Semi-supervised Subspace Learning Via Constrained Matrix Factorization
Viet-Hang Duong, Manh Quan Bui, Jia‐Ching Wang · 2021
This paper adopts the matrix factorization approach by improving the NMF model to build a semi-supervised learning framework (DCNMF) that integrates the linear discriminate analysis (LDA) and base cone volume constraints. The proposed DCNMF is a subspace learning model in which minimizing the basic cone volume of the learned subspace and reducing the data dimensionality so that the distance within-class samples is minimized and the distance between-class samples is maximized. The proposed method is evaluated by experiments on two cases of face recognition tasks, namely, various numbers of training data and different subspace dimensionalities.