Oberseminar Wissenschaftliches Rechnen (Jun.-Prof. Anne Wald, Georg-August-Universität Göttingen)

Abstract:

Many applied inverse problems are inherently dependent on time, for example due to a change in the physical setup or because of a motion or deformation of an investigated object. This timedependence often leads to complex physical models and, accordingly, to a high computational eort during the numerical solution of such inverse problems. In order to overcome these obstacles, we aim at using simplied forward operators and interpret the dependence on time as an inexactness in the model, which we incorporate into the regularization scheme. An extension of sequential subspace optimization techniques in combination with Kaczmarz' method towards inverse problems with an inexact forward model not only provides a regularized solution of such problems, but yields an additional speed-up in comparison to, e.g., the Landweber iteration. Dynamic computerized tomography, in particular nanoscale CT imaging of metal alloys, serves as an example to evaluate these methods.

via Zoom video conference (request the Zoom link from frank.werner@mathematik.uni-wuerzburg.de)
Alle Vorträge im Oberseminar Wissenschaftliches Rechnen