Friday, 5 June 2015

A Guest Presentation by Steve Blurton, Carsten Søren Nielsen, and Søren Kyllingsbæk, on visual identification.

The Cognitive Research Group is proud to present a talk by some research guests, Steve Blurton and Carsten Søren Nielsen (University of Copenhagen).

TITLE: A Poisson Random Walk Model for Response Time and Pure Accuracy Tasks

WHEN: Thursday 11th June 12-1pm.

WHERE: Keats Reading Room (AVLG17)

ABSTRACT: Based on a simple ‘what first comes to mind’ rule, a Theory of Visual Attention (TVA; Bundesen, 1990) has been successful in explaining human performance in pure accuracy tasks with non-confusable stimuli. However, for mutually confusable a ‘what has the most evidence’ rule is more suited (Kyllingsbæk et al., 2012). Based on this work we propose and test a common model of the time course of visual identification of mutually confusable single stimuli in two-alternative, response time and pure accuracy tasks. The central model assumption is that during the analysis of a single stimulus in the visual field, tentative evidence for one of two categorizations of the stimulus is generated by a Poisson process at a constant rate in such a way that a tentative categorization automatically counts against the other categorization. Visual identification is thus assumed to follow a simple random walk with exponential distributed interstep times. An identification is conclusively made if and when evidence reaches one of two thresholds. If a threshold is not reached before the analysis is stopped, then an informative guess will be made based on ‘what has the most evidence’. One important question that is to be addressed in an application of the model is whether it is possible to identify invariances of model parameters across conditions of pure accuracy task and speeded responses. With Poisson rate estimates being in the same range across conditions our common model provides a close fit to individual data on identification of Gabor patches in a two-alternative, response time and pure accuracy task.