NeurIPS 2020

SnapBoost: A Heterogeneous Boosting Machine


Meta Review

This paper describes a boosting framework where for each iteration the classifier is randomly selected from a given set of classifiers. The paper provides some theoretical guarantees, and an empirical evaluation with some good results. Judging the results, it is really the version that combines trees and a linear regressor over random Fourier features, that works best. I suppose this supports the original idea, but will also warrant a more detailed future investigation.