Robust Multiple Kernel Subspace Clustering Based on Low Rank Consensus Kernel Learning
Liang Qin, Xiaoqian Zhang, Liang Luo · Proceedings of the 2020 4th International Conference on Electronic Information Technology and Computer Engineering · 2020
Subspace clustering method based on kernel learning shows superior performance when dealing with nonlinear high dimensional data. However, the clustering performance of existing single kernel subspace clustering methods largely depends on the selected kernel function, and the low-rank structure of the data in the feature space after kernel mapping is not considered. In addition, the learned affinity matrix cannot maintain the block diagonal property, which may reduce the clustering performance. In this paper, we propose a robust multiple kernel subspace clustering based on low rank consensus kernel learning (MKLRSC) method for data clustering. Our model has three innovations: (1) The introduction of correntropy in the multiple kernel weighting strategy helps to learn the optimal consensus kernel. (2) In order to maintain the low-rank structure of input data, we impose a nuclear norm constraint on the optimal consensus kernel matrix. (3) Considering the block diagonal property of the affinity matrix, MKLRSC applies block diagonal constraint to the coefficient matrix. Compared with several state-of-the-art multiple kernel subspace clustering methods, experiments on three datasets confirm that MKLRSC achieves more competitive clustering results.