Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

VL-MIA

Benchmarks

Task NameDataset NameSOTA ResultTrend
Membership Inference Attack (TEXTLEN=32)VL-MIA TEXT
Accuracy78.2
33
Membership Inference AttackVL-MIA Text LLaMA Adapter TEXTLEN=64
AUC77.8
21
Membership Inference AttackVL-MIA Text LLaMA Adapter TEXTLEN=32
AUC82.8
21
Membership Inference AttackVL-MIA Text MiniGPT-4 TEXTLEN=64
AUC89.1
21
Membership Inference AttackVL-MIA Text MiniGPT-4 TEXTLEN=32
AUC82.4
21
Membership Inference AttackVL-MIA Text LLaVA TEXTLEN=64
AUC0.994
21
Membership Inference AttackVL-MIA Text LLaVA TEXTLEN=32
AUC81
21
Membership Inference Attack (TEXTLEN=64)VL-MIA TEXT
Accuracy96.3
18
Image Membership Inference AttackVL-MIA Flickr (test)
AUC0.63
14
Image Membership Inference AttackVL-MIA DALL-E (test)
AUC81.5
14
Membership Inference AttackVL-MIA DALL-E 1.0 (test)
TPR@5%FPR (Aug_KL)8.1
11
Membership Inference AttackVL-MIA Flickr 1.0 (test)
TPR@5%FPR (Aug_KL)8
11
Showing 12 of 12 rows