A Computational Model of the Visual Oddity Task
Andrew Lovett, Kate Lockwood, Kenneth D. Forbus · 2008
Understanding how high-level visual properties are computed is a central problem in perception. Oddity tasks, where participants must identify a stimulus that is distinct in some way from others in an array, provide a method for determining what features are being computed. We describe a computational model of oddity detection that models data by Dehaene et al. (2006) on perception of simple geometric shapes. It starts with virtually the same input stimuli as given to human subjects, and automatically constructs representations. Oddity detection is accomplished by analogical processing, using SME and SEQL. The simulation is able to perform the task, and moreover, provides some insight as to what makes one problem harder than another.