From: Improving deep neural network design with new text data representations
1-ofm vs. log(m) embedding | ||||||
---|---|---|---|---|---|---|
 | log-m | 1-of-m | ||||
Convolutional layer | Number of filters | Filter | Pool | Number of filters | Filter | Pool |
1 | 32 | \(3 \times 3\) | – | 32 | \(3 \times 3\) | \(2 \times 2\) |
2 | 32 | \(3 \times 3\) | – | 32 | \(3 \times 3\) | \(2 \times 2\) |
3 | 32 | \(3 \times 3\) | \(2 \times 2\) | 32 | \(3 \times 3\) | \(2 \times 2\) |
Dense layer | Number of neurons | Number of neurons | ||||
---|---|---|---|---|---|---|
1 | 512 | 1024 | ||||
2 | 512 | 1024 | ||||
3 | 2 | 2 |
Padded vs. non-padded | ||||||
---|---|---|---|---|---|---|
 | Deep padded | Padded/non-padded | ||||
Convolutional layer | Number of filters | Filter | Pool | Number of filters | Filter | Pool |
1 | 128 | \(3 \times 3\) | – | 128 | \(3 \times 3\) | – |
2 | 128 | \(3 \times 3\) | – | 128 | \(3 \times 3\) | – |
3 | 128 | \(3 \times 3\) | \(2 \times 2\) | 128 | \(3 \times 3\) | \(2 \times 2\) |
4 | 128 | \(3 \times 3\) | – | – | – | – |
5 | 128 | \(3 \times 3\) | – | – | – | – |
6 | 128 | \(3 \times 3\) | \(2 \times 2\) | – | – | – |
Dense layer | Number of neurons | Number of neurons | ||||
---|---|---|---|---|---|---|
1 | 512 | 1024 | ||||
2 | 512 | 1024 | ||||
3 | 2 | 2 |