Journal
NEUROCOMPUTING
Volume 275, Issue -, Pages 2855-2863Publisher
ELSEVIER
DOI: 10.1016/j.neucom.2017.11.061
Keywords
Feature selection; Co-regularized learning; Data distribution; Data reconstruction
Categories
Funding
- National Program on Key Basic Research Project [2013CB329304]
- National Natural Science Foundation of China [61502332, 61432011, 61222210]
- Natural Science Foundation of Tianjin [17JCZDJC30800]
Ask authors/readers for more resources
Unsupervised feature selection (UFS) is very challenging due to the lack of label information. Most UFS methods generate pseudo labels by spectral clustering, matrix factorization or dictionary learning, and convert UFS into a supervised feature selection problem. Generally, the features that can preserve the data distribution (i.e., cluster indicator matrix) should be selected. Under the matrix factorization framework, when the cluster indicator matrix is orthogonal, the cluster base (center) matrix can be used to select features that well reconstruct the data. However, almost all the UFS algorithms only select features that can either well preserve the cluster structure or reconstruct the data. In this paper, we propose a novel co-regularized unsupervised feature selection (CUFS) algorithm. Joint l(2,1)-norm co-regularization is imposed on multiple feature selection matrices to ensure that the selected features can both preserve data distribution and data reconstruction. Extensive experiments on benchmark datasets show that CUFS is superior to the state-of-the-art UFS algorithms. (c) 2017 Elsevier B.V. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available