Correcting Boundary Over-Exploration Deficiencies in Bayesian Optimization With Virtual Derivative Sign Observations

Abstract

Bayesian optimization (BO) is a global optimization strategy designed to find the minimum of an expensive black-box function, typically defined on a compact subset of R d , by using a Gaussian process (GP) as a surrogate model for the objective. Although currently available acquisition functions address this goal with different degree of success, an over-exploration effect of the contour of the search space is typically observed. However, in problems like the configuration of machine learning algorithms, the function domain is conservatively large and with a high probability the global minimum does not sit on the boundary of the domain. We propose a method to incorporate this knowledge into the search process by adding virtual derivative observations in the GP at the boundary of the search space. We use the properties of GPs to impose conditions on the partial derivatives of the objective. The method is applicable with any acquisition function, it is easy to use and consistently reduces the number of evaluations required to optimize the objective irrespective of the acquisition used. We illustrate the benefits of our approach in an extensive experimental comparison.

Efficient Acquisition Rules for Model-Based Approximate Bayesian Computation

“…in areas of high likelihood, leading to sub-optimal posterior estimation whenever the prior is also informative. Comparing Figures 5a and 5b shows that using the expintvar strategy also avoids unnecessary evaluations on the boundary, which is often undesired also in the Bayesian optimisation methods, see Siivola et al [2017] for a discussion. Curiously, the uniform sampling (unif rule) works well when prior information is strong.…”

Output-Weighted Optimal Sampling for Bayesian Experimental Design and Uncertainty Quantification

“…It has also been reported that US has a disproportionate tendency toward selecting points on the boundary of the search space, simply because the variance is often largest far away from regions where data has been collected. Whether or not boundary points are informative is still an open question [27,47].…”

Preferential Batch Bayesian Optimization

“…In practice, we sample q continuous draws from the posterior predictive distribution of the latent variable and select each batch location as a minimum of the corresponding sample. The problem with independent draws is over-exploration of the border (Siivola et al, 2018). If the uncertainty is big on the border or there are border minima, many draws from the same batch are more likely to have minimum exactly on the border resulting to inefficient use of samples.…”