In this chapter, we have covered what an RNN is and how to train one. We have seen that, in order to effectively model long-term dependencies and overcome training challenges, changes to a standard RNN are necessary, including additional information-across-time control mechanisms that are provided by GRU/LSTM units. We built such a network in Gorgonia.

In next chapter, we will learn how to build a CNN and how to tune some of the hyperparameters.