Gated Recurrent Unit
GRU is a lighter gated RNN with reset and update gates, often matching LSTM quality at lower cost on medium-length sequences.
Definition
Cho et al. introduced the GRU as a two-gate recurrence used in early neural machine translation. It has no separate cell state; the hidden state is both memory and output. Empirically it is close to LSTM on many tasks and cheaper to train.
Why it matters
GRU is the right mental model for “gated recurrence without LSTM ceremony.” If a paper says RNN in 2014–2016 NMT, it is often this.
Case
The original encoder–decoder NMT models used GRUs to compress a source sentence into a vector, then generate the target. Attention was added because that vector bottleneck lost the source.
How to read it
Treat GRU vs LSTM as a compute trade, not a religion. On long context both lose to attention; on short telemetry GRU is a sane baseline.
Key papers
Landmark references with real publisher or arXiv links — not token-overlap noise.
- Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine TranslationCho et al. · 2014 · EMNLPGRU + seq2seq.