Summary

RNNs are a special type of neural network that is well suited for sequential data such as time series, audio, video, and text. Research showed that RNNs have improved the performance of sequential data types when compared to other architectures such as FNNs and CNNs. The key to an RNN is the sequence memory state, which helps it store information from the previously analyzed state; this is good for sequential signal analysis and predictive analysis. In this chapter, we learned how RNNs are different from FNNs and CNNs. We understood the different types of RNNs and what makes them good for sequential data analysis by looking at a few examples. RNNs, as you may have noticed, are good for mapping a fixed or variable-sized input sequence to a fixed or variable-sized output; we have seen several examples to understand this.

We also looked at how RNNs can help with genomics tasks and understood the different architectural types of RNNs. Bidirectional RNN, LSTM, and GRU are variants...