Representational shifts during category learning

Abstract

Prototype and exemplar models form two extremes in a class of mixture model accounts of human category learning. This class of models allows flexible representations that can interpolate from simple prototypes to highly differentiated exemplar accounts. We apply one such framework to data that afford an insight into the nature of representational changes during category learning. While generally supporting the notion of a prototype-to-exemplar shift during learning, the detailed analysis suggests that the nature of the changes is considerably more complex than previous work suggests.