My goal is to understand and improve the algorithms that researchers can use to learn about the structure of an underlying system from data and reason about intervening in such a system.
Structure learning and hypothesis testing -- What we assume about the underlying causal structure affects how we understand data and the association between variables. I am interested in learning from data alone the invariant features of the underlying causal graph, with a particular interest in problems involve complex data types. In recent work, we showed how to learn local independence graphs from irregularly-sampled time series data.
Causal inference -- With knowledge of the causal graph, quantifying the effect of an intervention or counterfactual can be equally challenging. I am also motivated by treatment effect estimation problems and questions of transportability asking under what conditions a particular effect can be computed in a target population. Representative work in this direction include algorithms for individualized treatment effect estimation with discrete, continuous, and longitudinal treatments, and bounding counterfactuals in target domains.
Application of machine learning in health -- A move towards data-driven decision making in healthcare has the potential to revolutionize patient care. I am interested in general machine learning applications in healthcare, especially those that incorporate causality into their reasoning. Together with colleagues we have worked on several survival prediction algorithms in this area.