Retinal Adaptation to the Spatiotemporal Statistical Structure of Visual Input

Abstract

Under natural conditions, the statistical characteristics of sensory input vary considerably. For efficient representation, the sensory systems should adjust their coding rules. The best understood example is retinal adaptation to the mean light level. Recently it was also found that retina adapts to the contrast, the range of the light level variation. Visual input is not a single variable, but composed of many “pixels”, or multiple variables. Even when the mean and the contrast of these variables are constant, the statistical relationship among the variables can vary. Theory predicts that it is again advantageous to adjust coding rules. Since previous experiments suggested that the retinal behavior changes according to such statistical structures, we investigated the class of statistical structures that lead to response adjustments, and the profiles of those adjustments.