Table 2 LeNet-5 structure used for experiments
From: Neural network training with limited precision and asymmetric exponent
Item | Parameters |
|---|---|
1st convolutional layer | Input channels: 1 Output channels: 6 Kernel: 5 |
2nd convolutional layer | Input channels: 6 Output channels: 16 Kernel: 5 |
3rd convolutional layer | Input channels: 16 Output channels: 120 Kernel: 1 |
1st fully connected layer | Input: 120 Output: 84 |
2nd fully connected layer | Input: 84 Output: 10 |
Activation function (all layers) | TANH |
Classifier | Log_softmax |