The CMA-ES algorithm searches a fitness landscape by sampling from a multivariate normal distribution and updating its mean by taking a weighted average of the highest fitness candidate solutions. In this work, we explore the possibility of using Genetic Programming to evolve new mean-update selection methods that take into account information other than just raw fitness values. These results show that CMA-ES can be tuned to specific problem classes to achieve better results.
2019 Genetic and Evolutionary Computation Conference, GECCO 2019 (2019: Jul. 13-17, Prague, Czech Republic)
Keywords and Phrases
- Genetic Programming,
International Standard Book Number (ISBN)
Article - Conference proceedings
© 2019 Association for Computing Machinery (ACM), All rights reserved.
Samuel N. Richter, Michael G. Schoen and Daniel R. Tauritz. "Comparing Terminal Sets for Evolving CMA-ES Mean-Update Selection" GECCO 2019 Companion -- Proceedings of the 2019 Genetic and Evolutionary Computation Conference Companion
(2019) p. 326 - 327 ISSN: 2474-249X
Available at: http://works.bepress.com/daniel-tauritz/78/