Long short-term memory (LSTM) networks
Redes LSTM de memoria a largo plazo
- Date
- 1997 · confidence high
- Place
- Technical University of Munich, Germany, and IDSIA, Lugano, Switzerland
- Credited to
- Sepp Hochreiter and Jürgen Schmidhuber
- People
- 👤 Jürgen Schmidhuber1963–👤 Sepp Hochreiter1967–
- Field
- Knowledge & symbols
What it is
Recurrent network architecture with gates that remembers information over long sequences. It dominated speech recognition and machine translation until the transformer.
What it led to
- The speech recognition of Siri, Alexa and Google
- Neural machine translation (Google Translate, 2016)
- The transformer as its successor
Built on
- Backpropagation for training neural networks1986
- Vanishing gradient problem1991
- recurrent neural networks
- vanishing gradient problem
Made possible
- Gated recurrent unit2014
- neural speech recognition
- neural machine translation