2015 QPRC

Latin Hypercube Design-based Block Bootstrap for Computer Experiment Modeling

Abstract: Computer experiments are becoming increasingly important in science and Gaussian process (GP) models are widely used in the analysis of computer experiments. However the computational issue that hinders GP from broader application is generally recognized, especially for massive data observed on irregular grids. To overcome the computational issue, we introduce an efficient framework, including estimation and prediction, based on a novel experimental design based bootstrap method. The main challenge in GP modeling is that both estimation and prediction rely heavily on large correlation matrix operations, which are computationally intensive and often intractable for massive data. Using the idea of design-based data reduction, the proposed framework provides a consistent estimation for the parameters in, together with a dramatic reduction in computation. The finite-sample performance is examined through simulation studies. We illustrate the proposed method by a data center example based on tens of thousands of computer experiments generated from a computational fluid dynamics simulator.