Categorization as probability density estimation (Q1899553)

First, the basic structure of categorization models is described in terms of their representation, category access, and response selection assumptions. Next, we elaborate on the notion of categorization as probability density estimation. We discuss the differences between classifiers that know the true category pdfs and classifiers that must estimate these functions from the trial-by-trial feedback provided by the experimenter. It is shown that consistency is the key statistical property that guarantees two such classifiers will reasonably agree. The third section defines parametric and nonparametric pdf estimation and relates these statistical terms to the categorization process. The fourth section discusses parametric categorization models. In particular, it is shown that the prototype model and several decision-bound models are parametric. The fifth section shows that most popular exemplar models of categorization are nonparametric. In particular, we show that they are equivalent to a classifier that uses the minimum variance unbiased estimator of the category base rates and a nonparametric pdf estimator that is one of the most widely used estimators of professional statisticians. We also show that in most applications these models predict essentially optimal performance. Finally, the last section discusses implications of these results and suggests an alternative approach to studying human categorization performance.