FNet: A Two-Stream Model for Detecting Adversarial Attacks against 5G-Based Deep Learning Services

Table 7

Performance of normal images and their adversarial examples generated by CW on CIFAR-10.

Model		Method	Normal images		Adv images
			Precision	Recall	Precision	Recall
White model	VGG16	RGB-Net	0.912	0.856	0.843	0.903
		SRM-Net	0.539	1.000	0.000	0.000
		KDBU [32]	0.852	0.525	0.617	0.893
		FNet	0.913	0.922	0.908	0.898
Black model	ResNet	RGB-Net	0.916	0.856	0.840	0.906
		SRM-Net	0.544	1.000	0.000	0.000
		KDBU [32]	0.545	0.525	0.457	0.478
		FNet	0.883	0.922	0.901	0.855
	LeNet	RGB-Net	0.943	0.856	0.792	0.914
		SRM-Net	0.625	1.000	0.000	0.000
		KDBU [32]	0.589	0.525	0.330	0.390
		FNet	0.903	0.922	0.865	0.835