Abstract:Robust Principal Component Analysis (RPCA) and its associated non-convex relaxation methods constitute a significant component of matrix completion problems, wherein matrix factorization strategies effectively reduce dimensionality and enhance computational speed. However, some non-convex factorization forms lack theoretical guarantees. This paper proposes a novel strategy in non-convex quasi-norm representation, introducing a method to obtain weighted matrix quasi-norm factorization forms. Especially, explicit bilinear factor matrix factorization formulations for the weighted logarithmic norm and weighted Schatten-$q$ quasi norms with $q=1, 1/2, 2/3$ are provided, along with the establishment of corresponding matrix completion models. An Alternating Direction Method of Multipliers (ADMM) framework algorithm is employed for solving, and convergence results of the algorithm are presented.

What problem does this paper attempt to address?

The paper is primarily dedicated to addressing the problem of low-rank matrix decomposition in Matrix Recovery and proposes a new non-convex relaxation method to improve existing techniques. Specifically, the paper focuses on the problem of Robust Principal Component Analysis (RPCA) and its related non-convex relaxation methods. When dealing with data containing outliers, RPCA can decompose a given matrix into a low-rank part and a sparse part. The main contributions of the paper include: 1. **Theoretical Guarantees and Proposal of New Methods**: Existing non-convex factorization forms lack theoretical guarantees. This paper proposes a new strategy by introducing a weighted matrix norm factorization form in the non-convex quasi-norm representation. Specifically, it provides explicit bilinear factor matrix factorization formulas for the weighted log-norm and weighted Schatten-q norm (q=1, 1/2, 2/3). 2. **Algorithm Design and Convergence Analysis**: The paper designs an algorithm framework based on the Alternating Direction Method of Multipliers (ADMM) to solve the proposed model and provides convergence results for the algorithm. 3. **Exploration of Weighted Strategies**: The paper explores the theoretical models and effective algorithms when combining weighted strategies with non-convex representations (such as the Schatten-q norm or log-norm). This not only extends the theoretical foundation of the matrix recovery problem but also has broad application prospects. Through these contributions, the paper aims to improve the performance of existing RPCA methods, particularly in terms of efficiency and accuracy when handling large-scale datasets and complex matrix structures. Additionally, by introducing weighted factorization forms, it can effectively reduce computational costs and enhance the applicability and flexibility of the model.

Reweighted Quasi Norm Regularized Low-Rank Factorization for Matrix Robust PCA

Robust PCA Based on Adaptive Weighted Least Squares and Low-Rank Matrix Factorization

A NEW MODEL FOR SPARSE AND LOW-RANK MATRIX DECOMPOSITION

Randomized Truncated Pivoted QLP Factorization for Low-Rank Matrix Recovery

Robust PCA Using Generalized Nonconvex Regularization.

Robust Principal Component Analysis with Matrix Factorization

Schatten Capped P Regularization for Robust Principle Component Analysis.

Robust principal component analysis: A factorization-based approach with linear complexity

Robust Principal Component Analysis with Missing Data.

Strongly Convex Programming for Exact Matrix Completion and Robust Principal Component Analysis

Robust PCA via Nonconvex Rank Approximation

Bilinear Factor Matrix Norm Minimization for Robust PCA: Algorithms and Applications

Quadratic Regularization Projected Barzilai–Borwein Method for Nonnegative Matrix Factorization

Efficient Robust Matrix Factorization with Nonconvex Penalties.

Robust Principal Component Analysis via Truncated $L_{1-2}$ Minimization.

Double-Weighted Low-Rank Matrix Recovery Based on Rank Estimation.

Robust Matrix Completion Based on Factorization and Truncated-Quadratic Loss Function

A New Nonconvex Rank Approximation of RPCA

RES-PCA: A Scalable Approach to Recovering Low-Rank Matrices

An Alternating Minimization Method for Robust Principal Component Analysis

A Fast Factorization-based Approach to Robust PCA