Numerical issues in maximum likelihood parameter estimation for Gaussian process interpolation

This article investigates the origin of numerical issues in maximum likelihood parameter estimation for Gaussian process (GP) interpolation and investigates simple but effective strategies for improving commonly used open-source software implementations. This work targets a basic problem but a host of studies, particularly in the literature of Bayesian optimization, rely on off-the-shelf GP implementations. For the conclusions of these studies to be reliable and reproducible, robust GP implementations are critical.

Keywords

Gaussian process

Domains

Machine Learning [stat.ML] Computation [stat.CO]

Fichier principal

paper.pdf (345.41 Ko)

sbasak-gpmle-dcerg.pdf (257.2 Ko)

Origin	Files produced by the author(s)

Emmanuel Vazquez : Connect in order to contact the contributor

https://centralesupelec.hal.science/hal-03119528

Submitted on : Sunday, January 24, 2021-5:34:56 PM

Last modification on : Tuesday, July 16, 2024-11:56:03 AM

Long-term archiving on : Sunday, April 25, 2021-6:12:49 PM

Dates and versions

hal-03119528 , version 1 (24-01-2021)

hal-03119528 , version 2 (28-07-2021)

Licence

Attribution - NonCommercial - NoDerivatives

Identifiers

HAL Id : hal-03119528 , version 1
ARXIV : 2101.09747

Cite

Subhasish Basak, Sébastien Petit, Julien Bect, Emmanuel Vazquez. Numerical issues in maximum likelihood parameter estimation for Gaussian process interpolation. The 7th International Conference on machine Learning, Optimization and Data science - LOD 2021, Oct 2021, Grasmere, Lake District, England, United Kingdom. ⟨hal-03119528v1⟩