Backpropagation for training neural networks
Retropropagación para entrenar redes neuronales profundas
- Date
- 1986 · confidence high
- Place
- UC San Diego and Carnegie Mellon, United States
- Credited to
- David Rumelhart, Geoffrey Hinton and Ronald Williams (1986); Seppo Linnainmaa (1970) and Paul Werbos (1974) as precursors
- People
- 👤 David Rumelhart1942–2011👤 Geoffrey Hinton1947–👤 Paul Werbos1947–👤 Ron Williams1944–2004👤 Seppo Linnainmaa1945–
- Field
- Knowledge & symbols
⚠️ Disputed: the sources disagree.
What it is
Algorithm that efficiently computes how to adjust each weight of a multi-layer network by propagating the error backwards. It is the training engine of all deep learning.
Linnainmaa (1970) and Werbos (1974) described the method earlier; the 1986 paper popularised it for neural networks.
What it led to
- Connectionism and the revival of neural networks
- Deep learning from 2012 onwards
- Large language models