Membership Inference Attack on Yelp
94AUCProjRes
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ProjResTarget Model=Llama3-8B2026.04 | 94 | 72.16 | |
| ProjResTarget Model=Qwen2.5-14B2026.04 | 93.3 | 71.82 | |
| ProjResTarget Model=BERT-Base2026.04 | 89.4 | 46.32 | |
| ProjResTarget Model=GPT2-Large2026.04 | 88 | 36.6 | |
| FedMIATarget Model=BERT-Base2026.04 | 61.1 | — | |
| CosineTarget Model=BERT-Base2026.04 | 58.2 | — | |
| Gradient-diffTarget Model=BERT-Base2026.04 | 57.4 | — | |
| CosineTarget Model=GPT2-Large2026.04 | 55.8 | — | |
| Score-DiffTarget Model=BERT-Base2026.04 | 55.6 | — | |
| Score-RatioTarget Model=BERT-Base2026.04 | 55.3 | — | |
| FedMIATarget Model=GPT2-Large2026.04 | 54.9 | — | |
| Gradient-diffTarget Model=Llama3-8B2026.04 | 54.6 | — | |
| Gradient-diffTarget Model=Qwen2.5-14B2026.04 | 54.3 | — | |
| Fed-lossTarget Model=BERT-Base2026.04 | 54.1 | — | |
| CosineTarget Model=Llama3-8B2026.04 | 53.9 | — | |
| Score-DiffTarget Model=GPT2-Large2026.04 | 53.9 | — | |
| FedMIATarget Model=Llama3-8B2026.04 | 53.7 | — | |
| Score-RatioTarget Model=GPT2-Large2026.04 | 53.5 | — | |
| Score-RatioTarget Model=Qwen2.5-14B2026.04 | 53.4 | — | |
| Score-DiffTarget Model=Qwen2.5-14B2026.04 | 53.4 | — | |
| Score-RatioTarget Model=Llama3-8B2026.04 | 53.2 | — | |
| Score-DiffTarget Model=Llama3-8B2026.04 | 53.2 | — | |
| Gradient-diffTarget Model=GPT2-Large2026.04 | 53.1 | — | |
| FedMIATarget Model=Qwen2.5-14B2026.04 | 53 | — | |
| FTATarget Model=GPT2-Large2026.04 | 52.1 | — | |
| Fed-lossTarget Model=Llama3-8B2026.04 | 52 | — | |
| CosineTarget Model=Qwen2.5-14B2026.04 | 51.8 | — | |
| FTATarget Model=Qwen2.5-14B2026.04 | 51.5 | — | |
| Fed-lossTarget Model=GPT2-Large2026.04 | 50.9 | — | |
| Fed-lossTarget Model=Qwen2.5-14B2026.04 | 50.8 | — | |
| FTATarget Model=Llama3-8B2026.04 | 50 | — | |
| FTATarget Model=BERT-Base2026.04 | 44.6 | — |