Interdisciplinary Initiatives Program Round 11 - 2022

Project Investigators:

E.J. Chichilnisky, Neurosurgery and Ophthalmology
Dan Yamins, Psychology and Computer Science


Visual signaling begins in the retina, where parallel channels of information about the world are generated by many types of retinal ganglion cells. Recent work has revealed a wide diversity of these cell types in the mouse, and there is some evidence of many types in the primate. However, only a few primate cell types have been studied. Thus, we have little information about how the diverse retinal cell types and associated visual pathways shape vision in primates, including humans. This knowledge is fundamental to understanding, preserving, and restoring vision. Recent advances in our lab, using large-scale electrophysiological recordings, have produced a breakthrough, enabling us to probe the visual processing performed by ~15 additional cell types in both macaque and human retinas. These novel cell types exhibit striking and surprising spatial, temporal, and chromatic visual response properties, which we hypothesize reveal complex and behaviorally relevant functionality of a kind previously seen only in non-primate species. The potential implications for primate vision are therefore profound. Our large and unique archive of ~1,000 large-scale recordings from macaque retina, as well as new experiments in human and macaque retina, now allow us to analyze thousands of these novel cells. However, identifying and modeling the cells in this large dataset, and understanding their relationship to the human retina, will require a new interdisciplinary collaboration between an experimental neuroscientist and a machine learning expert. We will apply classification approaches using machine learning to the large dataset to determine the homologous and unique cell types in humans. We will probe the complex light response properties of novel and known cell types by using neural network models from contemporary machine learning. Finally, we will use these networks to design new experiments with precisely targeted stimuli. This interdisciplinary approach to response modeling, experimental design, and neural classification using modern machine learning methods will yield new insights into the diverse parallel pathways underlying human vision.