Mathematics Colloquium
Tom Cass
Imperial College
Title: Normalization effects on deep neural networks and deep learning for scientific problems
Date: Friday, May 2, 2025
Place and Time: Love 101, 3:05-3:55 pm
Abstract. The Chen-Fliess series is a fundamental object in rough path theory. Recently it has been used to provide effective summaries multimodal, irregularly sampled, ordered data in a way that can be effectively exploited computationally. At the core of this approach is the way in which it provides a convenient basis of function on the space of so-called unparameterised paths. This talk surveys this mathematical background and describes how it gives rise to practical uses. We draw connections between this foundation and the theory of rough paths, tools from probability such as random matrix theory and fundamental concepts in statistical learning such as kernel methods.