Image Classification on MNIST (test)
100AccuracyOverfitting on test
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Overfitting on testBackbone=LeNet5, Evaluation Protocol=Overfitting on test2022.10 | 100 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Overfitting on testBackbone=LeNet5, Evaluation Protocol=Overfitting on test, Config=All Layers2022.10 | 100 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Previous SOTA [76]2022.05 | 99.91 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Homogeneous Vector CapsulesModel Type=Ensemble, Year=20212020.01 | 99.87 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Homogeneous Vector CapsulesModel Type=Single Model, Year=20212020.01 | 99.83 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RMDLModel Type=Ensemble, Year=20182020.01 | 99.82 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCT-7/3x1# Params=3.76 M, MACs=1.19 G2021.04 | 99.82 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Efficient-CapsNetSize=0.16M2021.10 | 99.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet18# Params=11.18 M, MACs=0.04 G2021.04 | 99.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Wan et al.Training set size=Full2019.04 | 99.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DropConnectModel Type=Ensemble, Year=20132020.01 | 99.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet50# Params=23.53 M, MACs=0.08 G2021.04 | 99.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCT-6/3x1# Params=3.23 M, MACs=1.02 G2021.04 | 99.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Similar NetworkNumber of hidden layers=12019.01 | 99.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cireşan et al.Training set size=Full2019.04 | 99.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| APACModel Type=Single Model, Year=20152020.01 | 99.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Multi-Column Deep Neural NetworksModel Type=Single Model, Year=20122020.01 | 99.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet34# Params=21.29 M, MACs=0.08 G2021.04 | 99.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MobileNetV2/1.25# Params=3.47 M, MACs=0.01 G2021.04 | 99.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-Lite-7/4# Params=3.72 M, MACs=0.26 G2021.04 | 99.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Batch-Normalized Maxout Network in NetworkModel Type=Single Model, Year=20152020.01 | 99.76 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CVT-7/4# Params=3.72 M, MACs=0.25 G2021.04 | 99.76 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCT-7/3x2# Params=3.85 M, MACs=0.29 G2021.04 | 99.76 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Baseline CapsNetepochs=1000E2018.05 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DCNetepochs=50E2018.05 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Sabour et al.Training set size=Full2019.04 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Dynamic Routing Between CapsulesModel Type=Single Model, Year=20172020.01 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Lets keep it simpleModel Type=Single Model, Year=20162020.01 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MobileNetV2/1.0# Params=2.21 M, MACs=0.01 G2021.04 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MobileNetV2/2.0# Params=8.72 M, MACs=0.02 G2021.04 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CVT-6/4# Params=3.19 M, MACs=0.22 G2021.04 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCT-6/3x2# Params=3.33 M, MACs=0.25 G2021.04 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WaveMixAugmentation=TrivialAugment2022.05 | 99.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-Lite-6/4# Params=3.19 M, MACs=0.22 G2021.04 | 99.74 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CVT-6/8# Params=3.21 M, MACs=0.05 G2021.04 | 99.74 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-Lite-6/8# Params=3.22 M, MACs=0.06 G2021.04 | 99.73 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCT-4/3x2# Params=0.48 M, MACs=0.05 G2021.04 | 99.73 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DCNetepochs=50E, reconstruction_subnetwork=BR2018.05 | 99.72 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DCNetepochs=50E, reconstruction_subnetwork=NR2018.05 | 99.71 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DCNet++epochs=50E2018.05 | 99.71 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WaveMix-128/7Number of parameters (Million)=2.42022.03 | 99.71 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HingeOptimizer=ASAM, Architecture=M3-CNN2022.05 | 99.71 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VGG-5Size=3.65M2021.10 | 99.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FlexNet-16Size=0.67M2021.10 | 99.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MobileNetV2/0.5# Params=0.70 M, MACs=< 0.01 G2021.04 | 99.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CVT-7/8# Params=3.74 M, MACs=0.06 G2021.04 | 99.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCT-2/3x2# Params=0.28 M, MACs=0.04 G2021.04 | 99.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-18Number of parameters (Million)=11.22022.03 | 99.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-Lite-7/8# Params=3.74 M, MACs=0.06 G2021.04 | 99.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-Lite-7/16# Params=3.89 M, MACs=0.02 G2021.04 | 99.68 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HingeOptimizer=SGD, Architecture=M3-CNN2022.05 | 99.68 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Q-RUN#Params=1.36M2025.12 | 99.68 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Baseline CapsNetepochs=50E2018.05 | 99.