Image Classification on CIFAR-100 non-IID (test)
72.73Test Accuracy (Avg Best)FedCorr
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FedCorrp=0.7, alpha_Dir=10, Noise setting (rho, tau)=(0.4, 0), Number of trials=52022.04 | 72.73 | 1.02 | |
| FedCovaNoise Level ρ=0.4, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 67.63 | 0.0104 | |
| FedCovaNoise Level ρ=0.4, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 66.84 | 0.0073 | |
| FedProxp=0.7, alpha_Dir=10, Noise setting (rho, tau)=(0.4, 0), Number of trials=52022.04 | 65.72 | 1.3 | |
| FedAvgp=0.7, alpha_Dir=10, Noise setting (rho, tau)=(0.4, 0), Number of trials=52022.04 | 64.75 | 1.75 | |
| FedNedNoise Level ρ=0.4, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 64.67 | 0.004 | |
| FedNoRoNoise Level ρ=0.4, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 62.2 | 0.0047 | |
| FedNedNoise Level ρ=0.4, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 61.53 | 0.0209 | |
| FedCovaNoise Level ρ=0.6, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 60.67 | 0.0111 | |
| GPSLBackbone=ResNet-34, Severe non-IID setting (C, α)=C=2, α=3.0, Global batch size (B)=128, Number of participating clients (K)=642024.07 | 60.04 | — | |
| CLBackbone=ResNet-34, Severe non-IID setting (C, α)=C=2, α=3.0, Global batch size (B)=128, Number of participating clients (K)=642024.07 | 59.86 | — | |
| JointOptp=0.7, alpha_Dir=10, Noise setting (rho, tau)=(0.4, 0), Number of trials=52022.04 | 59.84 | 1.99 | |
| FedCorrNoise Level ρ=0.4, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 59.66 | 0.0093 | |
| RoFLp=0.7, alpha_Dir=10, Noise setting (rho, tau)=(0.4, 0), Number of trials=52022.04 | 59.31 | 4.14 | |
| FedNoRoNoise Level ρ=0.4, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 58.57 | 0.005 | |
| FedCovaNoise Level ρ=0.6, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 58.46 | 0.0158 | |
| FedAvgNoise Level ρ=0.4, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 57.7 | 0.0295 | |
| FedNed-Noise Level ρ=0.4, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 57.56 | 0.0074 | |
| FedNedNoise Level ρ=0.6, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 56.63 | 0.0045 | |
| SLTBackbone=DenseNet40, SFD/FedRolex constraint=1.02023.05 | 55.8 | — | |
| FedCovaNoise Level ρ=0.8, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 55.71 | 0.0082 | |
| FedCorrNoise Level ρ=0.4, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 55.6 | 0.0045 | |
| FedNed-Noise Level ρ=0.4, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 55.35 | 0.0038 | |
| FedNedNoise Level ρ=0.6, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 54.84 | 0.006 | |
| FedCorrNoise Level ρ=0.6, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 53.81 | 0.006 | |
| FedNed-Noise Level ρ=0.6, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 51.83 | 0.0086 | |
| Small modelBackbone=DenseNet40, SFD/FedRolex constraint=0.662023.05 | 51.8 | — | |
| FedAvgNoise Level ρ=0.4, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 51.21 | 0.0147 | |
| RoFLNoise Level ρ=0.4, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 50.71 | 0.0057 | |
| FedCorrNoise Level ρ=0.6, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 50.01 | 0.0098 | |
| FedCorrNoise Level ρ=0.8, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 49.68 | 0.0046 | |
| FedNedNoise Level ρ=0.8, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 49.51 | 0.0095 | |
| FedAvgModel=ResNet442025.12 | 49.46 | — | |
| RoFLNoise Level ρ=0.4, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 48.8 | 0.0016 | |
| SLTBackbone=DenseNet40, SFD/FedRolex constraint=0.662023.05 | 48.4 | — | |
| FedOLFModel=ResNet44, TOA setting=no TOA2025.12 | 48.15 | — | |
| ARFLp=0.7, alpha_Dir=10, Noise setting (rho, tau)=(0.4, 0), Number of trials=52022.04 | 48.03 | 4.39 | |
| FedNed-Noise Level ρ=0.6, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 46.37 | 0.0063 | |
| FedAvgModel=ResNet202025.12 | 46.01 | — | |
| SLTBackbone=DenseNet40, SFD/FedRolex constraint=0.332023.05 | 45.9 | — | |
| RoFLNoise Level ρ=0.6, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 45.58 | 0.0017 | |
| SLTModel=ResNet442025.12 | 43.73 | — | |
| FedNed-Noise Level ρ=0.8, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 43.44 | 0.0129 | |
| FedRolexBackbone=DenseNet40, SFD/FedRolex constraint=0.662023.05 | 42.9 | — | |
