Machine learning of visual object categorization: an application of the SUSTAIN model

Abstract

Formal models of categorization are psychological theories that try to describe the process of categorization in a lawful way, using the language of mathematics. Their mathematical formulation makes it possible for the models to generate precise, quantitative predictions. SUSTAIN (Love, Medin & Gureckis, 2004) is a powerful formal model of categorization that has been used to model a range of human experimental data, describing the process of categorization in terms of an adaptive clustering principle. Love et al. (2004) suggested a possible application of the model in the field of object recognition and categorization. The present study explores this possibility, investigating at the same time the utility of using a formal model of categorization in a typical machine learning task. The image categorization performance of SUSTAIN on a well-known image set is compared with that of a linear Support Vector Machine, confirming the capability of SUSTAIN to perform image categorization with a reasonable accuracy, even if at a rather high computational cost.