1. Computer Science Department & Center for the Neural Basis of Cognition, Carnegie Mellon University, Pittsburgh, PA 15213, USA
2. Present address: Center for Neural Science, New York University, New York, New York, USA (Y.K.); Electrical Engineering and Computer Science Department, Case Western University, Cleveland, Ohio, USA and Wissenschaftskolleg (Institute for Advanced Study) zu Berlin, Germany (M.S.L.).

Correspondence to: Yan Karklin^1,2Michael S. Lewicki^1,2 Correspondence and requests for materials should be addressed to Y.K. (Email: yan.karklin@nyu.edu) or M.S.L. (Email: michael.lewicki@case.edu).

Abstract

A fundamental function of the visual system is to encode the building blocks of natural scenes—edges, textures and shapes—that subserve visual tasks such as object recognition and scene understanding. Essential to this process is the formation of abstract representations that generalize from specific instances of visual input. A common view holds that neurons in the early visual system signal conjunctions of image features^{1, 2}, but how these produce invariant representations is poorly understood. Here we propose that to generalize over similar images, higher-level visual neurons encode statistical variations that characterize local image regions. We present a model in which neural activity encodes the probability distribution most consistent with a given image. Trained on natural images, the model generalizes by learning a compact set of dictionary elements for image distributions typically encountered in natural scenes. Model neurons show a diverse range of properties observed in cortical cells. These results provide a new functional explanation for nonlinear effects in complex cells^{3, 4, 5, 6} and offer insight into coding strategies in primary visual cortex (V1) and higher visual areas.

MORE ARTICLES LIKE THIS

These links to content published by NPG are automatically generated.