Research Article

Adversarial Sample Detection with Gaussian Mixture Conditional Generative Adversarial Networks

Table 3

Unsupervised scenarios for detecting adversarial samples.

ModelDatasetMethodFGSMUnsupervised scenario
BIMDeepFoolCWPGD

DenseNetCIFAR10Odds-testing45.2369.0158.3061.2997.93
GM-CGAN87.8973.6980.8178.1246.45
CIFAR100Odds-testing43.2265.2249.5347.6496.91
GM-CGAN98.0968.3474.4265.7241.09
SVHNOdds-testing56.1471.1167.8170.7199.25
GM-CGAN83.3573.5681.8680.1646.40

ResNetCIFAR10Odds-testing46.3259.8575.5857.5896.18
GM-CGAN96.6864.7479.4873.4998.80
CIFAR100Odds-testing38.2643.5261.1344.7493.73
GM-CGAN80.9681.2080.3567.5691.30
SVHNOdds-testing65.0970.3177.0572.1299.08
GM-CGAN94.9789.0494.7183.4197.52