Adaptive Doubly Trained Evolution Control for the Covariance Matrix Adaptation Evolution Strategy

0478629 - ÚI 2018 RIV DE eng C - Conference Paper (international conference)
Pitra, Zbyněk - Bajer, Lukáš - Repický, Jakub - Holeňa, Martin
Adaptive Doubly Trained Evolution Control for the Covariance Matrix Adaptation Evolution Strategy.
Proceedings ITAT 2017: Information Technologies - Applications and Theory. Aachen & Charleston: Technical University & CreateSpace Independent Publishing Platform, 2017 - (Hlaváčová, J.), s. 120-128. CEUR Workshop Proceedings, V-1885. ISBN 978-1974274741. ISSN 1613-0073.
[ITAT 2017. Conference on Theory and Practice of Information Technologies - Applications and Theory /17./. Martinské hole (SK), 22.09.2017-26.09.2017]
R&D Projects: GA ČR GA17-01251S
Grant - others:ČVUT(CZ) SGS17/193/OHK4/3T/14; GA MŠk(CZ) LO1611; GA MŠk(CZ) LM2010005
Institutional support: RVO:67985807
Keywords : black-box optimization * evolutionary optimization * surrogate modelling * Gaussian process * CMA-ES
OECD category: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
http://ceur-ws.org/Vol-1885/120.pdf

An area of increasingly frequent applications of evolutionary optimization to real-world problems is continuous black-box optimization. However, evaluating realworld black-box fitness functions is sometimes very timeconsuming or expensive, which interferes with the need of evolutionary algorithms for many fitness evaluations. Therefore, surrogate regression models replacing the original expensive fitness in some of the evaluated points have been in use since the early 2000s. The Doubly Trained Surrogate Covariance Matrix Adaptation Evolution Strategy (DTS-CMA-ES) represents a surrogate-assisted version of the state-of-the-art algorithm for continuous blackbox optimization CMA-ES. The DTS-CMA-ES saves expensive function evaluations through using a surrogate model. However, the model inaccuracy on some functions can slow-down the algorithm convergence. This paper investigates an extension of DTS-CMA-ES which controls the usage of the model according to the model’s error. Results of testing an adaptive and the original version of DTS-CMA-ES on the set of noiseless benchmarks are reported.
Permanent Link: http://hdl.handle.net/11104/0274762