The University of Arizona

Efficient Monte Carlo methods and machine learning-based surrogate modeling for uncertainty quantification

Efficient Monte Carlo methods and machine learning-based surrogate modeling for uncertainty quantification

Series: Modeling and Computation Seminar
Location: Math 402
Presenter: Yaning Liu, PhD, Lawrence Berkeley National Laboratory, California

Quantifying the uncertainty in mathematical models allows us to quantitatively describe our confidence in the model predictions in the presence of parametric uncertainty. In this talk, we will discuss three closely connected aspects of probabilistic uncertainty quantification, namely the forward propagation of model uncertainty, global sensitivity analysis and Bayesian parameter inversion. For the forward uncertainty propagation, we will show that a Monte Carlo algorithm exploiting the sensitivity derivatives and randomized quasi-random sequences can improve the convergence rate by orders of magnitude. For Bayesian parameter inversion, we use implicit particle filter, which is capable of outperforming and replacing Markov chain Monte Carlo. These methods all require a large number of model simulations and the use of high-fidelity models is thus computationally intractable. To increase the efficiency, we consider how different types of machine learning-based surrogate models can be incorporated into these methods. In describing the above methods, we will use a wide range of applications, including the quantification of uncertainty associated with wildland fire behavior, constructing a hybrid surrogate model for fine-resolution soil moisture modeling, an efficient framework of multi-fidelity surrogate modeling for river basin hydrology modeling, and combining implicit particle filter with surrogate modeling for a vadose zone hydrological model.  

(Refreshments will be served.)

Department of Mathematics, The University of Arizona 617 N. Santa Rita Ave. P.O. Box 210089 Tucson, AZ 85721-0089 USA Voice: (520) 621-6892 Fax: (520) 621-8322 Contact Us © Copyright 2017 Arizona Board of Regents All rights reserved