Image Classification on CIFAR-10 (Accuracy, Parameter Reduction)
97.92AccuracyLoRA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LoRABackbone=ViT-B/162026.04 | 97.92 | 860.28 | |
| CDWFBackbone=ViT-B/16, Parameter Budget (fmax)=0.052026.04 | 96.81 | 859.35 | |
| Full Fine-tuning (FL)Backbone=ViT-B/162026.04 | 96.33 | 1 | |
| Full Fine-tuning (FL)Backbone=ResNet-1012026.04 | 86.94 | 1 | |
| Full Fine-tuning (FL)Backbone=ResNet-182026.04 | 85.63 | 1 | |
| CDWFBackbone=ResNet-101, Parameter Budget (fmax)=0.052026.04 | 84.79 | 53.03 | |
| CDWFBackbone=ResNet-18, Parameter Budget (fmax)=0.052026.04 | 83.19 | 24.34 | |
| LoRABackbone=ResNet-1012026.04 | 81.74 | 65.12 | |
| LoRABackbone=ResNet-182026.04 | 77.43 | 21.14 |