Abstract:Utility preference robust optimization (PRO) has recently been proposed to deal with optimal decision-making problems where the decision maker's (DM's) preference over gains and losses is ambiguous. In this paper, we take a step further to investigate the case that the DM's preference is random. We propose to use a random utility function to describe the DM's preference and develop distributional utility preference robust optimization (DUPRO) models when the distribution of the random utility function is ambiguous. We concentrate on data-driven problems where samples of the random parameters are obtainable but the sample size may be relatively small. In the case when the random utility functions are of piecewise linear structure, we propose a bootstrap method to construct the ambiguity set and demonstrate how the resulting DUPRO can be solved by a mixed-integer linear program. The piecewise linear structure is versatile in its ability to incorporate classical non-parametric utility assessment methods into the sample generation of a random utility function. Next, we expand the proposed DUPRO models and computational schemes to address general cases where the random utility functions are not necessarily piecewise linear. We show how the DUPRO models with piecewise linear random utility functions can serve as approximations for the DUPRO models with general random utility functions and allow us to quantify the approximation errors. Finally, we carry out some performance studies of the proposed bootstrap-based DUPRO model and report the preliminary numerical test results. This paper is the first attempt to use distributionally robust optimization methods for PRO problems.

Discrete choice prox-functions on the simplex

Preference Robust Optimization with Quasi-Concave Choice Functions in Multi-Attribute Decision-Making: Characterization and Computation

A proximal average for prox-bounded functions

Optimal Design for Probit Choice Models with Dependent Utilities

Analysis of Discrete Choice Models: A Welfare-Based Framework

A Proximal-Gradient Method for Constrained Optimization

Convex optimization over a probability simplex

On Convergence Rates of Linearized Proximal Algorithms for Convex Composite Optimization with Applications.

Various Notions of Nonexpansiveness Coincide for Proximal Mappings of Functions

Proximal determination of convex functions

An extension of the proximal point algorithm beyond convexity

Gaining or losing perspective for convex multivariate functions on a simplex

Variational Analysis of Proximal Compositions and Integral Proximal Mixtures

Proximal Algorithms for a class of abstract convex functions

Distributional utility preference robust optimization models in multi-attribute decision making

A single cut proximal bundle method for stochastic convex composite optimization

Probabilistic risk aversion for generalized rank-dependent functions

Surrogate Maximization/minimization Algorithms and Extensions

Universal subgradient and proximal bundle methods for convex and strongly convex hybrid composite optimization

Single-Crossing Differences in Convex Environments