Most recent variational methods consist of bricks of very different computational complexity. A particular instance are parts containing a costly (e.g. imaging) operator whose evaluation is limited to a small number, and often is the crucial time factor in a practical setup. We use (accelerated) inexact forward-backward splittings and (accelerated) inexact primal-dual methods to separate these limiting parts from easier ones to provide the best possible result within a small budget of operator evaluations.

This presentation is part of Minisymposium “MS10 - Advanced optimization methods for image processing (2 parts)”
organized by: Marco Prato (University of Modena and Reggio Emilia) , Ignace Loris (Université Libre de Bruxelles) .

Authors:

Julian Rasch (Westfälische Wilhelms-Universität Münster)

Antonin Chambolle (Ecole Polytechnique)

Keywords:

inverse problems, numerical optimization