Black-box Membership Inference Attack on CIFAR-10 Most-memorized 1% forget samples
0.91AUCPGU (base)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PGU (base)Unlearning Algorithm=PGU, Defense Strategy=Base2025.11 | 0.91 | 0.201 | 0.511 | 0.706 | |
| Salun (base)Unlearning Algorithm=Salun, Defense Strategy=Base2025.11 | 0.91 | 0.129 | 0.321 | 0.52 | |
| BT (base)Unlearning Algorithm=BT, Defense Strategy=Base2025.11 | 0.902 | 0.119 | 0.295 | 0.582 | |
| PGU + WARPUnlearning Algorithm=PGU, Defense Strategy=WARP2025.11 | 0.875 | 0.16 | 0.431 | 0.663 | |
| BT + WARPUnlearning Algorithm=BT, Defense Strategy=WARP2025.11 | 0.865 | 0.113 | 0.275 | 0.537 | |
| Salun + WARPUnlearning Algorithm=Salun, Defense Strategy=WARP2025.11 | 0.826 | 0.107 | 0.264 | 0.487 | |
| SCRUB (base)Unlearning Algorithm=SCRUB, Defense Strategy=Base2025.11 | 0.71 | 0.086 | 0.227 | 0.397 | |
| NGP (base)Unlearning Algorithm=NGP, Defense Strategy=Base2025.11 | 0.649 | 0.058 | 0.157 | 0.277 | |
| SCRUB + WARPUnlearning Algorithm=SCRUB, Defense Strategy=WARP2025.11 | 0.61 | 0.041 | 0.119 | 0.213 | |
| NGP + WARPUnlearning Algorithm=NGP, Defense Strategy=WARP2025.11 | 0.598 | 0.015 | 0.082 | 0.206 | |
| SF (base)Unlearning Algorithm=SF, Defense Strategy=Base2025.11 | 0.518 | 0.089 | 0.034 | 0.079 | |
| SF + WARPUnlearning Algorithm=SF, Defense Strategy=WARP2025.11 | 0.501 | 0.006 | 0.026 | 0.068 |