Image Classification on CIFAR-10 (test) (Accuracy and Calibration)
95.67AccuracyProposed (RMD)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Proposed (RMD)Regularization=Relative Mahalanobis Distance2023.04 | 95.67 | 1.212 | |
| CRLRegularization=Correctness ranking loss2023.04 | 94.06 | 0.957 | |
| Linear ProbeInterpret.=false, Backbone=CLIP RN50, Training Setting=fully-supervised2026.01 | 88.8 | — | |
| VLG-CBMInterpret.=true, Backbone=CLIP RN50, Training Setting=fully-supervised2026.01 | 88.63 | — | |
| PCBM-ReDInterpret.=true, Backbone=CLIP RN50, Training Setting=fully-supervised2026.01 | 88.61 | — | |
| DN-CBMInterpret.=true, Backbone=CLIP RN50, Training Setting=fully-supervised2026.01 | 87.6 | — | |
| CDMInterpret.=true, Backbone=CLIP RN50, Training Setting=fully-supervised2026.01 | 86.5 | — | |
| Label-free CBMInterpret.=true, Backbone=CLIP RN50, Training Setting=fully-supervised2026.01 | 86.4 | — | |
| EDFIPC=500, Evaluation Protocol=Knowledge Distillation, Backbone=ConvNet-52024.10 | 86.1 | — | |
| EDFIPC=500, Evaluation Protocol=DD, Backbone=ConvNet-52024.10 | 84.8 | — | |
| PCBMInterpret.=true, Backbone=CLIP RN50, Training Setting=fully-supervised2026.01 | 84.5 | — | |
| DATMIPC=500, Evaluation Protocol=DD, Backbone=ConvNet-52024.10 | 83.5 | — | |
| OnDev-LCT-8/1β=5, num_clients=50, algorithm=FedAvg2024.01 | 82.42 | — | |
| OnDev-LCT-4/1β=5, num_clients=50, algorithm=FedAvg2024.01 | 80.87 | — | |
| OnDev-LCT-2/1β=5, num_clients=50, algorithm=FedAvg2024.01 | 80.55 | — | |
| OnDev-LCT-1/1β=5, num_clients=50, algorithm=FedAvg2024.01 | 79.71 | — | |
| OnDev-LCT-8/1β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 78.28 | — | |
| OnDev-LCT-2/1β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 78.2 | — | |
| OnDev-LCT-4/1β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 77.68 | — | |
| EDFIPC=50, Evaluation Protocol=DD, Backbone=ConvNet-52024.10 | 77.3 | — | |
| OnDev-LCT-1/1β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 76.93 | — | |
| ResNet-32β=5, num_clients=50, algorithm=FedAvg2024.01 | 76.17 | — | |
| ResNet-44β=5, num_clients=50, algorithm=FedAvg2024.01 | 76.1 | — | |
| DATMIPC=50, Evaluation Protocol=DD, Backbone=ConvNet-52024.10 | 76.1 | — | |
| ResNet-20β=5, num_clients=50, algorithm=FedAvg2024.01 | 74.84 | — | |
| CCT-4/2β=5, num_clients=50, algorithm=FedAvg2024.01 | 74.52 | — | |
| CCT-2/2β=5, num_clients=50, algorithm=FedAvg2024.01 | 73.45 | — | |
| ResNet-44β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 71.08 | — | |
| ResNet-32β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 70.5 | — | |
| CCT-4/2β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 69.82 | — | |
| ResNet-20β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 69.62 | — | |
| EDFIPC=50, Evaluation Protocol=Knowledge Distillation, Backbone=ConvNet-52024.10 | 69.5 | — | |
| CCT-2/2β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 68.87 | — | |
| OnDev-LCT-8/1β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 68.44 | — | |
| RDEDIPC=50, Evaluation Protocol=Knowledge Distillation, Backbone=ConvNet-52024.10 | 68.4 | — | |
| OnDev-LCT-2/1β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 68.25 | — | |
| OnDev-LCT-4/1β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 67.24 | — | |
| OnDev-LCT-1/1β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 66.18 | — | |
| MobileNetv2/0.5β=5, num_clients=50, algorithm=FedAvg2024.01 | 63.29 | — | |
| MobileNetv2/0.5β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 62.36 | — | |
| MobileNetv2/0.2β=5, num_clients=50, algorithm=FedAvg2024.01 | 60.84 | — | |
| ViT-Lite-2/8β=5, num_clients=50, algorithm=FedAvg2024.01 | 57.28 | — | |
| ViT-Lite-1/8β=5, num_clients=50, algorithm=FedAvg2024.01 | 57.05 | — | |
| CCT-4/2β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 56.05 | — | |
| MobileNetv2/0.2β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 55.97 | — | |
| ResNet-32β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 55.79 | — | |
| ResNet-44β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 55.78 | — | |
| CCT-2/2β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 55.67 | — | |
| ResNet-20β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 53.44 | — | |
| ViT-Lite-2/8β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 53.13 | — | |
| ViT-Lite-1/8β=0.5, num_clients=50, algorithm=FedAvg2024.01 | 52.94 | — | |
| EDFIPC=1, Evaluation Protocol=DD, Backbone=ConvNet-52024.10 | 47.8 | — | |
| DATMIPC=1, Evaluation Protocol=DD, Backbone=ConvNet-52024.10 | 46.9 | — | |
| MobileNetv2/0.5β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 46.59 | — | |
| MobileNetv2/0.2β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 44.3 | — | |
| ViT-Lite-2/8β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 36.85 | — | |
| ViT-Lite-1/8β=0.1, num_clients=50, algorithm=FedAvg2024.01 | 36.22 | — | |
| EDFIPC=1, Evaluation Protocol=Knowledge Distillation, Backbone=ConvNet-52024.10 | 28.1 | — | |
| RDEDIPC=1, Evaluation Protocol=Knowledge Distillation, Backbone=ConvNet-52024.10 | 23.5 | — |