in convex optimisation, often machine learning, gradients (partial derivatives) are used. In deep learning an approximation of the chain rule is used in backpropigation.