Table 6 Configurations of CNN, CWNN experiments.
From: Research on improved convolutional wavelet neural network
No. | Parameter type | Parameter name | CNN | CWNN |
|---|---|---|---|---|
1 | 1st to 5th layers | First type of NN | CPNN | CPNN |
2 | 2nd, 4th layers | Activation function of convolutional layer | Sigmoid | Sigmoid |
3 | 1st layer | Dimension of the 1st layer | \(28\times 28\) | \(28\times 28\) |
4 | 2nd layer | Dimension of 1st convolutional layer | \(28\times 28\) | \(28\times 28\) |
5 | 3rd layer | Dimension of 1st pooling layer | \(24\times 24\) | \(24\times 24\) |
6 | 2nd, 3rd layers | Number of features | \(6\) | \(6\) |
7 | 4th layer | Dimension of 2nd convolutional layer | \(12\times 12\) | \(12\times 12\) |
8 | 5th layer | Dimension of 2nd 1 pooling layer | \(8\times 8\) | \(8\times 8\) |
9 | 4th, 5th layers | Number of features | \(12\) | \(12\) |
10 | − 1st, − 2nd, − 3rd layers | Second type of NN | FCNN | WNN |
11 | − 3rd layer | Dimension of -3rd input layer | 192 | 192 |
12 | − 2nd layer | Dimension of -2nd hidden layer | None | \(50\) |
13 | − 2nd layer | Activation function of hidden layer | None | Wavelet |
14 | − 1st layer | Dimension of − 1st output layer | 10 | 10 |
15 | − 1st layer | Activation function of output layer | Sigmoid | Sigmoid |
16 | Hyperparameters | Learning rate \(\eta\) | 0.1 | 0.1 |
17 | Hyperparameters | Coefficient of inertia \(\mathrm{\alpha }\) | None | 0.2 |
18 | Hyperparameters | \(max\_SPs\) | 10 | 10 |
19 | Hyperparameters | \(max\_ACs\) | 6000 | 6000 |
20 | Hyperparameters | \(target\_err\) | 0.0000001 | 0.0000001 |
21 | Hyperparameters | \(BatchSize\) | 10 | 10 |
