"Learning with Subset Stacking" is a supervised learning algorithm that is based on training many local estimators on subsets of a given dataset, and then passing their predictions to a global estimator. You can find the details about LESS in our manuscript at
Learning in Subset Stacking (R version)