Continual Learning on CIFAR-10 Split
80.55Average AccuracyAdaptive Training
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Adaptive TrainingTraining Mode=Adaptive2026.02 | 80.55 | 19.51 | — | |
| Adaptive TrainingTraining Mode=Adaptive, Framework=Adaptive Hybrid Update Framework2026.02 | 80.55 | 19.51 | — | |
| NestedCapsule Oracletask identification=Oracle2026.03 | 79.5 | — | — | |
| NestedCapsule OracleContext Source=One-hot task ID2026.03 | 79.5 | 0 | — | |
| NestedCapsule Gradient Ctxcontext signal=gradient context2026.03 | 78.5 | — | — | |
| NestedCapsule + GradCtxContext Source=∇θL + routing2026.03 | 78.5 | 0 | 78.5 | |
| HyperNet Oracletask identification=One-hot task ID2026.03 | 77.7 | — | — | |
| HyperNet OracleContext Source=One-hot task ID2026.03 | 77.7 | 0 | 77 | |
| HyperNet Gradient Contextcontext signal=gradient context2026.03 | 77 | — | 77 | |
| HyperNet + GradCtxContext Source=∇θL2026.03 | 77 | 0 | 77 | |
| Frozen TrainingTraining Mode=Frozen2026.02 | 75.98 | 21.1 | — | |
| Frozen TrainingTraining Mode=Frozen2026.02 | 75.98 | 21.1 | — | |
| Initialized TrainingTraining Mode=Initialized2026.02 | 75.74 | 23.3 | — | |
| Initialized TrainingTraining Mode=Initialized2026.02 | 75.74 | 23.3 | — | |
| HyperNet Learned (batch)context signal=batch statistics, cosine similarity=> 0.9952026.03 | 54.4 | — | — | |
| HyperNet Learned (batch)Context Source=Batch pixel mean/var2026.03 | 54.4 | — | 0.5 | |
| SGD + Replaymode=Catastrophic forgetting2026.03 | 46 | — | 0 |