Backpropagation (MLP)

Geoffrey Hinton, David Rumelhart & Ronald Williams, 1986

Popularized by Hinton, Rumelhart, and Williams in 1986, backpropagation made training deep networks practical by efficiently computing gradients layer by layer. The visualization shows a 3-layer network: data flows forward (gold) through input, hidden, and output layers, then error gradients flow backward (coral) adjusting weights. Teal indicates settled, updated weights. Each iteration refines the network's predictions.