Machine learning · 1986 · David E. Rumelhart, Geoffrey E. Hinton & Ronald J. Williams
Learning Representations by Back-Propagating Errors
Popularize efficient layer-by-layer credit assignment with the chain rule so hidden units can learn useful internal representations.
The central move
Popularize efficient layer-by-layer credit assignment with the chain rule so hidden units can learn useful internal representations.
Why it had to exist
Multi-layer networks could represent complex functions, but adjusting early weights required assigning output error through every intermediate transformation.
Where it leads
Backpropagation → deep representation learning → convolutional and transformer models.