Imprecise Bayesian optimization

Julian Rodemann, Thomas Augustin

Abstract: Highlights•We study the effect of Gaussian process misspecification on Bayesian optimization (BO).•Prior mean parameters are found to have the highest impact on BO’s convergence.•We prove that prior mean parameter misspecification leads to linear regret bounds.•We propose a robust variant of BO that avoids prior mean parameter misspecification.•This is achieved by deploying imprecise Gaussian processes as surrogate models.