In this chapter, we dived deep into the inner workings of the LSTM network. We explored both the concepts and mathematical implementation related to these networks, understanding how information is processed in an LSTM cell and using short-term and long-term memory of events. We also saw why the network gets its name, being adept at conserving relevant cell states over very distant timesteps. While we discussed some variants to the architecture, such as the peephole connection, it is seldom seen in most common LSTM candidate scenarios. Although we executed our demonstrations with a simple time series dataset, we highly encourage you to implement this architecture to tackle other problems that you may already be familiar with (such as the IMDB sentiment classification dataset), and compare results with our earlier efforts.
LSTMs have really been shown to shine at natural...