 |
Center for Nonlinear Studies |
The past few years have seen an explosion of data from astrophysical observations of neutron stars, giving us an unprecedented window into the nature of the densest matter in the universe. A comprehensive analysis of this data using modern Bayesian inference methods involves a large number of evaluations of computationally expensive multi-physics simulations. In this talk, I will discuss how machine learning-based algorithms can act as efficient surrogate models for some expensive LANL models relevant for astrophysical analyses. The emphasis will be on building such surrogates with limited, scarce data as well as uncertainty quantification for the models' predictions. Finally, I will discuss how these surrogate models are applicable to a wide range of LANL problems such as the analysis of thermonuclear fusion experiments performed at the National Ignition Facility.
Host: Wesley Even and Vitaliy Gyrya (T-5)