Fetching the paper…

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