Wednesday, September 27

Invited Lecture: Choose your path wisely: iterative regularisation in a Bregman distance framework

Time: 8:00 - 9:00
Room: L1, Building L
Chair: Stephan Schmidt, Universität Würzburg

 

In this talk we present a generalisation of classical gradient descent that has become known in the literature as the so-called linearised Bregman iteration and — as the key novelty of this talk — apply it to minimise smooth but not necessarily convex objectives over a Banach space. We study the convergence of the proposed approach and present some numerical experiments for solving inverse imaging problems.

This is joint work with Martin Benning, Marta Betcke and Matthias Ehrhardt.

Carola-Bibiane Schönlieb
University of Cambridge