List your company
ResearchFormerly WizWand

Membership Inference Attack on SST

1AUCProjRes
0.36040.526450.69250.85855Apr 23, 2026
Updated 6mo ago

Evaluation results

MethodLinks
ProjResTarget Model=BERT-Base2026.04140.25
ProjResTarget Model=GPT2-Large2026.04138.89
ProjResTarget Model=Llama3-8B2026.04172.41
ProjResTarget Model=Qwen2.5-14B2026.04171.23
FedMIATarget Model=GPT2-Large2026.040.72—
FedMIATarget Model=BERT-Base2026.040.713—
CosineTarget Model=GPT2-Large2026.040.709—
Fed-lossTarget Model=BERT-Base2026.040.7—
CosineTarget Model=BERT-Base2026.040.665—
Fed-lossTarget Model=GPT2-Large2026.040.643—
Gradient-diffTarget Model=BERT-Base2026.040.616—
Score-RatioTarget Model=BERT-Base2026.040.612—
Score-DiffTarget Model=BERT-Base2026.040.596—
Score-RatioTarget Model=GPT2-Large2026.040.595—
Score-DiffTarget Model=GPT2-Large2026.040.594—
CosineTarget Model=Qwen2.5-14B2026.040.584—
Fed-lossTarget Model=Qwen2.5-14B2026.040.582—
Score-DiffTarget Model=Llama3-8B2026.040.58—
Score-RatioTarget Model=Llama3-8B2026.040.579—
Fed-lossTarget Model=Llama3-8B2026.040.57—
Score-DiffTarget Model=Qwen2.5-14B2026.040.57—
CosineTarget Model=Llama3-8B2026.040.567—
Score-RatioTarget Model=Qwen2.5-14B2026.040.558—
FedMIATarget Model=Qwen2.5-14B2026.040.557—
Gradient-diffTarget Model=Qwen2.5-14B2026.040.557—
Gradient-diffTarget Model=GPT2-Large2026.040.553—
Gradient-diffTarget Model=Llama3-8B2026.040.536—
FedMIATarget Model=Llama3-8B2026.040.494—
FTATarget Model=Llama3-8B2026.040.462—
FTATarget Model=Qwen2.5-14B2026.040.449—
FTATarget Model=GPT2-Large2026.040.412—
FTATarget Model=BERT-Base2026.040.385—