GradientFlow: Optimizing Network Performance for Large-Scale Distributed DNN Training | IEEE Journals & Magazine | IEEE Xplore