Convex nonnegative matrix factorization with manifold regularization.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Nonnegative Matrix Factorization (NMF) is widely used but requires nonnegative data.
• Standard NMF may not capture intrinsic geometric structures in data.
• Existing methods have limitations in handling mixed-sign data and preserving manifold structures.

Purpose of the Study:

• To propose a novel matrix factorization technique, Graph Regularized and Convex Nonnegative Matrix Factorization (GCNMF).
• To extend NMF capabilities to process mixed-sign data matrices.
• To incorporate manifold learning principles for improved data representation.

Main Methods:

• Introduced a graph regularized term into Convex Nonnegative Matrix Factorization (CNMF).
• Developed GCNMF to simultaneously learn basis and encoding vectors.
• Applied the method to clustering tasks on diverse datasets.

Main Results:

• GCNMF effectively processes mixed-sign data matrices, overcoming NMF limitations.
• The method preserves the intrinsic low-dimensional manifold structure of the data.
• Clustering experiments demonstrate the superior effectiveness of GCNMF on real-world datasets.

Conclusions:

• GCNMF offers a powerful extension to nonnegative matrix factorization.
• The technique enhances data representation by preserving geometric and manifold structures.
• GCNMF proves effective for clustering tasks involving both nonnegative and mixed-sign data.