495. Complexity of a Riemannian Direct-search algorithm
Invited abstract in session MD-1: Derivative-Free Optimization Methods for challenging applications: Handling Nonsmoothness and Constraints, stream Zeroth and first-order optimization methods.
Monday, 16:30-18:30Room: B100/1001
Authors (first author is the speaker)
| 1. | Bastien Cavarretta
|
| LAMSADE, Université Paris Dauphine-PSL | |
| 2. | Clément Royer
|
| LAMSADE, Université Paris Dauphine-PSL | |
| 3. | Florian Yger
|
| INSA Rouen, LITIS | |
| 4. | Florentin Goyens
|
| UCLouvain |
Abstract
Direct-search algorithms are derivative-free optimization techniques that operate by polling the variable space along specific directions forming a positive spanning set (PSS). We investigate the construction of PSSs when the variables are constrained to a Riemannian manifold, and polling must be performed along tangent directions. We show that projecting a PSS from the ambient space to the tangent space may lead to worse complexity guarantees than generating directions directly in the tangent space. Our numerical experiments illustrate the practical benefit of the latter construction.
Keywords
- Derivative-free optimization
- Complexity and efficiency of algorithms
Status: accepted
Back to the list of papers