FriSem

Satchel Grant, Ph.D. student with Professor Jay McClelland, Department of Psychology, Stanford University 

Title: Mechanistic Interpretations of Neural Isomorphisms: From computations to neurons

Abstract: Understanding the circuit mechanisms of neural systems is a central goal of sensory neuroscience. In this work, we show that a three-layer convolutional artificial neural network predicts retinal natural scene responses with an accuracy nearing experimental limits. We then show that the model’s internal structure is interpretable, as interneurons recorded separately and not modeled directly are highly correlated with model interneurons. Models fit to natural scene stimuli reproduce a diverse set of known response phenomena establishing the models' ethological relevance to natural visual computation. Lastly, we use a new approach to decompose the computations of model responses into the contributions of intermediate model computations, allowing generation and interpretation of new mechanistic hypotheses for how interneurons with different spatiotemporal responses are combined to generate neural phenomena. We use this method to propose a model for a neural phenomenon that, before this work, lacked an explanation. Our results demonstrate a unified and general approach to study the circuit mechanisms of ethological neural computations.