Machine Unlearning on TOFU (Metrics: Agg., Mem., Priv., Utility)
85Aggregation MetricRank DD
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Rank DDk=202026.05 | 85 | 80 | 81 | 0.95 | |
| RetrainConfig=Retain902026.05 | 78 | 53 | 98 | 1.03 | |
| Linear DDα=1.52026.05 | 78 | 56 | 95 | 1 | |
| Distill DDlr=4e-05, T=1.52026.05 | 73 | 51 | 89 | 0.95 | |
| NPOlr=4e-6, e=22026.05 | 67 | 57 | 68 | 0.82 | |
| RMUlr=8e-7, e=42026.05 | 67 | 60 | 74 | 0.69 | |
| ECOlr=1e-5, str=502026.05 | 65 | 69 | 66 | 0.62 | |
| GradDiff/GA-GDRlr=2e-6, e=32026.05 | 64 | 52 | 62 | 0.86 | |
| GradAscentlr=2e-6, e=32026.05 | 63 | 51 | 61 | 0.87 | |
| UNDIALlr=4e-6, e=102026.05 | 60 | 57 | 43 | 1.07 | |
| SimNPOlr=3e-6, e=32026.05 | 59 | 48 | 54 | 0.9 | |
| WHPlr=1e-5, α=3.02026.05 | 56 | 43 | 51 | 0.94 | |
| ULDlr=2e-32026.05 | 40 | 40 | 39 | 0.42 | |
| δ-Unlearninglr=7e-62026.05 | 28 | 78 | 69 | 0.13 | |
| TargetConfig=Full2026.05 | 2 | 1 | 38 | 1 | |
| LUNARlr=0.00012026.05 | 2 | 1 | 38 | 1 | |
| GUARDlr=1e-3, δ=0.32026.05 | 2 | 1 | 38 | 0.86 |