Quasi-Newton method for sparse matrix factorization with frobenius norm regularization

This paper considers quasi-Newton method for sparse matrix factorization (SMF) that incorporates Frobenius norm regularization to control overfitting and enhance generalization in data-driven applications. Sparse matrix factorization seeks to decompose a matrix into 2 matrices while promoting sparsity in one (or both) of the resulting factors, which is particularly useful in applications such as recommendation systems, signal processing, and dimensionality reduction. The Frobenius norm regularization is employed to penalize large parameter values, ensuring sparser factorization. The proposed quasi-Newton method, leveraging approximate second-order information, efficiently optimizes the objective function with significantly reduced computational overhead compared to full Newton methods. Experimental results on an example demonstrate the efficacy of the method in achieving high-quality sparse factorizations under different regularization parameter.

Text 120306.pdf - Published Version Download (470kB) Official URL or Download Paper: https://persama.org.my/images/Menemui_Matematik/20...

Actions (login required)

View Item