Data-driven inverse optimization with imperfect information (Q681497)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Data-driven inverse optimization with imperfect information
scientific article

    Statements

    Data-driven inverse optimization with imperfect information (English)
    0 references
    12 February 2018
    0 references
    Roughly speaking, the inverse optimization problem consists in selecting the objective function, within a certain given family of candidates, of an unknown parametric optimization problem when only a set of parameters and optimal decisions are known. More precisely, in this paper, the authors assume that an observer aims to learn the preferences of an agent who solves a parametric optimization problem depending on an exogenous signal. The observer knows the constraints imposed on the agent's actions but is unaware of his objective function. By monitoring a sequence of signals and corresponding actions, the observer seeks to identify an objective function that makes the observed actions optimal in the agent's optimization problem. The hope is that the solution of this inverse problem enables the observer to predict the agent's future actions in response to new signals. To do this, the authors introduce a new suboptimality loss, which quantifies the degree of suboptimality of an observed response under a given candidate objective, and formulate a distributionally robust inverse optimization problem that minimizes the worst-case risk of loss for any combination of a risk measure with a loss function, where the worst case is taken over all distributions in the Wasserstein ball. Tractable reformulations are given for distributionally robust inverse optimization problems over search spaces of linear and quadratic candidate objectives. Numerical results are also reported.
    0 references
    stochastic programming
    0 references
    convex programming
    0 references
    minimax problems
    0 references
    0 references
    0 references
    0 references
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references