Semi-supervised Power Iteration Clustering

Spectral clustering is of significance for many research areas, but high computation complexity restricted its power obviously. Even though Power Iteration Clustering(PIC) algorithm could speed up its process by random selection of the initial vector, it is still at the cost of accuracy reduction. Aiming at the problems above, we proposed one Semi-supervised Power Iteration Clustering(SPIC) algorithm that is based on semi-supervised learning and power iteration clustering(PIC). Several pairwise constraints could be added to modify the graph adjacency matrix while a new affinity matrix with partial supervision would be built. As opposed to standard spectral methods that largely depends on pairwise distances between points, our algorithm focuses on changing the part of the affinity matrix. At last, various synthetic datasets are collected, then several experiments are performed based on these datasets to validate the superiority of our proposed algorithm over the traditional algorithms.