Reducing dimension in Bayesian Optimization

Rodolphe Le Riche (1, 2, 3, 4, 5) , Adrien Spagnol (4, 6, 5, 3) , David Gaudrie (3, 7, 5, 4) , Sébastien da Veiga (8) , Victor Picheny (9)

1 LIMOS - Laboratoire d'Informatique, de Modélisation et d'Optimisation des Systèmes
2 CNRS - Centre National de la Recherche Scientifique
3 Mines Saint-Étienne MSE - École des Mines de Saint-Étienne
4 FAYOL-ENSMSE - Département Génie mathématique et industriel
5 FAYOL-ENSMSE - Institut Henri Fayol
6 Safran Tech
7 PSA - PSA Peugeot - Citroën
8 Safran Tech
9 PROWLER.io, 72 Hills Road, Cambridge

This talk was given at the LIMOS on July the 9th 2020 and was mainly intended for an audience of non specialists of Gaussian processes (GPs). The first slides (up to slide 12) about GPs and Bayesian Optimization should probably be skipped by readers already aware about them. However, the two research contributions on variable selection for optimization 1) by kernel methods and, 2) by penalized likelihood in the PCA space, may be of interest to some experts.

Mots clés: Global optimization; Dimension reduction; Gaussian process models; Bayesian optimization; Shape optimization; Global optimization

Domaines: Ingénierie, finance et science [cs.CE]; Optimisation et contrôle [math.OC]; Génie mécanique [physics.class-ph]

Rodolphe Le Riche, Adrien Spagnol, David Gaudrie, Sébastien da Veiga, Victor Picheny. Reducing dimension in Bayesian Optimization. LIMOS internal seminar, Jul 2020, Clerrmont-Ferrand, France. ⟨hal-02913882v1⟩