Premium
Robust efficient global optimisation via adaptive surrogate refinement
Author(s) -
Le Carrer Noémie,
Moens David,
Faes Matthias
Publication year - 2019
Publication title -
pamm
Language(s) - English
Resource type - Journals
ISSN - 1617-7061
DOI - 10.1002/pamm.201900474
Subject(s) - kriging , surrogate model , local optimum , mathematical optimization , particle swarm optimization , global optimization , black box , computer science , ellipsoid , bayesian optimization , function (biology) , bayesian probability , algorithm , mathematics , artificial intelligence , machine learning , physics , astronomy , evolutionary biology , biology
Through the combination of a given surrogate model (e.g. Kriging) and a procedure allowing to compute the local optima of a given black‐box function (e.g. tuned Particle swarm optimization), our algorithm produces refined estimates of the global optima of a multidimensional black‐box model. A confidence ellipsoid computed in a Bayesian way is associated to each estimate, and shown to be reliable. The setting is tested on the 2‐dimensional Branin function. Results outperform those given by two state‐of‐the‐art non‐parallelised efficient global optimisation methods.