ENBIS: European Network for Business and Industrial Statistics

Kriging models including functional information

21 September 2009, 15:45 – 16:05

Abstract

Submitted by

Bertrand Gauthier

Authors

B. Gauthier, N. Durrande, O. Roustant

Affiliation

EMSE

Abstract

Kriging models are more and more used to approximate the response of complex phenomena or expensive simulators. Those models are built from a design of experiments that consists of a finite set of points at which the response is evaluated. However, in some applications, the response can be known everywhere on a subspace or a contour (for instance, if the response is known on the boundary of the design region). In that case, such information can be seen as a function on the subspace, and the regular approach does not stand since an infinite number of observation points would be necessary to account for all the information.
In this work, we show that, for particular classes of kernels and contours, it is possible to construct kriging models considering both functional and discrete information. We prove that the model we propose can be seen as the limit of a kriging model conditionally to an infinite number of observations taken along the contour.

Return to programme