A Comparative Study of Local Search Algorithms for Correlation Clustering
Research output: Contribution to book/Conference proceedings/Anthology/Report › Conference contribution › Contributed › peer-review
Contributors
Abstract
This paper empirically compares four local search algorithms for correlation clustering by applying these to a variety of instances of the correlation clustering problem for the tasks of image segmentation, hand-written digit classification and social network analysis. Although the local search algorithms establish neither lower bounds nor approximation certificates, they converge monotonously to a fixpoint, offering a feasible solution at any time. For some algorithms, the time of convergence is affordable for all instances we consider. This finding encourages a broader application of correlation clustering, especially in settings where the number of clusters is not known and needs to be estimated from data.
Details
Original language | English |
---|---|
Title of host publication | Pattern Recognition |
Editors | Volker Roth, Thomas Vetter |
Publisher | Springer, Cham |
Pages | 103–114 |
ISBN (electronic) | 978-3-319-66709-6 |
ISBN (print) | 978-3-319-66708-9 |
Publication status | Published - 2017 |
Peer-reviewed | Yes |
Publication series
Series | Lecture Notes in Computer Science |
---|---|
Volume | 10496 |
ISSN | 0302-9743 |
External IDs
ORCID | /0000-0001-5036-9162/work/161888479 |
---|