SMASH seminar
SMASH seminar given by Prof. Primoz Skraba, titled "Topology and Universality with a View toward Deep Learning".
Abstract
Recently, there has been a recent trend in deep learning of using topological invariants to try and improve performance as well as try to understand how they work. I will motivate the main part of my talk by giving a few examples of how a popular topological feature, persistent homology, has been used in deep learning. In the main part of the talk, I will discuss recent advances in understanding the statistics of these invariant, through a form of universality in a similar spirit as the central limit theorem results in a Gaussian distribution for large collections of independent measurements. Finally, I will discuss briefly about the implications of this for future work.
The talk will also introduce persistent homology, so no prior knowledge will be assumed.