Speaker: Romain Couillet, professor at Centrale-Supélec, Saclay, France, researcher at Université Grenoble-Alpes, Gipsa lab, Grenoble, France.
Abstract: In this talk, I will present the recent advances at the intersection between random matrix theory and machine learning. We will in particular see that in the random matrix regime (i.e., for numerous and large data), many standard “small dimensional” intuitions collapse due to a curse of dimensionality, that random matrices manage to both understand and analyse, so to eventually propose new algorithms. Without going much into technical details, this talk will provide the key ideas under the change of paradigm along with practical outcomes on real datasets.