Improved Generalized Successive Projection Algorithm for Generalized Separable Non-negative Matrix Factorization
Mingyang Liu, Junhang Chen, Lingjiang Li, Zuyuan Yang · 2022 41st Chinese Control Conference (CCC) · 2022
Non-negative matrix factorization (NMF) is a widely used technique for dimensionality reduction, and generalized separable NMF (GSNMF) can learn the representation with better interpretability, as it decomposes the given matrix based on the row features and the column features at the same time. But in some cases, the GSNMF algorithm faces the 0-$K$problem, where only one perspective of feature can be developed. This paper modified the generalized successive projection scheme, and proposes an improved generalized successive projection algorithm (IGSPA) to avoid the 0-$K$problem. To verify the effectiveness of our method, we conduct extensive experiments on three commonly used face datasets. Compared with the existing methods, numerical experiments show that our method has superior performance.