Expected Improvements for the Asynchronous Parallel Global Optimization of Expensive Functions : Potentials and Challenges

LION 6 Conference Proceedings 2012
Jānis Januševskis, Rodolphe Le Riche, David Ginsbourger, Ramunas Girdziusas

Sequential sampling strategies based on Gaussian processes are now widely used for the optimization of problems involving costly simulations. But Gaussian processes can also generate parallel optimiza- tion strategies. We focus here on a new, parameter free, parallel expected improvement criterion for asynchronous optimization. An estimation of the criterion, which mixes Monte Carlo sampling and analytical bounds, is proposed. Logarithmic speed-ups are measured on 1 and 9 dimensional functions.

Keywords
asynchronous parallel global optimization
DOI
10.1007/978-3-642-34413-8_37
Hyperlink
http://www.emse.fr/~leriche/para_ei_lion6_4.pdf

Januševskis, J., Le Riche, R., Ginsbourger, D., Girdziusas, R. Expected Improvements for the Asynchronous Parallel Global Optimization of Expensive Functions : Potentials and Challenges. In: LION 6 Conference Proceedings: LION 6 Conference "Learning and Intelligent Optimization", France, Paris, 16 Jan-22 Feb., 2012. Paris: LION 6, 2012, pp.413-418. ISBN 978364234412.

Publication language
English (en)