Research map / Learning theory and algorithms / Matrix and tensor methods
A Constructive Approach to $L_0$ Penalized Regression
Jian Huang, Yuling Jiao, Yanyan Liu, Xiliang Lu
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 SDAR instead of MCP at low correlation | MCP · Simulation rho=0.2 | CPU time (s): 9.3e+2 -> 1.4e+0 | Improved |
| use SDAR instead of MCP | MCP · Simulation rho=0.4 | CPU time (s): 2.2e+2 -> 2.3e+0 | Improved |
| use SDAR instead of FoBa at higher correlation | FoBa · Simulation rho=0.4 | relative error: 1.0e-2 -> 6.0e-4 | Improved |
| use SDAR instead of FoBa at rho=0.6 | FoBa · Simulation rho=0.6 | relative error: 8.3e-3 -> 4.3e-4 | Improved |
| use SDAR instead of LARS | LARS · Simulation rho=0.6 | relative error: 3.0e-1 -> 4.3e-4 | Improved |
| use SDAR instead of GraDes | GraDes · Simulation rho=0.2 | relative error: 1.1e-3 -> 7.5e-4 | Improved |
| use SDAR instead of ASDAR (no adaptive support size) | ASDAR · Simulation rho=0.2 | CPU time (s): 8.4e+0 -> 1.4e+0 | Improved |
| use ASDAR instead of MCP, GraDes, FoBa, LARS for increasing sparsity K | ASDAR · Support recovery (n=500, p=1000) | probability of exact recovery: ASDAR good up to K=260; MCP fails K>110; GraDes and FoBa fail K>160 | Improved |
| use ASDAR when sample size is small | ASDAR · Support recovery (p=500, K=10) | probability of exact recovery: better at n=30 and 50 | Improved |
| use ASDAR as ambient dimension p grows | ASDAR · Support recovery (n=100, K=20) | probability of exact recovery: ASDAR and MCP higher than others as p increases | Improved |
| use ASDAR as correlation rho increases | ASDAR · Support recovery (n=150, p=500, K=25) | probability of exact recovery: generally better than others as rho rises from 0.05 to 0.95 | Improved |
| use SDAR instead of GraDes (IHT) for iteration count as K grows | GraDes · Iteration count (n=500, p=1000) | number of iterations: SDAR iterations vary little vs GraDes increasing almost sublinearly | Improved |
Closest papers by meaning
- Feature Adaptation for Sparse Linear Regression (NeurIPS 2023 · similarity 0.88)
- Robust Sparse Regression with Non-Isotropic Designs (NeurIPS 2024 · similarity 0.88)
- Harder, Better, Faster, Stronger Convergence Rates for Least-Squares Regression (JMLR 2017 · similarity 0.88)
- Rank-one Convexification for Sparse Regression (JMLR 2025 · similarity 0.88)
- Sharp Oracle Inequalities for Square Root Regularization (JMLR 2017 · similarity 0.88)
- Efficient Sublinear-Regret Algorithms for Online Sparse Linear Regression with Limited Observation (NeurIPS 2017 · 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.
