The United States Department of Energy (DOE) Laboratory System grew out of the federally-funded scientific developments of World War II. Today, the national laboratories comprise one of the world’s largest scientific research systems. Tackling areas such as environmental modeling, precision medicine, and global security, the DOE laboratories are at the forefront of scientific innovation and, thus, have access to unique research problems, data sets, and facilities. This minisymposium will showcase the many applications and innovations in UQ stemming from the challenges of the national lab environment.
LLNL-ABS-791303. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.
14:00
Design of Experiments for Agent Based Models
Kevin Quinlan | Lawrence Livermore National Laboratory | United States
Show details
Authors:
Kevin Quinlan | Lawrence Livermore National Laboratory | United States
Charles H. Tong | Lawrence Livermore National Laboratory | United States
Jim R. Leek | Lawrence Livermore National Laboratory | United States
Joshua G. Sherfield | Lawrence Livermore National Laboratory | United States
Agent Based Models (ABM) mimic the operation of a real or proposed system, such as the day-to-day operation of the stock market, the running of an assembly line in a factory, or the interactions on computer networks. Often the simulation models a large number of agents with multiple parameter settings each, but the agents may also fall into homogenous classes. Considering these challenges, we present design algorithms for this unique setting and more generally high dimensional problems. Our algorithm outperforms currently available software in terms of two commonly used space-filling criterion in most cases as well as in computation time. We also make suggestions for dealing with problems with multiple parameters per agent.
This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.
Release Number: LLNL-ABS-791177
14:30
- CANCELED - Boosting Carbon Capture with Sequential Design
James R. Gattiker | Los Alamos National Laboratory | United States
Show details
Authors:
James R. Gattiker | Los Alamos National Laboratory | United States
Christine Anderson-Cook | Los Alamos National Laboratory | United States
Sham Bhat | Los Alamos National Laboratory | United States
Carbon Capture Simulation for Industry Impact (CCSI2) is an applied modeling and analysis project sponsored by the US Department of Energy bringing together national laboratories, academia, and industry partners to accelerate the deployment of carbon capture technology. This talk will focus on the principles and application of Sequential Design of Experiments across elements of CCSI2, spanning from materials and chemistry research to large-scale engineered systems design, involving design of both computer experiments and experimental observations across scales. We present examples discussing target inferences, the science and engineering computational models used to estimate the utility of observations in those inferences, the theory of sequential design combining these resources with observations, and the process and results in the applied problem. The sequential design methodology is codified in the CCSI FOQUS toolset, an open-source resource customized and targeted to industry partners in this domain.
15:00
Scaling and Uncertainty in Model-Based Clustering of Medical Trajectories
George Ostrouchov | Oak Ridge National Laboratory and University of Tennessee | United States
Show details
Author:
George Ostrouchov | Oak Ridge National Laboratory and University of Tennessee | United States
Different patient response patterns over time to a treatment are of great interest in precision medicine. These response patterns can be used as phenotypes for the development of patient-specific treatments. We align a collection of patient measurement sequences by treatment date and cluster the measurement trajectories via model-based clustering. The models are temporal spline fits to the trajectories, which are computed by an EM-like algorithm. Uncertainty is used to determine the number of clusters. This talk describes the methodology, its implementation in R, and scaling with pbdR packages on a cluster system to make the computation time bearable for a very large cohort from the Million Veterans Project.
15:30
Implementing Bayesian UQ for Complex Systems
Adah Zhang | Sandia National Laboratories | United States
Show details
Authors:
Adah Zhang | Sandia National Laboratories | United States
Gabriel Huerta | Sandia National Laboratories | United States
At Sandia National Laboratories, Bayesian approaches have been applied to complex system situations. Two case-studies are: quantifying uncertainty for reliability estimates of highly reliable complex systems with sparse data, and quantifying behavioral uncertainty for digital representations of physical insults in a computer simulated digital system.
When complex system level reliability tests (pass/fail) are infeasible due to expense and time constraints, cheaper component level data can be rolled up in a reliability block diagram and used as a surrogate for the system level reliability estimate. When component level data is limited or highly reliable, additional information from historical data and experts are needed to help inform the component level uncertainty, which gets rolled up into the system level uncertainty estimate.
While the complex system level reliability data can be represented by a Beta-binomial distribution, the multivariate extension, Dirichlet-Multinomial distribution, can be used in the second case-study. This case study estimates uncertainty in proportions of digital message errors due to a physical insult and environmental parameters. When digital upsets are incorporated in a formal or complexity model along with uncertainty-informed simulation results, complex system performance and outcomes can be assessed.