Image Classification on CIFAR-10 (test) (Empirical Robustness)
83.57Clean AccuracyCAAT-LORA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CAAT-LORABackbone=Swin-B, Tuned params (M)=1.25, Tuned / Total (%)=1.442026.04 | 83.57 | 47.14 | 48.15 | 44.52 | |
| CAAT-ADAPTERBackbone=Swin-B, Tuned params (M)=1.15, Tuned / Total (%)=1.332026.04 | 82.89 | 46.62 | 47.72 | 43.95 | |
| FULLLORA-ATBackbone=Swin-B, Tuned params (M)=3.17, Tuned / Total (%)=3.662026.04 | 82.51 | 45.75 | 47.58 | 44.27 | |
| FULLBackbone=Swin-B, Tuned params (M)=86.69, Tuned / Total (%)=1002026.04 | 81.28 | 48.66 | 50.69 | 46.27 | |
| LoRA-16Backbone=Swin-B, Tuned params (M)=3.15, Tuned / Total (%)=3.632026.04 | 80.84 | 45.73 | 46.59 | 43.81 | |
| ADAPTER-32Backbone=Swin-B, Tuned params (M)=2.00, Tuned / Total (%)=2.312026.04 | 76.53 | 42.53 | 39.88 | 39.47 | |
| ADAPTER-8Backbone=Swin-B, Tuned params (M)=1.72, Tuned / Total (%)=1.982026.04 | 75.68 | 42.1 | 38.85 | 39.37 |