Parallel expected improvements for global optimization: summary, bounds and speed-up

Abstract : The sequential sampling strategies based on Gaussian processes are widely used for optimization of time consuming simulators. In practice, such computationally demanding problems are solved by increasing number of processing units. This has therefore induced extensions of sampling criteria which consider the framework of parallel calculation. This report further studies expected improvement criteria for parallel and asynchronous computations. A unified parallel asynchronous expected improvement criterion is formulated. Bounds and strategies for comparing criteria values at various design points are discussed. Finally, the impact of the number of available computing units on the performance is empirically investigated.
Document type :
Reports
Liste complète des métadonnées

Cited literature [13 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-00613971
Contributor : Janis Janusevskis <>
Submitted on : Monday, August 8, 2011 - 12:35:27 PM
Last modification on : Tuesday, October 23, 2018 - 2:36:08 PM
Document(s) archivé(s) le : Monday, November 12, 2012 - 3:15:17 PM

File

TechReportASyncrEIfinal_v4.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-00613971, version 1

Citation

Janis Janusevskis, Rodolphe Le Riche, David Ginsbourger. Parallel expected improvements for global optimization: summary, bounds and speed-up. 2011. ⟨hal-00613971⟩

Share

Metrics

Record views

384

Files downloads

266