ErrorCompensatedQuantizedSGDanditsApplicationstoLarge-scaleDistributedOptimizationJiaxiangWu1WeidongHuang1JunzhouHuang1TongZhang1AbstractTherehavebeenseveralworksattemptingtoimprovetheefficiencyofd...
AnAlternativeView:WhenDoesSGDEscapeLocalMinima?RobertKleinberg1YuanzhiLi2YangYuan1Abstractmoreiterationstoconverge,butfewergradientevaluationsperiteration.Therefore,forthestandardempiricalriskmin-S...