Strategyproof Mechanisms for Facility Location with Prediction Under the Maximum Cost Objective

Abstract: We study the mechanism design problem of facility location on a metric space in the learning-augmented framework, where mechanisms have access to an imperfect prediction of optimal facility locations. Our goal is to design strategyproof (SP) mechanisms to elicit agent preferences on the facility locations truthfully and, leveraging the given imperfect prediction, determine the facility location that approximately minimizes the maximum cost among all agents. In particular, we seek SP mechanisms whose approximation guarantees depend on the prediction errors -- achieve improved guarantees when the prediction is accurate (known as the \emph{consistency}), while still ensuring robust worst-case performance when the prediction is arbitrarily inaccurate (known as the \emph{robustness}). When the metric space is the real line, we characterize all deterministic SP mechanisms with consistency strictly less than 2 and bounded robustness: such mechanisms must be the MinMaxP mechanism, which returns the prediction location if it lies between the two extreme agent locations and, otherwise, returns the closest agent location to the prediction. We further show that, for any prediction error $\eta\ge 0$, while MinMaxP is $(1+\min(1, \eta))$-approximation, no deterministic SP mechanism can achieve a better approximation. In two-dimensional spaces with the $l_p$ metric, we analyze the approximation guarantees of a deterministic mechanism that runs MinMaxP independently on each coordinate, as well as a randomized mechanism that selects between two deterministic ones with specific probabilities. Finally, we discuss the group strategyproofness of the considered mechanisms.