Course: STAT 37796
Title: Topics in Machine Learning: Symmetries and Harmonic Analysis
Instructor(s): Risi Kondor
Teaching Assistant(s):
Class Schedule: Sec 1: TR 11:00 AM-12:20 PM in Stuart 102
Description: Many algorithms in machine learning and statistical inference have inherent symmetries. In other cases, symmetries need to be enforced on the algorithm explicitly from the outside. In both cases, a systematic study of the symmetries inevitably leads to considerations borrowed from group representation theory and harmonic analysis. In this course we will take a broad view of this topic spanning the range from the concept of exchangeability in probability and nonparametric statistics, via the implementation of symmetries in kernel methods such as Gaussian processes to the new and quickly developing field of equivariant neural networks. One area that where symmetries play a key role and we will specifically focus on is learning physical systems and developing scalable algorithms for large scale modeling for physics and chemistry. The course starts with a short introduction to representation theory and wavelets, of which no prior knowledge is required.
