Large-Scale Spectral Clustering Based on Representative Points

Algorithm 1

Representative point-based spectral clustering.

	Input: Data matrix
	;
	Output: clusters;
(1)	Produce the two-layer representative points using BKHK successively;
(2)	Construct the similarity matrixes B1 and B2 according to (6);
(3)	Construct the final similarity matrix A between original data and the second-layer representative points according to (7);
(4)	Compute D1 and D2: ; ;
(5)	Compute ;
(6)	Compute singular vectors of : , ;
(7)	Form the matrix ;
(8)	Run the k-means algorithm on the matrix Z to obtain the desired -way multipartitioning.