This presentation covered early ideas on learning to branch in column-generation-based exact optimization. It laid out the core challenge that later became central in my subsequent work: how to use data-driven branching guidance in a setting where branching affects both the restricted master problem and the structure of the generated columns.

In retrospect, this talk served as an earlier step toward the later two-stage learning-to-branch framework and helped shape the direction that eventually led to the corresponding journal paper.