This is known as a generative approach since by sampling from the joint distribution it is possible to generate synthetic examples of the feature vector x.
