Fakultät Informatik


Kolloq. Robert Gantner, topic: Computational Higher-Order Quasi-Monte Carlo Methods for Random Partial Differential Equations

Tuesday, 4th of July 2017, 1:00 pm FMI 00.09.038 (MI-Building, Campus Garching)

We consider forward and inverse uncertainty quantification of partial differential equations with distributed uncertain inputs. Focusing on the Bayesian approach to inverse problems together with suitable uncertainty parametrization, the problem reduces to the computation of high-dimensional integrals. Under certain sparsity conditions on the parametric input, the resulting integrands fulfill analogous conditions, allowing recently developed higher-order QMC methods to be applied. We focus in this talk on computational aspects of this problem, including single-level and multilevel formulations of the discrete Bayesian inverse problem, which are shown to outperform conventional Monte Carlo-based approaches. Numerical results confirming the theoretical improvement of the time-to-solution for a given error level are presented, and software for the high-performance evaluation of the considered QMC rules is commented on.

Robert Gantner is Ph.D. Student at the Seminar for Applied Mathematics at ETH Zürich under supervision of Christoph Schwab. He is working in the field of Monte Carlo methods and has already 10 publications in this field. Robert Gantner won the BGCE student paper prize 2017, which was granted at SIAM CSE conference 2017 in Atlanta.

contact person:
Michael Rippl
Phone: +
Email: ripplm(at)in.tum.de



« December - 2017 »