Sandia Labs FY22 Laboratory Directed Research & Development Annual Report

FY22 ANNUAL REPORT

REDUCING THE COST OF QUANTIFYING UNCERTAINTY USING MULTI-FIDELITY FUSION AND RESOURCE ALLOCATION.

Quantifying uncertainty in computationally expensive numerical simulations via repeated evaluation of the highest-fidelity model is often intractable. Approximations of the highest fidelity model (e.g., simplified physics, reduced order, or ML models) can be used to reduce the computational cost of quantifying uncertainty but can introduce significant error. This two-year project developed novel multi-fidelity methods that judiciously integrate limited high-fidelity data alongside data from multiple cheaper models to both: (1) construct accurate ML models; and (2) produce unbiased estimates of uncertainty. The team’s methods have repeatedly reduced the cost of quantifying uncertainty by 1-3 orders of magnitude when applied to applications spanning plasma physics to ice-sheet models and can increase the utility of ML models, by leveraging them alongside high-fidelity simulation to make validated predictions with quantified uncertainty. This project resulted in five publications and numerous invited and contributed talks to several high-profile societies. The team partnered with students, postdocs, and faculty at various universities, including Sandia National/Regional partner University of Michigan, Sandia Alliance partner Texas A&M University, Notre Dame, University of Southern California, University of Utah, and Michigan State University. (PI: John Davis Jakeman)

By using novel multi-fidelity methods that integrate limited high-fidelity data alongside data from multiple cheaper models, the team repeatedly reduced the cost of quantifying uncertainty by 1-3 orders of magnitude in various applications from plasma physics to ice-sheet models.

IDENTIFYING AND CHARACTERIZING DISINFORMATION RISKS IN NATIONAL SECURITY MISSIONS. Cyber-influencing via foreign disinformation campaigns is a fundamental challenge that poses a serious threat to national security missions, especially since the rise of social media platforms has lowered the adversary’s cost of operations.

mathematical formalism called the non-abelian Fourier transform. The software, which combines the accurate identification of adversarial tactics with digital media context, determines likely indicators of a foreign disinformation campaign, and is already impacting Sandia missions in emerging spaces. It will also support future early warning systems. (PI: Michael Brzustowicz)

This work expanded the quantitative analytic methods available to the national security community by producing software based on a

68