This looks useful. I wonder if the concept could be used to make predictions about data instead of just preparing it. ie, to interpolate likely missing values

This looks useful. I wonder if the concept could be used to make predictions about data instead of just preparing it. ie, to interpolate likely missing values like in collaborative filtering? There are probably several datasets that could be explained with this kind of scalable weak semi supervision