Title: Assisted Policy Search
Advisor: Professor Susan A. Murphy
Committee Members: Assistant Professor Shuheng Zhou, Professor Satinder Singh
Abstract: The estimation of optimal treatment policies using experimental or observational data is an important problem in causal inference and in health research. A number of Sequential Multiple Assignment Randomized Trials have been conducted and provide high quality data for use constructing treatment policies. In some cases, scientists propose a parameterized policy class, over which the optimal policy is searched. We would like to accomplish this goal by utilizing the structural nested mean models, which posit the treatment effect models at all stages, independently from the policy class that we will be searching from. We consider efficient approaches to conduct the policy search. Future work will involve statistical inference on the estimated optimal policy, and extension of this methodology to observational data.