Prehistory — From Neural Networks to the TransformerSepp Hochreiter, Jürgen Schmidhuber

Hochreiter and Schmidhuber publish Long Short-Term Memory

Published Long Short-Term Memory — Neural Computation 9(8) 1735–1780

Why it matters

Used gates to mitigate vanishing gradients, making it possible to learn long sequences. The standard sequence model before the Transformer

Source — Primary sources first; secondary reports are labelled “Press”

Officialbioinf.jku.athttps://www.bioinf.jku.at/publications/older/2604.pdf

Added in Week 35, 2026 · See that week

Nearby on the same axis

Gradient-based learning applied to document recognitionPressLearning long-term dependencies with gradient descent is difficultPressA Neural Probabilistic Language ModelOfficialHochreiter's TU Munich thesis identifies the vanishing gradient problemPress