Early-stopped gradient descent achieves the minimax-optimal classification error for Gaussian mixtures with label noise under fast-decaying covariance spectra, while interpolating classifiers can be exponentially worse.
Benign overfitting in linear regression.Proceedings of the National Academy of Sciences, 117(48):30063–30070, 2020
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
stat.ML 1years
2026 1verdicts
ACCEPT 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Minimax Optimal Early-Stopped Gradient Descent for Gaussian Mixture Classification
Early-stopped gradient descent achieves the minimax-optimal classification error for Gaussian mixtures with label noise under fast-decaying covariance spectra, while interpolating classifiers can be exponentially worse.