Batched High-dimensional Bayesian Optimization via Structural Kernel\n Learning

Abstract

Optimization of high-dimensional black-box functions is an extremely\nchallenging problem. While Bayesian optimization has emerged as a popular\napproach for optimizing black-box functions, its applicability has been limited\nto low-dimensional problems due to its computational and statistical challenges\narising from high-dimensional settings. In this paper, we propose to tackle\nthese challenges by (1) assuming a latent additive structure in the function\nand inferring it properly for more efficient and effective BO, and (2)\nperforming multiple evaluations in parallel to reduce the number of iterations\nrequired by the method. Our novel approach learns the latent structure with\nGibbs sampling and constructs batched queries using determinantal point\nprocesses. Experimental validations on both synthetic and real-world functions\ndemonstrate that the proposed method outperforms the existing state-of-the-art\napproaches.\n