Reviews: Union of Intersections (UoI) for Interpretable Data Driven Discovery and Prediction

This paper focuses on model selection and, to some extent, feature selection in large datasets with many features, of which only a small subset are assumed to be necessary for accurate prediction. The authors propose a general method by which model selection is performed by way of feature compression performed by taking the intersection of a multiple regularization parameters in an ensemble method, and then model estimation by taking a union over multiple outputs. A second contribution is found in the union operation in a model averaging step with a boosting/bagging flavor. Overall, I found the paper's method section well written and the idea proposed to be complete. The paper's experimental section was difficult to follow, but the results do seem to support the framework.