GRU
In plain terms
A streamlined recurrent network similar to LSTM.
Going deeper
Gated Recurrent Units handle sequences with fewer gates than LSTMs, often training faster with comparable performance. They're a lighter alternative.
Under the hood
Combines forget and input gates; simpler and sometimes more efficient than LSTM.
Examples
Sequence modeling, time series, NLP (pre-transformer), efficient RNNs.
Think of it like...
Like a lighter version of the same note-taker, faster but nearly as capable.