SueYeon Chung – Emergence of Object Manifolds in Deep Networks and the Brain
Abstract: Stimuli are represented in the brain by the collective
population responses of sensory neurons, and an object presented
under varying conditions gives rise to a collection of neural population
responses called an “object manifold.” Changes in the object
representation along a hierarchical sensory system are associated
with changes in the geometry of those manifolds. To study this, we
developed a statistical mechanical theory for the linear classification of
these object manifolds, connecting the geometry of object manifolds
with their perceptron capacity, as a measure of linear separability. Our
theory and its extensions provide a new framework for characterizing
high-dimensional population responses to objects or categories in
biological and artificial neural networks. We demonstrate results from
applying our method to neural networks for visual, auditory, and
language tasks. Exciting future work lies ahead as manifold
representations of the sensory world are ubiquitous in both biological
and artificial neural systems.
Bio: SueYeon Chung is a postdoctoral research scientist at the Center
for Theoretical Neuroscience at Columbia University, where she is
mentored by Larry Abbott. Prior to that, she was a Fellow in
Computation at the Department of Brain and Cognitive Sciences at
MIT, where she collaborated with Jim DiCarlo and Josh McDermott.
She obtained her PhD at Harvard University, advised by Haim
Sompolinsky.