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

Yi Zhou, Junjie Yang, Huishuai Zhang, Yingbin Liang, Vahid Tarokh

Browse the full ICLR paper archive.