Research map / Learning theory and algorithms / Matrix and tensor methods
Robust High-Dimensional Low-Rank Matrix Estimation: Optimal Rate and Data-Adaptive Tuning
Xiaolong Cui, Lei Shi, Wei Zhong, Changliang Zou
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 rank-based Wilcoxon loss instead of least-squares loss in nuclear-norm estimation | matrix lasso · matrix regression, Cauchy/log-normal errors | log Frobenius error | Improved |
| use rank-based loss versus l1 (LAD) loss under Gaussian errors | RML · matrix regression, Gaussian errors | log Frobenius error | Improved |
| use RML instead of robustified matrix lasso with robust CV tuning, Cauchy errors | matrix regression · Table 1 (I) Cauchy | squared Frobenius error: 0.105 -> <0.001 | Improved |
| use RML instead of robustified matrix lasso, Normal errors | matrix regression · Table 1 (I) Normal | squared Frobenius error: 0.114 -> 0.130 | Hurt |
| use RML instead of robustified matrix lasso, multivariate regression Cauchy | multivariate regression · Table 1 (II) Cauchy | squared Frobenius error: 0.076 -> 0.003 | Improved |
| use RML instead of robustified matrix lasso, matrix completion log-normal | matrix completion · Table 1 (III) Log-normal | squared Frobenius error: 0.121 -> 0.048 | Improved |
| pivotal tuning of RML instead of robust cross-validation, tuning time | matrix regression · Table 1 (I) Normal | total time (s): 344 -> 1.97 | Improved |
| pivotal tuning of RML instead of robust CV, matrix completion total time | matrix completion · Table 1 (III) Normal | total time (s): 370 -> 3.66 | Improved |
| use RML instead of robustified matrix lasso for rank recovery under heavy tails | multivariate regression · Table 1 (II) Log-normal | estimated rank: 21.83 -> 5 | Improved |
| use RML instead of fixed-rank factorization methods (SGD, SSM) with misspecified rank R=3,10 | matrix regression · Figure 2 | log Frobenius error | Improved |
| increase t-distribution degrees of freedom d with l1 loss versus RML (ablation) | multivariate regression · Figure 4 | Frobenius error | Mixed |
| use RML instead of l2 matrix lasso on gene data | Arabidopsis thaliana multivariate regression · Arabidopsis thaliana gene data | MSE: 0.609 -> 0.567 | Improved |
Closest papers by meaning
- On the Power of Truncated SVD for General High-rank Matrix Estimation Problems (NeurIPS 2017 · similarity 0.89)
- Low-Rank-Sparse Subspace Representation for Robust Regression (CVPR 2017 · similarity 0.88)
- Accelerating Ill-Conditioned Low-Rank Matrix Estimation via Scaled Gradient Descent (JMLR 2021 · similarity 0.88)
- Towards Faster Rates and Oracle Property for Low-Rank Matrix Estimation (ICML 2016 · similarity 0.88)
- Rank Overspecified Robust Matrix Recovery: Subgradient Method and Exact Recovery (NeurIPS 2021 · similarity 0.88)
- Guarantees of a Preconditioned Subgradient Algorithm for Overparameterized Asymmetric Low-rank Matrix Recovery (ICML 2025 · similarity 0.87)
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.
