Multilayer perceptron
Perceptrón multicapa
- Date
- c. 1986 · confidence low
- Place
- San Diego, California, United States
- Credited to
- David Rumelhart, Geoffrey Hinton and Ronald Williams
- Field
- Knowledge & symbols
⚠️ Disputed: the sources disagree.
What it is
An artificial neural network with several layers of fully connected neurons and nonlinear activation functions, able to learn patterns that a simple perceptron cannot linearly separate. It became practical once trained with the backpropagation algorithm and underlies deep learning.
The exact attribution is debated: the trainable multilayer network concept rests on the backpropagation algorithm described by Paul Werbos (1974) and popularised by Rumelhart, Hinton and Williams (1986); Wikidata mistakenly links this entity to Frank Rosenblatt, creator of the single-layer perceptron.
What it led to
- popularisation of backpropagation training in the 1980s
- theoretical basis of modern deep learning
- application to a wide range of pattern-recognition tasks