Thema der Dissertation:
Approximation methods for materials with discontinuities Thema der Disputation:
Kernel methods: From hard-margin SVM to operator estimation
Approximation methods for materials with discontinuities Thema der Disputation:
Kernel methods: From hard-margin SVM to operator estimation
Abstract: Kernel methods in machine learning provide a common framework for learning classification rules and real-valued prediction functions from data. Starting from the hard-margin support vector machine (SVM), this talk explains its formulation as a convex optimization problem and shows how the dual formulation motivates the kernel trick, which allows nonlinear classification without explicitly transforming the data. In this context, reproducing kernel Hilbert spaces (RKHSs) and the representer theorem are introduced. This approach is then adapted to regression and subsequently extended to the approximation of mappings between function spaces, with particular emphasis on learning solution operators associated with parameter-dependent partial differential equations.
Zeit & Ort
21.08.2026 | 10:30
Seminarraum 006
(Fachbereich Mathematik und Informatik, Königin-Luise-Str. 24-26, 14195 Berlin)