Image Classification Error Rate on MNIST (test)
0.0031Error RateRCNN-96
Evaluation results
| Method | Links | |
|---|---|---|
| RCNN-962015.11 | 0.0031 | |
| Deeply-supervised nets2015.11 | 0.0039 | |
| Network in Network2015.11 | 0.0045 | |
| Conv. Maxout+Dropout2015.11 | 0.0047 | |
| Stochastic Pooling2015.11 | 0.0047 | |
| Softmax BaselineArchitecture=3-layer, 256 neuron-wide, fully-connected network, Nonlinearity=GELU, Optimizer=Adam2016.10 | 0.0169 | |
| Base-Branching & Merging CNNw/HFCsYear=2020, Ensemble=true2021.01 | 0.16 | |
| Efficient-CapsNetYear=2021, Ensemble=true2021.01 | 0.16 | |
| RMDLNumber of RDLs=302018.05 | 0.18 | |
| RMDL: Random Multimodel Deep Learning for ClassificationYear=2018, Ensemble=true2021.01 | 0.18 | |
| RMDLNumber of RDLs=152018.05 | 0.21 | |
| Regularization of Neural Networks using DropConnectYear=2013, Ensemble=true2021.01 | 0.21 | |
| Multi-Column Deep Neural Networks for Image ClassificationYear=2012, Ensemble=true2021.01 | 0.23 | |
| CapsNetBackbone=Caps, # of params (Millions)=6.8, Training Arch.=GPU, Pre-trained on ImageNet=false2022.03 | 0.25 | |
| E-CapsNetBackbone=Caps, # of params (Millions)=0.2, Training Arch.=GPU, Pre-trained on ImageNet=false2022.03 | 0.26 | |
| Capsule VBBackbone=Caps, # of params (Millions)=0.2, Training Arch.=GPU, Pre-trained on ImageNet=false2022.03 | 0.3 | |
| AgglomeratorBackbone=Conv/MLP/Caps, # of params (Millions)=72, Training Arch.=GPU, Pre-trained on ImageNet=false2022.03 | 0.3 | |
| EM routingResolution=32x32, Architecture=4 capsule layers2018.05 | 0.32 | |
| VGGBackbone=Conv, # of params (Millions)=20, Training Arch.=GPU, Pre-trained on ImageNet=false2022.03 | 0.32 | |
| FREMResolution=32x32, Architecture=4 capsule layers2018.05 | 0.38 | |
| RMDLNumber of RDLs=92018.05 | 0.41 | |
| FRMSResolution=32x32, Architecture=4 capsule layers2018.05 | 0.42 | |
| Matrix-CapsNetBackbone=Caps, # of params (Millions)=0.3, Training Arch.=GPU, Pre-trained on ImageNet=false2022.03 | 0.44 | |
| RMDLNumber of RDLs=32018.05 | 0.51 | |
| PCANet-12018.05 | 0.62 | |
| Baseline CNNResolution=32x32, Architecture=4 capsule layers2018.05 | 0.65 | |
| gcForest2018.05 | 0.74 | |
| Deep L2-SVM2018.05 | 0.87 | |
| Maxout Network2018.05 | 0.94 | |
| BinaryConnect2018.05 | 1.29 | |
| ResNet-110Backbone=Conv, # of params (Millions)=1.7, Training Arch.=GPU, Pre-trained on ImageNet=false2022.03 | 2.1 |