ENBIS: European Network for Business and Industrial Statistics

Asymptotic optimal designs for Kriging with noisy observations

23 September 2009, 11:15 – 11:35

Abstract

Submitted by

Picheny Victor

Authors

V. Picheny, X. Bay, R.T. Haftka

Affiliation

Ecole des Mines de Saint Etienne

Abstract

This work aims at finding optimal strategies for the design of experiments for a kriging model when noisy observations are considered. In this case, it is often possible to choose a trade-off between the number of observations and the accuracy of the responses. Both numerical and theoretical studies are performed to determine optimal strategies, and show that the accuracy of the kriging model monotonically increases with the number of observations (even though the accuracy of each response is deteriorated). It is also found that optimal designs consists of an infinite number of observations (with infinite noise). In that case, we show that prediction variance and IMSE criterion can be expressed in simple form that can be used for comparison with realistic designs.

Return to programme