Hierarchical Multi-fidelity Surrogate Modeling

Abstract

AbstractMost of the existing multi-fidelity (MF) surrogates assume that high-fidelity (HF) models are generally more accurate than low-fidelity (LF) models but LF models are less expensive than HF models. In other words, the fidelity levels between the HF and LF models can be clearly identified, which is the core concept of hierarchical multi-fidelity surrogate modeling. The motivation of MF surrogate modeling is that a large number of inexpensive LF sampling points can be used to decrease the computational cost, while a limited number of expensive HF sampling points can be used to ensure the prediction accuracy of the surrogate model.