Neural Networks for Causal Inference: Nonparametric and Semiparametric Approaches

This thesis develops new statistical methods for estimating the effects of policies or treatments using modern artificial intelligence tools. In many real-world studies, researchers must adjust for many background factors, which makes reliable inference difficult. This thesis shows how deep neural networks can be combined with carefully designed statistical techniques to produce accurate and trustworthy estimates of treatment effects. It also proposes new ways to use data more efficiently when training these models. The results help bridge the gap between powerful machine-learning prediction methods and the rigorous standards needed for scientific and policy-relevant conclusions.