Low-rank Subspace Consistency Clustering

The Clustering algorithm realizes data partition by similarity relation among data, and similarity calculation often depends on practical feature expression. Traditional shallow feature representation learning models cannot capture the deep semantic concepts between data and often fail to meet the current requirements of big data applications due to high computational complexity. In recent years, the depth subspace clustering algorithm has solved the above problems to some extent. However, in the process of feature expression learning, the existing deep clustering algorithms overemphasize the data reconstruction ability of the algorithm but ignore the rich potential structure information. Therefore, a new representation learning method is proposed in this paper, which adopts the consistency of subspace and uses the similarity between data samples as the supervision to guide the representation learning so that the subspace structure after the original data space transformation is more stable and the feature expression is more robust. A large number of experiments on public data sets have proved the good performance of this method in clustering.