Research map / Learning theory and algorithms / Kernels and regression theory
Triple descent and the two kinds of overfitting: where & why do they appear?
Stéphane d'Ascoli, Levent Sagun, Giulio Biroli
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 |
|---|---|---|---|
| increase degree of nonlinearity of the activation (abs, ReLU, Tanh, linear ordering) (ablation) | random features model · noisy regression, SNR=0.2, P/D=10 | height of nonlinear peak at N=P: r=0 (abs), 0.5 (ReLU), ~0.92 (Tanh), 1 (linear) | Improved |
| increase degree of nonlinearity of the activation (ablation) | random features model · noisy regression, SNR=0.2, P/D=10 | height of linear peak at N=D | Hurt |
| replace Tanh by ReLU in student network (ablation) | teacher-student NN · noisy regression, SNR=0.2, D=196 | linear peak height: r~0.92 -> r=0.5 | Hurt |
| ensembling over 10 random feature or weight seeds (ablation) | random features and NN models · noisy regression | nonlinear peak of test loss | Improved |
| ensembling over 10 seeds (ablation) | random features and NN models · noisy regression | linear peak of test loss | No clear effect |
| add explicit regularization (ridge gamma or weight decay 0.05) (ablation) | random features and NN models · noisy regression | nonlinear peak of test loss: weight decay 0.05 | Improved |
| add explicit regularization (ridge gamma or weight decay 0.05) (ablation) | random features and NN models · noisy regression | linear peak of test loss | No clear effect |
| remove label noise (SNR to infinity) (ablation) | random features model · RF, sigma=ReLU, P/D=100 | linear peak | Improved |
| remove label noise (SNR to infinity) (ablation) | random features model · RF, sigma=ReLU, P/D=100 | nonlinear peak | No clear effect |
| use structured data (MNIST/CIFAR10) instead of random Gaussian inputs | random features model · MNIST downsampled, D=100 | location of peaks | No clear effect |
Closest papers by meaning
- Unraveling the Enigma of Double Descent: An In-depth Analysis through the Lens of Learned Feature Space (ICLR 2024 · similarity 0.89)
- Noisy Interpolation Learning with Shallow Univariate ReLU Networks (ICLR 2024 · similarity 0.88)
- Rethinking Benign Overfitting in Two-Layer Neural Networks (ICML 2025 · similarity 0.87)
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 2025 · similarity 0.87)
- Least Squares Regression Can Exhibit Under-Parameterized Double Descent (NeurIPS 2024 · similarity 0.86)
- Bias of Stochastic Gradient Descent or the Architecture: Disentangling the Effects of Overparameterization of Neural Networks (ICML 2024 · similarity 0.86)
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.
