Image Classification on Kuzushiji-MNIST original (test)
0.99Test Error RateVGG8B + Cutout (Predsim Loss)
Evaluation results
| Method | Links | |
|---|---|---|
| VGG8B + Cutout (Predsim Loss)Backbone=VGG8B, Parameters=7.3M, Training=Local Predsim Loss, Regularization=Cutout2019.01 | 0.99 | |
| PARN + MM (Global)Backbone=PreActResNet-18, Parameters=11M, Training=Global Backpropagation, Regularization=Manifold Mixup2019.01 | 1.17 | |
| VGG8B (Predsim Loss)Backbone=VGG8B, Parameters=7.3M, Training=Local Predsim Loss2019.01 | 1.36 | |
| VGG8B (Global)Backbone=VGG8B, Parameters=7.3M, Training=Global Backpropagation2019.01 | 1.53 | |
| PARN (Global)Backbone=PreActResNet-18, Parameters=11M, Training=Global Backpropagation2019.01 | 2.18 | |
| VGG8B (Similarity Matching Loss)Backbone=VGG8B, Parameters=7.3M, Training=Local Similarity Matching Loss2019.01 | 2.19 | |
| VGG8B (Prediction Loss)Backbone=VGG8B, Parameters=7.3M, Training=Local Prediction Loss2019.01 | 2.22 | |
| 3x1024 MLP (Global)Backbone=3x1024 MLP, Parameters=2.9M, Training=Global Backpropagation2019.01 | 5.99 | |
| 3x1024 MLP (Prediction Loss)Backbone=3x1024 MLP, Parameters=2.9M, Training=Local Prediction Loss2019.01 | 7.26 | |
| 3x1024 MLP (Predsim Loss)Backbone=3x1024 MLP, Parameters=2.9M, Training=Local Predsim Loss2019.01 | 7.33 | |
| 3x1024 MLP (Similarity Matching Loss)Backbone=3x1024 MLP, Parameters=2.9M, Training=Local Similarity Matching Loss2019.01 | 9.8 |