PIMS- SFU CS Distinguished Colloquium: Mike Eldred

  • Date: 03/23/2018
  • Time: 15:00
Mike Eldred, Sandia National Labs

Simon Fraser University


Multilevel-Multifidelity Approaches for Uncertainty Quantification and Design Under Uncertainty


In the simulation of complex physics, multiple model forms of varying fidelity and resolution are commonly available. In computational fluid dynamics, for example, common model fidelities include potential flow, inviscid Euler, Reynolds-averaged Navier-Stokes, and large eddy simulation, each potentially supporting a variety of spatio-temporal resolutions / discretizations. While we seek analysis results that are consistent with the highest fidelity and finest discretization, the computational cost of directly applying uncertainty quantification (UQ) in high random dimensions quickly becomes prohibitive. In this presentation, we will overview the development of multilevel-multifidelity (MLMF) approaches that combine information from multiple fidelities and resolutions in order to reduce the overall computational burden. We will focus primarily on forward propagation, and describe extensions for statistical inversion and design under uncertainty.


For forward UQ with MLMF, we are interested in both Monte Carlo sampling approaches that demonstrate robustness and scalability and in emulator-based approaches (e.g., polynomial chaos) that can exploit available structure (sparsity via compressed sensing, low rank via tensor train) to accelerate convergence. For inverse UQ, we leverage the latter MLMF emulators for accelerating the Markov-chain Monte Carlo sampling process and refine the emulators in regions of high posterior density. And for design under uncertainty, we employ multigrid optimization and recursive trust region model management to manage hierarchies of both simulation fidelity and statistical resolution. Performance of these different MLMF strategies will be described for both model problems and engineered systems such as integrated aircraft nozzles and scramjets.



Mike received his B.S. in Aerospace Engineering from Virginia Tech in 1989, his M.S.E. and Ph.D. in Aerospace Engineering from the University of Michigan in 1990 and 1993, and is currently a Distinguished Member of the Technical Staff in the Optimization and Uncertainty Quantification Department within the Center for Computing Research at Sandia National Laboratories.
Mike led the DAKOTA project for 15 years (1994-2009) and now leads algorithm research and development activities related to DAKOTA. Mike's research interests include uncertainty quantification, design under uncertainty, multifidelity modeling, surrogate-based optimization, and high-performance computing, with application to stockpile stewardship and energy initiatives through the NNSA ASC, DOE ASCR, and DOE SciDAC programs. A number of his publications are available on the DAKOTA web site.
Mike is an Associate Fellow of the American Institute of Aeronautics and Astronautics (AIAA) and a member of the Society for Industrial and Applied Mathematics (SIAM) and the United States Association for Computational Mechanics (USACM). Mike currently serves as an alumni member of the AIAA Nondeterministic Approaches Technical Committee and on the editorial board for the International Journal for Uncertainty Quantification.

Other Information: 

SFU's Big Data Hub, Rm 10900
Reception: 2:30 pm
Lecture: 3:00 pm