Image Classification on EMNIST Balanced (test)
91.06AccuracyWaveMix-128/7
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| WaveMix-128/7Number of parameters (Million)=2.42022.03 | 91.06 | — | — | — | |
| WaveMixAugmentation=TrivialAugment2022.05 | 91.06 | — | — | — | |
| Previous SOTA [73]2022.05 | 91.05 | — | — | — | |
| VGG-5 (Spinal FC)Size of Fully Connected Layer=4HL, 128 Neurons Per Layer, Epoch=200, Parameters=3.630M2020.07 | 90.73 | — | 91.05 | 0.1 | |
| VGG-5Size of Fully Connected Layer=1HL, 512 Neurons, Epoch=200, Parameters=3.646M2020.07 | 90.71 | — | 91.04 | — | |
| WaveMix-256/7Number of parameters (Million)=9.62022.03 | 90.36 | — | — | — | |
| ResNet-50Number of parameters (Million)=23.62022.03 | 89.76 | — | — | — | |
| ResNet-34Number of parameters (Million)=21.32022.03 | 89.17 | — | — | — | |
| ResNet-18Number of parameters (Million)=11.22022.03 | 89 | — | — | — | |
| CNN (Spinal FC)Size of Fully Connected Layer=6HL, 10 Neurons Per Layer, Epoch=8, Parameters=16.05k2020.07 | 82.57 | — | 83.21 | 17.66 | |
| CentralizedArchitecture=MLP (2 hidden layers, 64 neurons)2022.10 | 82.45 | — | — | — | |
| CNN (Spinal FC)Size of Fully Connected Layer=6HL, 8 Neurons Per Layer, Epoch=8, Parameters=13.82k2020.07 | 82.13 | — | 82.77 | 15.5 | |
| CNNSize of Fully Connected Layer=1HL, 50 Neurons, Epoch=8, Parameters=21.84k2020.07 | 78.99 | — | 79.61 | — | |
| DReS-FLArchitecture=PINN (2 hidden layers, 64 neurons), degree of gradient=82022.10 | 78.04 | — | — | — | |
| FedAvg-ISArchitecture=MLP (2 hidden layers, 64 neurons)2022.10 | 77.09 | — | — | — | |
| 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 | 73.12 | — | — | — | |
| FedAvgArchitecture=MLP (2 hidden layers, 64 neurons)2022.10 | 71.5 | — | — | — | |
| 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 | 66.87 | — | — | — | |
| 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 | 66.19 | — | — | — | |
| MCLmodel type=flexible models, presence of unlabeled samples=false, #n=131600, #f=784, #c=472020.01 | 65.41 | — | — | — | |
| 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 | 63.55 | — | — | — | |
| 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 | 63.55 | — | — | — | |
| SCAFFOLDArchitecture=MLP (2 hidden layers, 64 neurons)2022.10 | 55.15 | — | — | — | |
| Fwdmodel type=flexible models, presence of unlabeled samples=false, #n=131600, #f=784, #c=472020.01 | 18.21 | — | — | — | |
| 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 | 17.74 | — | — | — | |
| 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 | 17.25 | — | — | — | |
| 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 | 17.22 | — | — | — | |
| PCmodel type=flexible models, presence of unlabeled samples=false, #n=131600, #f=784, #c=472020.01 | 14.28 | — | — | — | |
| MCULmodel type=flexible models, presence of unlabeled samples=true, #n=131600, #f=784, #c=472020.01 | 6.77 | — | — | — | |
| GAmodel type=flexible models, presence of unlabeled samples=false, #n=131600, #f=784, #c=472020.01 | 4.25 | — | — | — | |
| MCLmodel type=flexible models, presence of unlabeled samples=true, #n=131600, #f=784, #c=472020.01 | 4.14 | — | — | — | |
| Fwdmodel type=flexible models, presence of unlabeled samples=true, #n=131600, #f=784, #c=472020.01 | 2.68 | — | — | — | |
| GAmodel type=flexible models, presence of unlabeled samples=true, #n=131600, #f=784, #c=472020.01 | 2.54 | — | — | — | |
| PCmodel type=flexible models, presence of unlabeled samples=true, #n=131600, #f=784, #c=472020.01 | 2.36 | — | — | — | |
| DenseNet100-12Activation setting=Base2018.11 | — | 8.81 | — | — | |
| DenseNet100-12Activation setting=Drop-Activation2018.11 | — | 8.9 | — | — | |
| PreResNet164Activation setting=Base2018.11 | — | 8.88 | — | — | |
| PreResNet164Activation setting=Drop-Activation2018.11 | — | 8.72 | — | — | |
| ResNet164Activation setting=Base2018.11 | — | 8.85 | — | — | |
| ResNet164Activation setting=Drop-Activation2018.11 | — | 8.82 | — | — | |
| ResNeXt29, 8*64Activation setting=Base2018.11 | — | 9.07 | — | — | |
| ResNeXt29, 8*64Activation setting=Drop-Activation2018.11 | — | 8.91 | — | — | |
| WRN28-10Activation setting=Base2018.11 | — | 8.97 | — | — | |
| WRN28-10Activation setting=Drop-Activation2018.11 | — | 8.72 | — | — |