NeurIPS 2020

Demystifying Orthogonal Monte Carlo and Beyond

Meta Review

The work presents theory behind the behavior of OMC methods based on negatively dependent random variables, which yields tighter bounds than MC, and the work also establishes uniform convergence bounds. The reviewers appreciated the theoretical development for OMC, the results appear new and carefully done. The work also proposed NOMC which has empirically promising results. Reviewers have concerns about the proposed NOMC, including missed connections with closely related literature, unclear connections with the theory of OMC established in the paper, no evidence of empirical evidence in downstream applications. The author response addressed some concerns, but some concerns persisted. The work will be strengthened by addressing these concerns.