Image Classification on CIFAR-10 (test) (Dirichlet & Pathological Accuracy)
88.85Accuracy (Dirichlet α=0.1)DFedPGP
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DFedPGPBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD, Local training epochs (shared part)=5, Local training epochs (personal part)=1, PFL Layer Partitioning=classifier as personal2024.05 | 88.85 | 85.61 | 91.26 | 87.12 | |
| FedRepBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD, PFL Layer Partitioning=classifier as personal2024.05 | 88.78 | 84.5 | 91.09 | 86.22 | |
| FedPerBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD, PFL Layer Partitioning=classifier as personal2024.05 | 88.57 | 84.06 | 90.94 | 86.97 | |
| FedBABUBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD, PFL Layer Partitioning=classifier as personal2024.05 | 87.79 | 83.26 | 91.28 | 83.9 | |
| Dis-PFLBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD2024.05 | 87.77 | 82.71 | 88.19 | 84.18 | |
| OSGPBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD2024.05 | 87.39 | 83.14 | 90.72 | 84.69 | |
| DFedAvgMBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD2024.05 | 86.94 | 82.49 | 90.23 | 85.26 | |
| FedAvgBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Client Sampling Ratio=0.1, Batch Size=128, Optimizer=SGD2024.05 | 84.17 | 79.66 | 85.04 | 81.18 | |
| DittoBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD2024.05 | 80.22 | 73.51 | 84.96 | 75.59 | |
| LocalBackbone=ResNet-18, Normalization=Group Normalization, Communication Rounds=500, Number of Clients=100, Batch Size=128, Optimizer=SGD2024.05 | 78.96 | 63.2 | 85.16 | 68.56 |