NMF with feature relationship preservation penalty term for clustering problems

作者：

Highlights：

• Add the relationship between the data scatter and that of the centroids to the cost function of the NMF model.

• The update of W becomes parameterizable.

• The convergence towards the cluster centers is improved.

• The proposed method outperforms the existing orthogonal NMF methods.

摘要

•Add the relationship between the data scatter and that of the centroids to the cost function of the NMF model.•The update of W becomes parameterizable.•The convergence towards the cluster centers is improved.•The proposed method outperforms the existing orthogonal NMF methods.

论文关键词：NMF,Orthogonal NMF,Clustering,Unsupervised learning,Low-rank matrix factorization,

论文评审过程：Received 4 February 2019, Revised 31 October 2020, Accepted 1 January 2021, Available online 7 January 2021, Version of Record 15 January 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.107814