Image Classification on CIFAR100 (Accuracy & Communication Floats)
68.1Test AccuracyFedBCGD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FedBCGDBackbone=ResNet-18, Target Accuracy=54%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 68.1 | 277 | |
| FedBCGD+Backbone=ResNet-18, Target Accuracy=54%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 65.1 | 154 | |
| FedBCGD+Backbone=VGG-19, Target Accuracy=45%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 63.6 | 176 | |
| Centralised SGDBackbone=ResNet-18, Target Accuracy=54%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 62.2 | — | |
| FedBCGDBackbone=VGG-11, Target Accuracy=48%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 62.2 | 107 | |
| FedAvgMBackbone=ResNet-18, Target Accuracy=54%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 61.8 | 398 | |
| FedBCGDBackbone=VGG-19, Target Accuracy=45%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 61.1 | 206 | |
| Centralised SGDBackbone=VGG-19, Target Accuracy=45%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 58.9 | — | |
| FedAdamBackbone=VGG-19, Target Accuracy=45%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 58.7 | 367 | |
| FedBCGD+Backbone=VGG-11, Target Accuracy=48%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 58.7 | 105 | |
| SCAFFOLDBackbone=VGG-19, Target Accuracy=45%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 58.3 | 556 | |
| FedDCBackbone=VGG-19, Target Accuracy=45%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 56.8 | 321 | |
| Centralised SGDBackbone=VGG-11, Target Accuracy=48%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 56.3 | — | |
| FedAvgMBackbone=VGG-19, Target Accuracy=45%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 56 | 403 | |
| FedBCGDBackbone=LeNet-5, Target Accuracy=40%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 55.7 | 77 | |
| FedBCGD+Backbone=LeNet-5, Target Accuracy=40%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 55.6 | 75 | |
| FedAvgBackbone=ResNet-18, Target Accuracy=54%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 54.2 | 927 | |
| FedAdamBackbone=ResNet-18, Target Accuracy=54%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 53.9 | — | |
| Centralised SGDBackbone=LeNet-5, Target Accuracy=40%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 53.7 | — | |
| FedDCBackbone=LeNet-5, Target Accuracy=40%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 53.2 | 302 | |
| SCAFFOLDBackbone=ResNet-18, Target Accuracy=54%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 52.3 | — | |
| FedAvgMBackbone=VGG-11, Target Accuracy=48%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 51.7 | 299 | |
| FedAdamBackbone=VGG-11, Target Accuracy=48%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 50.9 | 597 | |
| SCAFFOLDBackbone=LeNet-5, Target Accuracy=40%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 50.3 | 214 | |
| FedAvgBackbone=VGG-11, Target Accuracy=48%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 48.7 | 720 | |
| FedAvgMBackbone=LeNet-5, Target Accuracy=40%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 48.2 | 277 | |
| FedDCBackbone=VGG-11, Target Accuracy=48%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 48.2 | 956 | |
| SCAFFOLDBackbone=VGG-11, Target Accuracy=48%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 47.9 | — | |
| FedAvgBackbone=VGG-19, Target Accuracy=45%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 47.6 | 735 | |
| FedDCBackbone=ResNet-18, Target Accuracy=54%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 46.6 | — | |
| FedAdamBackbone=LeNet-5, Target Accuracy=40%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 46.2 | 391 | |
| FedAvgBackbone=LeNet-5, Target Accuracy=40%, Heterogeneity (rho)=0.6, Number of blocks (N)=5, Total communication floats=1000d2026.03 | 41.2 | 558 |