Effects of Covariance and Acquisition Functions on Bayesian Optimization using a Gaussian Process
Covariance and acquisition functions play important roles in Bayesian optimization using a Gaussian process.
However, there have been few studies on the effects of these functions on the performance of Bayesian optimization. In this
paper, we propose a modified sampling procedure for function evaluation and optimization by considering a new termination
condition. Then, we perform numerical experiments using three a synthetic two-dimensional function for analyzing the
effects of various combinations of covariance and acquisition functions on the performance of Bayesian optimization. We
found that the modified expected improvement acquisition function combined with the Matern covariance function generally
works well in terms of computational time and accuracy.
Keywords - Acquisition function, Bayesian optimization, covariance function, Gaussian process