McIntosh, L. T. | bioRxiv Science

bioRxiv2018

Deep learning models reveal internal structure and diverse computations in the retina under natural scenes

Understanding how the visual system encodes natural scenes is a fundamental goal of sensory neuroscience. We show here that a three-layer network model predicts the retinal response to natural scenes with an accuracy nearing the fundamental limits of predictability. The models internal structure is interpretable, in that model units are highly correlated with interneurons recorded separately and not used to fit the model. We further show the ethological relevance to natural visual processing of a diverse set of phenomena of complex motion encoding, adaptation and predictive coding. Our analysis uncovers a fast timescale of visual processing that is inaccessible directly from experimental data, showing unexpectedly that ganglion cells signal in distinct modes by rapidly (< 0.1 s) switching their selectivity for direction of motion, orientation, location and the sign of intensity. A new approach that decomposes ganglion cell responses into the contribution of interneurons reveals how the latent effects of parallel retinal circuits generate the response to any possible stimulus. These results reveal extremely flexible and rapid dynamics of the retinal code for natural visual stimuli, explaining the need for a large set of interneuron pathways to generate the dynamic neural code for natural scenes.

neuroscience↗

bioRxiv2017

Synchronous inhibitory pathways create both efficiency and diversity in the retina

Visual information is conveyed from the retina to the brain by a diverse set of retinal ganglion cells. Although they have differing nonlinear properties, nearly all ganglion cell receptive fields on average compute a difference in intensity across space and time using a region known as the classical or linear surround1,2, a property that improves information transmission about natural visual scenes3,4. The spatiotemporal visual features that create this fundamental property have not been quantitatively assigned to specific interneurons. Here we describe a generalizable causal approach using simultaneous intracellular and multielectrode recording to directly measure and manipulate the sensory feature conveyed by a neural pathway to a downstream neuron. Analyzing two inhibitory cell classes, horizontal cells and linear amacrine cells, we find that rather than transmitting different temporal features, the two inhibitory pathways act synchronously to create the salamander ganglion cell surround at different spatial scales. Using these measured visual features and theories of efficient coding, we computed a fitness landscape representing the information transmitted using different weightings of the two inhibitory pathways. This theoretical landscape revealed a ridge that maintains near-optimal information transmission while allowing for receptive field diversity. The ganglion cell population showed a striking match to this prediction, concentrating along this ridge across a wide range of positions using different weightings of amacrine or horizontal cell visual features. These results show how parallel neural pathways synthesize a sensory computation, and why this architecture achieves the potentially competing objectives of high information transmission of individual ganglion cells, and diversity among receptive fields.

neuroscience↗