AI Paper Atlas

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

NeurIPS 2020 · Learning theory and algorithms › Kernels and regression theory · approach: Double descent and overfitting · 27 citations (OpenAlex)

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 changedWhereMeasuredEffect
increase degree of nonlinearity of the activation (abs, ReLU, Tanh, linear ordering) (ablation)random features model · noisy regression, SNR=0.2, P/D=10height 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=10height of linear peak at N=DHurt
replace Tanh by ReLU in student network (ablation)teacher-student NN · noisy regression, SNR=0.2, D=196linear peak height: r~0.92 -> r=0.5Hurt
ensembling over 10 random feature or weight seeds (ablation)random features and NN models · noisy regressionnonlinear peak of test lossImproved
ensembling over 10 seeds (ablation)random features and NN models · noisy regressionlinear peak of test lossNo clear effect
add explicit regularization (ridge gamma or weight decay 0.05) (ablation)random features and NN models · noisy regressionnonlinear peak of test loss: weight decay 0.05Improved
add explicit regularization (ridge gamma or weight decay 0.05) (ablation)random features and NN models · noisy regressionlinear peak of test lossNo clear effect
remove label noise (SNR to infinity) (ablation)random features model · RF, sigma=ReLU, P/D=100linear peakImproved
remove label noise (SNR to infinity) (ablation)random features model · RF, sigma=ReLU, P/D=100nonlinear peakNo clear effect
use structured data (MNIST/CIFAR10) instead of random Gaussian inputsrandom features model · MNIST downsampled, D=100location of peaksNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy