DATA ASSIMILATION, PARAMETER LEARNING, AND MACHINE LEARNING FOR LARGE COMPLEX MODELS
Abstract
The goal of the research is to develop new tools to quantify uncertainty in complex, heterogeneous computer models of physical systems. These tools will be based around novel combinations of ideas from the fields of data assimilation and machine learning. The growth of data acquisition provides an opportunity to use data to deepen our understanding of the world. However, the pressure to describe large and complex phenomena also grows, leading to models with unknown parameters, systematic biases and uncertainties, leading to associated uncertainty in their predictions. Quantifying this uncertainty makes the predictions considerably more valuable. The US Navy has direct interest in many such models, including those arising in oceanography; many other DoD applications share these characteristics. The new tools that are developed will be tailored to the specifics of complex models comprising many interactingblack-box components, and will be designed to be scalable on emerging computer architectures. As such they will have direct impact on many problems of interest within DoD.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 13, 2019
- Source ID
- N000141912408
Entities
People
- Andrew M. Stuart
Organizations
- California Institute of Technology
- Office of Naval Research
- United States Navy