Conceptual
Deprecated

Superseded by the transformer: the recurrent encoder and decoder must run one timestep at a time, so training cannot be parallelised across the sequence, and attention removed that constraint entirely

Learn more →
Why study this historical topic?

Superseded by the transformer: the recurrent encoder and decoder must run one timestep at a time, so training cannot be parallelised across the sequence, and attention removed that constraint entirely

Recurrent Sequence-to-Sequence Translation with a Single Context Vector

the concrete system that defined neural machine translation for several years, and the one the transformer paper was written against