Type of backpropagation where we use a fixed number of training examples which is lees than the actual dataset to perform a calculation of the gradient

Mini-batch Gradient Descent.