Fig. 5. Architecture of a recurrent neural network (RNN).

At each time step, the model receives the current time slice data as input as well as the own output from the preceding time step. The features are captured in a truly sequential fashion.