“…1 provides an overview of existing approaches, and good reviews are provided by [10,54]. One method is to simply train a surrogate model of the continuous design variables for each of the combinations of the discrete variables, known as Category-Wise (CW) Kriging [45]. The problem with this approach is that for the optimization of expensive black-box functions there are often not enough samples available to construct an accurate surrogate model for each of the discrete variable combinations [53].…”