A multi-threaded parallel implementation of the MCSFilter optimization algorithm

Araújo, Leonardo Oliveira; Pacheco, Maria F.; Rufino, José; Fernandes, Florbela P.

http://hdl.handle.net/10198/29180

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
ICNAAM 2021_1.pdf		793.42 KB	Adobe PDF	Download

Send Feedback

Authors

Araújo, Leonardo Oliveira

Pacheco, Maria F.

Rufino, José

Fernandes, Florbela P.

Abstract(s)

The Multistart Coordinate Search Filter (MCSFilter) method is an optimization method able to find all minimizers (local and global) of a nonlinear and nonconvex problem. Reducing the time to find such minimizers may be of uttermost importance, specially in real-world applications, whether for economical reasons or due to other constrains of the processes under the control of optimization methods. In this paper, a first attempt to come up with a parallel version of the MCSFilter, tailored to multi-core machines, is described and discussed. Although the evaluation results indicate that this first parallel version ensures a moderate speedup, adequate to desktop-level systems, this work paves the way for increasingly efficient versions to be developed futurely.

Keywords

Derivative-free optimization Direct search Black box

URI

http://hdl.handle.net/10198/29180

Citation

Araújo, Leonardo Oliveira; Pacheco, Maria F.; Rufino, José; Fernandes, Florbela P. (2023). A multi-threaded parallel implementation of the MCSFilter optimization algorithm. In International Conference on Numerical Analysis and Applied Mathematics (ICNAAM). ISSN 0094-243X. 2849:1, p. 1-4