| FedNoRoNoise Level ρ=0.6, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 42.86 | 0.0016 | |
| HO-SFLOptimization Order=Hybrid2026.03 | 42.5 | — | |
| FedAvgNoise Level ρ=0.6, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 41.65 | 0.0432 | |
| FLSBackbone=ResNet-34, Severe non-IID setting (C, α)=C=2, α=3.0, Global batch size (B)=128, Number of participating clients (K)=642024.07 | 41.61 | — | |
| FedCovaNoise Level ρ=0.8, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 40.8 | 0.0078 | |
| FedOLFModel=ResNet44, TOA setting=TOA (s = 0.75)2025.12 | 40.72 | — | |
| NeFLModel=ResNet442025.12 | 40.54 | — | |
| Small modelBackbone=DenseNet40, SFD/FedRolex constraint=0.332023.05 | 40.5 | — | |
| DivideMixp=0.7, alpha_Dir=10, Noise setting (rho, tau)=(0.4, 0), Number of trials=52022.04 | 39.76 | 1.18 | |
| SFLOptimization Order=First Order2026.03 | 39.5 | — | |
| FedAvgNoise Level ρ=0.8, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 38.49 | 0.0237 | |
| RoFLNoise Level ρ=0.6, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 37.89 | 0.0028 | |
| FedOLFModel=ResNet20, TOA setting=no TOA2025.12 | 37.85 | — | |
| DepthFLModel=ResNet442025.12 | 37.82 | — | |
| FDBackbone=DenseNet40, SFD/FedRolex constraint=0.662023.05 | 36.9 | — | |
| FedOLFModel=ResNet20, TOA setting=TOA (s = 0.75)2025.12 | 36.04 | — | |
| FedNedNoise Level ρ=0.8, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 36.04 | 0.0033 | |
| ScaleFLModel=ResNet442025.12 | 35.51 | — | |
| RI-FedAvgNumber of Clients=10, Communication Rounds=1502026.02 | 33.73 | — | |
| FedNed-Noise Level ρ=0.8, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 32.65 | 0.006 | |
| NeFLModel=ResNet202025.12 | 32.04 | — | |
| FedAvgNoise Level ρ=0.6, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 31.58 | 0.0638 | |
| SCAFFOLDNumber of Clients=10, Communication Rounds=1502026.02 | 30.51 | — | |
| FedOLFModel=ResNet44, TOA setting=TOA (s = 0.5)2025.12 | 29.68 | — | |
| RI-FedAvgArchitecture=Architecture 1, Communication Rounds=502026.02 | 29.5 | — | |
| RoFLNoise Level ρ=0.8, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 29.4 | 0.0062 | |
| FedCorrNoise Level ρ=0.8, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 29.04 | 0.0061 | |
| FedDynNumber of Clients=10, Communication Rounds=1502026.02 | 28.58 | — | |
| FedProxNumber of Clients=10, Communication Rounds=1502026.02 | 27.94 | — | |
| CoCoFLModel=ResNet442025.12 | 27.56 | — | |
| FedAvgNumber of Clients=10, Communication Rounds=1502026.02 | 26.91 | — | |
| FedNoRoNoise Level ρ=0.6, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 26.7 | 0.0108 | |
| AdaptiveFLModel=ResNet442025.12 | 25.58 | — | |
| SLTModel=ResNet202025.12 | 25.04 | — | |
| FedOLFModel=ResNet20, TOA setting=TOA (s = 0.5)2025.12 | 24.93 | — | |
| DepthFLModel=ResNet202025.12 | 24.87 | — | |
| FedProxArchitecture=Architecture 1, Communication Rounds=502026.02 | 24.8 | — | |
| FedNoRoNoise Level ρ=0.8, Noise Level τ=0.5, Noise Type=Symmetric noise2026.03 | 24.43 | 0.0031 | |
| RI-FedAvgNumber of Clients=100, Communication Rounds=1502026.02 | 24.15 | — | |
| SCAFFOLDNumber of Clients=100, Communication Rounds=1502026.02 | 22.41 | — | |
| CoCoFLModel=ResNet202025.12 | 22.16 | — | |
| RI-FedAvgArchitecture=Architecture 2, Communication Rounds=502026.02 | 22 | — | |
| FedProxArchitecture=Architecture 2, Communication Rounds=502026.02 | 21 | — | |
| FedAvgArchitecture=Architecture 1, Communication Rounds=502026.02 | 21 | — | |
| DP-FedAvgNumber of Clients=10, Communication Rounds=1502026.02 | 20.75 | — | |
| AdaptiveFLModel=ResNet202025.12 | 20.51 | — | |
| FedRolexBackbone=DenseNet40, SFD/FedRolex constraint=0.332023.05 | 20 | — | |
| FedAvgArchitecture=Architecture 2, Communication Rounds=502026.02 | 19 | — | |
| FedAvgNoise Level ρ=0.8, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 18.27 | 0.0524 | |
| FeddropModel=ResNet202025.12 | 17.02 | — | |
| FedDynNumber of Clients=100, Communication Rounds=1502026.02 | 16.17 | — | |
| DP-FedAvgArchitecture=Architecture 1, Communication Rounds=502026.02 | 16 | — | |
| FjORDModel=ResNet442025.12 | 14.68 | — | |
| FedNoRoNoise Level ρ=0.8, Noise Level τ=0.7, Noise Type=Symmetric noise2026.03 | 14.13 | 0.0026 | |
| DP-FedAvgArchitecture=Architecture 2, Communication Rounds=502026.02 | 14 | — | |
| ScaleFLModel=ResNet202025.12 | 13.73 | — |