Research map / Learning theory and algorithms / Matrix and tensor methods
Rate Optimal Denoising of Simultaneously Sparse and Low Rank Matrices
Dan Yang, Zongming Ma, Andreas Buja
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 |
|---|---|---|---|
| increase sparsity block size from k=l=50 to k=100, l=200 | Iterative thresholding denoiser · Synthetic denoising, m=2000 n=1000 r=10 | average L2 loss: 1133.03 -> 3598.69 | Hurt |
| increase sparsity block size from k=l=50 to k=100, l=200 | Iterative thresholding denoiser · Synthetic denoising, m=2000 n=1000 r=10 | average L1 loss: 19056.47 -> 65099.19 | Hurt |
| vary sparsity level (k,l) with loss rescaled by theoretical rate | Iterative thresholding denoiser · Synthetic denoising, m=2000 n=1000 r=10 | rescaled L2 loss: 0.64, 0.60, 0.68, 0.63 | No clear effect |
| vary sparsity level (k,l) with loss rescaled by theoretical rate | Iterative thresholding denoiser · Synthetic denoising, m=2000 n=1000 r=10 | rescaled L1 loss: 1.08, 0.98, 1.23, 1.07 | No clear effect |
| increase noise standard deviation sigma from 0.2 to 2 | Iterative thresholding denoiser · Synthetic denoising, r=10, k=l=50 | L1 and L2 loss | Hurt |
| increase rank r from 1 to 10 | Iterative thresholding denoiser · Synthetic denoising, sigma=1, k=l=50 | L1 loss | Hurt |
| increase rank r from 1 to 10 | Iterative thresholding denoiser · Synthetic denoising, sigma=1, k=l=50 | L2 loss | Hurt |
| vary loss parameter q between 1 and 2 | Iterative thresholding denoiser · Synthetic denoising, sigma=1 | log loss vs 1/q | No clear effect |
| use data-based rank selection instead of known rank | Iterative thresholding denoiser · Synthetic denoising, r=10 | selected rank: r = 10 | Improved |
| increase spikiness parameter s from 0.1 to 1.1 | Iterative thresholding denoiser · Synthetic denoising, r=1, k=l=50, sigma=1 | L1 and L2 loss | Hurt |
| increase spikiness parameter s beyond 1.1 to 1.6 | Iterative thresholding denoiser · Synthetic denoising, r=1, k=l=50, sigma=1 | L1 and L2 loss | Improved |
Closest papers by meaning
- Matrix Denoising with Doubly Heteroscedastic Noise: Fundamental Limits and Optimal Spectral Methods (NeurIPS 2024 · similarity 0.87)
- Generalized SURE for optimal shrinkage of singular values in low-rank matrix denoising (JMLR 2017 · similarity 0.87)
- Accelerating Ill-Conditioned Low-Rank Matrix Estimation via Scaled Gradient Descent (JMLR 2021 · similarity 0.87)
- A Unified Variance Reduction-Based Framework for Nonconvex Low-Rank Matrix Recovery (ICML 2017 · similarity 0.87)
- Minimizing the Maximal Rank (CVPR 2016 · similarity 0.87)
- Guarantees of a Preconditioned Subgradient Algorithm for Overparameterized Asymmetric Low-rank Matrix Recovery (ICML 2025 · similarity 0.85)
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.
