gradient

Partial derivatives generalise the concept of the derivative to functions of multiple variables by quantifying the rate of change with respect to a single variable while holding the others constant.

Backpropagation is the algorithm used to train neural networks by computing the gradient of the loss function with respect to each parameter.