Divide-and-conquer based large-scale spectral clustering

Spectral clustering is one of the most popular methods. However, how to balance efficiency and effectiveness large-scale spectral with limited computing resources has not been properly solved for a long time. In this paper, we propose divide-and-conquer based method strike good between effectiveness. proposed method, landmark selection algorithm novel approximate similarity matrix approach are designed construct sparse within low computational complexities. Then results can be computed quickly through bipartite graph partition process. The achieves lower complexity than existing Experimental on ten datasets have demonstrated method. MATLAB code experimental available at https://github.com/Li-Hongmin/MyPaperWithCode.