Research map / Learning theory and algorithms / Matrix and tensor methods
Bootstrap-Based Regularization for Low-Rank Matrix Estimation
Julie Josse, Stefan Wager
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 |
|---|---|---|---|
| add iterated stable autoencoding (ISA) over single-pass SA (ablation) | stable autoencoder · Poisson noise (N=200) | MSE: 1.83 -> 1.13 | Improved |
| use ISA instead of rank-3 TSVD-k | truncated SVD · Poisson noise (N=200) | MSE: 2.62 -> 1.13 | Improved |
| use ISA instead of ASYMP under Poisson noise | ASYMP · Poisson noise (N=200) | MSE: 1.71 -> 1.13 | Improved |
| increase counts N from 200 to 2000 | ISA · Poisson noise | MSE: 1.13 -> 0.10 | Improved |
| gap between SA and ISA vanishes at high N (ablation) | stable autoencoder · Poisson noise (N=1000) | MSE: 0.25 vs 0.24 | No clear effect |
| learn new singular vectors with SA over plain SVD | SVD · Poisson noise (N=200) | RV coefficient for U: 0.29 -> 0.34 | Improved |
| add ISA on top of SA for U alignment (ablation) | SA · Poisson noise (N=800) | RV coefficient for U: 0.71 -> 0.79 | Improved |
| use ISA instead of ASYMP for rank estimation | ASYMP · Poisson noise (N=2000) | mean rank estimate (true 3): 4.07 -> 3.00 | Improved |
| use ISA instead of SVST at low SNR with high rank | SA/ISA · Gaussian noise (rank 100, SNR 0.5) | MSE: 1.000 -> 0.852 for SVST | Hurt |
| use SA and ISA in high-SNR Gaussian setting | TSVD-k · Gaussian noise (rank 100, SNR 2) | MSE: 0.152 -> 0.142 | Mixed |
| add ISA regularization to correspondence analysis (ablation) | correspondence analysis · Rotten Tomatoes sentiment | accuracy (%): 61.8 -> 67.0 | Improved |
| use correspondence analysis instead of document averaging | document averaging · Rotten Tomatoes sentiment | accuracy (%): 62.1 -> 61.8 | No clear effect |
Closest papers by meaning
- Accelerating Ill-Conditioned Low-Rank Matrix Estimation via Scaled Gradient Descent (JMLR 2021 · similarity 0.88)
- Robust High-Dimensional Low-Rank Matrix Estimation: Optimal Rate and Data-Adaptive Tuning (JMLR 2023 · similarity 0.88)
- Towards Faster Rates and Oracle Property for Low-Rank Matrix Estimation (ICML 2016 · similarity 0.87)
- Matrix Denoising with Doubly Heteroscedastic Noise: Fundamental Limits and Optimal Spectral Methods (NeurIPS 2024 · similarity 0.86)
- From Low Rank Gradient Subspace Stabilization to Low-Rank Weights: Observations, Theories, and Applications (ICML 2025 · similarity 0.86)
- Coherence-free Entrywise Estimation of Eigenvectors in Low-rank Signal-plus-noise Matrix Models (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.
