Representation Theory in Machine Learning

Representation theory, specifically, non-commutative Fourier transforms, have recently found several applications in learning algorithms. The talk will showcase several of these developments, including the theory of invariant image descriptors based on the non-commutative bispectrum, and various problems involving the symmetric group including multi-object tracking and ranking, as well as more classical problems, such as constructing graph invariants and solving the quadratic assignment problem.