Advancing Bayesian Optimization: The Mixed-Global-Local (MGL) Kernel and Length-Scale Cool Down

Abstract

Bayesian Optimization (BO) has become a core method for solving expensive black-box optimization problems. While much research focussed on the choice of the acquisition function, we focus on online length-scale adaption and the choice of kernel function. Instead of choosing hyperparameters in view of maximum likelihood on past data, we propose to use the acquisition function to decide on hyperparameter adaptation more robustly and in view of the future optimization progress. Further, we propose a particular kernel function that includes non-stationarity and local anisotropy and thereby implicitly integrates the efficiency of local convex optimization with global Bayesian optimization. Comparisons to state-of-the art BO methods underline the efficiency of these mechanisms on global optimization benchmarks.

Long version of accepted NIPS BayesOpt 2016 paper

References in corpus (5)

• Gaussian Process Optimization in the Bandit Setting: No Regret and Experimental Design
• Predictive Entropy Search for Efficient Global Optimization of Black-box Functions
• Bayesian Optimization with Exponential Convergence
• Heteroscedastic Treed Bayesian Optimisation
• Small ensembles of kriging models for optimization

Cited by in corpus (3)

• Optimization, fast and slow: optimally switching between local and Bayesian optimization
• Cautious Bayesian Optimization for Efficient and Scalable Policy Search
• No-Regret Bayesian Optimization with Unknown Hyperparameters