Abstract
Categorization refers to the process of mapping sensory inputs onto discrete concepts. Humans and other animals can readily learn arbitrary categories defined by low-level visual features such as hue, and behavioral studies indicate that such learning distorts perceptual sensitivity for the category-defining feature such that discrimination performance for physically similar exemplars from different categories is enhanced and discrimination performance for equally similar exemplars from the same category is reduced. These distortions could result from changes in how sensory neural populations selective for category-defining features encode information. Here, we tested this possibility by using noninvasive measurements of human brain activity (fMRI and EEG) to visualize and quantify population-level representations of oriented stimuli encoded by early visual cortical areas after participants had learned to classify these stimuli into discrete groups. Representations of orientation encoded by visual areas V1-V3 were systematically biased by category membership, as indicated by shifts in the representation away from the physical stimulus’ orientation and towards the center of the appropriate category. These shifts were strongest for orientations near the category boundary where they would be most beneficial for behavioral performance, predicted participants’ overt category judgments, and emerged within a few hundred milliseconds of stimulus onset. Collectively, these results suggest that categorizing a stimulus alters how that stimulus is represented at the earliest stages of the visual processing hierarchy, and may provide a physiological basis for distortions in perceptual sensitivity following category learning.
Data Availability All data and analytic software are publicly available on the Open Sciences Framework at https://osf.io/xzay8/
Significance Statement Category learning alters perceptual sensitivity by enhancing the discriminability of similar exemplars from different categories. These distortions could in part reflect changes in how sensory neural populations selective for category-defining features encode information. To test this possibility, we used multivariate analytical techniques to reconstruct and quantify representations of oriented stimuli after observers had learned to classify them into two discrete groups. Representations of orientation encoded by several early visual areas were systematically biased according to their category membership, with larger biases for orientations adjacent to the boundary that defined each category. This result suggests that categorizing a stimulus alters how that stimulus is represented at the earliest stages of the visual processing hierarchy.