Image Classification on CIFAR-10 40% flips Standard
66.23AccuracyREFINE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| REFINEBackbone=CNNs2025.05 | 66.23 | 0.9388 | 0.6625 | 43.94 | |
| AdapterBackbone=CNNs2025.05 | 65.78 | 0.9376 | 0.6581 | 45.2 | |
| LinearProbeBackbone=CNNs2025.05 | 65.54 | 0.9378 | 0.6561 | 42.82 | |
| LoRABackbone=CNNs2025.05 | 65.47 | 0.9374 | 0.6545 | 42.38 | |
| DANN-GateBackbone=CNNs2025.05 | 65.4 | 0.9353 | 0.6539 | 43.4 | |
| DistillBackbone=CNNs2025.05 | 57.01 | 0.9115 | 0.5674 | 34.84 | |
| NoTransBackbone=CNNs2025.05 | 56.05 | 0.9037 | 0.558 | 32.4 |