Adaptation of the Retinal Code: What the Eye Does Not Tell the Brain

Abstract

The optic nerve is a bottleneck for visual information: Nowhere else along the visual pathway is the entire scene represented by such a small number of neurons. Correspondingly, the retina can transmit to the brain only a tiny fraction of the raw visual information that reaches the photoreceptors, and discards the rest. Many aspects of retinal processing can be understood as clever strategies for extracting visual features that are novel and not already available to the brain, either based on earlier transmissions, or based on other parts of the image. “Center-surround” spatial processing, and “high-pass” temporal processing are both adaptations to the average statistical structure of natural scenes and suppress the redundant components of the image. Light adaptation and contrast adaptation serve to match retinal function to changing conditions of illumination, and, again, this entails a rejection of redundant information. Recent work shows that the retina can adapt dynamically to the spatial and temporal statistics of the visual environment, even under conditions of constant mean intensity and contrast. Under many different stimulus environments, this has the effect of enhancing novel stimulus features over predictable ones. Finally, I will speculate how this general strategy of novelty detection could be implemented by modifiable circuits in the inner retina.