Research map / Learning theory and algorithms / Matrix and tensor methods
Recovering PCA and Sparse PCA via Hybrid-(l1,l2) Sparse Sampling of Data Elements
Abhisek Kundu, Petros Drineas, Malik Magdon-Ismail
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 hybrid-(l1,l2) sampling instead of l1 sampling | sparse sketch for PCA · Stock PCA (2% non-zeros) | variance preserved: 85.09 % -> 97.07% | Improved |
| use hybrid-(l1,l2) sampling instead of Bernstein sampling | sparse sketch for PCA · Stock PCA (2% non-zeros) | variance preserved: 95.61 % -> 97.07% | Improved |
| use hybrid-(l1,l2) sampling instead of l1 or Bernstein sampling | sparse sketch for PCA · Digit PCA (9% non-zeros) | variance preserved: 98.47 % and 98.46 % vs 98.51% | No clear effect |
| use hybrid sampling PCA instead of Gaussian random projection PCA | Digit PCA · Digit (7% non-zeros) | variance preserved: 96.89% -> 97.94% | Improved |
| use hybrid-(l1,l2) sketch instead of uniform sketch for sparse PCA | Spasm sparse PCA r=40 · Gene | f(sp)/f(G sp): 0.15 -> 0.88 | Improved |
| use hybrid-(l1,l2) sketch instead of uniform sketch for sparse PCA | Spasm sparse PCA r=40 · TechTC TTC1 | f(sp)/f(G sp): 0.38 -> 0.99 | Improved |
| use hybrid-(l1,l2) sketch instead of uniform sketch for sparse PCA | Spasm sparse PCA r=40 · Digit | f(max)/f(G max): 1.01 vs 0.99 | No clear effect |
| use hybrid-(l1,l2) sketch instead of uniform sketch for sparse PCA | Spasm sparse PCA r=40 · Stock | f(max)/f(G max): 0.66 -> 1.00 | Improved |
| use optimal mixing alpha* instead of alpha = 1 (pure l1) (ablation) | hybrid-(l1,l2) sketch · Stock sparse PCA | f(Hsp,r) | Improved |
| use hybrid-(l1,l2) sampling instead of leverage score sampling | low-rank power-law matrix sparsification · Power-law matrix gamma=1.0, s/(k(m+n))=3 | spectral-norm error: 42% -> 8% | Improved |
| use hybrid-(l1,l2) sampling instead of leverage score sampling | rank-truncated A0.1 sparsification · A0.1, k=5, s/(k(m+n))=3 | spectral-norm error: 80% -> 25% | Improved |
| use hybrid-(l1,l2) sampling instead of l2 sampling with truncation epsilon=0.1 | sparse sketch · A0.1 synthetic | spectral-norm error | Hurt |
Closest papers by meaning
- Support Recovery in Sparse PCA with Incomplete Data (NeurIPS 2022 · similarity 0.91)
- A Kernel Random Matrix-Based Approach for Sparse PCA (ICLR 2019 · similarity 0.91)
- Efficient Sparse PCA via Block-Diagonalization (ICLR 2025 · similarity 0.90)
- Fast and Provable Algorithms for Sparse PCA with Improved Sample Complexity (ICML 2025 · similarity 0.90)
- Combinatorial Sparse PCA Beyond the Spiked Identity Model (ICML 2026 · similarity 0.88)
- Oja's Algorithm for Streaming Sparse PCA (NeurIPS 2024 · similarity 0.86)
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.
