Review for NeurIPS paper: SnapBoost: A Heterogeneous Boosting Machine

Strengths: Combining several learner classes has been a common technique in practical boosting and ensemble methods in general, since it ensures a better diversity among the base classifiers, hence better performance. While the empirical results shown in this paper are not surprising to any ensemble learning practitioner, the strength of this work resides in providing a full theoretical setting for understanding and analyzing heterogeneous base learners. To the best of my knowledge, HNBM is the first framework that provides a clear theoretical insight on heterogeneous learners which englobes several learning paradigms, from heterogeneous data/attributes, to multi-view/multi-source learning. This by itself makes this contribution of significant interest for all the ML community. In particular, HNBM opens up several research questions (different probability mass functions, theoretical aspects of diversity in ensemble learning, etc.).