April 25; 3 pm EDT
Speaker: Erin Acquesta, Sandia National Laboratories
Title: Managing Uncertainty and Credibility Assessments for Epidemiology Exascale Agent Based Models
The combination of ExaScale computing and scientific machine learning (SciML) are providing great potential for effective decision-making at scale for high-consequence and uncertain complex systems. While the potential has been realized across the computational modeling community, it also introduces new challenges in managing the many sources of uncertainties associated with the data, models, and computational methods. There is a 30-year history and leadership in verification, validation, and uncertainty quantification (VVUQ) established by the Department of Energy (DOE) National Nuclear Security Administration (NNSA) Labs to evaluate the credibility of computational models used in high consequence applications. In this presentation we will articulate why credibility is important and review the Sandia framework known as the Predictive Capability Maturity Model (PCMM). We will further the discussion focused on an epidemiology example that leverages neural network (NN) function approximations of the model-form error corrections in compartmental models. A valuable tool to reduce model discrepancy, these NN approximations introduce new challenges in evaluating VVUQ principles for hybrid models. We will highlight these challenges as we demonstrate the design of a SciML surrogate to an exascale agent-based model while highlighting opportunities to characterize and manage the uncertainties in the classic computational model, the NN model-form error corrections, data, and numerical inconsistencies.
Join Zoom Meeting
https://us06web.zoom.us/j/92756548524?pwd=cTFoRXIvNVN4dVFoaHEzK0pQQjhldz09 (Meeting ID: 927 5654 8524/Passcode: 934745)
Please forward to anyone who may be interested.