PIMS-SFU CSC Seminar: Akil Narayan

We present an algorithm for coupling inexpensive low-fidelity model simulations with high-fidelity simulation data of parameterized differential equations. The goal is to grab a "free lunch": simulation accuracy of the high-fidelity model with algorithmic complexity of only a simplified low-fidelity model. The procedure forms an approximation with sparsely available high-fidelity simluations, and is a simple linear algebraic construction with connections to kernel learning and column skeletonization of matrices. We discuss theoretical results that establish accuracy guarantees, and introduce recent analysis providing bounds on the error committed by the algorithm.