SC²S Colloquium - October 1, 2012
|Date:||October 8, 2012|
|Time:||1:30 pm, s.t.|
Leo Ng: Multifidelity Uncertainty Propagation for Optimization Under Uncertainty
In this seminar, I will first present a general overview of uncertainty quantification (UQ), including sources of uncertainties and types of UQ problems. I will then present my research in optimization under uncertainty. In optimization under uncertainty problems, computing the statistics that comprise the objective function and/or constraints many times during the search for the optimum can be computationally expensive. In many practical situations, low-fidelity models are available and can provide useful information about the outputs of the high-fidelity model at a lower cost. To take advantage of these low-fidelity models in a rigorous manner, I present a multifidelity approach to Monte Carlo estimation of parameters such as the mean and the variance of the high-fidelity model outputs based on the control variate variance reduction method. Numerical experiments have shown 84% reduction in computational cost relative to using the high-fidelity model alone.