Machine Unlearning on CIFAR-100 1.0 (test)
1ToWRetrain
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RetrainBackbone=ResNet-502024.06 | 1 | 52.044 | 99.966 | 73.26 | 0.635 | 0 | |
| NegGrad+ RUM (high → med → low)Backbone=ResNet-50, Strategy=Reverse Order2024.06 | 0.929 | 49.9 | 99.703 | 71.08 | 0.657 | 0.022 | |
| NegGrad+ RUMFBackbone=ResNet-50, Strategy=RUM (low -> med -> high)2024.06 | 0.921 | 50.789 | 98.413 | 68.82 | 0.576 | 0.059 | |
| L1-sparse RUM (high → med → low)Backbone=ResNet-50, Strategy=Reverse Order2024.06 | 0.908 | 53.967 | 98.732 | 66.963 | 0.591 | 0.044 | |
| L1-sparse RUMFBackbone=ResNet-50, Strategy=RUM (low -> med -> high)2024.06 | 0.883 | 52.444 | 97.269 | 64.317 | 0.602 | 0.033 | |
| Fine-tune RUMFBackbone=ResNet-50, Strategy=RUM (low -> med -> high)2024.06 | 0.784 | 61.767 | 96.706 | 63.09 | 0.542 | 0.093 |