The Annals of Statistics
- Ann. Statist.
- Volume 43, Number 3 (2015), 962-990.
Role of normalization in spectral clustering for stochastic blockmodels
Spectral clustering is a technique that clusters elements using the top few eigenvectors of their (possibly normalized) similarity matrix. The quality of spectral clustering is closely tied to the convergence properties of these principal eigenvectors. This rate of convergence has been shown to be identical for both the normalized and unnormalized variants in recent random matrix theory literature. However, normalization for spectral clustering is commonly believed to be beneficial [ Stat. Comput. 17 (2007) 395–416]. Indeed, our experiments show that normalization improves prediction accuracy. In this paper, for the popular stochastic blockmodel, we theoretically show that normalization shrinks the spread of points in a class by a constant fraction under a broad parameter regime. As a byproduct of our work, we also obtain sharp deviation bounds of empirical principal eigenvalues of graphs generated from a stochastic blockmodel.
Ann. Statist., Volume 43, Number 3 (2015), 962-990.
Received: October 2013
Revised: November 2014
First available in Project Euclid: 15 May 2015
Permanent link to this document
Digital Object Identifier
Mathematical Reviews number (MathSciNet)
Zentralblatt MATH identifier
Sarkar, Purnamrita; Bickel, Peter J. Role of normalization in spectral clustering for stochastic blockmodels. Ann. Statist. 43 (2015), no. 3, 962--990. doi:10.1214/14-AOS1285. https://projecteuclid.org/euclid.aos/1431695635
- Supplement to “Role of normalization in spectral clustering for stochastic blockmodels”. Because of space constraints we have moved some of the technical details to the supplementary material .