Balanced Policy Evaluation and Learning

Nathan Kallus

Neural Information Processing Systems 

We present a new approach to the problems of evaluating and learning personalized decision policies from observational data of past contexts, decisions, and outcomes.