562. Forward-backward type splitting algorithms with minimal lifting I
Invited abstract in session MB-8: Systematic and computer-aided analyses I: Analyses of proximal splittings methods & friends, stream Systematic and computer-aided analyses of optimization algorithms.
Monday, 10:30-12:30Room: B100/7007
Authors (first author is the speaker)
| 1. | Anton Åkerman
|
| Department of Automatic Control, Lund University | |
| 2. | Emanuele Naldi
|
| Mathematics, Università di Genova | |
| 3. | Enis Chenchene
|
| 4. | Pontus Giselsson
|
| Dept. of Automatic Control, Lund University |
Abstract
We consider forward-backward type splitting methods for solving monotone inclusion problems involving maximally monotone and cocoercive operators, and present a characterization of all such methods that are frugal, averaged nonexpansive and with minimal lifting. In cases where simpler counterexamples seem difficult, we derive necessary and sufficient conditions for nonexpansiveness using Performance Estimation.
Ultimately, we present a parameterization of this general algorithm class, based on which the second part of the talk will discuss practical performance and heuristics for (computer aided) algorithm design.
Keywords
- Computer-aided algorithm analysis
- First-order optimization
- Monotone inclusion problems
Status: accepted
Back to the list of papers