MaD Seminar: On the Ising perceptron

Speaker: Nike Sun

Location: 60 Fifth Avenue, Room 150
Videoconference link: https://nyu.zoom.us/j/92830550229

Date: Thursday, May 5, 2022

The perceptron is a toy model of a single-layer neural network that “stores” a collection of given patterns. The model with N nodes and M=N*alpha random patterns is related to a natural problem in high-dimensional probability, concerning the intersection of M random half-spaces with the discrete cube or sphere in N dimensions. We also consider a more general version of the problem, defined by bounded activation functions, which was introduced and studied by Talagrand. We will present some new techniques and recent results on these models. Based on joint works with Jian Ding, Erwin Bolthausen, Shuta Nakajima, and Changji Xu.