It's funny how the article just assumes PIM be for computing weights of a model without even mentioning it in the introduction.