In the field of data mining and machine learning, non-negative matrix factorization (NMF) has received extensive attention as an efficient method for data dimensionality reduction and feature representation.However, standard NMF is sensitive to noise or outliers and can handle only non-negative data, thus is prone to overfitting and reduced generalization and robustness when dealing with small sample size (SSS) problems where features and samples are imbalanced.To overcome these limitations and achieve better clustering performance, a robust NMF algorithm based on sparsity constraints and exponential graph regularization is proposed.The Semi-NMF is used to handle mixed-sign data, the norm is employed to mitigate the influence of noise and outliers, and the exponential graph regularization is introduced to preserve the distribution characteristics and geometric structure information of the data.Meanwhile, the combination of sparsity and orthogonality constraints is used to reduce the redundant information, enhance the independence of features and prevent overfitting.Comparitive experimental results on nine public datasets show that the clustering accuracy and robustness of the proposed algorithm are superior to those of the other 8 classic clustering algorithms, which verifys its effectiveness in small sample clustering tasks.
本文基于Semi-NMF框架,结合指数图正则化、稀疏性约束及正交性约束条件提出了一种基于稀疏约束与指数图正则化的鲁棒非负矩阵分解聚类算法(Robust NMF via Sparse Constraints and Exponential Graph Regularization, RNMFSCEG)。算法通过引入范数重构误差项显著提升了噪声及异常值的干扰,且所设计的指数图正则化项能够刻画数据流形的局部几何结构与全局分布特征,还通过施加稀疏性约束增强了特征子空间的可解释性。在Semi-NMF下,算法进一步拓展了对混合符号型数据的处理能力,同时在小样本SSS场景下也具有更好的适应性。
YaoX H, GuoM. Deep embedding clustering algorithm based on convolutional residual autoencoder [J/OL].Journal of Chongqing Technology and Business University (Natural Science Edition),2025-10-23.
SabaT, RehmanA, MujahidM, et al.Smart defense based on explainable stacked machine learning architecture for securing internet of health things with K-means clustering [J].Sci Rep, 2025, 15(1): 37241.
[8]
YanS K, DingH L, QiuJ Y, et al.Low-voltage distribution network topology identification method based on manifold density peak clustering [J].Journal of Electric Power Science and Technology, 2025, 40(4): 61-71.
LuxburgU.A tutorial on spectral clustering [J].Stat Comput, 2007, 17(4): 395-416.
[11]
FuX, GuZ L, QiB J, et al. Structural modal parameter automatic identification and order determination method based on clustering and neural networks [J/OL].Journal of Vibration Engineering,2025-08-06.
LeeD D, SeungH S.Learning the parts of objects by non-negative matrix factorization [J].Nature, 1999, 401(6755): 788-791.
[14]
ZhouJ.Research of SWNMF with new iteration rules for facial feature extraction and recognition [J].Symmetry, 2019, 11(3): 354.
[15]
QianY, TanC, DingD, et al.Fast and secure distributed nonnegative matrix factorization [J].IEEE Trans Knowl Data Eng, 2020, 34(2): 653-666.
[16]
KongD, DingC, HuangH.Robust nonnegative matrix factorization using L21-norm [C]//Proceedings of the 20th ACM International Conference on Information and Knowledge Management.New York: ACM, 2011: 673-682.
[17]
LeeD D, SeungH S.Algorithms for non-negative matrix factorization [C]//Advances in Neural Information Processing Systems.Cambridge: MIT Press, 2001: 556-562.
KuoB C, ChangK Y.Feature extractions for small sample size classification problem [J].IEEE Trans Geosci Remote Sens, 2007, 45: 756-764.
[30]
WangS J, ChenH L, PengX J, et al.Exponential locality preserving projections for small sample size problem [J].Neurocomputing, 2011, 74(17): 3654-3662.
DornaikaF, BosaghzadehA.Exponential local discriminant embedding and its application to face recognition [J].IEEE Trans Cybern, 2013, 43(3): 921-934.
[33]
ZhangT, FangB, TangY Y, et al.Generalized discriminant analysis: A matrix exponential approach [J].IEEE Trans Systems Man Cybernetics Part B, 2010, 40(1): 186-197.
[34]
DingC H Q, LiT, et al.Convex and semi-nonnegative matrix factorizations [J].IEEE Trans Pattern Anal Mach Intell, 2010, 32(1): 45-55.
DingC H Q, LiT, PengW, et al.Orthogonal nonnegative matrix tri-factorizations for clustering [C]//Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.New York: ACM, 2006: 126-135.