Machine Unlearning on Tiny-ImageNet 20 classes (forget)
66.08AccuracyOriginal
Evaluation results
| Method | Links | |
|---|---|---|
| OriginalEvaluation Protocol=Linear Probe2026.04 | 66.08 | |
| OriginalEvaluation Protocol=Output2026.04 | 66.02 | |
| OriginalEvaluation Protocol=NCC2026.04 | 65 | |
| Scrub with CMFEvaluation Protocol=Linear Probe2026.04 | 54.28 | |
| NegGrad+ with CMFEvaluation Protocol=Linear Probe2026.04 | 53.82 | |
| NegGrad+ with CMFEvaluation Protocol=NCC2026.04 | 51.3 | |
| Scrub with CMFEvaluation Protocol=NCC2026.04 | 50.42 | |
| Scrub with CMFEvaluation Protocol=Output2026.04 | 47.76 | |
| Salun with CMFEvaluation Protocol=Linear Probe2026.04 | 46.14 | |
| Retain-only RetrainEvaluation Protocol=NCC2026.04 | 44.3 | |
| NegGrad+ with CMFEvaluation Protocol=Output2026.04 | 41.42 | |
| Salun with CMFEvaluation Protocol=NCC2026.04 | 38.1 | |
| Random-label with CMFEvaluation Protocol=Linear Probe2026.04 | 36.58 | |
| Salun with CMFEvaluation Protocol=Output2026.04 | 32.74 | |
| Retain-only RetrainEvaluation Protocol=Linear Probe2026.04 | 30.36 | |
| Random-label with CMFEvaluation Protocol=NCC2026.04 | 28.24 | |
| UNSIR with CMFEvaluation Protocol=Linear Probe2026.04 | 23.02 | |
| Random-label with CMFEvaluation Protocol=Output2026.04 | 22.2 | |
| UNSIR with CMFEvaluation Protocol=NCC2026.04 | 16.72 | |
| UNSIR with CMFEvaluation Protocol=Output2026.04 | 14.44 | |
| Retain-only RetrainEvaluation Protocol=Output2026.04 | 0 |