Rapid Robust Principal Component Analysis: CUR Accelerated Inexact Low Rank Estimation Article Swipe

View

Related Concepts

Robust principal component analysis Principal component analysis Computer science Block matrix Rank (graph theory) Algorithm Iterated function Dimension (graph theory) Acceleration Matrix decomposition Low-rank approximation Dimensionality reduction Mathematical optimization Robustness (evolution) Reduction (mathematics) Block (permutation group theory) Sparse matrix Mathematics Artificial intelligence Gene Geometry Biochemistry Combinatorics Classical mechanics Mathematical analysis Pure mathematics Gaussian Chemistry Eigenvalues and eigenvectors Quantum mechanics Physics Hankel matrix

HanQin Cai , Keaton Hamm , Longxiu Huang , Jiaqi Li , Tao Wang ·

YOU? · · 2020 · Open Access · · DOI: https://doi.org/10.1109/lsp.2020.3044130 · OA: W3093365368

Robust principal component analysis (RPCA) is a widely used tool for\ndimension reduction. In this work, we propose a novel non-convex algorithm,\ncoined Iterated Robust CUR (IRCUR), for solving RPCA problems, which\ndramatically improves the computational efficiency in comparison with the\nexisting algorithms. IRCUR achieves this acceleration by employing CUR\ndecomposition when updating the low rank component, which allows us to obtain\nan accurate low rank approximation via only three small submatrices.\nConsequently, IRCUR is able to process only the small submatrices and avoid\nexpensive computing on the full matrix through the entire algorithm. Numerical\nexperiments establish the computational advantage of IRCUR over the\nstate-of-art algorithms on both synthetic and real-world datasets.\n