Collaborative Research: Multifidelity Uncertainty Quantification Through Model Ensembles and Repositories

Stanford University, Stanford CA

Investigators

Abstract

Numerical simulations are increasingly used in clinical research and practice for diagnosis and treatment planning in cardiovascular disease, creating new demand for reliable simulation and analysis tools. Quantification of uncertainty in these simulations is crucial to increased clinical adoption but has previously been largely disregarded due to its excessive computational cost and complexity. To address these challenges, the project leverages a new class of multi-fidelity Monte Carlo estimators for direct and inverse problems, designed to mitigate computational complexity through the solution of a large number of inexpensive low-fidelity surrogates. It demonstrates the proposed approach in full-scale clinical problems including multiple uncertainty sources at a reasonable computational budget. The project’s main objective is to create an end-to-end advanced cyberinfrastructure ecosystem for uncertainty quantification (direct problem) and parameter estimation (inverse problem) in cardiovascular models incorporating realistic sources of uncertainty, able to leverage arbitrary low-fidelity models through advanced Monte Carlo estimators, while drastically reducing computational cost and complexity. The project’s interdisciplinary team is synergizing computational modeling, cardiovascular physiology, UQ and open-source software towards making UQ tractable in full-scale 3D cardiovascular simulations, leveraging multi-fidelity estimators for the solution of both direct and inverse problems. The project will produce seamless cyberinfrastructure linking two well-regarded open-source packages, Dakota and SimVascular, with sizable user communities. The project is creating new cyberinfrastructure ecosystems for large-scale UQ tasks. Dissemination to industry/academia is performed through SimVascular, a leading open-source platform for cardiovascular modeling. It will leverage SimVascular and the proposed multi-fidelity estimators to create hands-on teaching material for graduate and undergraduate courses. Although the project focuses on cardiovascular modeling, its results are directly applicable to other engineering problems. The PIs will organize minisymposia and workshops at national conferences. They will lead outreach activities to local K-12 schools to attract girls and underrepresented minority (URM) students to STEM. The PIs will mentor URM summer students through the SURF program and women students through the Women in Mathematics, Scientific Computing and Engineering (WiMSCE) group at Stanford. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.