A learning-based exploration approach is proposed to escape from the basins of attraction of converged-to optima, by selecting on what is termed the interestingness of a solution. This interestingness is based on the modeling error made by a surrogate model that is trained on all solutions encountered earlier during the search. Compared to multiple standard optimization runs, a learning-guided restart scheme that alternates between a quality optimization phase and an exploration phase directed by interestingness finds solutions that are more diverse and of higher quality. © 2013 Springer-Verlag.
CITATION STYLE
Reehuis, E., Olhofer, M., Sendhoff, B., & Bäack, T. (2013). Learning-guided exploration in airfoil optimization. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 8206 LNCS, pp. 505–512). https://doi.org/10.1007/978-3-642-41278-3_61
Mendeley helps you to discover research relevant for your work.