Image Classification on CIFAR-10.1
98Top-1 AccWiSE-FT
Evaluation Results
| Method | Links | |
|---|---|---|
| WiSE-FTalpha=optimal, fine-tuning strategy=end-to-end2021.09 | 98 | |
| WiSE-FTalpha=0.5, fine-tuning strategy=end-to-end2021.09 | 97.6 | |
| CLIPmode=fine-tuned, fine-tuning strategy=end-to-end2021.09 | 95.9 | |
| WATT-PBackbone=ViT-L/142024.06 | 92.97 | |
| CLIPmode=zero-shot2021.09 | 92.5 | |
| TENTBackbone=ViT-L/142024.06 | 92.22 | |
| WATT-SBackbone=ViT-L/142024.06 | 92.1 | |
| TPTBackbone=ViT-L/14, Batch Size=322024.06 | 91.32 | |
| CLIPBackbone=ViT-L/142024.06 | 91.2 | |
| CLIPArTTBackbone=ViT-L/142024.06 | 91.02 | |
| CLIPArTTBackbone=ViT-B/162024.06 | 88.72 | |
| TENTBackbone=ViT-B/162024.06 | 88.52 | |
| WATT-SBackbone=ViT-B/162024.06 | 88.1 | |
| WATT-PBackbone=ViT-B/162024.06 | 87.9 | |
| WATT-PMechanism=Parallel MTWA2024.06 | 87.78 | |
| TENT2024.06 | 87.6 | |
| WATT-SMechanism=Sequential MTWA2024.06 | 86.98 | |
| CLIPARTT2024.06 | 86.35 | |
| CLIPBackbone=ViT-B/162024.06 | 84 | |
| TPTBackbone=ViT-B/16, Batch Size=322024.06 | 83.75 | |
| CLIP2024.06 | 83.25 | |
| TPTBatch Size=322024.06 | 81.8 | |
| PERADAAlgorithm Type=pFL, Trained Params=Adapter, Evaluation Protocol=Global Model, Knowledge Distillation=True2023.02 | 62.5 | |
| FEDDF (w/ KD)Algorithm Type=generic FL, Trained Params=Full, Evaluation Protocol=Global Model, Knowledge Distillation=True2023.02 | 60.95 | |
| PERADA W/o KDAlgorithm Type=pFL, Trained Params=Adapter, Evaluation Protocol=Global Model, Knowledge Distillation=False2023.02 | 57.6 | |
| FEDDYNAlgorithm Type=generic FL, Trained Params=Full, Evaluation Protocol=Global Model2023.02 | 56.3 | |
| DITTOAlgorithm Type=pFL, Trained Params=Full, Evaluation Protocol=Global Model2023.02 | 55.25 | |
| FEDAVGAlgorithm Type=generic FL, Trained Params=Full, Evaluation Protocol=Global Model2023.02 | 54.95 | |
| FEDPROXAlgorithm Type=generic FL, Trained Params=Full, Evaluation Protocol=Global Model2023.02 | 54.75 | |
| APFLAlgorithm Type=pFL, Trained Params=Full, Evaluation Protocol=Global Model2023.02 | 53.6 | |
| PFEDMEAlgorithm Type=pFL, Trained Params=Full, Evaluation Protocol=Global Model2023.02 | 52.55 | |
| PERADAPersonalized Params=Adapter2023.02 | 47.25 | |
| DITTOPersonalized Params=Full model2023.02 | 42.72 |