Abstract
This study focuses on mixed-variable black-box optimization (MV-BBO),addressing continuous, integer, and categorical variables. Many real-worldMV-BBO problems involve dependencies among these different types of variables,requiring efficient methods to optimize them simultaneously. Recently,stochastic optimization methods leveraging the mechanism of the covariancematrix adaptation evolution strategy have shown promising results inmixed-integer or mixed-category optimization. However, such methods cannothandle the three types of variables simultaneously. In this study, we proposeCatCMA with Margin (CatCMAwM), a stochastic optimization method for MV-BBO thatjointly optimizes continuous, integer, and categorical variables. CatCMAwM isdeveloped by incorporating a novel integer handling into CatCMA, amixed-category black-box optimization method employing a joint distribution ofmultivariate Gaussian and categorical distributions. The proposed integerhandling is carefully designed by reviewing existing integer handlings andfollowing the design principles of CatCMA. Even when applied to mixed-integerproblems, it stabilizes the marginal probability and improves the convergenceperformance of continuous variables. Numerical experiments show that CatCMAwMeffectively handles the three types of variables, outperformingstate-of-the-art Bayesian optimization methods and baselines that simplyincorporate existing integer handlings into CatCMA.