<ul><li>This study focuses on an extension of contextual stochastic linear optimization involving inequality constraints with uncertain parameters predicted by a machine learning model.</li><li>Contextual uncertainty sets constructed using methods like conformal prediction are used to handle constraint uncertainty.</li><li>The study introduces the 'Smart Predict-then-Optimize with Robust Constraints' (SPO-RC) loss, a feasibility-sensitive adaptation measuring decision error of predicted objective parameters.</li><li>Experiments conducted on fractional knapsack and alloy production problem instances show that SPO-RC+ effectively deals with constraints uncertainty, and combining truncation with importance reweighting can further enhance performance.</li></ul>

Smart Surrogate Losses for Contextual Stochastic Linear Optimization with Robust Constraints

Discover more