J. M. Morales González, A. Esteban Pérez
We deal with stochastic programs conditional on some covariates, where the only information on the relationship between these covariates and the input parameters of the stochastic program is reduced to a finite data sample of their joint distribution. We take advantage of the close link between the notion of trimmings of a probability measure and the partial mass transportation problem to construct a data-driven Distributionally Robust Optimization framework that hedges the decision maker against the intrinsic error in the process of inferring conditional information from limited joint data. We show that our approach naturally leads to distributionally robust versions of some local nonparametric predictive methods, such as Nadaraya-Watson kernel regression and K-nearest neighbors.
Keywords: Distributionally Robust Optimization, Trimmings, Side information, Partial Mass Transportation
Scheduled
GT04 Multivariate Analysis and Classification V. Mathematical Optimization and Data Science
June 8, 2022 12:40 PM
Cloister room