Continual Learning on Rotated MNIST
99.66AccuracyModel Zoo
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Model ZooReplay=true, Single Epoch=false2021.06 | 99.66 | — | |
| Multi-Head (multi-task)2021.06 | 99.66 | — | |
| IsolatedReplay=false, Single Epoch=false2021.06 | 99.64 | — | |
| Joint trainingEpisodes=20, Backbone=Neural network with 2 hidden layers (256 neurons)2022.12 | 90.84 | 0 | |
| OGDReplay=false, Single Epoch=true2021.06 | 88.32 | — | |
| GEMReplay=true, Single Epoch=true2021.06 | 86.07 | — | |
| MC-SGDReplay=true, Single Epoch=true2021.06 | 82.63 | — | |
| Rebasin /w replay (Ours)Episodes=20, Backbone=Neural network with 2 hidden layers (256 neurons), Examples per class=5, Alpha=0.82022.12 | 78.14 | 0.12 | |
| Stable-SGDReplay=false, Single Epoch=true2021.06 | 70.8 | — | |
| A-GEMEpisodes=20, Backbone=Neural network with 2 hidden layers (256 neurons), Examples per class=52022.12 | 68.47 | 0.28 | |
| LwFEpisodes=20, Backbone=Neural network with 2 hidden layers (256 neurons)2022.12 | 61.86 | 0.29 | |
| EWCEpisodes=20, Backbone=Neural network with 2 hidden layers (256 neurons)2022.12 | 59.92 | 0.34 | |
| FinetuneEpisodes=20, Backbone=Neural network with 2 hidden layers (256 neurons)2022.12 | 46.28 | 0.52 |