Implementing LSTMs in Code Video (how do layers feed into each other)


How does a 64 Bidirectional LSTM feed into a 32 Bidirectional LSTM? Is it only using the last 32 outputs of the original 64 outputs of the first LSTM?

In keras how do the forward and backward passes get fed into the next layer? Do the forward passes act as input to the next forward pass and the backward pass get fed into the next backward pass (like in the screen shot below) or do they get combined into one number then that one number is input into the next layer “above it”?

Does this help?

So if I understand that post correctly then when you give two inputs to a Bidirectional LSTM then one is used for forward and the other is used for backward like the diagram attached shows?

You provide 1 input to the bidirectional layer. Internally, the layer traverses the input from left to right and right to left.

Thank you I get it now!