High-dimensional efficient global optimization using both random and supervised embeddings
Rémy Priem, Nathalie Bartoli, Youssef Diouane, Sylvain Dubreuil, Paul Maxime Valentin Saves · 2023
View Video Presentation: https://doi.org/10.2514/6.2023-4448.vid Bayesian optimization (BO) is one of the most powerful strategies to solve expensive black-box optimization problems. However, BO methods are conventionally used for optimization problems of small dimension because of the curse of dimensionality. In this paper, to solve high dimensional optimization problems, we propose to incorporate linear embedding subspaces of small dimension to efficiently perform the optimization. An adaptive learning strategy for these linear embeddings is carried out in conjunction with the optimization. The resulting BO method, named EGORSE, combines in an adaptive way both random and supervised linear embeddings. EGORSE has been compared to state-of-the-art algorithms and tested on academic examples with a number of design variables ranging from 10 to 600. The obtained results show the high potential of EGORSE to solve high-dimensional black-box optimization problems, both in terms of CPU time and number of calls to the expensive black-box.