Class Unlearning on CIFAR100 (Membership Inference & Unlearning Accuracy)
2.4MIA Success Rate (No Attack)retrain
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| retrainMIA Strategy=Shadow-model-based2025.04 | 2.4 | 75.4 | 0 | 0 | 1.4 | 0 | 1.4 | — | — | |
| SalUnMIA Strategy=Shadow-model-based2025.04 | 22.1 | 73.89 | 4.13 | 61.57 | 78.8 | 85.16 | 88.8 | — | — | |
| RLMIA Strategy=Shadow-model-based2025.04 | 26.9 | 69.96 | 3.2 | 51.53 | 68.8 | 80.5 | 79 | — | — | |
| FTMIA Strategy=Shadow-model-based2025.04 | 30.6 | 67.64 | 0.48 | 99.28 | 97.7 | 99.89 | 99.2 | — | — | |
| RetrainBackbone=ResNet342026.06 | 38 | — | — | — | — | — | — | 37 | 0 | |
| l1-sparseMIA Strategy=Shadow-model-based2025.04 | 40.2 | 70.7 | 1.3 | 99.77 | 92.5 | 99.91 | 94.5 | — | — | |
| FedUPBackbone=ResNet342026.06 | 67 | — | — | — | — | — | — | 37 | 0 | |
| IUMIA Strategy=Shadow-model-based2025.04 | 84.8 | 66.42 | 53.37 | 99.93 | 100 | 99.93 | 100 | — | — | |
| FUSEDBackbone=ResNet342026.06 | 91 | — | — | — | — | — | — | 36 | 0 | |
| OriginalMIA Strategy=Shadow-model-based2025.04 | 99.08 | 75.25 | 100 | — | — | — | — | — | — |