Machine Unlearning on Tiny-ImageNet 200 classes (train test)
99.92Acctr (Residual)FT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FTDrm=true, Dfg=false2026.01 | 99.92 | 18.73 | 74.25 | 16.17 | 3.73 | 8.02 | |
| RLDrm=true, Dfg=true2026.01 | 99.91 | 1.52 | 74.43 | 0.33 | 0 | 0.62 | |
| SalUnDrm=true, Dfg=true2026.01 | 99.88 | 2.57 | 74.71 | 1 | 0 | 0.9 | |
| retrainedDrm=true, Dfg=false2026.01 | 99.67 | 0 | 75.44 | 0 | 0 | 0 | |
| pretrainedDrm=true, Dfg=true2026.01 | 99.63 | 99.75 | 74.54 | 74.83 | 96.63 | — | |
| DELETEDrm=true, Dfg=false2026.01 | 99.13 | 13.32 | 73.31 | 6.5 | 1.85 | 4.88 | |
| LOFTDrm=only Σ_rm^pre, Dfg=only Σ_fg^pre2026.01 | 98.59 | 1.48 | 74.27 | 0.5 | 0 | 0.85 | |
| SalUnDrm=false, Dfg=true2026.01 | 93.16 | 13.55 | 67.49 | 10 | 3.45 | 8.29 | |
| RLDrm=false, Dfg=true2026.01 | 92.94 | 9.98 | 67.17 | 7.5 | 1.83 | 6.85 | |
| L2ULDrm=false, Dfg=true2026.01 | 92.69 | 4.43 | 67.69 | 2 | 2.57 | 4.75 | |
| COUN+RLDrm=false, Dfg=true2026.01 | 92.56 | 10.07 | 67.85 | 5.67 | 2.28 | 6.54 | |
| BTDrm=false, Dfg=true2026.01 | 91.7 | 7.72 | 66.35 | 5.5 | 1.4 | 6.34 | |
| GADrm=false, Dfg=true2026.01 | 88.4 | 23.12 | 65.27 | 18.82 | 19.95 | 16.67 |