Accelerated Alternating Projections for Robust Principal Component Analysis
About
We study robust PCA for the fully observed setting, which is about separating a low rank matrix $\boldsymbol{L}$ and a sparse matrix $\boldsymbol{S}$ from their sum $\boldsymbol{D}=\boldsymbol{L}+\boldsymbol{S}$. In this paper, a new algorithm, dubbed accelerated alternating projections, is introduced for robust PCA which significantly improves the computational efficiency of the existing alternating projections proposed in [Netrapalli, Praneeth, et al., 2014] when updating the low rank factor. The acceleration is achieved by first projecting a matrix onto some low dimensional subspace before obtaining a new estimate of the low rank matrix via truncated SVD. Exact recovery guarantee has been established which shows linear convergence of the proposed algorithm. Empirical performance evaluations establish the advantage of our algorithm over other state-of-the-art algorithms for robust PCA.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Principal Component Estimation | Gaussian setting with mean-shift and covariance-shift contamination synthetic (test) | Largest PC Alignment (%)9.74 | 112 | |
| Principal Component Analysis | Synthetic Gaussian Data | Runtime (ms)86 | 12 | |
| Principal Component Analysis | Synthetic Gaussian data d=900, n=1000 (test) | Runtime (ms)79 | 9 | |
| Principal Component Alignment | Gaussian Mixture ($d=900, n=10^3, \pi_1=5\%$) | Alignment (%)8.4 | 7 | |
| Principal Component Alignment | Gaussian Mixture ($d=900, n=10^3, \pi_1=10\%$) | Alignment8.75 | 7 | |
| Principal Component Alignment | Gaussian Mixture ($d=900, n=10^3, \pi_1=15\%$) | PCA Alignment7.47 | 7 | |
| Principal Component Alignment | Gaussian Mixture ($d=900, n=10^3, \pi_1=20\%) | Alignment (%)7.74 | 7 |