SkewOrthogonalConvolutionsSahilSingla1SoheilFeizi1Abstractadversarialrobustness(Cisse´etal.,2017;Szegedyetal.,2014)andinterpretablegradients(Tsiprasetal.,2018).TheTrainingconvolutionalneuralnetwor...
EvolvingAttentionwithResidualConvolutionsYujingWang1YamingYang2JiangangBai12MingliangZhang12JingBai2JingYu3CeZhang4GaoHuang5YunhaiTong1Abstract8079.6379.1Transformerisaubiquitousmodelfornaturallan-...
Time-awareLargeKernelConvolutionsVasileiosLioutas1YuhongGuo1Abstractetal.,2019;Radfordetal.,2019),automaticsummarization(Paulusetal.,2018;Fanetal.,2018;Celikyilmazetal.,Todate,moststate-of-the-arts...
PDO-eConvs:PartialDifferentialOperatorBasedEquivariantConvolutionsZhengyangShen1LingshenHe2ZhouchenLin2JinwenMa1AbstractasignificantadvantageofCNNsisthattheyareshiftequiv-ariant:shiftinganimageandt...
Memory-OptimalDirectConvolutionsforMaximizingClassificationAccuracyinEmbeddedApplicationsAlbertGural1BorisMurmann1Abstract(IoT),themachinelearningmodelistrainedonaserverandsubsequentlydeployedacros...
EmergingConvolutionsforGenerativeNormalizingFlowsEmielHoogeboom1RiannevandenBerg2MaxWelling13Abstractin2in2Generativeflowsareattractivebecausetheyadmitin1out2=exactlikelihoodoptimizationandefficien...
HighPerformanceZero-MemoryOverheadDirectConvolutionsJiyuanZhang1FranzFranchetti1TzeMengLow1AbstractPerformancenormalizedtoOpenBLASGEMMonAMDPileDriverThecomputationofconvolutionlayersindeep4.0GHz,4/...
WarpedConvolutions:EfficientInvariancetoSpatialTransformationsJoa˜oF.Henriques1AndreaVedaldi1Abstractacrossspace)andlesscomputation(duetotheirlimitedsupport)comparedtoafully-connectedlayer.Itsloca...