A new spectral clustering algorithm for large training sets

Author

Prieto, Ramon ; Jiang, Jing ; Choi, Chi-Ho

Author_Institution

Dept. of Electr. Eng., Stanford Univ., CA, USA

Volume

1

fYear

2003

fDate

2-5 Nov. 2003

Firstpage

147

Abstract

A new algorithm for spectral clustering is depicted. This algorithm can cluster a large number of samples (in the order of 50000 samples) that would be impossible to cluster with current approaches. It is characterized by a complexity that is significantly lower than the cubic complexity that characterizes the calculation of the eigenvectors of a matrix. It\´s based on a "clustering of clusters" technique, that combines the use of k-means and spectral clustering. Additionally, this method includes the use of expectation maximization (EM) clustering with axis smoothing that is shown to improve the separation in the spectral domain for high values of the scaling parameter σ².

Keywords

computational complexity; eigenvalues and eigenfunctions; learning (artificial intelligence); matrix algebra; pattern clustering; axis smoothing; clustering of clusters technique; complexity; expectation maximization clustering; k-means clustering; matrix eigenvectors; spectral clustering algorithm; training sets; Clustering algorithms; Clustering methods; Eigenvalues and eigenfunctions; Fuzzy sets; Gaussian distribution; Laplace equations; Machine learning algorithms; Smoothing methods; Speech recognition; Tin;

fLanguage

English

Publisher

ieee

Conference_Titel

Machine Learning and Cybernetics, 2003 International Conference on

Print_ISBN

0-7803-8131-9

Type

conf

DOI

10.1109/ICMLC.2003.1264460

Filename

1264460