336. A derivative-free algorithm based on resilient positive spanning sets
Invited abstract in session MC-1: Strategies to Improve Zeroth-Order Optimization Methods, stream Zeroth and first-order optimization methods.
Monday, 14:00-16:00Room: B100/1001
Authors (first author is the speaker)
| 1. | Sébastien Kerleau
|
| LAMSADE, Université Paris Dauphine PSL | |
| 2. | Clément Royer
|
| LAMSADE, Université Paris Dauphine-PSL |
Abstract
Positive spanning sets (PSSs) are very useful in DFO algorithms. Optimization methods based on PSSs typically favor those with the best cosine measure, yet this metric does not fully account for their structure: in particular, it does not reflect the spanning capabilities of their subsets. This talk focuses on a particular class of PSSs, namely PkSSs, that remain positively spanning when losing elements. After showing how to construct PkSSs whose subsets all possess a large cosine measure, I present a derivative-free algorithm based on parallel computing and resilient to stragglers.
Keywords
- Black-box optimization
- Derivative-free optimization
Status: accepted
Back to the list of papers