Membership Inference Attack on Clinicalnotes (test)
0.8843AUCHT-MIA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HT-MIATarget Model=Qwen-3-0.6B, Fine-tuned=true2026.01 | 0.8843 | 68.61 | 38.39 | |
| RatioTarget Model=Qwen-3-0.6B, Fine-tuned=true2026.01 | 0.8681 | 64.27 | 28.42 | |
| HT-MIATarget Model=LLaMA-3.2-1B, Fine-tuned=true2026.01 | 0.8659 | 70.57 | 42.53 | |
| RatioTarget Model=LLaMA-3.2-1B, Fine-tuned=true2026.01 | 0.8642 | 70.7 | 38.83 | |
| SPV-MIATarget Model=Qwen-3-0.6B, Fine-tuned=true2026.01 | 0.8359 | 48.2 | 13.3 | |
| LossTarget Model=LLaMA-3.2-1B, Fine-tuned=true2026.01 | 0.8212 | 58.95 | 31.48 | |
| LossTarget Model=Qwen-3-0.6B, Fine-tuned=true2026.01 | 0.81 | 51.39 | 26.4 | |
| PACTarget Model=Qwen-3-0.6B, Fine-tuned=true2026.01 | 0.8001 | 48.48 | 22.51 | |
| Min-K++Target Model=LLaMA-3.2-1B, Fine-tuned=true2026.01 | 0.8 | 52.01 | 21.68 | |
| ZlibTarget Model=LLaMA-3.2-1B, Fine-tuned=true2026.01 | 0.7998 | 53.7 | 26.63 | |
| PACTarget Model=LLaMA-3.2-1B, Fine-tuned=true2026.01 | 0.7982 | 51.13 | 21.43 | |
| ZlibTarget Model=Qwen-3-0.6B, Fine-tuned=true2026.01 | 0.7838 | 45.81 | 24.07 | |
| Min-K++Target Model=Qwen-3-0.6B, Fine-tuned=true2026.01 | 0.7649 | 37.54 | 5.36 | |
| LowercaseTarget Model=LLaMA-3.2-1B, Fine-tuned=true2026.01 | 0.7511 | 43.69 | 19.9 | |
| LowercaseTarget Model=Qwen-3-0.6B, Fine-tuned=true2026.01 | 0.7332 | 39.73 | 18.91 | |
| HT-MIATarget Model=GPT-2, Fine-tuned=true2026.01 | 0.6978 | 32.23 | 5.24 | |
| RatioTarget Model=GPT-2, Fine-tuned=true2026.01 | 0.644 | 20.75 | 2.01 | |
| Min-K++Target Model=GPT-2, Fine-tuned=true2026.01 | 0.6336 | 20.27 | 2.79 | |
| LossTarget Model=GPT-2, Fine-tuned=true2026.01 | 0.6238 | 19.97 | 2.23 | |
| SPV-MIATarget Model=GPT-2, Fine-tuned=true2026.01 | 0.6233 | 17.3 | 1.5 | |
| ZlibTarget Model=GPT-2, Fine-tuned=true2026.01 | 0.5954 | 16.93 | 1.74 | |
| PACTarget Model=GPT-2, Fine-tuned=true2026.01 | 0.5886 | 15.04 | 1.35 | |
| LowercaseTarget Model=GPT-2, Fine-tuned=true2026.01 | 0.5651 | 13.93 | 1.7 | |
| SPV-MIATarget Model=LLaMA-3.2-1B, Fine-tuned=true2026.01 | 0.5266 | 11.2 | 1.2 |