An adaptive psychophysical method for subject classification

In psychophysical experiments, one's goal is usually to measure some continuous parameter hypothesized to determine the statistical properties of a subject's responses. Methods are well developed that adaptively manipulate stimulus characteristics in such a way that the reliability of the parameter estimate is maximized. However, such methods are inapplicable in situations in which the goal is to assign subjects to discrete categories, rather than to measure a continuous parameter. This paper introduces a technique that is directly applicable to efficient categorization and that adaptively manipulates stimulus characteristics in such a way that the information obtained from each trial is maximized. This technique is based on the principle of minimum estimated expected entropy, whereby stimulus parameters on each trial are chosen in order to minimize the estimated expected entropy of the a posteriori probability distribution that expresses how likely a subject is to belong to each of a group of mutually exclusive categories. A sample implementation of the technique--the classification of infant subjects according to their audiograms--is then described and evaluated via computer simulation.