Multi-label classification on CIFAR-10 multi-label 28 (beta=0.5, gamma=0.5)
92.2macro AUCCentralized
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Centralizedbeta (Dirichlet concentration parameter)=0.5, gamma (class-presence ratio)=0.5, Backbone=ResNet-18, Communication rounds=100, Local epochs=1, Batch size=32, Optimizer=AdamW2025.09 | 92.2 | 61.3 | 90.04 | 62.02 | |
| FedNCA-MLbeta (Dirichlet concentration parameter)=0.5, gamma (class-presence ratio)=0.5, Backbone=ResNet-18, Communication rounds=100, Local epochs=1, Batch size=32, Optimizer=AdamW2025.09 | 87.55 | 48.17 | 87 | 48.61 | |
| SphereFedbeta (Dirichlet concentration parameter)=0.5, gamma (class-presence ratio)=0.5, Backbone=ResNet-18, Communication rounds=100, Local epochs=1, Batch size=32, Optimizer=AdamW2025.09 | 83.63 | 42.58 | 83.5 | 43.18 | |
| FedLGTbeta (Dirichlet concentration parameter)=0.5, gamma (class-presence ratio)=0.5, Backbone=ResNet-18, Communication rounds=100, Local epochs=1, Batch size=32, Optimizer=AdamW2025.09 | 83.52 | 43.6 | 83.36 | 44.03 | |
| FedCurvbeta (Dirichlet concentration parameter)=0.5, gamma (class-presence ratio)=0.5, Backbone=ResNet-18, Communication rounds=100, Local epochs=1, Batch size=32, Optimizer=AdamW2025.09 | 82.53 | 39.96 | 82.1 | 40.34 | |
| SCAFFOLDbeta (Dirichlet concentration parameter)=0.5, gamma (class-presence ratio)=0.5, Backbone=ResNet-18, Communication rounds=100, Local epochs=1, Batch size=32, Optimizer=AdamW2025.09 | 82.51 | 39.98 | 82.08 | 40.26 | |
| FedProxbeta (Dirichlet concentration parameter)=0.5, gamma (class-presence ratio)=0.5, Backbone=ResNet-18, Communication rounds=100, Local epochs=1, Batch size=32, Optimizer=AdamW2025.09 | 82.45 | 39.22 | 81.77 | 39.66 | |
| FedAvgbeta (Dirichlet concentration parameter)=0.5, gamma (class-presence ratio)=0.5, Backbone=ResNet-18, Communication rounds=100, Local epochs=1, Batch size=32, Optimizer=AdamW2025.09 | 82.29 | 39.47 | 81.48 | 40.26 |