no code implementations • 26 Jun 2019 • Kosuke Haruki, Taiji Suzuki, Yohei Hamakawa, Takeshi Toda, Ryuji Sakai, Masahiro Ozawa, Mitsuhiro Kimura
Large-batch stochastic gradient descent (SGD) is widely used for training in distributed deep learning because of its training-time efficiency, however, extremely large-batch SGD leads to poor generalization and easily converges to sharp minima, which prevents naive large-scale data-parallel SGD (DP-SGD) from converging to good minima.