NeurIPS 2020

Learning Causal Effects via Weighted Empirical Risk Minimization


Meta Review

The paper applies weighted ERM to learn the causal functionals given the graph. The proposed framework converts the causal effect into weighted distribution without conditional ignorability assumption. We agree that learning causal effect without ignorability is very interesting and could lead to further advances in causal inference. But, the clarity of the theoretical analysis and the experimental results received much criticism during the discussions. Overall, the theoretical contributions are strong, but the paper needs a lot of work to improve its clarity. Also, the material from the rebuttal should be incorporated into the main paper to strengthen the motivation and the connections with prior work.