Gradient Descent with Early Stopping is Provably Robust to Label Noise for Overparameterized Neural Networks.

Mingchen Li, Mahdi Soltanolkotabi, Samet Oymak

Browse the full AISTATS paper archive.