Backpropagation is a training method for neural networks that calculates how much each model weight contributed to an error and uses that information to adjust the weights.
It is a foundation of modern deep learning because it enables layered models to improve through repeated examples and optimization steps.