back-propagation

A fundamental algorithm used for training neural networks, where gradients of loss are propagated backward through the network to update weights.