Research map / Learning theory and algorithms / Kernels and regression theory
Signal Strength Estimation in Logistic Regression Using Data Splitting
Weihao Li, Jun S. Liu
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 sigma*-based proxy instead of kappa^2 alpha*^2-based proxy for signal strength estimation (ablation) | data-splitting kappa estimator · Identity covariance, delta=3, kappa=3, p=800 | |kappa - kappa_hat|: 0.322 -> 0.155 | Improved |
| use true v_j instead of estimated v_j in confidence intervals (ablation) | adjusted CIs under general covariance · General covariance, delta=3, kappa=1.226, p=100 | 95% CI coverage: 0.983 -> 0.950 | Improved |
| increase sample ratio delta from 3 to 6 (larger n) (ablation) | data-splitting kappa estimator · General covariance, estimated v_j | 95% CI coverage: 0.983 -> 0.973 (p=100, kappa about 1.2) | Improved |
| increase dimension p from 100 to 800 (ablation) | data-splitting kappa estimator (sigma*-based) · Identity covariance, delta=3, kappa=2 | |kappa - kappa_hat|: 0.303 -> 0.106 | Improved |
| increase signal strength kappa from 2 to 4 (delta=3, p=800) (ablation) | data-splitting kappa estimator (sigma*-based) · Identity covariance, delta=3 | |kappa - kappa_hat|: 0.106 -> 0.244 | Hurt |
| use regularization-free MLE-based proxy kappa^2 alpha*^2 + sigma*^2 (as in SLOE) instead of separate proxies (ablation) | signal strength estimation · Generalized ridge, delta=2 | identifiability of kappa | Hurt |
| Replace MLE with generalized ridge regularization weighted by sample covariance | logistic regression estimator · Separable data (delta=3, kappa=3) | estimator existence: 0.95 of datasets separable at identity covariance | Improved |
| heavy-tailed Student-t design with df below 4 instead of Gaussian (ablation) | generalized ridge theory · Universality experiment (p=500) | match to theoretical squared error | Hurt |
| use Student-t design with df 4 or greater instead of Gaussian (ablation) | generalized ridge theory · Universality experiment (p=500) | match to theoretical squared error | No clear effect |
| compare data splitting (DS) with SLOE and ProbeFrontier on non-separable data | signal strength estimators · Identity covariance, non-separable data | |kappa - kappa_hat| | No clear effect |
Closest papers by meaning
- The Impact of Regularization on High-dimensional Logistic Regression (NeurIPS 2019 · similarity 0.85)
- A Provably Accurate Randomized Sampling Algorithm for Logistic Regression (AAAI 2024 · similarity 0.84)
- More Efficient Estimation for Logistic Regression with Optimal Subsamples (JMLR 2019 · similarity 0.84)
- TIC-TAC: A Framework For Improved Covariance Estimation In Deep Heteroscedastic Regression (ICML 2024 · similarity 0.82)
- Finite-Sample Performance of Gradient Descent in Logistic Regression with Gaussian Design (NeurIPS 2026 · similarity 0.82)
- Risk and cross validation in ridge regression with correlated samples (ICML 2025 · 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.
