Estimating the form of a decision maker's preference function and converging towards preferred solutions

Preference functions have been widely used to scalarize multiple objectives. Various forms such as linear, quasiconcave, or general monotone have been assumed. In this article, we consider a general family of functions that can take a variety of forms and has properties that allow for estimating the form efficiently. We exploit these properties to estimate the form of the function and converge towards a preferred solution(s). We develop the theory and algorithms to efficiently estimate the parameters of the function that best represent a decision maker's preferences. This in turn facilitates fast convergence to preferred solutions. We demonstrate on a variety of experiments that the algorithms work well both in estimating the form of the preference function and converging to preferred solutions.