Skip to Main content Skip to Navigation

Model selection and adaptive sampling in surrogate modeling: Kriging and beyond

Abstract : Surrogate models are used to replace an expensive-to-evaluate function to speed-up the estimation of a feature of a given function (optimum, contour line, …). Three aspects of surrogate modeling are studied in the work:1/ We proposed two surrogate model selection algorithms. They are based on a novel criterion called the penalized predictive score. 2/ The main advantage of probabilistic approach is that it provides a measure of uncertainty associated with the prediction. This uncertainty is an efficient tool to construct strategies for various problems such as prediction enhancement, optimization or inversion. We defined a universal approach for uncertainty quantification that could be applied for any surrogate model. It is based on a weighted empirical probability measure supported by cross-validation sub-models’ predictions.3/ We present the so-called Split-and-Doubt algorithm that performs sequentially both feature estimation and dimension reduction.
Document type :
Complete list of metadatas
Contributor : Florent Breuil <>
Submitted on : Tuesday, January 5, 2021 - 2:36:28 PM
Last modification on : Friday, January 8, 2021 - 3:07:00 AM


Files produced by the author(s)


  • HAL Id : tel-03097719, version 1


Malek Ben Salem. Model selection and adaptive sampling in surrogate modeling: Kriging and beyond. Modeling and Simulation. UNIVERSITE DE LYON, 2018. English. ⟨NNT : 2018LYSEM006⟩. ⟨tel-03097719⟩



Record views


Files downloads