Class Unlearning on Lacuna-10
1.67Test ErrorRetrain
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| RetrainBackbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 1.67 | 100 | 0 | 55.67 | 0.09 | 0 | 0 | 2.62 | |
| FinetuneBackbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 1.67 | 0 | 0 | 56.33 | 0.27 | 0 | 0 | 1.25 | |
| OriginalBackbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 1.7 | 0 | 0 | 58 | 0.21 | 0 | 0 | 1.63 | |
| SCRUBBackbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 1.96 | 100 | 0 | 50.33 | 0.34 | 0 | 0 | 2.62 | |
| SCRUB+RBackbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 1.96 | 100 | 0 | 50.33 | 0.34 | 0 | 0 | 2.62 | |
| NegGrad+Backbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 2 | 14.27 | 0 | 54.33 | 0 | 0 | 0.74 | 2.05 | |
| CF-kBackbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 2.07 | 0 | 0 | 52.33 | 0.14 | 0 | 0 | 2.05 | |
| EU-kArchitecture=ResNet, Application=User Privacy (UP)2023.02 | 2.48 | 7.71 | 0 | 54.33 | — | — | — | — | |
| RetrainArchitecture=ResNet, Application=User Privacy (UP)2023.02 | 2.52 | 100 | 0 | 55 | — | — | — | — | |
| Bad-TBackbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 2.56 | 38.65 | 0.37 | 63.33 | 0.09 | 0.18 | 36.8 | 2.49 | |
| NegGrad+Architecture=ResNet, Application=User Privacy (UP)2023.02 | 2.74 | 9.48 | 0 | 53.67 | — | — | — | — | |
| OriginalArchitecture=ResNet, Application=User Privacy (UP)2023.02 | 2.81 | 0 | 0 | 56 | — | — | — | — | |
| CF-kArchitecture=ResNet, Application=User Privacy (UP)2023.02 | 2.81 | 0 | 0 | 56 | — | — | — | — | |
| RetrainBackbone=ResNet, Application=Removing Biases (RB)2023.02 | 2.87 | 99.75 | 0 | — | 0.34 | 0 | 0.56 | — | |
| CF-kBackbone=ResNet, Application=Removing Biases (RB)2023.02 | 2.89 | 0 | 0 | — | 0.22 | 0 | 0 | — | |
| FinetuneBackbone=ResNet, Application=Removing Biases (RB)2023.02 | 3.02 | 0 | 0 | — | 0.37 | 0 | 0 | — | |
| FinetuneArchitecture=ResNet, Application=User Privacy (UP)2023.02 | 3.04 | 0 | 0 | 54.67 | — | — | — | — | |
| OriginalBackbone=ResNet, Application=Removing Biases (RB)2023.02 | 3.07 | 0 | 0 | — | 0.26 | 0 | 0 | — | |
| EU-kBackbone=ResNet, Application=Removing Biases (RB)2023.02 | 3.19 | 4.06 | 0.01 | — | 0.17 | 0.02 | 0.83 | — | |
| SCRUBArchitecture=ResNet, Application=User Privacy (UP)2023.02 | 3.26 | 99.9 | 0.15 | 54.33 | — | — | — | — | |
| SCRUB+RArchitecture=ResNet, Application=User Privacy (UP)2023.02 | 3.26 | 99.9 | 0.15 | 54.33 | — | — | — | — | |
| FisherBackbone=ResNet, Application=Removing Biases (RB)2023.02 | 3.33 | 100 | 0 | — | 0.54 | 0 | 0 | — | |
| Bad-TArchitecture=ResNet, Application=User Privacy (UP)2023.02 | 3.37 | 67.6 | 1.06 | 58 | — | — | — | — | |
| Bad-TBackbone=ResNet, Application=Removing Biases (RB)2023.02 | 3.37 | 67.6 | 1.06 | — | 0.5 | 0.47 | 24.26 | — | |
| NegGrad+Backbone=ResNet, Application=Removing Biases (RB)2023.02 | 3.41 | 14.9 | 0 | — | 0.17 | 0 | 1.78 | — | |
| SCRUBBackbone=ResNet, Application=Removing Biases (RB)2023.02 | 3.69 | 100 | 0.28 | — | 0.36 | 0.23 | 0 | — | |
| EU-kBackbone=All-CNN, Unlearning Type=Class Unlearning, Application Context=User Privacy (UP)2023.02 | 4.15 | 62.08 | 0.81 | 52.67 | 1.22 | 0.53 | 44.26 | 3.68 |