Improving Experimental Uncertainty via Pre-built Gaussian Process Emulators

Kelly Moran, Duke University

Photo of Kelly Moran

Analyzing and extracting knowledge from modern experiments has become the rate-limiting step in the scientific process. This project aims to accelerate knowledge and discovery from experimental scientific facilities by combining computer and statistical science to produce an adaptive methodology and toolset that will analyze data and augment decision-making so the scientist can optimize experiments in real time. The team is developing this capability in the context of dynamic compression experiments, an area of core mission importance for Los Alamos National Laboratory and one in the midst of substantial data generation rate increases. These dynamic compression experiments consist of a multi-dimensional input parameter space (some of which is estimated, some of which is set by the experimenter) leading to a multi-dimensional output space. Inputs the experimenter sets include such parameters as time delay of X-ray probe pulse and angle of X-rays relative to shock. Those that must be estimated include shock pressure, material strength and crystal orientations. The measured outputs include velocimetry, diffraction and imaging. Specifically, this component of the project focuses on improving experimental uncertainty via pre-built Gaussian process emulators. We automate the construction of emulators so they can be built in advance of experimentation and used quickly in later analyses. The key extension to current methodology is the inclusion of emulation to facilitate accurate experiment calibration: An emulator is used to determine the distribution of physics parameters that best match the data. We account for uncertainty from the Markov-Chain Monte Carlo (MCMC) fitting process as well as from the noisy data generation process.

Abstract Author(s): Kelly Moran