Skip to main content
NASA Advanced Supercomputing (NAS) Division World-class modeling, simulation, and supercomputing in support of the NASA mission

Multi-Fidelity Modeling Under Uncertainty

AMS Seminar Series

Speaker: Ramin Bostanabad, University of California, Irvine

Presentation

Seminar Slide Deck (PDF-9.8MB)

Abstract

Design of many engineered systems such as advanced materials relies on solving complex optimization problems. This complexity stems from multiple sources such as lack of high-fidelity experimental data, inaccuracy and bias of simulations (due to, e.g., missing physics), high cost of data acquisition, and existence of categorical variables that give rise to combinatorially expanding design spaces. To holistically address these challenges, we propose a multi-fidelity Bayesian optimization (MFBO) framework for designing materials composition under uncertainty. Unlike existing MFBO methods, our framework (1) can handle high-dimensional input spaces that have categorical variables, (2) can leverage an “arbitrary” number of low-fidelity data sources with different sampling costs, and (3) can predict which low-fidelity sources are trustworthy. We will demonstrate the advantages of our framework over existing technologies using a host of analytic and engineering examples.

Biography

Photo of Ramin Bostanabad
Ramin Bostanabad is an assistant professor in the Mechanical and Aerospace Engineering Department at the University of California, Irvine (UCI). He is the founding director of PMACS lab, the recipient of 2021 NASA Early Career Faculty award, the chair of the ASME student chapter at UCI, and an editorial board member of the SMO Journal. Bostanabad earned his Ph.D. in 2019 from Northwestern University where his works were recognized with a number of awards including Terminal Year Fellowship, Martin Outstanding doctoral Fellowship, Predictive Science and Engineering Design Fellowship, and Walter P. Murphy Fellowship. At UCI, his groups’ research area is at the interface of engineering design, applied statistics, computational materials science, scientific machine learning, and computational mechanics. Bostanabad’s research has so far been supported by NSF, Toyota motor company, UC Fees program, ACRC, and ARPA-E.