The trick is to make sure the recursive dependency stays linear, that's how you ...

_0ffh 6 months ago | parent | context | favorite | on: Were RNNs all we needed?

The trick is to make sure the recursive dependency stays linear, that's how you enable parallel training.