Scalable Gaussian Process Regression for Prediction of Material Properties

Abstract

Gaussian process regression (GPR) is a non-parametric approach that can be used to make predictions based on a set of known points. It has been widely employed in recent years on a variety of problems. However the Gaussian process regression algorithm performs matrices inversions and the computational time can be extensive when accessing large training datasets. This is of critical importance when on-line learning and regression analyses are carried out on real-time applications. In this paper we propose a novel strategy, utilizing batch query processing and co-clustering, to achieve a scalable and efficient Gaussian process regression. The proposed strategy is applied to a real application involving the prediction of materials properties. Comprehensive tests have been conducted on two published properties data sets and the results demonstrate the high accuracy and efficiency of our new approach.