LSTM
Also called Long Short-Term Memory
A recurrent network with gates that decide what to remember and forget, so it can hold information longer.
Think of it like
A notepad with an eraser and a highlighter — you keep the important lines and clear the rest as you go.
Example
Before transformers, LSTMs powered much of machine translation and speech recognition for years.
How it actually works
The LSTM added input, forget, and output gates plus a protected cell state, giving gradients a stable path through time and letting the network retain information across longer spans than a plain RNN. It was the dominant sequence model of the 2010s. Transformers eventually beat it on scale and parallelism, but the LSTM’s gating ideas still echo in newer architectures.
For product teams
The reliable sequence model of the pre-transformer era — historically pivotal, now rarely the default.
For engineers
A gated recurrent unit with a cell state and forget/input/output gates that mitigate vanishing gradients.
Related
- Recurrent Neural Network — A gated recurrent neural network.
- GRU — A simpler cousin is the GRU.
- Transformer — Superseded by the transformer.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