Dataset

Wiki

Threshold Values

Threshold = 0.0001

Threshold = 0.11

Threshold = −0.08

Metrics

Accuracy

Precision

Recall

Accuracy

Precision

Recall

Accuracy

Precision

Recall

Attacks

Models

Self-calibrated Probabilistic Variation MIA

Mistral 7B

92.20

86.51

100.00

50.00

50.00

100

97.40

100

94.80

LLaMA 7B

89.70

82.92

100.00

50.00

50.00

100.00

65.90

100.00

31.80

LLaMA 13B

92.30

90.91

94.00

50.20

50.10

100.00

53.90

100.00

7.80

Mixtral 8x7B

94.70

90.42

100.00

50.10

50.05

100.00

78.20

100.00

56.40