December 2022
Volume 22, Issue 14
Open Access
Vision Sciences Society Annual Meeting Abstract  |   December 2022
Optimizing the classification of observers into distinct and diverse categories
Author Affiliations
  • Nicolaas Prins
    University of Mississippi
Journal of Vision December 2022, Vol.22, 4397. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Nicolaas Prins; Optimizing the classification of observers into distinct and diverse categories. Journal of Vision 2022;22(14):4397.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Adaptive psychophysical methods typically optimize the efficiency with which quantitative characteristics of sensory/perceptual performance (e.g., location or slope of the psychometric function, PF) can be estimated. However, when the objective is to classify observers into distinct categories, each typified by a different set of values of the PF parameters, these methods may be sub-optimal because they do not directly optimize classification accuracy per se. A notable exception is the method proposed by Cobo-Lewis (1997, Percept Psychophys, 59, 989) which explicitly optimizes the classification of observers into distinct categories. However, this method has the limitation that fixed values must be assumed for the parameters of the psychometric functions that typify each category. Here I propose a method that combines aspects of the psi-marginal method (Prins, 2013, J. Vis., 13, 3) and the method proposed by Cobo-Lewis. As in Cobo-Lewis, the method has the direct goal of minimizing the entropy (i.e., uncertainty) in the discrete category membership parameter. However, unlike Cobo-Lewis, categories of observers are not specified by fixed, archetypal PFs but instead by prior distributions over the PF parameters, allowing the specification of distinct, yet internally diverse, observer categories. Because the to-be-minimized entropy is calculated based on the posterior distribution in which all parameters except category membership have been marginalized, the method seeks to optimize the estimation of category membership only and will not attempt to optimize the estimation of the PF parameters per se. Simulations indicate that when categorical populations differ with respect to the slope of the PF the proposed method can reach 90% categorization accuracy in as few as one half of the number of trials required by a method that explicitly optimizes estimation of PF parameters. More modest results are obtained when categories differ only with respect to the location parameter of the PF.


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.