Causal Inference and Representation Learning with Observational Data

The dramatically growing availability of observational data is being witnessed in various domains of science and technology, which facilitates the study of causal inference. Compared with randomized controlled trials, causal inference from observational data has become an appealing research direction owing to a large amount of available data and low budget requirement for its collection. In particular, the success of representation learning inspires advanced methods for learning causal effects with observational data. However, some issues around the causal effect estimation are still challenging, such as missing counterfactual outcomes, treatment selection bias, lack of interpretability and explainability, inclusion of various covariate types, hidden confounders, difficulty of continual learning for incrementally available observational data, etc. This dissertation provides a comprehensive review of existing causal inference methods and proposes several novel approaches based on representation learning to solve these issues.