Machine Learning at the Flatiron Institute Seminar: Francesca Mignacco

Title: Statistical physics insights into the dynamics of learning algorithms

Abstract: In this talk, I will consider prototypical learning problems that are amenable to an exact characterization. I will show how dynamical mean-field theory from statistical physics can be used to derive an effective low-dimensional description of the network performance and the learning dynamics of multi-pass SGD. I will discuss how different sources of algorithmic noise affect the performance.