Multifidelity data fusion in convolutional encoder/decoder networks

Lauren Partin, Gianluca Geraci, Ahmad A. Rushdi, Michael S. Eldred, Daniele E. Schiavazzi

Abstract: Highlights•We consider a U-Net architecture trained on heterogeneous data combining high-fidelity and low-fidelity examples.•Modular encoder/decoder architectures generate flexible surrogate models for high- and low-dimensional inputs and outputs.•We use DropBlock layers to characterize uncertainty in the proposed multifidelity network.