GradientDisaggregation:BreakingPrivacyinFederatedLearningbyReconstructingtheUserParticipantMatrixMaximilianLam1Gu-YeonWei1DavidBrooks1VijayJanapaReddi1MichaelMitzenmacher1Abstractorwearables,andmai...
Accuracy,Interpretability,andDifferentialPrivacyviaExplainableBoostingHarshaNori1RichCaruana1ZhiqiBu2JudyHanwenShen3JanardhanKulkarni1AbstractWhileGAMsweretraditionallyfitusingsmoothlow-orderspline...
SharpCompositionBoundsforGaussianDifferentialPrivacyviaEdgeworthExpansionQinqingZheng1JinshuoDong1QiLong1WeijieSu1Abstracttheprioralgorithms’output,anadversarycanadaptivelyanalyzethedatasettocause...
ScalableDifferentialPrivacywithCertifiedRobustnessinAdversarialLearningNhatHaiPhan1MyT.Thai2HanHu1RuomingJin3TongSun4DejingDou56Abstracttrainingdatatoconsistofbenignexamplesandadversar-ialexamplesc...
PrivateReinforcementLearningwithPACandRegretGuaranteesGiuseppeVietri1BorjaBalle2AkshayKrishnamurthy3StevenWu1Abstractingdataisavailablebeforehand.Whilethesetechniquescoveralargenumberofapplications...
OptimalDifferentialPrivacyCompositionforExponentialMechanismsJinshuoDong12DavidDurfee2RyanRogers2Abstractsuchashyperparameterselection(LiuandTalwar,2019),privatePCA(Chaudhurietal.,2013),syntheticda...
Context-AwareLocalDifferentialPrivacyJayadevAcharya1K.A.Bonawitz2PeterKairouz2DanielRamage2ZitengSun13Abstractversion,localdifferentialPrivacy(LDP)(Definition1),pro-videscontext-freePrivacyguarante...
BayesianDifferentialPrivacyforMachineLearningAlekseiTriastcyn1BoiFaltings1AbstractDespitenotableadvances,differentiallyprivateMLstillsuffersfromtwomajorproblems:(a)utilitylossduetoTraditionaldiffer...
AlleviatingPrivacyAttacksviaCausalLearningShrutiTople1AmitSharma1AdityaV.Nori1Abstractsuchattackcalledmembershipinferencerevealswhetheradatasamplewaspresentinthetrainingdataset(ShokriMachinelearnin...
PoissonSubsampledRenyiDifferentialPrivacyYuqingZhu1Yu-XiangWang1Abstract2016)fordifferentiallyprivatedeeplearning.NoisySGDiterativelyupdatesthemodelparametersasfollows:Weconsidertheproblemof"privac...
LowLatencyPrivacyPreservingInferenceAlonBrutzkus1OrenElisha2RanGilad-Bachrach3Abstractdictionservice,whichshouldbecapableofoperatingontheencrypteddatawithouthavingaccesstotherawdata.Whenapplyingmac...
FormalPrivacyforFunctionalDatawithGaussianPerturbationsArdalanMirshani1MatthewReimherr1AleksandraSlavkovic1Abstractdata,withthecorrespondingbranchofstatisticscalledfunc-tionaldataanalysis,FDA.FDAis...
ImprovingthePrivacyandAccuracyofADMM-BasedDistributedAlgorithmsXueruZhang1MohammadMahdiKhalili1MingyanLiu1Abstractinteractiveprocessoflocalcomputationandmessagepass-ing,whichideallyshouldresultinal...
ImprovingtheGaussianMechanismforDifferentialPrivacy:AnalyticalCalibrationandOptimalDenoisingBorjaBalle1Yu-XiangWang23AbstracttheGaussianmechanismisthebuildingblockofprivateempiricalriskminimization...
ThePriceofDifferentialPrivacyforOnlineLearningNamanAgarwal1KaranSingh1Abstractthefullinformationandpartialinformation(bandit)set-tings.ThisresultimprovestheknownbestregretboundsWedesigndifferential...
Pain-FreeRandomDifferentialPrivacywithSensitivitySamplingBenjaminI.P.Rubinstein1FrancescoAlda`2Abstracttentrequirementistheneedtoboundglobalsensitivity—aLipschitzconstantofthetarget,non-privatefun...