Contrasting predictive coding and representational accounts of perceptual organization, stimulus predictability, and attentional modulation in the visual cortical hierarchy.

A predictive coding (PC) framework assumes that the visual cortex generates and tests hypotheses about the likely causes of retinal stimulation. In doing so, it requires computation of the error between the predicted model and the sensory data. By contrast, a representational framework assumes that the visual cortex builds a surface representation by engaging horizontal and feedback connections that form excitatory feedback loops between cortical areas, without requiring error computation. I review studies on contour integration, illusory contours (ICs), figure-ground segregation, temporal and spatial predictability, and attentional modulation that involve single- and multi-unit recordings of neural activity in the early stages of visual processing, specifically to search for evidence of error computation. In agreement with the representational framework, many studies revealed that the V1-V4 complex forms an excitatory feedback loop that enhances the representation of the global contour or figure and suppresses the background across multiple levels of the visual hierarchy. By contrast, several studies on ICs and figure-ground segregation revealed evidence for error computation as hypothesized by PC. Taken together, the available neuronal data associated with visual processing and attentional modulation do not unequivocally support either of the proposed frameworks.