Task-incremental learning on CIFAR-100/20 (test)
91.62Avg Top-1 AccuracyUpper Bound
Evaluation results
| Method | Links | |
|---|---|---|
| Upper BoundArchitecture=ResNet-18/32021.03 | 91.62 | |
| EFTArchitecture=ResNet-18/3, a=8, b=0, parameter_growth=+6.7%2021.03 | 90.17 | |
| EFTArchitecture=ResNet-18/3, a=4, b=0, parameter_growth=+5.2%2021.03 | 89.25 | |
| SupSup (GG) TransferArchitecture=ResNet-18/32021.03 | 88.52 | |
| SupSup (GG)Architecture=ResNet-18/32021.03 | 86.45 | |
| BatchE (GG)Architecture=ResNet-18/32021.03 | 79.75 | |
| Separate HeadsArchitecture=ResNet-18/32021.03 | 70.6 |