Algorithm 1: DCRA.

Input: training data EEG, EOG, Mahalanobis distance M

Output: DCRA model parameters

1    Training layer 1 GRU;

2    Training layer 2 GRU;

3    Training layer 3 GRU;

4    Update DCRA parameters using Formula (5) using gradient descent;

5    Repeat steps 1–4 until the model converges.