Lifted branching: Learning to improve branching strategies

2026-08-04

PRE

Simon Renard *

Quentin Louveaux

Bernard Fortz

DOI：10.1016/j.ejor.2026.07.046

Original

Original request for help

Save

Abstract

En 中文

• Novel ML framework iteratively improves existing MILP branching strategies. • Lifted branching yields smaller search trees and keeps decision times low. • Learned models can outperform state-of-the-art standard and imitation methods. • The proposed approach excels when problems share structural similarities.

Keywords:

Branch-and-bound

Integer programming

Machine learning in OR

Lifted branching: Learning to improve branching strategies

Abstract

Journal

Organization

Cited Papers

Citing Papers