Bidirectional RNN (BiRNN)

Appears in 1 paper

Two recurrent networks — one reading left to right, one right to left — whose hidden states are concatenated at each position.

As used in Paper 07 — Neural Machine Translation by Jointly Learning to Align and Translate →

Two recurrent networks — one reading left to right, one right to left — whose hidden states are concatenated at each position. Each position thus reflects context from both directions. Used in Bahdanau's encoder to give each source word a full-sentence representation.