Adaptation of large-scale neural populations to natural scene statistics

The idea that sensory neurons exploit the statistical structure of natural stimuli to efficiently transmit information has been a guiding principle in neuroscience for over half a century. This conceptual framework, known as the efficient coding hypothesis, has provided successful theoretical accounts of sensory coding across species and sensory systems with the retina being the paramount example. The spatial organisation of the retina has, however been assumed to be uniform and translation invariant. Here we challenge this view. By combining theory and novel experimental approaches, we demonstrate that large populations of retinal neurons exploit the inhomogeneous structure of natural scenes across the visual field to increase the efficiency of neural coding.