Deep Multi-fidelity Gaussian Processes
This work addresses a limitation in multi-fidelity modeling for complex systems, offering a solution for scenarios with discontinuous correlations, which is incremental but improves upon classical methods.
The paper tackled the problem of modeling discontinuous cross-correlations in multi-fidelity systems, developing a novel framework that combines multi-fidelity Gaussian processes with deep neural networks to handle such discontinuities effectively, as demonstrated on standard benchmark problems.
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.