Friday, October 1, 2021 - 11:30 in V2-210/216
Coupling Stochastic Gradient Methods with Mesh Refinement for PDE Constrained Optimization under Uncertainty
A talk in the BI.discrete Workshop series by
Winnifried Wollner from Darmstadt
| Abstract: |
In this talk we consider the optimization of a convex and smooth functional subject to a PDE constraint with uncertain coefficients. For the solution a stochastic gradient method is utilized to avoid detailed sampling of the random coefficients in every iteration. We will interpret the inexact PDE-solution, and thus gradient evaluation, as a bias term in the stochastic gradient method. We can then provide an a-priori coupling between iteration number and discretization accuracy to obtain the convergence rates known for the stochastic gradient method with exact, Unbiased, gradient evaluation. Numerical examples will complement the theoretical investigations.
This is joint work with Caroline Geiersbach (WIAS, Berlin).
Attendance is only possible after registration with the organizers and with
3G-certificate.
Within the CRC this talk is associated to the project(s): A7, B7 |
Back
