Representational Issues in Neural Systems: Example from a Neural Network Model of Set-Shifting Paradigm Experiments

Abstract

Experiments in the Set-Shifting Paradigm (SSP) are used to test subjects’ ability to acquire attentional sets and manipulate them. In normal primates, acquisition of discriminations requiring an intra-dimensional shift (IDS) is superior to that requiring an extra-dimensional shift (EDS). Further, damage to the prefrontal cortex is seen to impair EDS performance. We propose a bias theory to account for the IDS-EDS asymmetry. Uniform bias developed for exemplars from the relevant dimension makes it difficult to subsequently acquire exemplars from the previously irrelevant dimension. A neural network model embodying this theory replicates the experimental results. Prefrontal cortex is hypothesised to be involved in reward-based category learning which in turn facilitates EDS performance. Basic processes in neural systems that can support the formation of potentially higher-level representations are pointed out.