Research map / Learning theory and algorithms / Matrix and tensor methods
Bilinear Factor Matrix Norm Minimization for Robust PCA: Algorithms and Applications
Fanhua Shang, James Sheung-Chak Cheng, Yuanyuan Liu, Zhi-Quan Tom Luo, Zhouchen Lin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use bilinear factor norm penalties (S+L)_2/3 instead of convex nuclear norm RPCA | RPCA · Corrupted synthetic matrices (noise 0.5, 20% outliers) 500x500 | RSE: 0.1265 -> 0.0453 | Improved |
| use (S+L)_2/3 instead of Unifying (bilinear spectral penalty) | Unifying · Corrupted synthetic matrices (noise 0.5, 20% outliers) 1000x1000 | RSE: 0.0443 -> 0.0318 | Improved |
| use (S+L)_2/3 instead of WNNM for speed | WNNM · Corrupted synthetic matrices (noise 0.5, 20% outliers) 5000x5000 | time (s): 24,370.85 -> 128.11 | Improved |
| use (S+L)_2/3 instead of RPCA on 10,000x10,000 matrices | RPCA · Corrupted synthetic matrices (noise 0.5, 20% outliers) 10000x10000 | time (s): 20,321.36 -> 487.46 | Improved |
| use (S+L)_1/2 vs (S+L)_2/3 on corrupted matrices | (S+L)_1/2 · Corrupted synthetic matrices (noise 0.5, 20% outliers) 500x500 | RSE: 0.0469 -> 0.0453 | Hurt |
| add low-rank regularizer (bilinear norm penalties) to plain matrix factorization LMaFit (ablation) | LMaFit · Corrupted synthetic matrices (noise 0.5, 20% outliers) 1000x1000, noise 0.2 | RSE and F-measure | Improved |
| use bilinear norm penalties instead of Schatten quasi-norm LpSq at high missing ratio | LpSq · Corrupted matrices, 80% missing, 5% outliers | RSE | Improved |
| use bilinear norm penalties instead of LpSq with sufficient observations | LpSq · Corrupted matrices, low missing ratio | RSE | No clear effect |
| use (S+L)_1/2 instead of LpSq for text removal | LpSq · Text removal (256x256) | RSE: 0.1097 -> 0.0396 | Improved |
| use (S+L) methods instead of LpSq for text removal speed | LpSq · Text removal (256x256) | time (s): 77.65 -> 1.21 | Improved |
| use D-N/F-N penalties instead of nuclear norm solvers for inpainting | nuclear norm solvers · Image inpainting (85% missing) | PSNR | Improved |
| use our methods instead of RegL1 for moving object detection | RegL1 · Surveillance videos (5 sequences) | F-measure | Improved |
Closest papers by meaning
- Partial Sum Minimization of Singular Values in Robust PCA: Algorithm and Applications (TPAMI 2016 · similarity 0.90)
- Robust Matrix Factorization by Majorization Minimization (TPAMI 2017 · similarity 0.90)
- Projected Robust PCA with Application to Smooth Image Recovery (JMLR 2022 · similarity 0.89)
- Structured Low-Rank Matrix Factorization: Global Optimality, Algorithms, and Applications (TPAMI 2019 · similarity 0.89)
- Informed Non-Convex Robust Principal Component Analysis With Features (AAAI 2018 · similarity 0.88)
- Robust Formulation for PCA: Avoiding Mean Calculation With L2,p-norm Maximization (AAAI 2018 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
