Towards robust second order optimization methods on low rank matrix and tensor varieties (Valentin Khrulkov -- Skoltech, Moscow)

13.10.2017 14:15 – 15:00

Use of the traditional Riemannian Newton method on low rank matrix varieties faces the difficulties of singular points and of ill-behaved Hessian. Similar problems exist for low rank tensor varieties (e.g Tensor Trains with bounded ranks). We will discuss how one can deal with such problems using the concept of desingularization, implementation of the robust second order optimization scheme for the matrix varieties and possible generalizations of this approach to tensors.

Lieu

Room 624, Séminaire d'analyse numérique

Organisé par

Section de mathématiques

Intervenants

Valentin Khrulkov, Skoltech

entrée libre

Classement

Catégorie: Séminaire

Mots clés: analyse numérique