GRU
Also called Gated Recurrent Unit
A streamlined LSTM with fewer gates — nearly as capable, a bit cheaper and simpler.
Think of it like
A lighter backpack with one fewer pocket that still carries almost everything you need.
Example
A team on a tight compute budget picks a GRU over an LSTM for a sequence task and loses little accuracy.
How it actually works
The GRU merges the LSTM’s gates into two (update and reset) and drops the separate cell state, making it lighter and faster to train while often matching LSTM performance. It was a popular choice when efficiency mattered. Like the LSTM, it belongs to the recurrent era that transformers largely displaced, though it remains a clean baseline for small sequence problems.
For product teams
A lighter alternative to the LSTM from the same era — simple and efficient, but no longer the go-to.
For engineers
A recurrent unit with update/reset gates and no separate cell state, cheaper than an LSTM at similar quality.
Related
- LSTM — A simpler LSTM.
- Recurrent Neural Network — A gated recurrent neural network.
- Transformer — Overtaken by the transformer.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome