Defending against Deep-Learning-Based Flow Correlation Attacks with Adversarial Examples

Table 4

The protection success rate of various adversarial examples against flow correlation attacks with the same bandwidth overhead. PSR presents the protection success rate.

Method	Bandwidth overhead ()	DeepCorr (PSR) (%)	RAPTOR (PSR) (%)	Pearson (PSR) (%)	Cosine (PSR) (%)
FGSM	0.20	71.2	93.8	97.5	97.2
C&W	0.20	97.4	93.5	97.2	96.4
Deepfool	0.20	93.6	92.7	96.9	96.2
BIM	0.20	87.5	95.6	97.2	95.8