Single-class Unlearning on Tiny ImageNet
100Accuracy ForgottenOriginal Model
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Original Model2025.03 | 100 | 99.98 | 66 | 64.15 | — | 99.4 | |
| Influence Unlearn2025.03 | 12.2 | 99.36 | 2 | 60.3 | 62.09 | 1.6 | |
| Boundary Expand2025.03 | 8.6 | 98.89 | 2 | 58.88 | 61.33 | 0 | |
| Negative Gradient2025.03 | 7.8 | 95.36 | 2 | 56.24 | 59.87 | 3 | |
| Saliency Unlearn2025.03 | 6.6 | 98.8 | 2 | 59.67 | 61.76 | 0.6 | |
| Bad Teacher2025.03 | 5.8 | 98.32 | 2 | 59.02 | 61.41 | 0.2 | |
| Boundary Shrink2025.03 | 5.2 | 97.49 | 2 | 57.85 | 60.77 | 0.6 | |
| Random Label2025.03 | 3.2 | 97.58 | 2 | 58.84 | 61.31 | 0.2 | |
| Learn to Unlearn2025.03 | 1 | 85.57 | 0 | 50.73 | 57.37 | 0.2 | |
| Ours2025.03 | 0.4 | 99.94 | 0 | 62.15 | 64.02 | 0 | |
| Retrain Model2025.03 | 0 | 99.98 | 0 | 64.42 | 65.2 | 0 |