Research map / Learning theory and algorithms / Kernels and regression theory
Double Trouble in Double Descent: Bias and Variance(s) in the Lazy Regime
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Increase feature count beyond the interpolation threshold | Random features regression · Random features, overparametrized regime | test error | Helped |
| Increase feature count beyond the interpolation threshold | Random features regression · Random features, bias and sampling variance | test error | No clear effect |
| Average predictions from independently initialized estimators (ablation) | Random features regression · Random features, interpolation threshold | test error | Helped |
| Average predictions from independently initialized estimators (ablation) | Random features regression · Random features, P<N | test error | Hurt |
| Average predictions from independently initialized estimators (ablation) | Random features regression · Random features, sampling variance | test error | No clear effect |
| Train independently initialized estimators on disjoint data splits (ablation) | Random features regression · Random features, noisy labels | test error | Helped |
| Train independently initialized estimators on disjoint data splits (ablation) | Random features regression · Random features, reduced effective dataset | test error | Mixed |
| Increase ridge regularization (ablation) | Random features regression · Random features, interpolation threshold | test error | Helped |
| Optimize ridge regularization over λ (ablation) | Random features regression · Random features, varying P/N | test error | Helped |
| Use K→∞ ensembling rather than an optimally regularized single estimator (ablation) | Random features regression · Random features, varying P/N | test error | Helped |
| Double the number of features (ablation) | Random features regression · Random features, varying dataset size | test error | Mixed |
| Ensemble 20 independently initialized networks (ablation) | 5-layer network on binary 10-PCA CIFAR-10, α=100 · Binary 10-PCA CIFAR-10 | misclassification error | Helped |
Closest papers by meaning
- Understanding Double Descent Requires A Fine-Grained Bias-Variance Decomposition (NeurIPS 2020 · similarity 0.92)
- Unraveling the Enigma of Double Descent: An In-depth Analysis through the Lens of Learned Feature Space (ICLR 2024 · similarity 0.91)
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 2025 · similarity 0.90)
- Noisy Interpolation Learning with Shallow Univariate ReLU Networks (ICLR 2024 · similarity 0.88)
- Simplicity Bias in Overparameterized Machine Learning (AAAI 2024 · similarity 0.88)
- Bias of Stochastic Gradient Descent or the Architecture: Disentangling the Effects of Overparameterization of Neural Networks (ICML 2024 · 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.