Membership Inference Attack on Clinicalnotes fine-tuned (test)
70.7TPR@FPR=0.1Ratio
Evaluation results
| Method | Links | |
|---|---|---|
| RatioTarget Model=LLaMA-3.2-1B2026.01 | 70.7 | |
| HT-MIATarget Model=LLaMA-3.2-1B2026.01 | 70.57 | |
| HT-MIATarget Model=Qwen-3-0.6B2026.01 | 68.61 | |
| RatioTarget Model=Qwen-3-0.6B2026.01 | 64.27 | |
| LossTarget Model=LLaMA-3.2-1B2026.01 | 58.95 | |
| ZlibTarget Model=LLaMA-3.2-1B2026.01 | 53.7 | |
| Min-K++Target Model=LLaMA-3.2-1B2026.01 | 52.01 | |
| LossTarget Model=Qwen-3-0.6B2026.01 | 51.39 | |
| PACTarget Model=LLaMA-3.2-1B2026.01 | 51.13 | |
| PACTarget Model=Qwen-3-0.6B2026.01 | 48.48 | |
| SPV-MIATarget Model=Qwen-3-0.6B2026.01 | 48.2 | |
| ZlibTarget Model=Qwen-3-0.6B2026.01 | 45.81 | |
| LowercaseTarget Model=LLaMA-3.2-1B2026.01 | 43.69 | |
| LowercaseTarget Model=Qwen-3-0.6B2026.01 | 39.73 | |
| Min-K++Target Model=Qwen-3-0.6B2026.01 | 37.54 | |
| HT-MIATarget Model=GPT-22026.01 | 32.23 | |
| RatioTarget Model=GPT-22026.01 | 20.75 | |
| Min-K++Target Model=GPT-22026.01 | 20.27 | |
| LossTarget Model=GPT-22026.01 | 19.97 | |
| SPV-MIATarget Model=GPT-22026.01 | 17.3 | |
| ZlibTarget Model=GPT-22026.01 | 16.93 | |
| PACTarget Model=GPT-22026.01 | 15.04 | |
| LowercaseTarget Model=GPT-22026.01 | 13.93 | |
| SPV-MIATarget Model=LLaMA-3.2-1B2026.01 | 11.2 |