Machine Unlearning on CIFAR-10 SUBCLASS (deer) (forget/retain)
0KL Divergence (Test)pθr
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| pθrBackbone=dinov2-mlp12026.05 | 0 | — | 0 | 99.6 | 98 | 100 | |
| LDA-MixBackbone=dinov2-mlp12026.05 | 0.01 | 0.01 | 0.11 | 99.7 | 100 | 5.1 | |
| LDA-2CBackbone=dinov2-mlp12026.05 | 0.01 | 0.01 | 0.08 | 99.7 | 99.9 | 4.7 | |
| SCRUBBackbone=dinov2-mlp12026.05 | 0.01 | 2.3 | 0.11 | 99.6 | 99 | 6.2 | |
| FTBackbone=dinov2-mlp12026.05 | 0.01 | 0.01 | 0.13 | 99.7 | 99.8 | 17 | |
| pθBackbone=dinov2-mlp12026.05 | 0.02 | — | 0.22 | 99.7 | 100 | — | |
| LDABackbone=dinov2-mlp12026.05 | 0.02 | 0.02 | 0.18 | 99.6 | 100 | 4.8 | |
| DIR-2CBackbone=dinov2-mlp12026.05 | 0.02 | 0.02 | 0.15 | 99.6 | 99.2 | 4.7 | |
| GA+FTBackbone=dinov2-mlp12026.05 | 0.16 | 0.16 | 1.7 | 98 | 81.4 | 18 | |
| DIRBackbone=dinov2-mlp12026.05 | 0.65 | 0.75 | 6.6 | 90.5 | 10.3 | 4.6 | |
| RL+FTBackbone=dinov2-mlp12026.05 | 0.66 | 0.76 | 6.7 | 90.5 | 10.4 | 4.1 | |
| SalUnBackbone=dinov2-mlp12026.05 | 1 | 1 | 10 | 88.4 | 1 | 4.4 | |
| GABackbone=dinov2-mlp12026.05 | 13 | 1,338 | 46 | 45.5 | 0 | 2 |