R Deep Learning Essentials - Second Edition

Long short term memory model

LSTMs are designed to learn long-term dependencies. Similar to RNNs, they are chained and have four internal neural network layers. They split the state into two parts, where one part manages short-term state and the other adds long-term state. LSTMs have gates which control how memories are stored. The input gate controls which part of the input should be added to the long-term memory. The forget gate controls the part of long-term memory that should be forgotten. The final gate, the output gate, controls which part of the long-term memory should be in the output. This is a brief description of LSTMs – a good reference for more details is http://colah.github.io/posts/2015-08-Understanding-LSTMs/.

The code for ...