Poster on "Combination of Kriging models for Bayesian optimization in high-dimension" - Université Clermont Auvergne Accéder directement au contenu
Poster De Conférence Année : 2023

Poster on "Combination of Kriging models for Bayesian optimization in high-dimension"

Résumé

Kriging metamodeling (also called Gaussian Process regression) is a popular approach to predict the output of a function based on few observations. The Kriging method involves length-scale hyperparameters whose optimization is essential to obtain an accurate model and is typically performed using maximum likelihood estimation (MLE). However, for high-dimensional problems, the hyperparameter optimization is problematic due to the shape of the likelihood function, to the exponential growth of the search space with the dimension, and to over-fitting issues when there are too few observations. It often fails to provide correct hyperparameter values. This poster presents a new method based on a combination of Kriging sub-models with fixed length-scales. Contrarily to other approaches, it does not rely on dimension reduction techniques and it provides a closed-form expression for the model. We present a recipe to determine a suitable range for the sub-models length-scales based on the designs and on the employed kernel. We also compare different approaches to compute the weights in the combination. We show for a 50-dimensional test problem that our combination provides a more accurate surrogate model than the classical Kriging approach using MLE.
Fichier non déposé

Dates et versions

hal-04028577 , version 1 (14-03-2023)

Identifiants

  • HAL Id : hal-04028577 , version 1

Citer

Tanguy Appriou. Poster on "Combination of Kriging models for Bayesian optimization in high-dimension". COSEAL Workshop 2023, Mar 2023, Paris, France. ⟨hal-04028577⟩
33 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More