[Back]


Talks and Poster Presentations (with Proceedings-Entry):

P. Berger, T. Dittrich, G. Matz:
"Clustering on Dynamic Graphs based on Total Variation";
Talk: 2019 International Conference on Sampling Theory and Applications (SampTA), Bordeaux (invited); 07-08-2019 - 07-12-2019; in: "Proceedings of the 13th International Conference on Sampling Theory and Applications", IEEE, (2019), 4 pages.



English abstract:
We consider the problem of multiclass clustering on dynamic graphs. At each time instant, the proposed algorithm performs local updates of the clusters in regions of nodes whose cluster affiliation is uncertain and may change. These local cluster updates are carried out through semi-supervised multiclass total variation (TV) based clustering. The resulting optimization problem is shown to be directly connected to a minimum cut and thus very well suited to capture local changes in the cluster structure. We propose an ADMM based algorithm for solving the TV minimization problem. Its per iteration complexity scales linearly with the number of edges present in the local areas under change and linearly with the number of clusters. We demonstrate the usefulness of our approach by tracking several objects in a video with static background.


Electronic version of the publication:
https://publik.tuwien.ac.at/files/publik_283819.pdf


Created from the Publication Database of the Vienna University of Technology.