Sequential working of LSTMs

Long Short-Term Memory Unit (LSTM) cells are nothing but slightly more advanced architectures compared to Recurrent Networks. LSTMs can be thought of as a special kind of Recurrent Neural Networks with the capabilities of learning long-term dependencies that exist in sequential data. The main reason behind this is the fact that LSTMs contain memory and are able to store and update information within their cells unlike Recurrent Neural Networks.

Getting ready

The main components of a Long Short-Term Memory unit are as follows:

• The input gate
• The forget gate
• The update gate

Each of these gates is made up of a sigmoid layer followed by a pointwise multiplication operation. The sigmoid layer outputs...