NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:7294
Title:Variance Reduction in Bipartite Experiments through Correlation Clustering

As pointed out by the reviewers, these are the strengths and weaknesses of the paper: STRENGTHS The paper proposes a method for bipartite experiment design building upon previous work on graph cluster randomization. It presents a novel optimization function and a heuristic to solve it. The method is shown to lead to balanced partitioning using empirical evaluation. The paper is well-written and easy to follow, and the reviewers had a positive discussion about it. FOR IMPROVEMENT The main concerns that need to be addressed to strengthen this paper include insufficient experimental comparisons (R1,R3), unclear connection made to optimal experimental design (R1), unclear relationship between clustering quality and estimate variance (R1).