Research map / Optimization / Nonconvex and smooth optimization
Expanding Boundaries of Gap Safe Screening
Cassio F. Dantas, Emmanuel Soubies, Cédric Févotte
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 the new tighter local strong concavity bound (Algorithm 2) instead of the earlier bound | Gap Safe screening for sparse KL, SPIRAL · NIPS papers KL, lambda/lambda_max=1e-1, eps 1e-5 | average speedup: 2.77 -> 8.81 | Improved |
| use the new tighter strong concavity bound instead of the earlier bound | Screening with CoD solver · NIPS papers KL, lambda/lambda_max=1e-1, eps 1e-5 | average speedup: 6.71 -> 17.03 | Improved |
| add iterative refinement of the strong concavity bound (Algorithm 3 over Algorithm 2) | Gap Safe screening, KL divergence · NIPS papers KL | average speedup: 8.81 -> 8.75 | No clear effect |
| add iterative refinement of the strong concavity bound (Algorithm 3 over Algorithm 2) (ablation) | Gap Safe screening, beta=1.5 with MU · Urban hyperspectral, lambda/lambda_max<=1e-2 | average speedup | Improved |
| add iterative refinement (Algorithm 3 over Algorithm 2) (ablation) | Gap Safe screening, logistic regression with CoD · Leukemia, low regularization (1e-3) | average speedup | Improved |
| use local strong concavity bound (Algorithm 2) instead of global bound (Algorithm 1) (ablation) | Gap Safe screening, logistic regression · Leukemia, lambda/lambda_max<1.2e-2 | average speedup | Improved |
| initialize Algorithm 3 with the global bound instead of the local bound (ablation) | Algorithm 3, logistic regression · Leukemia, lambda/lambda_max=1e-3 | average speedup | No clear effect |
| add safe screening to MU solver | MU for beta=1.5 · synthetic m=10, n=20, lambda/lambda_max=1e-2 | average speedup | Improved |
| decrease convergence tolerance eps_gap from 1e-5 to 1e-7 | screening speedup, KL CoD · NIPS papers KL, lambda/lambda_max=1e-1 | average speedup: 17.03 -> 19.70 | Improved |
Closest papers by meaning
- Gap Safe Screening Rules for Sparsity Enforcing Penalties (JMLR 2017 · similarity 0.89)
- Simultaneous Safe Screening of Features and Samples in Doubly Sparse Modeling (ICML 2016 · similarity 0.88)
- GAP Safe Screening Rules for Sparse-Group Lasso (NeurIPS 2016 · similarity 0.87)
- Safe screening rules for L0-regression from Perspective Relaxations (ICML 2020 · similarity 0.87)
- Dynamic Sasvi: Strong Safe Screening for Norm-Regularized Least Squares (NeurIPS 2021 · similarity 0.87)
- Robust Sparse Regression with Non-Isotropic Designs (NeurIPS 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.
