LSTM (Long Memory)

โ† Back to the article

An LSTM is an RNN with gates that help it remember important things for longer.

What it does

An LSTM is an RNN with gates that help it remember important things for longer.

Gate: A valve for data flow โ€ข Cell State: Long-term storage

How it works

  • Use a forget gate to drop unneeded information.
  • Use an input gate to save useful new information.
  • Carry important memory forward.

Uses and applications

  • A text app can remember the start of a long sentence.
  • A weather system can remember older weather patterns.
  • A phone keyboard can use earlier words to guess the next word.

In short

It is like a diary that keeps important events and forgets tiny details.

Related methods

  • RNN (Recurrent Net): The simpler recurrent network that LSTM improves.
  • Transformer: Replaced recurrence with attention in many tasks.
  • Prophet: Forecasts time series with trend and seasonal components.

Back to Neural networks and deep learning ยท All topics

The interactive demonstration and the Python walkthrough for this topic need JavaScript.