Robust Principal Component Analysis with Non-Sparse Errors
arXiv:1902.08735
Abstract
We show that when a high-dimensional data matrix is the sum of a low-rank matrix and a random error matrix with independent entries, the low-rank component can be consistently estimated by solving a convex minimization problem. We develop a new theoretical argument to establish consistency without assuming sparsity or the existence of any moments of the error matrix, so that fat-tailed continuous random errors such as Cauchy are allowed. The results are illustrated by simulations.
References in corpus (8)
- The Augmented Lagrange Multiplier Method for Exact Recovery of Corrupted Low-Rank Matrices
- Quantum state tomography via compressed sensing
- Statistical analysis of factor models of high dimension
- Consistency of trace norm minimization
- Matrix Completion with Noisy Entries and Outliers
- Statistically Optimal and Computationally Efficient Low Rank Tensor Completion from Noisy Entries
- Cauchy Principal Component Analysis
- On Approximation Guarantees for Greedy Low Rank Optimization