Continual Learning on UCIT (Avg)
89.69ImageNet-R AccuracyOctopus
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| OctopusMethod Category=Regularization-based, Replay=No, Historical task proxy approximation=true2026.05 | 89.69 | 91.34 | 61.04 | 60.18 | 68.09 | 57.61 | 71.33 | |
| Multi-taskReplay=No2026.05 | 89.53 | 92.2 | 62.7 | 61.86 | 70.4 | 58.49 | 72.53 | |
| Vanilla RehearsalReplay=Yes2026.05 | 88.71 | 86.13 | 54.15 | 61.76 | 70.53 | 58.13 | 69.9 | |
| OctopusMethod Category=Regularization-based, Replay=No2026.05 | 88.41 | 93.3 | 61.65 | 59.91 | 66.32 | 56.87 | 71.08 | |
| HiDe-LLaVAMethod Category=Architecture-based, Replay=No2026.05 | 85.7 | 92.7 | 54.1 | 66.87 | 59.12 | 55.15 | 68.94 | |
| O-LoRAMethod Category=Regularization-based, Replay=No2026.05 | 75.26 | 86.73 | 55.86 | 58.47 | 57.38 | 53.52 | 64.54 | |
| MoELoRAMethod Category=Architecture-based, Replay=No2026.05 | 64.49 | 82.42 | 49.54 | 56.87 | 56.35 | 58.34 | 61.33 | |
| LwFMethod Category=Regularization-based, Replay=No2026.05 | 55.6 | 79.86 | 53.23 | 54.87 | 56.51 | 56.34 | 59.4 | |
| EWCMethod Category=Regularization-based, Replay=No2026.05 | 54.23 | 80.13 | 53.14 | 55.06 | 57.52 | 55.94 | 59.34 | |
| Sequential FinetuneReplay=No2026.05 | 49.31 | 78.4 | 50.48 | 53.44 | 55.53 | 57.95 | 57.52 | |
| L2PMethod Category=Architecture-based, Replay=No2026.05 | 41.52 | 82.32 | 51.98 | 52.21 | 43.16 | 52.77 | 53.99 | |
| Zero-shotReplay=No2026.05 | 17.2 | 52.16 | 41.07 | 18.7 | 18.63 | 46.99 | 32.46 |