Membership Inference Attack on CoLA
100AUCProjRes
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ProjResTarget Model=BERT-Base2026.04 | 100 | 33.68 | |
| ProjResTarget Model=GPT2-Large2026.04 | 100 | 43.68 | |
| ProjResTarget Model=Llama3-8B2026.04 | 100 | 75.75 | |
| ProjResTarget Model=Qwen2.5-14B2026.04 | 100 | 68.63 | |
| FedMIATarget Model=BERT-Base2026.04 | 74.8 | — | |
| CosineTarget Model=BERT-Base2026.04 | 74.5 | — | |
| Score-RatioTarget Model=GPT2-Large2026.04 | 69.6 | — | |
| Fed-lossTarget Model=BERT-Base2026.04 | 68.5 | — | |
| Fed-lossTarget Model=GPT2-Large2026.04 | 68.4 | — | |
| Score-RatioTarget Model=BERT-Base2026.04 | 67.4 | — | |
| Score-DiffTarget Model=GPT2-Large2026.04 | 67.1 | — | |
| CosineTarget Model=GPT2-Large2026.04 | 64.1 | — | |
| Score-DiffTarget Model=BERT-Base2026.04 | 63.6 | — | |
| Gradient-diffTarget Model=GPT2-Large2026.04 | 62.4 | — | |
| Gradient-diffTarget Model=Qwen2.5-14B2026.04 | 59.3 | — | |
| FedMIATarget Model=GPT2-Large2026.04 | 59.1 | — | |
| Gradient-diffTarget Model=Llama3-8B2026.04 | 56.9 | — | |
| Gradient-diffTarget Model=BERT-Base2026.04 | 56.5 | — | |
| FedMIATarget Model=Llama3-8B2026.04 | 54.5 | — | |
| FedMIATarget Model=Qwen2.5-14B2026.04 | 53 | — | |
| FTATarget Model=Qwen2.5-14B2026.04 | 51.8 | — | |
| FTATarget Model=GPT2-Large2026.04 | 50 | — | |
| FTATarget Model=BERT-Base2026.04 | 49 | — | |
| CosineTarget Model=Qwen2.5-14B2026.04 | 46.4 | — | |
| Score-RatioTarget Model=Llama3-8B2026.04 | 45.1 | — | |
| CosineTarget Model=Llama3-8B2026.04 | 45 | — | |
| Score-DiffTarget Model=Llama3-8B2026.04 | 44.7 | — | |
| Score-DiffTarget Model=Qwen2.5-14B2026.04 | 44.5 | — | |
| Fed-lossTarget Model=Llama3-8B2026.04 | 44.4 | — | |
| Fed-lossTarget Model=Qwen2.5-14B2026.04 | 44.4 | — | |
| FTATarget Model=Llama3-8B2026.04 | 43.8 | — | |
| Score-RatioTarget Model=Qwen2.5-14B2026.04 | 42.8 | — |