NeurIPS 2020

Reconsidering Generative Objectives For Counterfactual Reasoning

Meta Review

The reviewers appreciated the contribution and clarity of the paper. In the discussion the main things the reviewers wanted to see is: (a) Better clarification of the discussion on identification, with clear examples where latent variable identification is possible; (b) Some updated comparisons with CFR-WASS or other modern methods based on IHDP and ACIC; (c) An ablation study showing how each new component helps. If the authors could add these things to the final version it would further strengthen the paper. I vote to accept.