FILTRA:RethinkingSteerableCNNbyFilterTransformBoLi1QiliWang1GimHeeLee2Abstractputishard-bakedtotransformaccordinglywhentheinputreflectsorrotates.AplentyofworksdevelopthisideaSteerableCNNimposesthep...
PENNI:PrunedKernelSharingforEfficientCNNInferenceShiyuLi1EdwardHanson1HaiLi1YiranChen1Abstractlayer(Zagoruyko&Komodakis,2016),ormorebranches(Szegedyetal.,2015).Amajordrawbackofna¨ıvelyin-Although...
ApproximatedOracleFilterPruningforDestructiveCNNWidthOptimizationXiaohanDing1GuiguangDing1YuchenGuo1JungongHan2ChenggangYan3Abstractincludestheusageofnon-linearities(ReLU,sigmoidornone),downsamplin...
GradientDescentLearnsOne-hidden-layerCNN:Don’tbeAfraidofSpuriousLocalMinimaSimonS.Du1JasonD.Lee2YuandongTian3Barnaba´sPo´czos1AartiSingh1Abstractfully.WhysuchsimplemethodsinlearningDCNNissuc-ces...