Convergence Analysis of Affinity Propagation

Authors:
Jian Yu;Caiyan Jia
Affiliations:
Department of Computer Science, Beijing Jiaotong University, Beijing, P.R. China 100044;Department of Computer Science, Beijing Jiaotong University, Beijing, P.R. China 100044
Venue:
KSEM '09 Proceedings of the 3rd International Conference on Knowledge Science, Engineering and Management
Year:
2009

Citing 4
Cited 0

Data clustering: a review

ACM Computing Surveys (CSUR)
Clustering by soft-constraint affinity propagation

Bioinformatics
Finding image exemplars using fast sparse affinity propagation

MM '08 Proceedings of the 16th ACM international conference on Multimedia
Acquiring critical light points for illumination subspaces of face images by affinity propagation clustering

PCM'07 Proceedings of the multimedia 8th Pacific Rim conference on Advances in multimedia information processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Recently, Frey & Dueck proposed a novel clustering algorithm named affinity propagation (AP), which has been shown to be powerful as it costs much less time and reaches much lower error. However, its convergence property has not been studied in theory. In this paper, we focus on convergence property of the algorithm. The properties of the decision matrix when the affinity propagation algorithm converges are given, and the criterion that affinity propagation without the damping factor oscillates is obtained. Based on such results, we point out that damping factor might be important to alleviate oscillation of the affinity propagation, but it is not necessary to add a tiny amount of noise to a similarity matrix.