Article
Engineering, Electrical & Electronic
Liang Ji, Peng Song, Wenjing Zhang
Summary: This paper proposes a novel approach for cross-database facial expression recognition, which uses similarity and dissimilarity information obtained from labels to guide matrix factorization and feature transfer learning, and employs graph regularization to mitigate distribution shift across databases. Extensive experiments on several benchmarks demonstrate the effectiveness of the proposed method compared to state-of-the-art transfer learning algorithms.
DIGITAL SIGNAL PROCESSING
(2022)
Article
Computer Science, Information Systems
Guifang Zhang, Jiaxin Chen
Summary: This study proposed a non-negative matrix factorization algorithm NMF_ASGR, which combines sparse representation and manifold learning to obtain an l(1) sparse robust graph. This sparse robust graph can adaptively discover the potential manifold structure of the data and has strong robustness to noise and outliers.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Xiangguang Dai, Keke Zhang, Juntang Li, Jiang Xiong, Nian Zhang, Huaqing Li
Summary: This paper proposes a robust semi-supervised non-negative matrix factorization method, RSNMF, for image clustering. By adding a weighted constraint on the noise matrix and incorporating manifold learning, better clustering performance on datasets contaminated by outliers and noise can be achieved. Additionally, utilizing discrete hashing learning method to constrain the learned subspace leads to a binary subspace from the original data.
COMPLEX & INTELLIGENT SYSTEMS
(2022)
Article
Physics, Multidisciplinary
Chunchun Chen, Wenjie Zhu, Bo Peng
Summary: This study proposes a novel approach, called Differentiated Graph regularized Non negative Matrix Factorization (DGNMF), for improving the accuracy of semi-supervised community detection. By leveraging the paired constraints between network nodes and constructing a differentiated graph, the proposed method shows significant improvements compared to existing NMF-based methods in community detection accuracy.
PHYSICA A-STATISTICAL MECHANICS AND ITS APPLICATIONS
(2022)
Article
Pharmacology & Pharmacy
Mei-Neng Wang, Yu Li, Li-Lan Lei, De-Wu Ding, Xue-Jun Xie
Summary: In this study, a method called GNMFDMA was proposed to predict potential associations between drugs and miRNAs. The method combined graph Laplacian regularization with non-negative matrix factorization. GNMFDMA outperformed existing methods, obtaining an AUC of 0.9193 in cross-validation. Case studies on three common drugs verified the reliability and efficiency of GNMFDMA.
FRONTIERS IN PHARMACOLOGY
(2023)
Article
Computer Science, Information Systems
Yugen Yi, Shumin Lai, Shicheng Li, Jiangyan Dai, Wenle Wang, Jianzhong Wang
Summary: This paper proposes a new unsupervised dimensionality reduction method called RRNMF-MAGL, which can extract robust low-dimensional features and adaptively learn a manifold structure to reflect the data distribution. It constructs a robust non-negative matrix factorization (RNMF) with L21-norm loss function to alleviate the influence of outliers and noises. A multi-constraint adaptive graph learning (MAGL) model is then designed based on the low-dimensional features to reduce computational complexity and avoid the adverse influence of redundant information. Multiple constraints (sparsity and locality) are incorporated into graph learning to enhance the discrimination ability of the graph structure. To preserve the local geometric structure information of the data, a graph Laplacian regularization (GLR) is employed. The proposed iterative update strategy optimizes RRNMF-MAGL and its convergence is verified through numerical experiments. Extensive experiments on eight publicly available datasets demonstrate the superior effectiveness and robustness of RRNMF-MAGL.
INFORMATION SCIENCES
(2023)
Article
Biochemistry & Molecular Biology
Junmin Zhao, Yuanyuan Ma, Lifang Liu
Summary: In this paper, a new algorithm LVSNMF is introduced which simultaneously utilizes Laplacian and Vicus matrices to capture both global and local structure patterns. Experimental results demonstrate its superior performance in biological data analysis.
