Activity Number:
|
571
- Emerging Issues in Uncertainty Quantification for Computer Experiments
|
Type:
|
Topic Contributed
|
Date/Time:
|
Thursday, August 6, 2020 : 3:00 PM to 4:50 PM
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
Abstract #312990
|
|
Title:
|
Sequential Design of High-Dimensional Multifidelity Computer Models
|
Author(s):
|
Bledar Konomi* and Pulong Ma and Georgios Karagiannis
|
Companies:
|
University of Cincinnati and The Statistical and Applied Mathematical Sciences Institute and Durham University
|
Keywords:
|
Co-kriging ;
Augmented hierarchically nested desig
|
Abstract:
|
Computer models are often able to run at different levels of fidelity, sophistication, or resolution. As high fidelity runs are usually more expensive, collecting data by simulating the model at different fidelity levels is preferred for a given budget of resources. Predictive uncertainty of the statistical emulator combined with the computational complexity have usually been used to guide subsequent experimental runs from different fidelity computer models. The present sequential experimental designs in the multifidelity framework have been built upon the unrealistic assumptions of univariate computer model output and hierarchically nested design. In this work, we propose to extend sequential design procedure for multifidelity computer models to account for high-dimensional output and non-hierarchically nested design. The proposed sequential design will be used to produce more accurate emulators for physics-based computer models of storm surge.
|
Authors who are presenting talks have a * after their name.