November 10, 2020

12:00 pm / 1:15 pm


Recorded Seminar:
SueYeon Chung
Postdoctoral Research Scientist 
at the Centerfor TheoreticalNeuroscience 
at Columbia University    

Title: Emergence of Object Manifolds in Deep Networks and the Brain

Abstract: Stimuli are represented in the brain by the collective populationresponses of sensory neurons, and an object presented under varying conditionsgives rise to a collection of neural population responses called an ?objectmanifold.? Changes in the object representation along a hierarchical sensorysystem are associated with changes in the geometry of those manifolds. To studythis, we developed a statistical mechanical theory for the linearclassification of these object manifolds, connecting the geometry of objectmanifolds with their perceptron capacity, as a measure of linear separability.Our theory and its extensions provide a new framework for characterizinghigh-dimensional population responses to objects or categories in biologicaland artificial neural networks. We demonstrate results from applying our methodto neural networks for visual, auditory, and language tasks. Exciting futurework lies ahead as manifold representations of the sensory world are ubiquitousin both biological and artificial neural systems.

Bio:SueYeon Chung is a postdoctoral research scientist at the Center forTheoretical Neuroscience at Columbia University,where she is mentored by LarryAbbott. Prior to that, she was a Fellow in Computation at the Department of Brainand Cognitive Sciences at MIT, whereshe collaborated with Jim DiCarlo and JoshMcDermott. She obtained her PhD at Harvard University, advised by HaimSompolinsky.