SGD Converges to Global Minimum in Deep Learning via Star-convex Path

Open in new window