Sparsity and Compressed Coding in Sensory Systems

Victor J. Barranca, Gregor Kovacic, Douglas Zhou, David Cai

Abstract: Author Summary In forming a mental percept of the surrounding world, sensory information is processed and transmitted through a wide array of neuronal networks of various sizes and functionalities. Despite, and perhaps because of, this, sensory systems are able to render highly accurate representations of stimuli. In the retina, for example, photoreceptors transform light into electric signals, which are later processed by a significantly smaller network of ganglion cells before entering the optic nerve. How then is sensory information preserved along such a pathway? In this work, we put forth a possible answer to this question using compressed sensing, a recent advance in the field of signal processing that demonstrates how sparse signals can be reconstructed using very few samples. Through model simulation, we discover that stimuli can be recovered from ganglion-cell dynamics, and demonstrate how localized receptive fields improve stimulus encoding. We hypothesize that organisms have evolved to utilize the sparsity of stimuli, demonstrating that compressed sensing may be a universal information-processing framework underlying both information acquisition and retention in sensory systems.