The use of stimuli with increasingly naturalistic properties has become critical to advance our understanding of vision. Many studies demonstrate that simple artificial stimuli (e.g. sinusoidal gratings and white noise) fail to engage nonlinearities that profoundly alter responses in the retina, lateral geniculate nucleus (LGN), and primary visual cortex (V1). A recent and striking example comes from the use of naturalistic ?flow? stimuli, which engage robust responses in V1 that are not predicted from responses to gratings. This gap in understanding motivates the development of a stimulus ensemble and analysis framework that produces a quantitative understanding of visual processing to increasingly naturalistic stimuli and the nonlinearities that they engage. Our objective is to understand how flow stimuli are processed from retina through visual cortex. To meet this goal, we will make neural population recordings in retina (Aims 1 & 3), LGN (Aims 1 & 3) and V1 (Aim 3) using matched experimental conditions and a unified theoretical/modeling framework to map the transformations that occur across these stages of visual processing. Our central hypothesis is that V1 transforms a discrete and heavily light-level-de- pendent retinal representation of natural stimuli into a continuous (uniform) representation that is relatively in- variant to changes in the mean luminance. This invariance places a strong constraint on the class of nonlineari- ties that transform retinal responses to those observed in LGN and V1. We test this hypothesis in three aims: (1) determine early visual processing (retina & LGN) of naturalistic flow stimuli; (2) develop an encoding manifold to capture the population activity at each processing stage and transforms from one stage to the next; (3) test the ability of the manifold description to predict the impact of light adaptation on processing flow stimuli from retina to V1.
Aim 1 will yield a matched experimental dataset to an interesting and novel class of ecologically-relevant stimuli.
Aim 2 will yield a quantitative framework by which to understand the transformations that occur between retina, LGN, and V1.
Aim 3 will provide a platform for globally perturbing the output of the retina by switching from photopic to mesopic and scotopic conditions, and thereby compare predictions of our model to measured changes in LGN and V1 activity. The primary significance of this research is that it will provide a computationally and experimentally unified framework for understanding the transformations that occur in the processing of stim- uli across multiple stages of visual processing. The major innovations are (1) presenting visual stimuli for retinal recordings that are matched to eye movements and pupil dynamics in alert animals; (2) creating a novel analysis framework that captures the responses of neurons at all three levels and the inter-level transformations to in- creasingly complex stimuli; (3) utilizing light adaptation as a method of perturbing retinal output to test our model and the stability (invariance) of LGN and V1 responses to adapting retinal signals. The expected outcome is a data-driven model of the processing from retina to LGN and V1 that generalizes from starlight to sunlight.
Restoring vision to the blind likely requires understanding how retinal signals are communicated to the brain and how these signals are transformed in the thalamocortical pathway. This project aims to acquire an understanding of these transformations in the context of complex and more naturalistic visual stimuli.
