DoWeActuallyNeedDenseOver-Parameterization?In-TimeOver-ParameterizationinSparseTrainingShiweiLiu1LuYin1DecebalConstantinMocanu12MykolaPechenizkiy1AbstractFigure1.Asthefigureproceeds,weperformanOver...
AConvergenceTheoryforDeepLearningviaOver-ParameterizationZeyuanAllen-Zhu1YuanzhiLi23ZhaoSong456Abstract1IntroductionDeepneuralnetworks(DNNs)havedemon-Neuralnetworkshavedemonstratedagreatsuccessinst...