Abstract:Non-negative matrix factorization (NMF) is becoming increasingly popular in many research fields due to its particular properties of semantic interpretability and part-based representation. Sparseness constraints are usually imposed on the NMF problems in order to achieve potential features and sparse representation. These constrained NMF problems are usually reformulated as regularization models to solve conveniently. However, the regularization parameters in the regularization model are difficult to tune and the frequently used sparse-inducing terms in the regularization model generally have bias effects on the induced matrix and need an extra restricted isometry property (RIP). This paper proposes a multiobjective sparse NMF paradigm which refrains from the regularization parameter issues, bias effects, and the RIP condition. A novel multiobjective memetic algorithm is also proposed to generate a set of solutions with diverse sparsity and high factorization accuracy. A masked projected gradient local search scheme is specially designed to accelerate the convergence rate. In addition, a priori knowledge is also integrated in the algorithm to reduce the computational time in discovering our interested region in the objective space. The experimental results show that the proposed paradigm has better performance than some regularization algorithms in producing solutions with different degrees of sparsity as well as high factorization accuracy, which are favorable for making the final decisions.

First results on uniqueness of sparse non-negative matrix factorization

Extended sparse nonnegative matrix factorization

Non-Negative Matrix Factorization Revisited: Uniqueness and Algorithm for Symmetric Decomposition

Non-negative Matrix-Set Factorization

A Sparse Algorithm for Non-Negative Matrix Factorization Based on Projection Gradient Method

Sparse nonnegative matrix factorization applied to microarray data sets

Multiobjective Sparse Non-Negative Matrix Factorization

An Inherently Nonnegative Latent Factor Model for High-Dimensional and Sparse Matrices from Industrial Applications.

Nonlocal Low-Rank Nonnegative Tensor Factorization for Hyperspectral Unmixing

Incremental Nonnegative Matrix Factorization with Sparseness Constraint for Image Representation.

A Discrete-Time Neurodynamic Approach to Sparsity-Constrained Nonnegative Matrix Factorization

Nonnegative matrix factorization with piecewise smoothness constraint for hyperspectral unmixing

Region-Based Structure Preserving Nonnegative Matrix Factorization for Hyperspectral Unmixing

Log-based Sparse Nonnegative Matrix Factorization for Data Representation

Nonconvex Nonseparable Sparse Nonnegative Matrix Factorization for Hyperspectral Unmixing

Non-negative matrix factorization and deconvolution as dual simplex problem

A unified statistical approach to non-negative matrix factorization and probabilistic latent semantic indexing

Kernel Nonnegative Matrix Factorization Without the Curse of the Pre-image - Application to Unmixing Hyperspectral Images

Constraint-Relaxation Approach for Nonnegative Matrix Factorization: A Case Study

Linear Projection-based Non-negative Matrix Factorization