Machine Unlearning on CIFAR-100 (50% forget set)
0.3875Average GapRandom Labeling
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Random LabelingBackbone Architecture=ResNet182025.03 | 0.3875 | — | — | — | |
| SalUnBackbone=Vision Transformer2025.03 | 0.33 | — | 1.49 | 0 | |
| FinetuningBackbone Architecture=ResNet182025.03 | 0.32 | — | — | — | |
| NegGrad+Backbone Architecture=ResNet182025.03 | 0.315 | — | — | — | |
| Bad TeacherBackbone=Vision Transformer2025.03 | 0.305 | — | 0.58 | 0.0001 | |
| UNSIRBackbone=Vision Transformer2025.03 | 0.2925 | — | 6.25 | 0.0001 | |
| SCRUBBackbone=Vision Transformer2025.03 | 0.2225 | — | 0.62 | 0 | |
| SSDBackbone=Vision Transformer2025.03 | 0.2225 | — | 1.29 | 0 | |
| LoTUSBackbone Architecture=ResNet182025.03 | 0.1725 | — | — | — | |
| LoTUSBackbone=Vision Transformer2025.03 | 0.1725 | — | 0.57 | 0 | |
| NegGrad+Backbone Architecture=ViT2025.03 | 0.06 | — | — | — | |
| Bad TeacherBackbone=ResNet182025.03 | 0.0575 | — | 15.04 | 0 | |
| FinetuningBackbone Architecture=ViT2025.03 | 0.04 | — | — | — | |
| UNSIRBackbone=ResNet182025.03 | 0.0375 | — | 18.33 | 0 | |
| SalUnBackbone=ResNet182025.03 | 0.0375 | — | 38.08 | 0 | |
| SCRUBBackbone=ResNet182025.03 | 0.035 | — | 16.82 | 0 | |
| SSDBackbone=ResNet182025.03 | 0.035 | — | 18.69 | 0 | |
| Random LabelingBackbone Architecture=ViT2025.03 | 0.025 | — | — | — | |
| LoTUSBackbone Architecture=ViT2025.03 | 0.0225 | — | — | — | |
| LoTUSBackbone=ResNet182025.03 | 0.0225 | — | 13.79 | 0 |