Example-driven Parametrisations for Bayesian Shape Optimisation
The authors propose Bayesian Optimisation by Local Deformation (BOLD) to improve shape design for AI Radio components like aerofoils and cavities. BOLD learns a linear, interpretable search space by applying Principal Component Analysis (PCA) to deformations between existing designs. This method is more sample-efficient than hand-crafted methods and can explore shapes beyond the limitations of pre-defined parameterizations.