In my understanding, it is not multiple regression. Only one regression takes place here. In machine leaning, each column in the training set is defined as a variable. This algorithm should work on a dataset having multiple variables. That's why I called it multivariate linear regression. I hope it makes sense.