Complete Defense Framework to Protect Deep Neural Networks against Adversarial Examples

Table 8

Performance of the combination of the two detectors vs. the integrated detector [12] and feature squeezing detector [33] (%).

Detector	TPR								FPR
	FGSM	R-FGSM	BIM	UAP	DeepFool	CW_UT	CW_T	Average
The proposed detector	99.8	99.9	99.9	99.5	94.7	94.6	95.1	97.6	6.3
Integrated detector [12]	99.5	99.6	99.5	96.9	96.3	96.1	96.4	97.7	7.1
Feature squeezing detector [33]	56.3	52.3	67.9	88.1	90.4	95.3	97.1	78.2	8.2