This talk explains how learning can be integrated into branch-price-and-cut algorithms for solving vehicle routing problems exactly. It focuses on the technical difficulty of branching in column-generation-based solvers, where variables in the restricted master problem change dynamically and standard learning-to-branch ideas are harder to apply directly.

I first presented this line of work at the 2023 INFORMS Annual Meeting and later gave an invited version at the 2025 UK AI/ML Symposium and Nontechnical Workshop. Across both talks, the central message was the same: learning can improve branching efficiency without giving up exactness.