Continual Learning on TinyImageNet 25-split
58.28ACCAdam-NSCL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Adam-NSCLBackbone=ResNet-18, Requires storing previous data=false2021.03 | 58.28 | -6.05 | |
| MEGABackbone=ResNet-18, Requires storing previous data=true2021.03 | 57.12 | -5.9 | |
| LwFBackbone=ResNet-18, Requires storing previous data=false2021.03 | 56.57 | -11.19 | |
| A-GEMBackbone=ResNet-18, Requires storing previous data=true2021.03 | 53.32 | -7.68 | |
| EWCBackbone=ResNet-18, Requires storing previous data=false2021.03 | 52.33 | -6.17 | |
| OWMBackbone=ResNet-18, Requires storing previous data=false2021.03 | 49.98 | -3.64 | |
| MASBackbone=ResNet-18, Requires storing previous data=false2021.03 | 47.96 | -7.04 | |
| SIBackbone=ResNet-18, Requires storing previous data=false2021.03 | 45.27 | -4.45 | |
| GD-WILDBackbone=ResNet-18, Requires storing previous data=true2021.03 | 42.74 | -31.58 | |
| MUC-MASBackbone=ResNet-18, Requires storing previous data=false2021.03 | 41.18 | -4.03 | |
| Layerwise Optimization by Gradient DecompositionTraining Epochs=3 epochs2021.05 | 39.5 | — | |
| Layerwise Optimization by Gradient DecompositionTraining Epochs=1 epoch2021.05 | 38.3 | — | |
| GEMTraining Epochs=3 epochs, Memory Size=2562021.05 | 37.3 | — | |
| A-GEMTraining Epochs=3 epochs, Memory Size=2562021.05 | 36.9 | — | |
| tinyER+Training Epochs=1 epoch2021.05 | 36.5 | — | |
| S-GEMTraining Epochs=3 epochs, Memory Size=2562021.05 | 36.1 | — | |
| IndependentTraining Epochs=3 epochs2021.05 | 35.9 | — | |
| InstAParamBackbone=ResNet-18, Requires storing previous data=false2021.03 | 34.64 | -10.05 | |
| GEMTraining Epochs=1 epoch, Memory Size=2562021.05 | 34.2 | — | |
| S-GEMTraining Epochs=1 epoch, Memory Size=2562021.05 | 33.8 | — | |
| A-GEMTraining Epochs=1 epoch, Memory Size=2562021.05 | 33.5 | — | |
| GDumbTraining Epochs=1 epoch, Training Schedule=Online Task Incremental Disjoint2021.05 | 32.1 | — | |
| iCARLTraining Epochs=3 epochs2021.05 | 28.9 | — | |
| SingleTraining Epochs=3 epochs2021.05 | 28.7 | — | |
| IndependentTraining Epochs=1 epoch2021.05 | 26.9 | — | |
| iCARLTraining Epochs=1 epoch2021.05 | 26.5 | — | |
| SingleTraining Epochs=1 epoch2021.05 | 25.7 | — | |
| EWCTraining Epochs=3 epochs2021.05 | 25.4 | — | |
| EWCTraining Epochs=1 epoch2021.05 | 22.2 | — |