Gaussian Process Single-Index Models as Emulators for Computer Experiments

Summary: The authors reveal a Gaussian process (GP) single-index model (SIM) that makes an ideal emulator for some types of computer experiments, outperforming the usual separable GP regression model. The authors drastically simplify, reinterpret and generalize a previously proposed fully Bayesian GP-SIM combination. The favorable performance of this model is demonstrated with both a synthetic data set and a computer experiment using real data.