Cycle-Consistent Adversarial GAN: The Integration of Adversarial Attack and Defense

Table 4

Classification accuracy of adversarial examples generated by different methods transferred between different models on MNIST. (FGSM) means that we train the CycleAdvGAN with the adversarial examples crafted by FGSM.

Target		Source
		A
		FGSM	BIM	PGD
					FGSM	BIM	PGD
B	Without	46.530	28	20.8	34	24	18.8
	Defense with	98.2	97.3	96.8