Deep recurrent neural networks¶

Most RNN consists of 3 blocks of parameters.

  1. From input to hidden

  2. From previous hidden state to the next hidden state

  3. From hidden state to the output

There are multiple options how to implement deep RNN.

  • we can have a hierarchy of hidden states

  • we can add an extra step between hidden states

  • we can have skip connection between hidden states