Complete Defense Framework to Protect Deep Neural Networks against Adversarial Examples

Table 7

Performance of the minor alteration detector (%).

Operation		ri	re	rc	ra	max
Threshold		0.747	0.682	0.574	0.564	0.993
TPR	DeepFool	91.8	93.4	90.8	91.5	94.7
	CW_T	92.3	94.1	91.2	92.4	94.6
	CW_UT	92.7	94.1	91.3	92.7	95.0
	Overall	92.3	93.9	91.1	92.2	94.8
FPR	Legitimate	5.7	6.1	5.5	6.0	4.7