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

Open in new window