What is Jacobian matrix and why is it critical in deep learning?

Jacobian represents all partial derivatives of vector-valued functions.

Jacobian generalizes gradients to vector outputs. It describes how small changes in input affect each output dimension, forming the basis of backpropagation.