Machine Unlearning on CIFAR-100 ResNet18 (test)
77.69Retained Accuracy (Acc_r)Original Model
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Original ModelBackbone=ResNet182025.11 | 77.69 | 92 | 52 | 0.6 | 0.78 | 0.68 | 62 | |
| POUR-PBackbone=ResNet18, Data Requirement=Methods on forget set only, Modify encoder representations=False2025.11 | 77.65 | 0 | 100 | — | — | — | 62 | |
| FCSBackbone=ResNet18, Data Requirement=Methods requiring the retain set2025.11 | 76.81 | 2 | 97 | 0.61 | 0.78 | 0.68 | 55 | |
| FinetuneBackbone=ResNet18, Data Requirement=Methods requiring the retain set2025.11 | 76.32 | 0 | 99 | 0.57 | 0.76 | 0.67 | 54 | |
| Retrained ModelBackbone=ResNet182025.11 | 76.28 | 0 | 100 | 1 | 1 | 1 | — | |
| POUR-DBackbone=ResNet18, Data Requirement=Methods on forget set only2025.11 | 73.44 | 1 | 95 | 0.57 | 0.76 | 0.65 | 46 | |
| Boundary ShrinkBackbone=ResNet18, Data Requirement=Methods on forget set only2025.11 | 68.87 | 4 | 88 | 0.55 | 0.72 | 0.62 | 49 | |
| Boundary ExpandBackbone=ResNet18, Data Requirement=Methods on forget set only2025.11 | 66.47 | 13 | 79 | 0.55 | 0.74 | 0.63 | 42 | |
| DELETEBackbone=ResNet18, Data Requirement=Methods on forget set only2025.11 | 64.67 | 8 | 81 | 0.51 | 0.67 | 0.58 | 60 | |
| Random LabelBackbone=ResNet18, Data Requirement=Methods on forget set only2025.11 | 61.98 | 11 | 76 | 0.4 | 0.53 | 0.46 | 49 | |
| Gradient AscentBackbone=ResNet18, Data Requirement=Methods on forget set only2025.11 | 50.46 | 6 | 69 | 0.44 | 0.44 | 0.44 | 50 |