Backpropagation is the algorithm that computes, efficiently and exactly, how much each weight in a neural network contributed to the output error: propagating t
The algorithm that computes, efficiently and exactly, how much each weight in a neural network contributed to the output error, propagating the error signal backward through the layers so gradient descent knows how to adjust every parameter.
It is the mathematical engine of deep learning: it is what makes training networks with billions of weights computationally feasible at all. Without efficient backpropagation, modern large models could not be trained.