Event Details

The brain uses its prior knowledge of the world to help resolve ambiguities in the sensory information. We study the neural mechanisms of this disambiguation process from a machine learning perspective in humans and monkeys, in the specific context of a categorization task. Subjects learn, from scratch, precisely defined categories of naturalistic objects. We systematically vary the category information, and study its effect on neural and behavioral responses.  We find that neural responses in various regions in the temporal lobe and the visual corticostriatal loop parallel the learning‐dependent disambiguation at the perceptual level. In monkeys, the behavior of individual neurons and neuronal ensembles is often consistent with theoretical predictions, and reliably predicts the animal’s performance. Often, what the neurons learn and retain about one pair of categories can have unexpected effects on the categorization of objects along an orthogonal shape axis.