SPECTRE:DefendingAgainstBackdoorAttacksUsingRobustStatisticsJonathanHayase1WeihaoKong1RaghavSomani1SewoongOh1AbstractaccuracyonpoisonedtestexamplesStartingwiththeseminalworkof(Guetal.,2017),thereha...
CRFL:CertifiablyRobustFederatedLearningagainstBackdoorAttacksChulinXie1MinghaoChen2Pin-YuChen3BoLi1AbstractCRFLTrainingCRFLTestingFederatedLearning(FL)asadistributedlearn-ModelUpdatesParameteringpa...
BackdoorScanningforDeepNeuralNetworksthroughK-ArmOptimizationGuangyuShen1YingqiLiu1GuanhongTao1ShengweiAn1QiulingXu1SiyuanCheng1ShiqingMa2XiangyuZhang1Abstractdoor)attackexploitstheuninterpretabili...