par Loris, Ignace
Référence Applied Inverse Problems Conference (25-29/05/2015: Helsinki, Finland)
Publication Non publié, 2015-05-29
Communication à un colloque
Résumé : Inverse problems are often cast in the form of a non-smooth optimization problem involving an objective function that is the combination of a data misfit term and regularizing penalty or constraint. In this talk I will discuss iterative algorithms for the numerical solution of such problems. The algorithms are based on the availability of the proximal operator of the non-smooth part of the objective function. In particular, I will discuss convergence when the proximal operator can only be calculated approximately. Verifiable criteria for such an inexact computation are also given in some cases of practical interest.This presentation is based on joint work with S. Bonettini, F. Porta and M. Prato.