Chapter 12 Multilayer RNN

Hi Everyone,

in Chapter 12 there is a pictorial representation of a multilayer RNN, which I’ve been trying to rebuild.

I am a little confused about the green dotted arrows. In the previous representaions green arrows were embeddings. But here an embedding wouldn’t make much sense since the output of the previous linear layers isn’t categorial. So are those dotted green arrows supposed to be linear layers?


To my understanding the arrows are simply a representation of the weights. In that sense whether those are embedding weights or other it shouldn’t make much difference.

Hi ml0101,
I am pretty sure the green arrow represents the hidden state of the first RNN. It is used as the input sequence to the next layer, which has its own hidden state and output calculation.


Thanks for your replies. Got it to run now :slight_smile: