DynamicalIsometryandaMeanFieldTheoryofCNNs:HowtoTrain10,000-LayerVanillaConvolutionalNeuralNetworksLechaoXiao12YasamanBahri12JaschaSohl-Dickstein1SamuelS.Schoenholz1JeffreyPennington1AbstractFigure...
DynamicRegretofStronglyAdaptiveMethodsLijunZhang1TianbaoYang2RongJin3Zhi-HuaZhou1Abstractincurredbythelearnerandthatofthebestfixeddecisioninhindsight,i.e.,Tocopewithchangingenvironments,recentde-ve...
DynamicEvaluationofNeuralSequenceModelsBenKrause1EmmanuelKahembwe1IainMurray1SteveRenals1Abstracthavebeenshowntohaveproblemslearningtoreproducesequenceelements(Marcus,2001;Prickett,2017).IntheWeexp...
DifferentiallyPrivateIdentityandEquivalenceTestingofDiscreteDistributionsMaryamAliakbarpour1IliasDiakonikolas2RonittRubinfeld13Abstractequivalenceoftwodistributions(twosampletesting),andindependenc...
DesignofExperimentsforModelDiscriminationHybridisingAnalyticalandData-DrivenApproachesSimonOlofsson1MarcPeterDeisenroth12RuthMisener1AbstractAgency(EMA)handletheseapplications,respectively.TheFDAre...
DelayedImpactofFairMachineLearningLydiaT.Liu1SarahDean1EstherRolf1MaxSimchowitz1MoritzHardt1Abstractvantagedgroupsinthepopulation(ExecutiveofficeofthePresident,2016;Barocas&Selbst,2016).Consequentl...
DeepModelsofInteractionsAcrossSetsJasonHartford1DevonRGraham1KevinLeyton-Brown1SiamakRavanbakhsh1AbstractR.ThecanonicalrepresentationofsuchafunctionisamatrixX∈RN×M;ofcourse,wewantXn,m=xforeachWeu...
DecompositionofUncertaintyinBayesianDeepLearningforEfficientandRisk-sensitiveLearningStefanDepeweg12Jose´MiguelHerna´ndez-Lobato3FinaleDoshi-Velez4SteffenUdluft1AbstractInthisworkweshowhowtoperfo...
Data-DependentStabilityofStochasticGradientDescentIljaKuzborskij1ChristophH.Lampert2Abstractticeonemightnotevenreachaminimum,yetneverthelessobservesexcellentperformance.Weestablishadata-dependentno...
ConvolutionalImputationofMatrixNetworksQingyunSun1MengyuanYan2DavidDonoho3StephenBoyd2AbstractFigure1:(a)OriginalMRIimageframes(oracle).(b)Oursampledandcorruptedobservation.(c)Recoveredim-Amatrixne...
Convergenceguaranteesforaclassofnon-convexandnon-smoothoptimizationproblemsKoulikKhamaru1MartinJ.Wainwright12Abstractwellasreferencestherein.Accordingly,recentyearshavewitnessedanexplosionofresearc...
ConstrainingtheDynamicsofDeepProbabilisticModelsMarcoLorenzi1Mauriziofilippone2Abstractmenttoprovideamorepreciseandrealisticdescriptionofnaturalphenomena.Forexample,monotonicityoftheinter-Weintrodu...
ConditionalNoise-ContrastiveEstimationofUnnormalisedModelsCiwanCeylan1MichaelU.Gutmann2Abstractxi∈Xareindependentlysampledfromtheunknowndatadistributionpd.Unnormalisedmodelsoutputnon-negativeManyp...
CompositeFunctionalGradientLearningofGenerativeAdversarialModelsRieJohnson1TongZhang2Abstractinaterealdatafromgenerateddata.Mathematically,GANThispaperfirstpresentsatheoryforgenerativesolvesthefoll...
ClusteringSemi-RandomMixturesofGaussiansPranjalAwasthi1AravindanVijayaraghavan2Abstract1894;Dasgupta,1999;Arora&Kannan,2001;Vempala&Wang,2004;Dasgupta&Schulman,2007).Gaussianmixturemodels(GMM)areth...
CharacterizingImplicitBiasinTermsofOptimizationGeometrySuriyaGunasekar1JasonLee2DanielSoudry3NathanSrebro1Abstractassociatedhyperparametercanchangetheimplicitbias.Forexample,Wilsonetal.(2017)showed...
CharacterizingandLearningEquivalenceClassesofCausalDAGsunderInterventionsKarrenD.Yang1AbigailKatcoff1CarolineUhler1Abstracttheircausesareknownasperfect(orhard)interventions(Eberhardtetal.,2005).Und...
BoundsontheApproximationPowerofFeedforwardNeuralNetworksMohammadMehrabi1AslanTchamkerten2MansoorI.Yousefi2AbstractDeterminingthecapacityofaneuralnetworkswithapiece-wiselinearactivationfunctiontypic...
BoundingandCountingLinearRegionsofDeepNeuralNetworksThiagoSerra1ChristianTjandraatmadja1SrikumarRamalingam2Abstract1010WeinvestigatethecomplexityofdeepneuralNumberoflinearregions108networks(DNN)tha...
BayesianOptimizationofCombinatorialStructuresRicardoBaptista1MatthiasPoloczek2AbstractWepresentanovelalgorithmforthisproblem,BayesianOptimizationofCombinatorialStructures(BOCS),thatisTheoptimizatio...