TABLE 3:
The architecture of rich model in MIMIC-III, which includes a total of 51.6k parameters.
| Layer | Type | Hyper-parameters | Activation |
|---|---|---|---|
| 1 | Split | n_seg=6 | |
| 2 | Convolution1D | n_filter=64, kernel_size=4, stride=1 | ReLU |
| 3 | Convolution1D | n_filter=64, kernel_size=4, stride=1 | ReLU |
| 4 | AveragePooling1D | ||
| 5 | LSTM | hidden_units=64 | ReLU |
| 6 | PositionAttention | ||
| 7 | Dense | hidden_units=n_classes | Linear |
| 8 | Softmax |