Nonlinear spectral methods in machine learningMS32

I will give an overview of our work on nonlinar eigenproblems ranging from exact relaxations of combinatorial problems as nonlinear eigenproblems to our recent work on Perron-Frobenius theory of multi-homogeneous mappings related to spectral problems of tensors and hypergraphs. This also leads to a nonlinear spectral method with which one can train a certain neural network globally optimal with a linear convergence rate.

This presentation is part of Minisymposium “MS32 - Nonlinear Spectral Theory and Applications (part 1)
organized by: Aujol Jean-Francois (University of Bordeaux) , Gilboa Guy (Electrical Engineering Department, Technion) .

Matthias Hein (University of Tuebingen)