Jacobian Matrix

A matrix describing the first-order partial derivatives of a multivariate vector-valued function. In deep learning it is used to analyze the sensitivity of model outputs to inputs, and it is commonly applied in adversarial attack and defense research.