Class-incremental learning on Split CIFAR-10
81.6AccuracyLayerwise Optimization by Gradient Decomposition
Evaluation Results
| Method | Links | |
|---|---|---|
| Layerwise Optimization by Gradient DecompositionTraining Epochs=3 epochs2021.05 | 81.6 | |
| S-GEMTraining Epochs=3 epochs, Memory Size=2562021.05 | 81.5 | |
| A-GEMTraining Epochs=3 epochs, Memory Size=2562021.05 | 81 | |
| GEMTraining Epochs=3 epochs, Memory Size=2562021.05 | 80.8 | |
| Layerwise Optimization by Gradient DecompositionTraining Epochs=1 epoch2021.05 | 79 | |
| tinyER+Training Epochs=1 epoch2021.05 | 77.5 | |
| GEMTraining Epochs=1 epoch, Memory Size=2562021.05 | 76.8 | |
| S-GEMTraining Epochs=1 epoch, Memory Size=2562021.05 | 75.9 | |
| A-GEMTraining Epochs=1 epoch, Memory Size=2562021.05 | 75.8 | |
| GDumbTraining Epochs=1 epoch, Training Schedule=Online Task Incremental Disjoint2021.05 | 74.2 | |
| IndependentTraining Epochs=3 epochs2021.05 | 71.9 | |
| Conditional Channel Gated NetworksMemory type=generative2020.03 | 70.06 | |
| SingleTraining Epochs=3 epochs2021.05 | 68.4 | |
| EWCTraining Epochs=3 epochs2021.05 | 67.8 | |
| iCARLTraining Epochs=3 epochs2021.05 | 66.5 | |
| iCARLTraining Epochs=1 epoch2021.05 | 63.9 | |
| ER-ACE + InfluenceCLtraining epochs=502023.04 | 60.57 | |
| EWCTraining Epochs=1 epoch2021.05 | 59.8 | |
| SingleTraining Epochs=1 epoch2021.05 | 57.7 | |
| ER-ACEtraining epochs=502023.04 | 56.86 | |
| DGMwMemory type=generative2020.03 | 56.21 | |
| InfluenceCLtraining epochs=502023.04 | 52.81 | |
| DGMaMemory type=generative2020.03 | 51.75 | |
| BiCtraining epochs=502023.04 | 51.49 | |
| iCaRLtraining epochs=502023.04 | 47.87 | |
| IndependentTraining Epochs=1 epoch2021.05 | 43.3 |