Abstract: Spectral clustering algorithm is a popular data clustering method. It uses eigen-decomposition technictue to extract the cigenvectors of the affinity matrix. But the method is infcasible for largcscale data set because of the store and computational problem. Motivated by the property of symmetric matrix, in this paper each column of the affinity matrix was used as the input sample for the iterative algorithm. hhe eigcnvectors of the affinity matrix could be iteratively computed. The space complexity of proposed method was only O(n)，the time complexity was reduced to O(pkm). The effectiveness of proposed method was validated from experimental results.

Key words: Spectral algorithm, Affinity matrix, I_argcscale data set, Eigen-decomposition