Deep Learning Gradient Descent with Early Stopping is Provably Robust to Label Noise for Overparameterized Neural Networks (סקירה) מיכאל ארליכסון, 24/06/2021