Active Learning-Based Semi-supervised Spectral Clustering Algorithm

Abstract

Semi-supervised learning is one of the hottest research topics in the Machine Learning. The performance of semi-supervised clustering depends on the quality of supervision information, so it is necessary to actively learn high quality supervision information. An active learning algorithm based on pair-wise constraints with error correction is proposed in this paper. The algorithm searches the pair-wise constraints information which clustering algorithm can’t find, and try its best to reduce connections between this constraint information, which is used in the spectral clustering. Utilizing supervised information adjust the distance matrix in the spectral clustering, and sort the distances. The algorithm makes the learning can study actively when the learning receives the data without flags by the two-way search method, and get better clustering result with less constraints. Meanwhile, the algorithm reduces the computational complexity of the semi-supervised algorithms based on constraints and resolves the singular problem of the pair-wise constraints in the clustering process. Experimental results on UCI benchmark data sets and artificial data set states clearly the performance of the algorithm is better than other compared algorithms, and the performance of algorithm is better than the ones of the spectral clustering which randomly selects the supervision information.