Date: Wednesday, February 23, 2022
Location: 3096 East Hall (5:15 PM to 6:15 PM)
Title: The Perceptron Capacity model for Spin Glasses
Abstract: The Perceptron capacity model is a mean field model to study spin glass, and is fundamental in the theory of neural networks. In this talk, we will discuss its underlying mathematical problem, which concerns measuring typically how much of the discrete cube {1, 1}^N (or the unit sphere) is left when one intersects the set with many random halfspaces. We first introduce notations, phrase the question in the language of statistical mechanics, then describe results in the special case where both the number of random halfspaces and the dimension N are large.
Files:
Speaker: Han Le
Institution: University of Michigan
Event Organizer:
