Séance Séminaire

Séminaire de Probabilités et Statistique

Statistical guarantees for generative models

In this talk, I will introduce a convenient framework for studying (adversarial) generative models from a statistical perspective. It consists in modeling the generative device as a smooth transformation of the unit hypercube of a dimension that is much smaller than that of the ambient space and measuring the quality of the generative model by means of an integral probability metric. In the particular case of integral probability metric defined through a smoothness class, I will present risk bounds quantifying the role of various parameters. In particular, they will clearly show the impact of dimension reduction on the error of the generative model.

The talk is based on the paper co-authored by N. Schreuder and V.E. Brunel: http://proceedings.mlr.press/v132/schreuder21a.html

Séminaire en salle 109 (IMAG, bâtiment 9), en présentiel.
Également retransmis sur zoom : https://umontpellier-fr.zoom.us/j/94087408185