Table 2 Architecture of the multi-task; single task models differ only in output size of the final fully connected layer (1 output unit instead of 10)
Type | Filter size | # Filters | Dilation | Batchnorm | Activation |
|---|---|---|---|---|---|
1D average pool | 10 | – | – | – |  |
1D CNN | 5 | 75 | 2 | 1D | ReLU |
1D CNN | 5 | 150 | 2 | 1D | ReLU |
1D CNN | 5 | 225 | 2 | 1D | ReLU |
1D CNN | 5 | 300 | 2 | 1D | ReLU |
Flatten | – | – | – |  |  |
Fully connected | – | 800 | – | – | ReLU |
Dropout (0.2 probability) | – | – | – | – | – |
Fully connected | – | 200 | – | – | ReLU |
Dropout (0.2 probability) | – | – | – | – | – |
Fully connected | – | 10 (multi-tasking) 1 (single tasking) | – | – | – |