Causal Inference for Social and Engineering Systems

What will happen to Y if we do A? A variety of meaningful social and engineering questions can be formulated this way: a patient's health they are given new therapy? country's economy policy-makers legislate tax? data center's latency congestion control protocol is used? We explore how answer such counterfactual using observational data-which increasingly available due digitization pervasive sensors-and/or very limited experimental data. The two key challenges are: (i) prediction in the presence latent confounders; (ii) estimation with modern datasets which high-dimensional, noisy, sparse. framework introduce connecting causal inference tensor completion. In particular, represent various potential outcomes (i.e., counterfactuals) interest through an order-3 tensor. theoretical results presented Formal identification establishing under what missingness patterns, confounding, structure on recovery unobserved possible. Introducing novel estimators recover these proving finite-sample consistent asymptotically normal. Finally, discuss connections between matrix/tensor completion time series analysis reinforcement learning; believe could serve as basis forecasting, building data-driven simulators for learning.