Stochastic gradient descent is a method of performing gradient descent in which the parameters of the algorithm are updated for every sample in the training set.