Spectral techniques in matrix completion
schedule le lundi 14 janvier 2019 de 17h00 à 18h00
Organisé par : G. Conchon--Kerjan, F. Coppini, B. Dembin
Intervenant : Simon Coste (LPSM)
Lieu : Sophie Germain - Salle 1016
Sujet : Spectral techniques in matrix completion
This talk is an introduction to matrix completion. In matrix completion problems, there is a hidden matrix P which we want to recover, but we do not have access to it: we can only observe a small proportion of the entries, which is often random in the applications. The question is to craft efficient procedures to reconstruct P as accurately as possible from those revealed entries, only assuming some information on P (small rank, symmetries) and on the set revealed entries (sparsity, type of randomness). In this view, spectral techniques have proven to be very powerful. I will present these techniques, their limitations, and the many links they share with the theory of random matrices, random graphs, sparsifiers and numerical mathematics.