Federated Unlearning Evaluation on CIFAR-10 (MIA, Utility, PPS, CFI)
54.3MIA AccuracyKNOT-adapt.
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| KNOT-adapt.Backbone=FedRep2026.06 | 54.3 | 80.6 | 88.6 | 85.2 | |
| Mimir-adapt.Backbone=FedRep, partial personalization-aware components=true2026.06 | 53.7 | 80.1 | 89.4 | 86.1 | |
| FedRecovery-adapt.Backbone=FedRep2026.06 | 52.5 | 79.3 | 87.3 | 83.8 | |
| FUSED-adapt.Backbone=FedRep, partial personalization-aware components=true2026.06 | 52.4 | 81.3 | 90.5 | 87.1 | |
| ZeroFU-adapt.Backbone=FedRep, partial personalization-aware components=true2026.06 | 52.1 | 81.8 | 91.2 | 87.9 | |
| pFedULBackbone=FedRep2026.06 | 51.1 | 84.2 | 97.5 | 96.3 | |
| PGA-adapt.Backbone=FedRep2026.06 | 50.4 | 74.1 | 83.8 | 79.5 | |
| RetrainBackbone=FedRep2026.06 | 50.2 | 84.9 | 100 | 100 |