Pick-to-Learn Calibration of an MPC Policy for an Origin-to-Destination Flight Problem

What the math gives to ML

The paper contributes a scenario-compression mechanism for calibrating a parameterized MPC policy under uncertain future realizations. Pick-to-Learn selects a small compression set of informative scenarios, calibrates the policy using that set, and attaches a finite-sample probabilistic certificate to future constraint satisfaction. In the reported experiment, 400 wind scenarios are reduced to 2 informative scenarios while certifying a violation risk of 4.8 percent at confidence 1 minus 10 to the power of minus 5. The transferable neural-network mechanism is to select a small, data-dependent set of difficult environments, perturbations, or constraint cases for tuning architecture and optimizer hyperparameters while retaining an explicit scenario-based bound on future constraint violations.