Learning Optimizes Decision Templates in the Human Visual Cortex

Abstract

Summary Translating sensory information into perceptual decisions is a core challenge faced by the brain. This ability is understood to rely on weighting sensory evidence in order to form mental templates of the critical differences between objects. Learning is shown to optimize these templates for efficient task performance [1–4], but the neural mechanisms underlying this improvement remain unknown. Here, we identify the mechanisms that the brain uses to implement templates for perceptual decisions through experience. We trained observers to discriminate visual forms that were randomly perturbed by noise. To characterize the internal stimulus template that observers learn when performing this task, we adopted a classification image approach (e.g., [5–7]) for the analysis of both behavioral and fMRI data. By reverse correlating behavioral and multivoxel pattern responses with noisy stimulus trials, we identified the critical image parts that determine the observers’ choice. Observers learned to integrate information across locations and weight the discriminative image parts. Training enhanced shape processing in the lateral occipital area, which was shown to reflect size-invariant representations of informative image parts. Our findings demonstrate that learning optimizes mental templates for perceptual decisions by tuning the representation of informative image parts in higher ventral cortex.

Attention in the real world: toward understanding its neural basis

“…This is due in part to the learning of features that optimally distinguish the target from the distracter, as reflected in efficient neural representations of target-diagnostic features after training [86]. Additionally, targets are easier to find when the identity or spatial arrangement of the distracter set is familiar [87].…”

Characteristic sounds facilitate object search in real-life scenes

“…Previous visual-only studies have already made a point about the differences in how attention is distributed in naturalistic, real life scenes space compared to simple artificial search displays typically used in psychophysical studies (e.g., Peelen & Kastner, 2014, for a review;Henderson & Hayes, 2017). Given that experience and repetition tends to facilitate visual search (Shiffrin & Schneider, 1977;Evans, Georgian-Smith, Tambouret, Birdwell, & Wolfe, 2013;Kuai, Levi, & Kourtzi, 2013), another important difference could lie in our experience (and hence, predictability) with natural scenes, compared to laboratory displays. In addition, humans can extract abundant information from natural scenes (gist) at a glance, quickly building up expectations about spatial properties and object relationships (Biederman, Mezzanotte, & Rabinowitz, 1982;Greene & Oliva, 2009;Peelen, Fei-Fei, & Kastner, 2009;MacEvoy & Epstein, 2011).…”

Perceptual learning to discriminate the intensity and spatial location of nociceptive stimuli

“…Perceptual learning is typically obtained in an experimental context by providing feedback on discrimination performance, on a trial-by-trial basis1. Several mechanisms underlying perceptual learning have been proposed, from refined encoding of sensory information to improved decision making2345. Interestingly, learning is not necessarily confined to a specific stimulus feature: learning to discriminate one of the features that defines a sensory stimulus (e.g., the orientation of a visual stimulus) sometimes transfers, or generalizes, to a different feature of the same stimulus (e.g., the contrast of the same visual stimulus)67.…”