LAUSR

The presence of stimulus-unrelated covariability in neural responses poses a challenge to models of perception and behavior. Here we review advances in theory and experiments that highlight how noise correlations are implicated in perceptual processes. We focus on the structure of correlations and variations in it to reveal the computations performed by the cortex. We argue that to discover the functional role of correlations in cortical computations, and make predictions about them, we need to harness hierarchical encoding models. We propose that in hierarchical models the source of noise correlations is dual: inference of both task-related and higher-level perceptual variables contribute to their emergence, which is reflected in the task- and stimulus-dependence of the correlation structure.