[v1]GoingDeeperwithConvolutions,6.67%testerror,http://arxiv.org/abs/1409.4842[v2]BatchNormalization:AcceleratingDeepNetworkTrainingbyReducingInternalCovariateShift,4.8%testerror,http://arxiv.org/abs/1502.03167[v3]RethinkingtheInceptionArchitectureforComputerVision,3.5%testerror,http://arxiv.org/