Fast parallel kriging-based stepwise uncertainty reduction with application to the identification of an excursion set

Abstract : Stepwise Uncertainty Reduction (SUR) strategies aim at constructing a sequence of sampling points for a function f : Rd → R, in such a way that the residual uncertainty about a quantity of interest becomes small. In the context of Gaussian Process-based approximation of computer experiments, these strategies have been shown to be particularly efficient for the problem of estimating the volume of excursion of a function f above a threshold. However, these strategies remain difficult to use in practice because of their high computational complexity, and they only deliver at each iteration a single point to evaluate. In this paper we introduce parallel sampling criteria, which allow selecting several sampling points simultaneously. Such criteria are of particular interest when the function f is expensive to evaluate and many CPUs are available. We also manage to drastically reduce the computational cost of these strategies using closed form expressions. We illustrate their performances in various numerical experiments, including a nuclear safety test case.
Complete list of metadatas

Cited literature [22 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-00641108
Contributor : David Ginsbourger <>
Submitted on : Thursday, April 19, 2012 - 1:09:40 PM
Last modification on : Friday, April 6, 2018 - 8:24:15 AM
Long-term archiving on : Friday, July 20, 2012 - 2:35:24 AM

File

Fast_parallel_SUR.pdf
Files produced by the author(s)

Identifiers

Collections

Citation

Clément Chevalier, Julien Bect, David Ginsbourger, Emmanuel Vazquez, Victor Picheny, et al.. Fast parallel kriging-based stepwise uncertainty reduction with application to the identification of an excursion set. Technometrics, Taylor & Francis, 2014, 56 (4), pp.455-465. ⟨10.1080/00401706.2013.860918⟩. ⟨hal-00641108v2⟩

Share

Metrics

Record views

1485

Files downloads

936