Giovanni Marchetti: A Cubist Overview of the Mathematics of Machine Learning

Abstract

In this talk, we will overview (some of) the mathematical background of contemporary machine learning. First, we will introduce the basic notions, such as neural networks and their learning process. Second, we will consider function spaces parametrized by deep neural networks. In the spirit of cubism, we will discuss these spaces from various mathematical perspectives, including functional analysis, statistics, and (algebraic) geometry.