Complete Defense Framework to Protect Deep Neural Networks against Adversarial Examples

Table 9

Performance of ResGN optimized by FGSM adversarial examples with strength .

Training set	Testing set
FGSM	FGSM
	80.6			86.3	89.9	93.6
	R-FGSM
	78.8			81.4	87.2	92.3
	BIM
	79.3			90.6	86.8	91.2
	UAP	DeepFool	CW_UT	CW_T	Legitimate	Average
	90.4	90.6	91.7	90.4	98.1	87.6