Image Classification on EMNIST
98.9AccuracyFLAME
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| FLAMEAttack=Adaptive-MD2025.12 | 98.9 | — | — | — | — | — | — | |
| ABBR FLAMEAttack=Adaptive-MD2025.12 | 98.9 | — | — | — | — | — | — | |
| FLAMEAttack=Adaptive-AT2025.12 | 98.9 | — | — | — | — | — | — | |
| ABBR FLAMEAttack=Adaptive-AT2025.12 | 98.9 | — | — | — | — | — | — | |
| IndividualBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=12024.05 | 94.67 | — | — | — | — | — | — | |
| COSMOSalpha=5, heterogeneity=Dirichlet, round=final2026.05 | 93.9 | — | — | — | — | — | — | |
| COSMOSdata heterogeneity=Dirichlet, alpha=100, performance type=Final-round2026.05 | 93.8 | — | — | — | — | — | — | |
| COMETdata heterogeneity=Dirichlet, alpha=100, performance type=Final-round2026.05 | 93.3 | — | — | — | — | — | — | |
| EMR-MERGINGBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=302024.05 | 92.03 | — | — | — | — | — | — | |
| Ours (mobile)Params (M)=2.25, Cost (Days)=0.942025.02 | 91.48 | — | — | — | — | — | — | |
| Ours (tiny)Params (M)=0.40, Cost (Days)=0.362025.02 | 91.2 | — | — | — | — | — | — | |
| FedCTdata heterogeneity=Dirichlet, alpha=100, performance type=Final-round2026.05 | 91.2 | — | — | — | — | — | — | |
| WaveMixParams (M)=2.42, Cost (Days)=manual2025.02 | 91.06 | — | — | — | — | — | — | |
| FedCTalpha=5, heterogeneity=Dirichlet, round=final2026.05 | 90.6 | — | — | — | — | — | — | |
| FedHB-Mixtures=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 88.97 | — | — | — | — | — | — | |
| FedHB-NIWs=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 88.84 | — | — | — | — | — | — | |
| FedPops=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 88.4 | — | — | — | — | — | — | |
| Fed-Proxs=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 88.39 | — | — | — | — | — | — | |
| FedBEs=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 88.37 | — | — | — | — | — | — | |
| FedEMs=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 88.32 | — | — | — | — | — | — | |
| pFedBayess=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 88.12 | — | — | — | — | — | — | |
| Fed-Avgs=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 87.92 | — | — | — | — | — | — | |
| COMETalpha=5, heterogeneity=Dirichlet, round=final2026.05 | 87 | — | — | — | — | — | — | |
| DFCA-GIFramework Type=DFL, Number of Clients (N)=1002025.10 | 85.7 | — | — | — | — | — | — | |
| IFCAFramework Type=CFL, Number of Clients (N)=1002025.10 | 85.7 | — | — | — | — | — | — | |
| FedHB-Mixtures=5, f=0.2, τ=12023.05 | 85.58 | — | — | — | — | — | — | |
| FedHB-NIWs=5, f=0.2, τ=12023.05 | 85.4 | — | — | — | — | — | — | |
| DFCA-LIFramework Type=DFL, Number of Clients (N)=1002025.10 | 85.3 | — | — | — | — | — | — | |
| FedMDdata heterogeneity=Dirichlet, alpha=100, performance type=Final-round2026.05 | 85.3 | — | — | — | — | — | — | |
| Fed-Avgs=5, f=0.2, τ=12023.05 | 85.27 | — | — | — | — | — | — | |
| Fed-Proxs=5, f=0.2, τ=12023.05 | 85.27 | — | — | — | — | — | — | |
| FedPops=5, f=0.2, τ=12023.05 | 85.27 | — | — | — | — | — | — | |
| FedBEs=5, f=0.2, τ=12023.05 | 85.24 | — | — | — | — | — | — | |
| FedEMs=5, f=0.2, τ=12023.05 | 85.21 | — | — | — | — | — | — | |
| DFCA-LIk=4, N=100, Setup=DFL2025.10 | 85.1 | — | — | — | — | — | — | |
