Research map / Learning theory and algorithms / Matrix and tensor methods
Global and Quadratic Convergence of Newton Hard-Thresholding Pursuit
Shenglong Zhou, Naihua Xiu, Hou-Duo Qi
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 NHTP instead of HTP (restricted Newton step) on very sparse compressive sensing | HTP · Compressive sensing Ex.2, n=25000, s=ceil(0.05n) | CPU time (s): 36.93 -> 6.58 | Improved |
| use NHTP instead of HTP for recovery accuracy | HTP · Compressive sensing Ex.2, n=25000, s=ceil(0.05n) | average absolute error: 5.94e-14 -> 2.86e-14 | Improved |
| use NHTP instead of NIHT for recovery accuracy | NIHT · Compressive sensing Ex.2, n=25000, s=ceil(0.05n) | average absolute error: 1.01e-6 -> 2.86e-14 | Improved |
| use NHTP instead of greedy GP/OMP when solution is less sparse | GP · Compressive sensing Ex.2, n=25000, s=ceil(0.05n) | average absolute error: 1.24e-03 -> 2.86e-14 | Improved |
| use NHTP instead of GP/OMP when solution is very sparse | OMP · Compressive sensing Ex.2, n=25000, s=ceil(0.01n) | average absolute error: 1.12e-14 -> 2.47e-15 | No clear effect |
| use NHTP instead of other solvers on success rate, Gaussian matrix, s=22 | other CS solvers · Compressive sensing Ex.1, n=256, m=ceil(n/4) | success rate: 90% vs less than 40% | Improved |
| use NHTP instead of GraSP for time at s=ceil(0.05n) | GraSP · Sparse logistic regression Ex.3, n=40000 | CPU time (s): 619.5 -> 25.41 | Improved |
| use NHTP instead of IIHT at s=ceil(0.05n) | IIHT · Sparse logistic regression Ex.3, n=40000, s=ceil(0.05n) | logistic loss: 4.92e-4 -> 5.81e-7 | Improved |
| use NHTP instead of IIHT at s=ceil(0.01n) | IIHT · Sparse logistic regression Ex.3, n=40000, s=ceil(0.01n) | logistic loss: 1.21e-1 -> 2.05e-1 | Hurt |
| use NHTP instead of GraSP on correlated data time | GraSP · Sparse logistic regression Ex.4, n=40000, s=ceil(0.05n) | CPU time (s): 721.0 -> 22.46 | Improved |
| use NHTP instead of IIHT when sample size is large | IIHT · Sparse logistic regression Ex.4, n=256, s=ceil(0.05n) | logistic loss: m/n >= 0.2 | Hurt |
| use NHTP instead of IIHT when sample size is small | IIHT · Sparse logistic regression Ex.4, n=256, s=ceil(0.05n) | logistic loss: m/n < 0.2 | Improved |
Closest papers by meaning
- Gradient Hard Thresholding Pursuit (JMLR 2018 · similarity 0.93)
- Sparse Convex Optimization via Adaptively Regularized Hard Thresholding (ICML 2020 · similarity 0.91)
- Exact Recovery of Hard Thresholding Pursuit (NeurIPS 2016 · similarity 0.91)
- A Fast and Provable Algorithm for Sparse Phase Retrieval (ICLR 2024 · similarity 0.87)
- The Price of Sparsity: Sufficient Conditions for Sparse Recovery using Sparse and Sparsified Measurements (NeurIPS 2025 · similarity 0.86)
- Gradient Compressed Sensing: A Query-Efficient Gradient Estimator for High-Dimensional Zeroth-Order Optimization (ICML 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.
