NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:8728
Title:Machine Learning Estimation of Heterogeneous Treatment Effects with Instruments

This paper solves an open problem, giving for the first time a provable method for efficiently using arbitrary ML algorithms for estimating heterogeneous effects in the instrumental variable scenario. Substantial experiments on several large datasets very nicely tie together theory to practice, in important and meaningful application areas. The reviewers have several proposal for making the paper clearer, which I trust the authors will follow. An important issue that must be clarified is the role of the monotonicity assumption given informally in the second paragraph of the paper.