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

Table 6

Classification accuracy rates of adversarial examples recovered by which trained with different datasets generated by certain method. (FGSM) means that we train the CycleAdvGAN with the adversarial examples crafted by FGSM.

Source		Target
		FGSM	BIM	PGD	Clean
	FGSM		93.5	92.3	98.3
	BIM	94.6		95.2	98.4
	PGD	95.6	94.5		97.8