Keywords

global motion perception, psychophysics, mixture model, random-dot kinematograms

Abstract

Mixture models are well known in cognitive psychology, less so in vision. Are there cases where the data allow clear testing as to whether different strategies are employed in a task? Most psychophysical measurements manipulate a single staircase variable to map out a monotonic increasing function, but if performance is limited by different mechanisms over the range of the variable, classical fitting could be inappropriate. We present a data set and analyses that confirm the presence of two visual strategies addressing the same task, with the choice of strategies depending on the staircase variable. In a net-motion discrimination task, stimuli were random-dot kinematograms with a fixed percent coherence, and the contrast ratio between signal and noise dots was the staircase variable. When signal and noise dots have similar contrast, the most effective strategy is to average all local motion vectors across the display. However, low contrast ratios enable a selective-tracking strategy in which attending to the dim signal dots makes them easier to detect, which creates a positive feedback loop driving the signal dot contrast further down until a second threshold is reached.

Start Date

11-5-2016 4:25 PM

End Date

11-5-2016 4:50 PM

Share

COinS
 
May 11th, 4:25 PM May 11th, 4:50 PM

A mixture model demonstrates use of distinct strategies in a global motion direction task

Mixture models are well known in cognitive psychology, less so in vision. Are there cases where the data allow clear testing as to whether different strategies are employed in a task? Most psychophysical measurements manipulate a single staircase variable to map out a monotonic increasing function, but if performance is limited by different mechanisms over the range of the variable, classical fitting could be inappropriate. We present a data set and analyses that confirm the presence of two visual strategies addressing the same task, with the choice of strategies depending on the staircase variable. In a net-motion discrimination task, stimuli were random-dot kinematograms with a fixed percent coherence, and the contrast ratio between signal and noise dots was the staircase variable. When signal and noise dots have similar contrast, the most effective strategy is to average all local motion vectors across the display. However, low contrast ratios enable a selective-tracking strategy in which attending to the dim signal dots makes them easier to detect, which creates a positive feedback loop driving the signal dot contrast further down until a second threshold is reached.