Machine Unlearning on ImageNet-1K 1.0 (train test)
80.25Acc (Train, Retained)retrained
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| retrainedBackbone=ResNet50, Access to Drm=true, Access to Dfg=false2026.01 | 80.25 | 0 | 75.72 | 0 | 0 | — | — | |
| LOFTBackbone=ResNet50, Access to Drm=only Sigma_pre_rm, Access to Dfg=only Sigma_pre_fg2026.01 | 79.96 | 0 | 69.4 | 0 | 1.65 | 0.94 | — | |
| pretrainedBackbone=ResNet50, Access to Drm=true, Access to Dfg=true2026.01 | 77.92 | 83.46 | 76.15 | 81.33 | — | — | — | |
| DELETEBackbone=ResNet50, Access to Drm=false, Access to Dfg=true2026.01 | 72.46 | 0.95 | 71.9 | 2 | 3.64 | 17.82 | — | |
| RLBackbone=ResNet50, Access to Drm=false, Access to Dfg=true2026.01 | 72.11 | 1.87 | 71.53 | 5.33 | 4.88 | 14.4 | — | |
| BTBackbone=ResNet50, Access to Drm=false, Access to Dfg=true2026.01 | 72.07 | 1.87 | 71.49 | 5.33 | 4.9 | 16.84 | — | |
| GABackbone=ResNet50, Access to Drm=false, Access to Dfg=true2026.01 | 71.87 | 0 | 71.42 | 0 | 3.17 | 9.32 | — | |
| SalUnBackbone=ResNet50, Access to Drm=false, Access to Dfg=true2026.01 | 71.56 | 2.67 | 70.84 | 4.67 | 5.23 | 27.73 | — | |
| COUN+RLBackbone=ResNet50, Access to Drm=false, Access to Dfg=true2026.01 | 70.04 | 1.36 | 69.72 | 4.67 | 5.56 | 15.24 | — |