Purchase this article with an account.
Bharathi Jagadeesh, Yan Liu, Nicolas Brunet; Implicit measurement of uncertainty during classification of ambiguous photographs. Journal of Vision 2008;8(6):1040. doi: https://doi.org/10.1167/8.6.1040.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Experiments on perception often require decisions about difficult to discriminate stimuli, even when the observer is uncertain. Measurements of uncertainty, in addition to conventional measurement of performance can be useful in understanding perceptual experience. In this report we detail an implicit measurement of uncertainty in non-human primates that reflects the difficulty of discriminating stimuli and seems related to the strength of the neural signal in a discrimination task. We examined behavior and neural responses in inferotemporal cortex in two monkeys during performance of a two-alternative-forced-choice delayed match to sample task (2AFC-DMS task). The task required classification of a sample image as one of two choices; the monkey reported the choice by making a saccade to one of two choice images. Two of the sample images were identical to the two choices, but other samples were difficult to classify because they were altered versions of the choices. Performance with altered samples followed a predictable pattern, in which samples identical to the choices were easiest to classify and performance worsening as the dissimilarity of the samples from the choices increased. To obtain reward, monkeys had to saccade to the correct choice and maintain fixation on that choice for 500 ms. During this period, monkeys aborted over 60% of responses in incorrect easy trials, indicating an awareness that they had chosen incorrectly. The proportion of aborted incorrect trials decreased systematically as a function of the difficulty of the task, indicating decreasing awareness of incorrect responses. We sorted neural responses based on this measure of uncertainty - trials where the monkey aborted trials, compared to completed trials. Neural response selectivity for the sample images was higher in completed trials than aborted trials. These data suggest that the monkeys were aware of uncertainty, and that the uncertainty reflects the neural selectivity for the sample image.
This PDF is available to Subscribers Only