Training of deep residual networks with stochastic MG/OPT

Re-assign community