Universität Augsburg
|
Professor Dr. André Uschmajew
Max-Planck-Institut für Mathematik in den Naturwissenschaften
spricht am
Montag, 22. Januar 2018
um
16 Uhr s.t.
im
Raum 1007 (L1)
über das Thema:
Abstract: |
Low-rank tensor methods can be used for solving high dimensional equations and optimization problems (if the involved data admits low-rank structure). One family of methods are rank truncated iterations, in which classical iterative methods are combined with rank truncation in order to keep the dimensionality manageable. The low-rank truncation step can be seen as a perturbation to the iterative method and one may try to analyze the convergence properties from this viewpoint. One question that arises is about the worst case error of low-rank tensor approximation. For the simplest case of rank-one approximation this leads to the interesting question of finding the optimal norm constant between spectral and Frobenius norm of a tensor. This seemingly analytic question has a deep connection to algebra. We present recent results on this problem related to the concept of orthogonal tensors. |
Hierzu ergeht herzliche Einladung. |
Prof. Dr. Daniel Peterseim |