Adaptive Selection of Decomposed Function Information Sources for Rapid Neural Networks

What is it about?

Multi-fidelity methods rely on inexpensive low-fidelity information sources to supplement an expensive high-fidelity model. However, in practice, viable low-fidelity information sources are not always available. To address this problem, we propose a method for selecting decomposed functions of the high-fidelity model for use as low-fidelity emulators. This method outperforms the single-fidelity method in multiple tests, including a hypersonic vehicle design example.

Featured Image