FRONTIERS IN MOLECULAR BIOSCIENCES
(2021)
Article
Engineering, Electrical & Electronic
Xiao Zheng, Chujie Zhang, Cheng Wan
Summary: MicroRNAs (miRNAs) play a crucial role in the occurrence and development of various human diseases, and predicting their associations accurately remains a challenging scientific topic. A new matrix completion algorithm based on non-negative matrix factorization (NMFMC) is proposed in this translation, which effectively improves the prediction accuracy of miRNA-disease associations.
Article
Medicine, Research & Experimental
Mei-Neng Wang, Xue-Jun Xie, Zhu-Hong You, De-Wu Ding, Leon Wong
Summary: In this study, a method called weighted graph regularized collaborative non-negative matrix factorization (WNMFDDA) was proposed to predict drug-disease associations. The method achieved high AUC values in cross-validation and outperformed other prediction methods. Case studies further confirmed its predictive performance. The experimental results demonstrated that WNMFDDA is an effective method for drug-disease association prediction.
JOURNAL OF TRANSLATIONAL MEDICINE
(2022)
Article
Computer Science, Artificial Intelligence
Mei-Neng Wang, Zhu-Hong You, Lei Wang, Li-Ping Li, Kai Zheng
Summary: In this study, a new computational approach LDGRNMF was developed to predict lncRNA-disease associations, considering similarity calculation based on Gaussian interaction profile kernel and disease semantic information, as well as nearest known neighbor interaction profiles weighting to reconstruct association matrix. LDGRNMF outperformed other methods with an AUC of 0.8985 in cross-validation experiments, and showed high accuracy in predicting potential associations in case studies for stomach cancer, breast cancer, and lung cancer.
Article
Mathematics
Minghua Wan, Mingxiu Cai, Guowei Yang
Summary: In this paper, a new algorithm named robust exponential graph regularized non-negative matrix factorization (REGNMF) is proposed to address the possible singular matrix problem in GNMF. By adding a matrix exponent to the regularizer, the singular matrix is transformed into a non-singular matrix. The REGNMF method is iteratively solved using a multiplicative non-negative updating rule. Experimental results on various datasets show that the proposed method is more significant.
Article
Computer Science, Artificial Intelligence
Ping Deng, Fan Zhang, Tianrui Li, Hongjun Wang, Shi-Jinn Horng
Summary: Clustering remains a challenging research hotspot in data mining. This paper proposes a biased unconstrained non-negative matrix factorization (BUNMF) model to improve the clustering performance. The model modifies the update rules and adds bias, and introduces three different activation functions for iteration updates.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Automation & Control Systems
Yuheng Jia, Hui Liu, Junhui Hou, Sam Kwong
Summary: A new semisupervised model is proposed in this paper, which can simultaneously learn the similarity matrix and supervisory information to generate better clustering performance. By fully utilizing supervisory information and propagating through pairwise constraints, an informative similarity matrix is obtained.
IEEE TRANSACTIONS ON CYBERNETICS
(2021)
Article
Computer Science, Artificial Intelligence
Qingming Kong, Jianyong Sun, Zongben Xu
Summary: This paper proposes a novel community detection method, called joint symmetric non-negative matrix factorization model. The method characterizes the attribute homogeneity and topology similarities of an attribute network in a unified framework and imposes an orthogonal constraint on the factor matrix to improve the accuracy of nodes' affiliations to communities. Furthermore, it utilizes a novel multi-order graph regularization and an eigenvector centrality-based strategy to enhance adjacency information. Experimental results demonstrate that the proposed method outperforms existing state-of-the-art methods in community detection tasks in most benchmark complex networks.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Mathematical & Computational Biology
Jiyang Yu, Baicheng Pan, Shanshan Yu, Man-Fai Leung
Summary: Non-negative matrix tri-factorization (NMTF), as an extension of NMF, provides more degrees of freedom, but is affected by noise and lacks the learning of geometric information. Therefore, a novel robust capped norm dual hyper-graph regularized NMTF method (RCHNMTF) is proposed. RCHNMTF handles extreme outliers, exploits intrinsic geometric information, and improves clustering performance by adding orthogonality constraints. Experiments on seven datasets demonstrate the robustness and superiority of RCHNMTF.
MATHEMATICAL BIOSCIENCES AND ENGINEERING
(2023)