Membership Inference Attack on WikiText-103 (test)
0.782AUCSAMA
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SAMATarget Model=LLaDA-8B-Base2026.01 | 0.782 | 0.415 | 0.108 | 0.018 | |
| RatioTarget Model=LLaDA-8B-Base2026.01 | 0.584 | 0.178 | 0.028 | 0.003 | |
| LowercaseTarget Model=LLaDA-8B-Base2026.01 | 0.536 | 0.136 | 0.017 | 0.002 | |
| SecMITarget Model=LLaDA-8B-Base2026.01 | 0.528 | 0.13 | 0.015 | 0.001 | |
| ZLIBTarget Model=LLaDA-8B-Base2026.01 | 0.525 | 0.128 | 0.015 | 0.002 | |
| PIATarget Model=LLaDA-8B-Base2026.01 | 0.524 | 0.127 | 0.014 | 0.001 | |
| Min-K%Target Model=LLaDA-8B-Base2026.01 | 0.522 | 0.125 | 0.016 | 0.001 | |
| ReCallTarget Model=LLaDA-8B-Base2026.01 | 0.521 | 0.124 | 0.014 | 0.001 | |
| Min-K%++Target Model=LLaDA-8B-Base2026.01 | 0.519 | 0.123 | 0.014 | 0.001 | |
| LossTarget Model=LLaDA-8B-Base2026.01 | 0.518 | 0.122 | 0.014 | 0.001 | |
| CON-RecallTarget Model=LLaDA-8B-Base2026.01 | 0.517 | 0.121 | 0.013 | 0.001 | |
| NeighborTarget Model=LLaDA-8B-Base2026.01 | 0.508 | 0.113 | 0.011 | 0.001 | |
| BoWsTarget Model=LLaDA-8B-Base2026.01 | 0.501 | 0.105 | 0.01 | 0 |