ConvNet
In 1989, LeCun et al. [1] presented a new architecture known as the convolutional neural network with ConvNet.
Network features:
- 4 layers (2 convolutional + 2 fully connected)
- use tanh as activation function
Architecture
Parameters
| Layer | Activation Shape | Filter Shape | Weights | Biases | Parameters |
|---|---|---|---|---|---|
| Input | 1x16x16 | - | 0 | 0 | 0 |
| Conv1 | 12x8x8 | 5x5 | 300 | 12 | 312 |
| Conv2 | 12x4x4 | 5x5 | 3,600 | 12 | 3,612 |
| FC1 | 30x1x1 | - | 5,760 | 30 | 5,790 |
| FC2 | 10x1x1 | - | 300 | 10 | 310 |
| Total | 10,024 |
See [2] for explanations about parameters calculations.
According to the authors [1], there are 9760 parameters …