Gated Recurrent Unit (GRU)
Definition
A GRU is a recurrent neural network component with gates to control information flow and memory.
GRUs use update and reset gates to retain or discard relevant past information over sequences. They are simpler in structure than classical LSTMs.
Simply explained
Like a memory with two controllers: one decides what to keep, the other what can be reset.
Why it matters in practice
Relevant for time series and sequence models, especially as a compact alternative to LSTM; however, other architectures dominate many modern language tasks.
Example from business
A model processes sensor data as a time series and uses a GRU to capture patterns across multiple time steps.
Limits and pitfalls
- Equating GRU and LSTM; assuming long dependencies without benchmarks; not comparing modern Transformer or State-Space models.
Origin & history
Cho et al. described in 2014 in the RNN Encoder-Decoder a gated recurrent unit from which the widely used GRU emerged today.
Use in mid-sized companies
- Possible application: compact time series models for sensor, demand, or state data.
Frequently asked questions
Is a GRU an LSTM?
No, both are gated RNNs but structured differently. Is GRU always faster? Often more compact, but actual runtime depends on implementation and hardware.
Sources
- Learning Phrase Representations using RNN Encoder-Decoder · Cho et al. (2014), Learning Phrase Representations using RNN Encoder-Decoder
Sources checked on 10.10.2026.
Editorially reviewed on: