Decoder

Appears in 2 papers

The second half of the seq2seq architecture.

As used in Paper 06 — Sequence to Sequence Learning with Neural Networks →

The second half of the seq2seq architecture. An RNN/LSTM that takes

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

The half of the seq2seq model that generates the target language sequence. In Bahdanau's model, the decoder receives a fresh context vector at each step (via attention) rather than a single fixed vector at the start.