Bayesian active learning for electromagnetic structure design
Jixiang Qing, Nicolas Knudde, Ivo Couckuyt, Domenico Spina, Tom Dhaene · 2020
A novel design framework based on Bayesian active learning is presented in this contribution. The proposed approach allows one to identify a set of design configurations satisfying the chosen specification. In particular, the entropy search-based active learning strategy, which relies on a Gaussian Process model, is able to minimize the number of time-consuming computer simulations or expensive design trials necessary to reach this goal. A suitable application example validates the proposed method.