Abstract:As a classical data representation method, nonnegative matrix factorization (NMF) can well capture the global structure information of the observed data, and it has been successfully applied in many fields. It is generally known that the local manifold structures will have a better effect than the global structures in image recognition and clustering. The local structure information can well be preserved by the neighbor graph in the manifold learning methods. The traditional neighbor graph constructed relies heavily on the original observed data. However, the original data usually contain a lot of noise and outliers in practical application, which results in obtaining inaccurate neighbor graph, and ultimately leads to performance degradation. How to get the ideal local structure information becomes more and more important. By combing the manifold learning into NMF, we propose an adaptive graph regularized nonnegative matrix factorization (AGNMF). In AGNMF, the neighbor graph is obtained by adaptive iteration. Both the global information and the local manifold can be well captured in AGNMF, and the better data representation can be obtained. A large number of experiments on different data sets show that our AGNMF has good clustering ability.

What problem does this paper attempt to address?

This paper aims to solve the problem that Nonnegative Matrix Factorization (NMF) fails to fully preserve the local structure information of data in data representation. Specifically, although the traditional NMF method can well capture the global structure information of data, in image recognition and clustering tasks, local manifold structure information is often more effective than global structure information. However, constructing local structure information depends on the neighborhood graph, and the traditional neighborhood graph construction method is very sensitive to the original observed data, which usually contains a large amount of noise and outliers, resulting in an inaccurate neighborhood graph and ultimately affecting performance. To overcome these problems, the author proposes Adaptive Graph Regularized Nonnegative Matrix Factorization (AGNMF). In AGNMF, the neighborhood graph is obtained through an adaptive iterative method, and at the same time, it combines the advantages of NMF and graph regularization. It can not only capture the global information of data but also well preserve the local manifold structure information, thereby achieving better data representation effects. The main contributions and innovation points of the paper include: 1. Obtaining both Nonnegative Matrix Factorization (NMF) and the neighborhood graph at the same time. Compared with the classical NMF and GNMF, AGNMF has better representation ability for the original data and can achieve better clustering performance. 2. Proposing an iterative scheme for AGNMF and proving through experimental results that the algorithm has better clustering effects. Through a large number of experimental verifications, the clustering performance of AGNMF on multiple datasets is better than other methods. Especially on image datasets, AGNMF can better preserve the local structure information of data, thereby obtaining better clustering results.

Adaptive graph regularized nonnegative matrix factorization for data representation

Data representation using robust nonnegative matrix factorization for edge computing

Feature Extraction Via Multi-View Non-Negative Matrix Factorization with Local Graph Regularization.

Adversarial Graph Regularized Deep Nonnegative Matrix Factorization for Data Representation

Graph Regularized Nonnegative Matrix Factorization with Sparse Coding

Graph regularized non-negative matrix factorization with sparse coding

Graph Regularized Sparse Non-Negative Matrix Factorization for Clustering

Multiple Graph Adaptive Regularized Semi-Supervised Nonnegative Matrix Factorization with Sparse Constraint for Data Representation

Non-negative Matrix Factorization on Manifold

Semi-supervised non-negative matrix tri-factorization with adaptive neighbors and block-diagonal learning

Convergence Analysis of Graph Regularized Non-Negative Matrix Factorization

Discriminative Orthogonal Nonnegative Matrix Factorization with Flexibility for Data Representation

Neighborhood Preserving Non-Negative Tensor Factorization For Image Representation

Robust non-negative matrix factorization with multiple correntropy-induced hypergraph regularizer

Parameter-less Auto-weighted Multiple Graph Regularized Nonnegative Matrix Factorization for Data Representation.

Local Learning Regularized Nonnegative Matrix Factorization.

Robust Structured Nonnegative Matrix Factorization for Image Representation.

Rank-constrained nonnegative matrix factorization for data representation

Feature Weighted Non-negative Matrix Factorization

Semi-supervised Convex Nonnegative Matrix Factorizations with Graph Regularized for Image Representation

Multiple Graph Regularized NMF for Hyperspectral Unmixing