Abstract
Undertaking causal inference with observational data is extremely usefulacross a wide range of domains including the development of medical treatments,advertisements and marketing, and policy making. There are two main challengesassociated with undertaking causal inference using observational data:treatment assignment heterogeneity (i.e., differences between the treated anduntreated groups), and an absence of counterfactual data (i.e. not knowing whatwould have happened if an individual who did get treatment, were instead tohave not been treated). We address these two challenges by combining structuredinference and targeted learning. To our knowledge, Targeted VariationalAutoEncoder (TVAE) is the first method to incorporate targeted learning intodeep latent variable models. Results demonstrate competitive and state of theart performance.