Deep Multi-fidelity Gaussian Processes

TL;DR

This paper introduces a new multi-fidelity Gaussian Process framework that effectively models complex, discontinuous correlations across different fidelity levels, surpassing traditional AR(1) Co-kriging methods.

Contribution

It combines Gaussian Processes with deep neural networks to handle discontinuities in multi-fidelity modeling, extending beyond classical methods.

Findings

Effective on benchmark problems resembling complex high- and low-fidelity outputs

Handles general discontinuous cross-correlations among systems

Outperforms classical AR(1) Co-kriging in complex scenarios

Abstract

We develop a novel multi-fidelity framework that goes far beyond the classical AR(1) Co-kriging scheme of Kennedy and O'Hagan (2000). Our method can handle general discontinuous cross-correlations among systems with different levels of fidelity. A combination of multi-fidelity Gaussian Processes (AR(1) Co-kriging) and deep neural networks enables us to construct a method that is immune to discontinuities. We demonstrate the effectiveness of the new technology using standard benchmark problems designed to resemble the outputs of complicated high- and low-fidelity codes.