Continual Learning on CIFAR-100 (10-split)
92.35ACCDualPrompt+Soft-TF
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DualPrompt+Soft-TF2024.11 | 92.35 | — | 2.98 | — | — | — | |
| UBsetting=10-task2024.03 | 91.99 | — | — | — | — | — | |
| CPromptsetting=10-task2024.03 | 87.82 | — | — | 92.53 | 5.06 | — | |
| DualPromptsetting=10-task2024.03 | 86.61 | — | — | 90.82 | 5.86 | — | |
| DualPrompt2024.11 | 86.61 | — | 5.86 | — | — | — | |
| ESNsetting=10-task2024.03 | 86.42 | — | — | 91.65 | 6.08 | — | |
| ESN2024.11 | 86.42 | — | 6.08 | — | — | — | |
| L2Psetting=10-task2024.03 | 86.38 | — | — | 91.45 | 5.88 | — | |
| CODApromptsetting=10-task2024.03 | 85.73 | — | — | 91.03 | 7.13 | — | |
| CODA-Prompt2024.11 | 85.73 | — | 7.13 | — | — | — | |
| L2P2024.11 | 83.38 | — | 5.88 | — | — | — | |
| MultitaskNetwork=multi-head2022.11 | 79.58 | — | — | — | — | — | |
| MultitaskNetwork=multi-head, Runs=5-run average2022.11 | 79.58 | — | — | — | — | — | |
| CPrompt2024.11 | 77.14 | — | 5.97 | — | — | — | |
| CUBERNetwork=multi-head2022.11 | 75.54 | 0.13 | — | — | — | — | |
| CUBERNetwork=multi-head, Runs=5-run average, epsilon_2=0.02022.11 | 75.54 | 0.13 | — | — | — | — | |
| TRGPNetwork=multi-head2022.11 | 74.46 | -0.9 | — | — | — | — | |
| TRGPNetwork=multi-head, Runs=5-run average2022.11 | 74.46 | -0.9 | — | — | — | — | |
| Adam-NSCLBackbone=ResNet-18, Store previous data=false2021.03 | 73.77 | -1.6 | — | — | — | — | |
| HATNetwork=multi-head2022.11 | 73.21 | 0 | — | — | — | — | |
| GPMNetwork=multi-head2022.11 | 72.48 | -0.9 | — | — | — | — | |
| GPMNetwork=multi-head, Runs=5-run average2022.11 | 72.48 | -0.9 | — | — | — | — | |
| ER_ResNetwork=multi-head2022.11 | 72.24 | -7 | — | — | — | — | |
| HATNetwork=multi-head, Runs=5-run average2022.11 | 72.06 | 0 | — | — | — | — | |
| ER ResNetwork=multi-head, Runs=5-run average2022.11 | 71.73 | -6 | — | — | — | — | |
| GD-WILDBackbone=ResNet-18, Store previous data=true2021.03 | 71.27 | -18.24 | — | — | — | — | |
| EWCBackbone=ResNet-18, Store previous data=false2021.03 | 70.77 | -2.83 | — | — | — | — | |
| LwFBackbone=ResNet-18, Store previous data=false2021.03 | 70.7 | -6.27 | — | — | — | — | |
| OWMBackbone=ResNet-18, Store previous data=false2021.03 | 68.89 | -1.88 | — | — | — | — | |
| EWCNetwork=multi-head, Runs=5-run average2022.11 | 68.8 | -2 | — | — | — | — | |
| EWCNetwork=multi-head2022.11 | 68.3 | -2 | — | — | — | — | |
| MASBackbone=ResNet-18, Store previous data=false2021.03 | 66.93 | -4.03 | — | — | — | — | |
| A-GEMNetwork=multi-head2022.11 | 64.16 | -14 | — | — | — | — | |
| A-GEMNetwork=multi-head, Runs=5-run average2022.11 | 63.98 | -15 | — | — | — | — | |
| MUC-MASBackbone=ResNet-18, Store previous data=false2021.03 | 63.73 | -3.38 | — | — | — | — | |
| SIBackbone=ResNet-18, Store previous data=false2021.03 | 60.57 | -5.17 | — | — | — | — | |
| Joint trainingEpisodes=20, Backbone=Multi-head ResNet-182022.12 | 60.48 | — | 0 | — | — | — | |
| MEGABackbone=ResNet-18, Store previous data=true2021.03 | 54.17 | -2.19 | — | — | — | — | |
| Rebasin /w replay (Ours)Episodes=20, Backbone=Multi-head ResNet-18, Examples per class=5, Alpha=0.82022.12 | 51.34 | — | 7 | — | — | — | |
| OWMNetwork=multi-head, Runs=5-run average2022.11 | 50.94 | -30 | — | — | — | — | |
| EWCEpisodes=20, Backbone=Multi-head ResNet-182022.12 | 50.5 | — | 24 | — | — | — | |
| OWMNetwork=multi-head2022.11 | 50.44 | -30 | — | — | — | — | |
| A-GEMBackbone=ResNet-18, Store previous data=true2021.03 | 49.57 | -1.13 | — | — | — | — | |
| GEMBackbone=ResNet-18, Store previous data=true2021.03 | 49.48 | 2.77 | — | — | — | — | |
| InstAParamBackbone=ResNet-18, Store previous data=false2021.03 | 47.84 | -11.92 | — | — | — | — | |
| A-GEMEpisodes=20, Backbone=Multi-head ResNet-18, Examples per class=52022.12 | 44.42 | — | 36 | — | — | — | |
| LwFEpisodes=20, Backbone=Multi-head ResNet-182022.12 | 41.43 | — | 51 | — | — | — | |
| FinetuneEpisodes=20, Backbone=Multi-head ResNet-182022.12 | 35.41 | — | 49 | — | — | — | |
| KAN-CL+bbEWCBackbone regularization=bbEWC, Classification head=KAN-CL2026.05 | 27 | — | 1 | — | — | — | |
| EWCType=Standard global EWC2026.05 | 24 | — | 3 | — | — | — | |
| SIType=Synaptic intelligence2026.05 | 22 | — | 8 | — | — | — | |
| MLP+bbEWCBackbone=SmallCNNBackbone, Classification head=MLP, Backbone regularization=bbEWC2026.05 | 20 | — | 37 | — | — | — | |
| KAN-CL (head only)Backbone regularization=None, Classification head=KAN-CL2026.05 | 15 | — | 17 | — | — | — | |
| FinetuneRegularization=None2026.05 | 14 | — | 21 | — | — | — |