Deep learning
Backpropagation
Backpropagation applies the chain rule to compute gradients through a computation graph.
How it works
It moves from the loss toward earlier operations while reusing intermediate derivatives. The result is one gradient for every trainable parameter.
A concrete example
Illustrative: if dy/dh = 3 and dh/dw = 0.4, then dy/dw = 1.2.
Do not confuse it with Gradient descent
backpropagation computes gradients; gradient descent applies updates.