reimplementing-ml-papers
LeNet in Keras
See implementation notes which includes explanation of the various implementation versions, and additional details, such as an extension of the custom Subsampling layer.
| Approach | Description | Input pixel range |
Pooling layer |
Activation | Learning rate |
Notebook |
|---|---|---|---|---|---|---|
| v1 | Basic implementation | [0, 1] | MaxPool2D | tanh | 0.001 |    |
| v2 | Custom layer, activation |
[0, 1] | Subsampling | scaled tanh |
0.001 | |
| v3 | Custom layer, activation, scaling, learning rate |
[-0.1, 1.175] | Subsampling | scaled tanh |
schedule | |