The Central role of Bayesian Optimization in Efficient Learning

Abstract

Learning is an essential goal of statistical methods. It can be based on active experimentation or, on more passive, observational data. In this paper we introduce Bayesian optimization learning which involves a step-by-step directed approach combining expected improvement and prediction variance reduction. These steps can involve a search through historical data bases documenting past experiments or through new experiments. This methodology results in fast and efficient learning permitting researchers, engineers or scientists to reach optimal conditions in products and processes at unprecedent speed. The paper provides a background on Bayesian optimization learning and a case study describing the application. A concluding section provides an outlook on the potential impact of Bayesian optimization learning which bridges artificial intelligence and machine learning with statistics.

Keywords: Bayesian optimization learning, Box Hunter and Hunter, Statistical learning, Design of experiments, Sequential learning