Continual Learning on CIFAR-100 (Avg, Last, Throughput)
69.48Avg AccuracyMEMO + C-Flat Turbo
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MEMO + C-Flat TurboBackbone=ResNet-34, Training Protocol=trained from scratch2026.04 | 69.48 | 59.33 | 1,372.5 | |
| MEMO + C-FlatBackbone=ResNet-34, Training Protocol=trained from scratch2026.04 | 69 | 59.29 | 569.1 | |
| MEMOBackbone=ResNet-34, Training Protocol=trained from scratch2026.04 | 68.49 | 57.05 | 1,873.2 | |
| iCaRL + C-Flat TurboBackbone=ResNet-18, Training Protocol=trained from scratch2026.04 | 59.84 | 42.84 | 1,750.1 | |
| iCaRL + C-Flat TurboBackbone=ResNet-34, Training Protocol=trained from scratch2026.04 | 59.75 | 42.34 | 960.6 | |
| iCaRL + C-FlatBackbone=ResNet-34, Training Protocol=trained from scratch2026.04 | 59.55 | 42.09 | 359.8 | |
| iCaRL + C-FlatBackbone=ResNet-18, Training Protocol=trained from scratch2026.04 | 59.45 | 42.47 | 686.3 | |
| iCaRLBackbone=ResNet-18, Training Protocol=trained from scratch2026.04 | 59.13 | 41.23 | 2,333.3 | |
| iCaRLBackbone=ResNet-34, Training Protocol=trained from scratch2026.04 | 58.8 | 41.26 | 1,250.4 | |
| MEMO + C-Flat TurboBackbone=ResNet-18, Training Protocol=trained from scratch2026.04 | 50.51 | 32.24 | 1,891.9 | |
| MEMO + C-FlatBackbone=ResNet-18, Training Protocol=trained from scratch2026.04 | 49.98 | 30.76 | 886.1 | |
| MEMOBackbone=ResNet-18, Training Protocol=trained from scratch2026.04 | 48.63 | 29.19 | 2,413.8 |