We investigate the problem of estimating the causal effect of a treatment on individual subjects from observational data; this is a central problem in various application domains, including healthcare, social sciences, and online advertising. We first develop a theoretical foundation of causal inference for individualized treatment effects based on information theory. Next, we use this theory, to construct an information-optimal Bayesian causal inference algorithm. This algorithm embeds the potential outcomes in a vector-valued reproducing kernel Hilbert space and uses a multi-task Gaussian process prior over that space to infer the individualized causal effects. We show that our algorithm significantly outperforms the state-of-the-art causal inference algorithms. The talk will conclude with a discussion of the impact of this work on precision medicine and clinical trials.
STS – Statistical scalability