518. Forward-backward algorithms with deviations
Invited abstract in session MC-8: Systematic and computer-aided analyses II: Systematic algorithmic design approaches, stream Systematic and computer-aided analyses of optimization algorithms.
Monday, 14:00-16:00Room: B100/7007
Authors (first author is the speaker)
| 1. | Sebastian Banert
|
| Uni Bremen |
Abstract
We compare different variations of forward-backward-type algorithms for minimising a sum of two functions or finding a zero of the sum of two monotone operators. The focus will be on algorithms with, what we call, deviations or steering vectors. Such algorithms allow for a degree of freedom in the dimension of the optimisation variable instead of only scalar parameters, potentially leading to a greater adaptability to specific problem classes, for example by deep learning. We will compare different approaches to introduce deviations by their performance guarantees, expressivity, and numerical performance. This talk is based on joint work with Jevgenija Rudzusika, Ozan Öktem, Jonas Adler, Hamed Sadeghi, Pontus Giselsson, and Oskar Bircks.
Keywords
- Computational mathematical optimization
- Data driven optimization
Status: accepted
Back to the list of papers