Table 4. Pre- and post-defense accuracy... | Emerald Publishing

Table 4.

Pre- and post-defense accuracy of attack model on CIFAR-100

Attack method	Metric	No. defended	Non-member mask	Member mask
Ml-leaks	Accuracy	0.9134	0.5000	0.5396
	Precision	0.9033	0.0000	0.5206
	Recall	0.9702	0.0000	0.9999
	F1-score	0.9355	–	0.6847
BlindMI	Accuracy	0.8830	0.5010	0.5031
	Precision	0.8119	0.5005	0.5039
	Recall	0.9969	1.0000	0.4033
	F1-score	0.8958	0.6671	0.4480

Attack method	Metric	No. defended	Non-member mask	Member mask
Ml-leaks	Accuracy	0.9134	0.5000	0.5396
	Precision	0.9033	0.0000	0.5206
	Recall	0.9702	0.0000	0.9999
	F1-score	0.9355	–	0.6847
BlindMI	Accuracy	0.8830	0.5010	0.5031
	Precision	0.8119	0.5005	0.5039
	Recall	0.9969	1.0000	0.4033
	F1-score	0.8958	0.6671	0.4480

Source: Authors’ own data, using the Ml-leaks described by Salem et al. and BlindMI described by Hui et al.