CNN Architectures Trained With MNIST
Gabriel Falcao edited this page Jan 15, 2023
·
3 revisions
| input: 1x28x28 (grey-level image) | ||
| Convolutional Layer | kernel: 1x20x5x5 | stride=1; padding=0 |
| Batch Normalization | features: 20 | |
| ReLU (non-linearity) | ||
| Average Pooling | kernel: 2x2 | stride=2 |
| Convolutional Layer | kernel: 20x50x5x5 | stride=1; padding=0 |
| Batch Normalization | features: 50 | |
| ReLU (non-linearity) | ||
| Average Pooling | kernel: 2x2 | stride=2 |
| reshape: 50x4x4 => 800x1 | ||
| Fully Connected Layer | kernel: 800x500 | |
| Batch Normalization | features: 500 | |
| ReLU (non-linearity) | ||
| Fully Connected Layer | kernel: 500x250 | |
| Batch Normalization | features: 500 | |
| ReLU (non-linearity) | ||
| Fully Connected Layer | kernel: 250x10 | |