Disputation von Sven Kurras am 22.03.2017, 11:00 Uhr, ZBH Raum 16

22. März 2017, von Reinhard Zierke

Einladung zur hochschulöffentlichen Disputation von
Herrn Sven Kurras

Titel der Dissertation:
Variants of the Graph Laplacian with Applications in Machine Learning

Einladung zur hochschulöffentlichen Disputation
im Rahmen des Promotionsverfahrens von

Herrn Sven Kurras

Titel der Dissertation:
Variants of the Graph Laplacian with Applications in Machine Learning

Abstract:

Structural properties of a graph often correspond to algebraic properties of an associated graph matrix, such as the adjacency matrix or the graph's Laplacian matrix. This thesis studies three variants of the Laplacian and derives novel approaches to clustering, bias correction, and multi-scale analysis of graphs. The defense talk puts its focus on correlation clustering, that is clustering of a graph whose edge weights encode similarities and dissimilarities by positive and negative weights, respectively. By allowing for negative edge weights, most results that are known for the standard setting of a pure similarity-weighted graph need to be re-considered. In particular, the "trivial" lowest eigenvector of the Laplacian turns out to be no longer trivial, but to become the core ingredient of a novel spectral approach to correlation clustering. This re-interpretation is presented during the talk, and the derived algorithm is shown to compete with state-of-the-art.

Datum und Uhrzeit: Mittwoch, 22. März 2017 um 11:00 Uhr
Ort: Zentrum für Bioinformatik, Raum 16, Bundestraße 43, 20146 Hamburg

Betreuerin: Prof. Dr. Ulrike von Luxburg

Prof. Dr. Matthias Rarey
Vorsitzender des Fach-Promotionsausschusses