Image Classification on MNIST original (test)
99.72AccuracyResNet101
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ResNet1012025.08 | 99.72 | — | |
| ResNet182025.08 | 99.7 | — | |
| DEF-CNNInput size=32 x 322026.02 | 99.67 | — | |
| BaselineInput Size=28 × 282024.04 | 99.57 | — | |
| CNNInput size=32 x 322026.02 | 99.55 | — | |
| Baseline2025.08 | 99.55 | — | |
| ICPNet2025.08 | 99.49 | — | |
| ConvNeXt-B2025.08 | 99.45 | — | |
| ORNInput size=32 x 322026.02 | 99.42 | — | |
| ORNInput Size=32 × 322024.04 | 99.42 | — | |
| VGG162025.08 | 99.41 | — | |
| MambaOut-B2025.08 | 99.37 | — | |
| LBP-CNNInput Size=28 × 282024.04 | 99.36 | — | |
| LBP-CNNInput Size=28 × 282024.04 | 99.34 | — | |
| MAX-CNNInput Size=28 × 282024.04 | 99.31 | — | |
| G-CNNInput size=28 x 282026.02 | 99.27 | — | |
| G-CNNInput Size=28 × 282024.04 | 99.27 | — | |
| RotEqNetInput size=28 x 282026.02 | 99.26 | — | |
| RotEqNetInput Size=28 × 282024.04 | 99.26 | — | |
| H-NetInput size=32 x 322026.02 | 99.19 | — | |
| H-NetInput Size=32 × 322024.04 | 99.19 | — | |
| FedCILLocal training iteration (T)=400, Global communication round (R)=200, Mini-batch size (B)=32, Backbone=3-layer CNN, Optimizer=Adam, Learning rate=1e-42023.02 | 99.13 | — | |
| SB-CNNInput Size=28 × 282024.04 | 99.08 | — | |
| RIC-CNNInput size=32 x 322026.02 | 99.02 | — | |
| ST-CNNInput Size=28 × 282024.04 | 99.02 | — | |
| GD-CNNInput Size=28 × 282024.04 | 98.88 | — | |
| Resilientinvariance constraint level=0.12023.06 | 98.86 | — | |
| MAX-CNNInput Size=28 × 282024.04 | 98.85 | — | |
| Constrainedinvariance constraint level=0.12023.06 | 98.74 | — | |
| Swin-B2025.08 | 98.62 | — | |
| CONV32+GIDInput size=32 x 322026.02 | 98.58 | — | |
| Unconstrained2023.06 | 98.45 | — | |
| Augerino2023.06 | 98.44 | — | |
| E(2)-CNNInput size=29 x 292026.02 | 98.14 | — | |
| E(2)-CNNInput Size=29 × 292024.04 | 98.14 | — | |
| FedProx+DGRLocal training iteration (T)=400, Global communication round (R)=200, Mini-batch size (B)=32, Backbone=3-layer CNN, Optimizer=Adam, Learning rate=1e-42023.02 | 97.55 | — | |
| FedAvg+DGRLocal training iteration (T)=400, Global communication round (R)=200, Mini-batch size (B)=32, Backbone=3-layer CNN, Optimizer=Adam, Learning rate=1e-42023.02 | 97.46 | — | |
| B-CNNInput size=28 x 282026.02 | 97.4 | — | |
| B-CNNInput Size=32 × 322024.04 | 97.4 | — | |
| FedProx+ACGAN ReplayLocal training iteration (T)=400, Global communication round (R)=200, Mini-batch size (B)=32, Backbone=3-layer CNN, Optimizer=Adam, Learning rate=1e-42023.02 | 97.38 | — | |
| ViT-L-162025.08 | 97.23 | — | |
| FedAvg+ACGAN ReplayLocal training iteration (T)=400, Global communication round (R)=200, Mini-batch size (B)=32, Backbone=3-layer CNN, Optimizer=Adam, Learning rate=1e-42023.02 | 97.13 | — | |
| GA-CNNInput size=28 x 282026.02 | 95.67 | — | |
| FedLwF-2TLocal training iteration (T)=400, Global communication round (R)=200, Mini-batch size (B)=32, Backbone=3-layer CNN, Optimizer=Adam, Learning rate=1e-42023.02 | 75.61 | — | |
| FedProxLocal training iteration (T)=400, Global communication round (R)=200, Mini-batch size (B)=32, Backbone=3-layer CNN, Optimizer=Adam, Learning rate=1e-42023.02 | 72.84 | — | |
| FedAvgLocal training iteration (T)=400, Global communication round (R)=200, Mini-batch size (B)=32, Backbone=3-layer CNN, Optimizer=Adam, Learning rate=1e-42023.02 | 72.28 | — | |
| Baseline CNNtime(s)=16.4, params(%)=100.002017.01 | — | 0.73 | |
| BN onlyTraining Data=Original MNIST, Data Augmentation=None2020.02 | — | 0.5 | |
| BN onlyTraining Data=Original MNIST, Data Augmentation=Standard2020.02 | — | 0.357 | |
| BN+DO 0.25/0.25Training Data=Original MNIST, Data Augmentation=None2020.02 | — | 0.403 | |
| BN+DO 0.25/0.25Training Data=Original MNIST, Data Augmentation=Standard2020.02 | — | 0.315 | |
| BN+DO 0.4/0.2Training Data=Original MNIST, Data Augmentation=None2020.02 | — | 0.395 | |
| BN+DO 0.4/0.2Training Data=Original MNIST, Data Augmentation=Standard2020.02 | — | 0.331 | |
| BN+KF 3,6Training Data=Original MNIST, Data Augmentation=None2020.02 | — | 0.307 | |
| BN+KF 3,6Training Data=Original MNIST, Data Augmentation=Standard2020.02 | — | 0.276 | |
| BN+KF 6Training Data=Original MNIST, Data Augmentation=None2020.02 | — | 0.305 | |
| BN+KF 6Training Data=Original MNIST, Data Augmentation=Standard2020.02 | — | 0.281 | |
| ORN-4(None)time(s)=7.9, params(%)=15.912017.01 | — | 0.63 | |
| ORN-4(ORAlign)time(s)=8.1, params(%)=15.912017.01 | — | 0.57 | |
| ORN-4(ORPooling)time(s)=8, params(%)=7.952017.01 | — | 0.59 | |
| ORN-8(None)time(s)=17.5, params(%)=31.412017.01 | — | 0.79 | |
| ORN-8(ORAlign)time(s)=17.8, params(%)=31.412017.01 | — | 0.59 | |
| ORN-8(ORPooling)time(s)=17.9, params(%)=12.872017.01 | — | 0.66 | |
| STN(affine)time(s)=18.5, params(%)=100.402017.01 | — | 0.61 | |
| STN(rotation)time(s)=18.7, params(%)=100.392017.01 | — | 0.66 | |
| TIPooling(x8)time(s)=126.7, params(%)=100.00, Note=Under augmentation2017.01 | — | 0.97 |