Research map / Learning theory and algorithms / Matrix and tensor methods
Rate-Optimal Subspace Estimation on Random Graphs
Zhixin Zhou, Fan Zhou, Ping Li, Cun-Hui Zhang
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 soft instead of hard singular value thresholding when the regularization constant cannot be tuned well | Singular value thresholding estimator of M · Simulation, n1=n2=1000, r=3 | estimation loss | Improved |
| Use hard instead of soft thresholding with a well-chosen regularization constant, measured in operator norm | Singular value thresholding estimator of M · Simulation, n1=n2=1000, r=3 | operator norm loss | Improved |
| Use soft instead of hard thresholding, measured in Frobenius norm | Singular value thresholding estimator of M · Simulation, n1=n2=1000, r=3 | Frobenius norm loss | Improved |
| Add degree regularization (varying constant c from 0.2 to 1) to hard thresholding (ablation) | Algorithm 1 (hard SVT) · Simulation, p=0.01 | operator and Frobenius loss | Improved |
| Increase graph density p from 0.01 to 0.03 and 0.05 | Degree regularization effect · Simulation, n1=n2=1000, r=3 | loss: 0.01, 0.03, 0.05 | Hurt |
| Add regularization to soft thresholding (ablation) | Algorithm 2 (soft SVT) · Simulation, n1=n2=1000, r=3 | operator and Frobenius loss | No clear effect |
| Set regularization constant to 0.8 rather than 0.7 or 0.9 | Algorithm 2 (soft SVT) · Simulation, p=0.01 | loss: 0.8 vs 0.7 and 0.9 | Improved |
| Use regularization constant above 0.6 for soft thresholding | Algorithm 2 (soft SVT) · Simulation, p=0.03 and 0.05 | loss: greater than 0.6 | No clear effect |
| Vary regularization constant for subspace estimation | Algorithm 3 (singular space estimation) · Simulation, n1=n2=2000, r=3 | operator norm loss | No clear effect |
| Set regularization constant around 0.5 | Algorithm 3 (singular space estimation) · Simulation, n1=n2=2000, r=3 | Frobenius norm loss: around 0.5 | Improved |
Closest papers by meaning
- Optimal Estimation and Completion of Matrices with Biclustering Structures (JMLR 2016 · similarity 0.89)
- Schatten Norms in Matrix Streams: Hello Sparsity, Goodbye Dimension (ICML 2020 · similarity 0.86)
- Thy Friend is My Friend: Iterative Collaborative Filtering for Sparse Matrix Estimation (NeurIPS 2017 · similarity 0.86)
- Optimal community detection in dense bipartite graphs (NeurIPS 2025 · similarity 0.85)
- Distributed Least Squares in Small Space via Sketching and Bias Reduction (NeurIPS 2024 · similarity 0.85)
- Optimal Sketching for Residual Error Estimation for Matrix and Vector Norms (ICLR 2024 · similarity 0.84)
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.
