Visualize: Figure 11.
Description: " Hidden layer activations of a trained GRU network while processing different sequences.
The input labels, along with the mode (addition/subtraction) at every point in time are printed left of
the activation values."