| FedPAs=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 85.1 | — | — | — | — | — | — | |
| pFedBayess=5, f=0.2, τ=12023.05 | 84.65 | — | — | — | — | — | — | |
| Fed-BABUs=5, f=0.2, τ=12023.05 | 84.33 | — | — | — | — | — | — | |
| FedMDalpha=5, heterogeneity=Dirichlet, round=final2026.05 | 84.1 | — | — | — | — | — | — | |
| DFCA-LIFeature Skew Consistency=Inconsistent, alpha=0.9, N=200, Learning Paradigm=DFL2025.10 | 84 | — | — | — | — | — | — | |
| IFCAk=4, N=100, Setup=CFL2025.10 | 84 | — | — | — | — | — | — | |
| DFCA-LIFeature Skew Consistency=Consistent, alpha=0.9, N=200, Learning Paradigm=DFL2025.10 | 83.9 | — | — | — | — | — | — | |
| IFCAFeature Skew Consistency=Inconsistent, alpha=0.9, N=200, Learning Paradigm=CFL2025.10 | 83.8 | — | — | — | — | — | — | |
| DFCA-GIk=4, N=100, Setup=DFL2025.10 | 83.6 | — | — | — | — | — | — | |
| FedPAs=5, f=0.2, τ=12023.05 | 83.2 | — | — | — | — | — | — | |
| Fed-BABUs=5, f=0.2, tau=1, Evaluation Protocol=Personalisation2023.05 | 83.09 | — | — | — | — | — | — | |
| DFCA-LIFeature Skew Consistency=Consistent, alpha=0.7, N=200, Learning Paradigm=DFL2025.10 | 82.9 | — | — | — | — | — | — | |
| DFCA-LIFeature Skew Consistency=Consistent, alpha=0.8, N=200, Learning Paradigm=DFL2025.10 | 82.9 | — | — | — | — | — | — | |
| IFCAFeature Skew Consistency=Consistent, alpha=0.9, N=200, Learning Paradigm=CFL2025.10 | 82.9 | — | — | — | — | — | — | |
| DFCA-GIFeature Skew Consistency=Consistent, alpha=0.9, N=200, Learning Paradigm=DFL2025.10 | 82.8 | — | — | — | — | — | — | |
| IFCAFeature Skew Consistency=Consistent, alpha=0.7, N=200, Learning Paradigm=CFL2025.10 | 82.4 | — | — | — | — | — | — | |
| IFCAFeature Skew Consistency=Consistent, alpha=0.8, N=200, Learning Paradigm=CFL2025.10 | 82.4 | — | — | — | — | — | — | |
| IFCAFeature Skew Consistency=Inconsistent, alpha=0.7, N=200, Learning Paradigm=CFL2025.10 | 82.3 | — | — | — | — | — | — | |
| IFCAFeature Skew Consistency=Inconsistent, alpha=0.8, N=200, Learning Paradigm=CFL2025.10 | 82.3 | — | — | — | — | — | — | |
| DFCA-GIFeature Skew Consistency=Inconsistent, alpha=0.9, N=200, Learning Paradigm=DFL2025.10 | 82 | — | — | — | — | — | — | |
| ANNHidden Layers=200-200, Epochs=302018.05 | 81.77 | 81.77 | — | — | — | — | — | |
| DFCA-LIFeature Skew Consistency=Inconsistent, alpha=0.7, N=200, Learning Paradigm=DFL2025.10 | 81.5 | — | — | — | — | — | — | |
| DFCA-LIFeature Skew Consistency=Inconsistent, alpha=0.8, N=200, Learning Paradigm=DFL2025.10 | 81.5 | — | — | — | — | — | — | |
| DFCA-GIFeature Skew Consistency=Consistent, alpha=0.7, N=200, Learning Paradigm=DFL2025.10 | 80.5 | — | — | — | — | — | — | |
| DFCA-GIFeature Skew Consistency=Consistent, alpha=0.8, N=200, Learning Paradigm=DFL2025.10 | 80.5 | — | — | — | — | — | — | |
| DFCA-GIFeature Skew Consistency=Inconsistent, alpha=0.7, N=200, Learning Paradigm=DFL2025.10 | 79.8 | — | — | — | — | — | — | |
| DFCA-GIFeature Skew Consistency=Inconsistent, alpha=0.8, N=200, Learning Paradigm=DFL2025.10 | 79.8 | — | — | — | — | — | — | |
| FedSPDFramework Type=DFL, Number of Clients (N)=1002025.10 | 79.7 | — | — | — | — | — | — | |
| Spiking MLP (eRBP)Hidden Layers=200-200, Epochs=302018.05 | 78.17 | 78.17 | — | — | — | — | — | |
| PUMAPClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 77.5 | — | — | — | — | — | — | |
| UMAPClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 74.7 | — | — | — | — | — | — | |
| UDRNClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 73.5 | — | — | — | — | — | — | |
| DFedAvgMFramework Type=DFL, Number of Clients (N)=1002025.10 | 73.5 | — | — | — | — | — | — | |
| GRAEClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 72.4 | — | — | — | — | — | — | |
| tSNEClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 72.2 | — | — | — | — | — | — | |
| DFedAvgMFeature Skew Consistency=Inconsistent, alpha=0.7, N=200, Learning Paradigm=DFL2025.10 | 69.3 | — | — | — | — | — | — | |
| DFedAvgMFeature Skew Consistency=Inconsistent, alpha=0.8, N=200, Learning Paradigm=DFL2025.10 | 69.3 | — | — | — | — | — | — | |
| DFedAvgMFeature Skew Consistency=Inconsistent, alpha=0.9, N=200, Learning Paradigm=DFL2025.10 | 69.1 | — | — | — | — | — | — | |
| DFedAvgMFeature Skew Consistency=Consistent, alpha=0.9, N=200, Learning Paradigm=DFL2025.10 | 68.3 | — | — | — | — | — | — | |
| DFedAvgMFeature Skew Consistency=Consistent, alpha=0.7, N=200, Learning Paradigm=DFL2025.10 | 68.2 | — | — | — | — | — | — | |
| DFedAvgMFeature Skew Consistency=Consistent, alpha=0.8, N=200, Learning Paradigm=DFL2025.10 | 68.2 | — | — | — | — | — | — | |
| DLMEevaluation_protocol=linear SVM2022.07 | 65.7 | — | — | — | — | — | — | |
| UMAPevaluation_protocol=linear SVM2022.07 | 58.8 | — | — | — | — | — | — | |
| RegMeanBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=302024.05 | 48.67 | — | — | — | — | — | — | |
| tSNEevaluation_protocol=linear SVM2022.07 | 42 | — | — | — | — | — | — | |
| PHAevaluation_protocol=linear SVM2022.07 | 41.6 | — | — | — | — | — | — | |
| PUMevaluation_protocol=linear SVM2022.07 | 38.4 | — | — | — | — | — | — | |
| IVISClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 36.7 | — | — | — | — | — | — | |
| ivisevaluation_protocol=linear SVM2022.07 | 19 | — | — | — | — | — | — | |
| AdaMergingBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=302024.05 | 18.02 | — | — | — | — | — | — | |
| Task ArithmeticBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=302024.05 | 11.05 | — | — | — | — | — | — | |
| No DefenseAttack=Adaptive-MD2025.12 | 10 | — | — | — | — | — | — | |
| No DefenseAttack=Adaptive-AT2025.12 | 10 | — | — | — | — | — | — | |
| Weight AveragingBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=302024.05 | 7.73 | — | — | — | — | — | — | |
| Ties-MergingBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=302024.05 | 5.61 | — | — | — | — | — | — | |
| Adam2025.12 | — | — | — | — | — | — | 96.9 | |
| APFL2022.11 | — | — | 88.4 | 89.44 | — | — | — | |
| DITTO2022.11 | — | — | 89.08 | 91.3 | — | — | — | |
| FABAFramework=ABBR, Attack=DBA2025.12 | — | — | — | — | 10 | 98.7 | — | |
| FABAFramework=ABBR, Attack=Label Flipping2025.12 | — | — | — | — | — | 98.9 | — | |
| FEDAVG2022.11 | — | — | 90.14 | — | — | — | — | |
| FEDAVGFT2022.11 | — | — | 89.57 | 90.14 | — | — | — | |
| FEDREP2022.11 | — | — | 89.95 | 89.77 | — | — | — | |
| FLAMEFramework=ABBR, Attack=DBA2025.12 | — | — | — | — | 10 | 98.7 | — | |
| FLAMEFramework=ABBR, Attack=Label Flipping2025.12 | — | — | — | — | — | 98.9 | — |