Backpropagation
Also: Fehlerrückführung
The method a neural network uses to calculate how much each weight contributed to the error.
After a prediction, the error is propagated backwards from the output through all layers of the network. This tells each weight in which direction it must be adjusted to reduce the error.
Together with gradient descent, backpropagation is at the core of training practically all modern neural networks.
General information, not legal or professional advice.