and an approximation to punishment training were reproducible. Avoidance training and omission training were not possible due to the model failing to deal with the non-occurrence of expected events. The biological plausibility of model and implementation are discussed in the light of the findings reported in this dissertation which, though only a beginning, nevertheless provides an unusually complete examination of the plausibility of the model and a strong experimental basis for its assessment. Some suggestions for improvements to the model are given.