Constrained Gaussian process regression: an adaptive approach for the estimation of hyperparameters and the verification of constraints with high probability - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Journal of Machine Learning for Modeling and Computing Année : 2021

Constrained Gaussian process regression: an adaptive approach for the estimation of hyperparameters and the verification of constraints with high probability

Résumé

This paper focuses on the Gaussian Process regression (GPR) of non-linear functions subject to multiple linear constraints, such as boundedness, monotonicity or convexity. It presents an algorithm allowing to optimize, in a concerted way, the statistical moments of the Gaussian process used for the regression, and the position of a reduced number of points where the constraints are required to hold, such that the constraints are verified in the whole input space, with high probability, at a reasonable computational cost. After having presented the theoretical bases and the numerical implementation of this algorithm, this paper illustrates its efficiency though the analysis of several test functions of increasing dimensions.
Fichier principal
Vignette du fichier
JMLMC_KrigeageContraintPerrinDaVeiga.pdf (622.98 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03419062 , version 1 (08-11-2021)

Identifiants

Citer

Guillaume Perrin, S. da Veiga. Constrained Gaussian process regression: an adaptive approach for the estimation of hyperparameters and the verification of constraints with high probability. Journal of Machine Learning for Modeling and Computing, 2021, 2 (2), pp.55-76. ⟨10.1615/JMachLearnModelComput.2021039837⟩. ⟨hal-03419062⟩
17 Consultations
215 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More