Categorization as nonparametric Bayesian density estimation

Abstract

<jats:title>Abstract</jats:title> <jats:p>The authors apply the state of the art techniques from machine learning and statistics to reconceptualize the problem of unsupervised category learning, and to relate it to previous psychologically motivated models, especially Anderson's rational analysis of categorization. The resulting analysis provides a deeper understanding of the motivations underlying the classic models of category representation, based on prototypes or exemplars, as well as shedding new light on the empirical data. Exemplar models assume that a category is represented by a set of stored exemplars, and categorizing new stimuli involves comparing these stimuli to the set of exemplars in each category. Prototype models assume that a category is associated with a single prototype and categorization involves comparing new stimuli to these prototypes. These approaches to category learning correspond to different strategies for density estimation used in statistics, being nonparametric and parametric density estimation respectively.</jats:p>