This is the practice of training another copy model from an existing model. Done by training off sample inputs and learning from the outputs of the original