The objective function is a fundamental construct that underpins applications of optimization methods. This project will develop a novel computational theory of subjective functions; that is, objective functions that map decision variables to variables that estimate how desirable decisions are for an individual in a context. The main goal of the project is a computational theory and workflow for creating subjective functions which yield accurate and robust estimates of how 'good' alternative decisions are for an individual person. Our hypothesis that subjective functions could be achieved if we find a way to amortize both the forward (prediction) and inverse (model parameter fitting) simulations of users. The project will study amortization as a method that would allow fast execution end-to-end from observation data to desirability estimates. Results are evaluated in four challenging demonstrator cases.

2023

2027