Surrogate modeling approximation using a mixture of experts based on EM joint estimation

Dimitri Bettebghor 1 Nathalie Bartoli Stéphane Grihon 1 Joseph Morlier 2 Manuel Samuelides 3
1 Airbus (FRANCE)
2 ICA - Institut Clément Ader
3 DMIA - Département de Mathématiques, Informatique, Automatique
Abstract : An automatic method to combine several local surrogate models is presented. This method is intended to build accurate and smooth approximation of discontinuous functions that are to be used in structural optimization problems. It strongly relies on the Expectation-Maximization (EM) algorithm for Gaussian mixture models (GMM). To the end of regression, the inputs are clustered together with their output values by means of parameter estimation of the joint distribution. A local expert is then built (linear, quadratic, artificial neural network, moving least squares) on each cluster. Lastly, the local experts are combined using the Gaussian mixture model parameters found by the EM algorithm to obtain a global model. This method is tested over both mathematical test cases and an engineering optimization problem from aeronautics and is found to improve the accuracy of the approximation.
Keywords : EM clustering Gaussian mixture models Mixture of experts
Document type :
Journal articles
Complete list of metadatas
https://hal.archives-ouvertes.fr/hal-01852300
Contributor : Open Archive Toulouse Archive Ouverte (oatao) <>
Submitted on : Wednesday, August 1, 2018 - 12:18:50 PM
Last modification on : Wednesday, May 29, 2019 - 6:00:04 PM
Long-term archiving on : Friday, November 2, 2018 - 1:47:42 PM

File

Morlier_4265.pdf
Publisher files allowed on an open archive

Identifiers

Collections

INSA-TOULOUSE | ICA | INSA-GROUPE | UNIV-TLSE3

Citation

Dimitri Bettebghor, Nathalie Bartoli, Stéphane Grihon, Joseph Morlier, Manuel Samuelides. Surrogate modeling approximation using a mixture of experts based on EM joint estimation. Structural and Multidisciplinary Optimization, Springer Verlag (Germany), 2011, 43 (2), pp.243-259. ⟨10.1007/s00158-010-0554-2⟩. ⟨hal-01852300⟩

Export

BibTeX TEI DC DCterms EndNote

Share

Metrics

Record views

30

Files downloads

18

Contact

support.ccsd.cnrs.fr
hal.support@ccsd.cnrs.fr
Logo CCSD