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-34Number of parameters (Million)=21.32022.03 | 99.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EXACTOptimizer=SGD, Architecture=M3-CNN2022.05 | 99.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EXACTOptimizer=ASAM, Architecture=M3-CNN2022.05 | 99.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FAN#Params=1.86M2025.12 | 99.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-Lite-6/16# Params=3.36 M, MACs=0.02 G2021.04 | 99.66 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UndefendedBackbone=NiN2022.02 | 99.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Baseline CapsNetepochs=1000E, reconstruction_subnetwork=NR2018.05 | 99.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WaveMix-256/7Number of parameters (Million)=9.62022.03 | 99.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cross-entropyOptimizer=ASAM, Architecture=M3-CNN2022.05 | 99.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cross-entropyOptimizer=SGD, Architecture=M3-CNN2022.05 | 99.64 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-12/16# Params=85.63 M, MACs=0.43 G2021.04 | 99.63 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CNN#Params=1.86M2025.12 | 99.63 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UndefendedBackbone=ResNet182022.02 | 99.61 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DSNNo. of parameters=350 K2015.07 | 99.61 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HingeOptimizer=Adam, Architecture=M3-CNN2022.05 | 99.61 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet18 + VGG Ensemble2019.05 | 99.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Network in Network2021.10 | 99.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WholeIPC=Full, Ratio=100%2025.12 | 99.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NONEBackbone=NiN2022.02 | 99.58 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cross-entropyOptimizer=Adam, Architecture=M3-CNN2022.05 | 99.58 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EXACTOptimizer=Adam, Architecture=M3-CNN2022.05 | 99.58 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ST-RSBPNetwork structure=15C5-P2-40C5-P2-300, Time steps=400, Neurons=26K, Params=607K2021.09 | 99.57 | — | — | — | — | — | — | — | — | 99.62 | — | — | 0.04 | — | |
| NONEBackbone=ResNet182022.02 | 99.57 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PreActResNet-182019.05 | 99.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-50Number of parameters (Million)=23.62022.03 | 99.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Highway NetworkNo. of layers=10-layer, width=32, No. of parameters=151 K2015.07 | 99.55 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MaxoutNo. of parameters=420 K2015.07 | 99.55 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PointCNN2018.01 | 99.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Network in Network2018.01 | 99.53 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IDE-LIFNetwork structure=64C5s (F64C5), Time steps=30, Neurons=13K, Params=229K2021.09 | 99.53 | — | — | — | — | — | — | — | — | 99.59 | — | — | 0.04 | — | |
| TeacherData Size=50K~100K, Teacher Architecture=VGG13, Student Architecture=VGG112022.05 | 99.52 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PSTRN-SModel=LeNet-5, Compression Method=TR, Compression Ratio=6.5x2021.07 | 99.51 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADMM-based TT-formatModel=LeNet-5, Compression Method=TT, Compression Ratio=8.3x2021.07 | 99.51 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SGDModel compression status=Raw2018.01 | 99.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TSSL-BPNetwork structure=15C5-P2-40C5-P2-300, Time steps=5, Neurons=26K, Params=607K2021.09 | 99.5 | — | — | — | — | — | — | — | — | 99.53 | — | — | 0.02 | — | |
| KIP ConvNetIPC=50, Augmentation=true2021.07 | 99.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TeacherData Size=50K~100K, Teacher Architecture=ResNet32, Student Architecture=ResNet82022.05 | 99.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PointNet++2018.01 | 99.49 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HM2-BPNetwork structure=15C5-P2-40C5-P2-300-10, Data augmentation=true2018.05 | 99.49 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IDE-IFNetwork structure=64C5s (F64C5), Time steps=30, Neurons=13K, Params=229K2021.09 | 99.49 | — | — | — | — | — | — | — | — | 99.55 | — | — | 0.04 | — | |
| MLData Size=50K~100K, Teacher Architecture=ResNet32, Student Architecture=ResNet82022.05 | 99.49 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADMM-based TT-formatModel=LeNet-5, Compression Method=TT, Compression Ratio=17.9x2021.07 | 99.48 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AMData Size=10K, Teacher Architecture=VGG13, Student Architecture=VGG112022.05 | 99.47 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CTM (DC)Drop Clause=true2021.05 | 99.45 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CTM (DC)Drop Clause=true2021.05 | 99.45 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MEKD (hard)Data Size=10K, Teacher Architecture=VGG13, Student Architecture=VGG112022.05 | 99.45 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Spiking CNN (converted)Conversion type=Converted from a trained ANN2018.05 | 99.44 | — | — | — | — | — | — | — | — | — | — | — | — | — |