Representation Learning in Underspecified Problems via Decision Feedback

Decision support systems are characterized by underspecified problems in which an executive decision maker needs access to simplified information. By definition, optimal presentation of information is unknown a priori. Over a sequence of decisions, we can refine a model of information presentation, transitioning from a generative model to a maximum a posteriori solution. This project will analyze theoretically and empirically how to construct such systems for real world problems including decision making from images and video.