Unsupervised feature selection via self-paced learning and low-redundant regularization

Much more attention has been paid to unsupervised feature selection nowadays due the emergence of massive unlabeled data. The distribution samples and latent effect training a learning method using in effective order need be considered so as improve robustness method. Self-paced is an considering samples. In this study, proposed by integrating framework self-paced subspace learning. Moreover, local manifold structure preserved redundancy features constrained two regularization terms. L 2 , 1 / -norm applied projection matrix, which aims retain discriminative further alleviate noise Then, iterative presented solve optimization problem. convergence proved theoretically experimentally. compared with other state art algorithms on nine real-world datasets. experimental results show that can performance clustering methods outperform algorithms.