Research map / Learning theory and algorithms / Kernels and regression theory
Sparsity and Error Analysis of Empirical Feature-Based Regularization Schemes
Xin Guo, Jun Fan, Ding-Xuan Zhou
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 RKPCA with l1 penalty instead of dense RLS | RKPCA (Gaussian kernel) · Gaussian kernel simulation (n=100) | non-zero coefficients (%): 100(0)% -> 3.5(2.0)% | Improved |
| use RKPCA with l1 penalty instead of RLS, accuracy | RKPCA (Gaussian kernel) · Gaussian kernel simulation (n=100) | oracle RMSE: 0.012(0.005) -> 0.013(0.006) | No clear effect |
| use RKPCA l1 penalty instead of RLS | RKPCA (Sobolev kernel) · Sobolev kernel simulation, tau=1 (n=100) | RKHS distance to f_rho: 0.801(0.185) -> 0.685(0.318) | Improved |
| use RKPCA l1 penalty instead of RLS | RKPCA (Sobolev kernel) · Sobolev kernel simulation, tau=1 (n=1000) | RKHS distance to f_rho: 0.570(0.114) -> 0.434(0.085) | Improved |
| raise concave exponent q from 1/3 to 1 (ablation) | RKPCA (Sobolev kernel) · Sobolev kernel simulation, tau=4.5 (n=100) | RKHS distance to f_rho: 1.254(0.912) -> 0.950(0.155) | Improved |
| raise concave exponent q from 1/3 to 1 (ablation) | RKPCA (Sobolev kernel) · Sobolev kernel simulation, tau=1 (n=1000) | RKHS distance to f_rho: 0.533(0.368) -> 0.434(0.085) | Improved |
| lower concave exponent q from 1 to 1/3 (ablation) | RKPCA (Sobolev kernel) · Sobolev kernel simulation, tau=4.5 (n=100) | non-zero coefficients (%): 26.6(10.6)% -> 17.8(11.6)% | Improved |
| use SCAD penalty instead of l1 penalty (ablation) | RKPCA (Gaussian kernel) · Gaussian kernel simulation (n=100) | oracle RMSE: 0.013(0.006) -> 0.011(0.005) | No clear effect |
| use SCAD penalty instead of l1 penalty (ablation) | RKPCA (Gaussian kernel) · Gaussian kernel simulation (n=1000) | non-zero coefficients (%): 0.4(0.4)% -> 0.4(0.2)% | No clear effect |
| use RLS with string kernel K3 instead of NN-align | RLS · IEDB HLA-peptide binding (14 alleles, average) | AUC: 0.7982 -> 0.80260 | Improved |
| use RKPCA l1 penalty instead of RLS | RKPCA · IEDB HLA-peptide binding (14 alleles, average) | AUC: 0.80260 -> 0.80351 | Improved |
| use RKPCA l1 penalty instead of RLS | RKPCA · IEDB HLA-peptide binding (14 alleles, average) | RMSE: 0.21100 -> 0.21101 | No clear effect |
Closest papers by meaning
- Sparse Representer Theorems for Learning in Reproducing Kernel Banach Spaces (JMLR 2024 · similarity 0.83)
- Solving L1-regularized SVMs and Related Linear Programs: Revisiting the Effectiveness of Column and Constraint Generation (JMLR 2022 · similarity 0.83)
- Random Smoothing Regularization in Kernel Gradient Descent Learning (JMLR 2024 · similarity 0.83)
- Functional Regularization for Representation Learning: A Unified Theoretical Perspective (NeurIPS 2020 · similarity 0.83)
- Learning Optimal Linear Regularizers (ICML 2019 · similarity 0.83)
- Optimal Rates for Vector-Valued Spectral Regularization Learning Algorithms (NeurIPS 2024 · similarity 0.81)
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.
