Comparison of Ordinal and Metric Gaussian Process Regression as Surrogate Models for CMA Evolution Strategy

0477789 - ÚI 2018 RIV US eng C - Conference Paper (international conference)
Pitra, Z. - Bajer, L. - Repický, J. - Holeňa, Martin
Comparison of Ordinal and Metric Gaussian Process Regression as Surrogate Models for CMA Evolution Strategy.
GECCO 2017. Proceedings of the Genetic and Evolutionary Computation Conference Companion. New York: ACM, 2017, s. 1764-1771. ISBN 978-1-4503-4939-0.
[GECCO 2017. Genetic and Evolutionary Computation Conference. Berlin (DE), 15.07.2017-19.07.2017]
R&D Projects: GA ČR GA17-01251S
Grant - others:GA MŠk(CZ) LO1611; ČVUT(CZ) SGS17/193/OHK4/3T/14; GA MŠk(CZ) LM2010005
Institutional support: RVO:67985807
Keywords : black-box optimization * evolutionary optimization * surrogate modelling * Gaussian-process regression
OECD category: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

In this paper, Gaussian processes are studied in connection with the state-of-the-art method for continuous black-box optimization CMA-ES. To combine them with the CMA-ES is challenging because CMA-ES invariance with respect to order preserving transformations suggests ordinal regression, whereas Gaussian process continuity suggests metric regression. Results of testing ordinal and metric Gaussian process regression, the former in 14 different settings, combined with the CMA-ES on noiseless benchmarks of the COCO platform are reported.
Permanent Link: http://hdl.handle.net/11104/0274013