Saved in:
Bibliographic Details
Main Authors: Strang, Paul, Alès, Zacharie, Bissuel, Côme, Juan, Olivier, Kedad-Sidhoum, Safia, Rachelson, Emmanuel
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.19348
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912664010096640
author Strang, Paul
Alès, Zacharie
Bissuel, Côme
Juan, Olivier
Kedad-Sidhoum, Safia
Rachelson, Emmanuel
author_facet Strang, Paul
Alès, Zacharie
Bissuel, Côme
Juan, Olivier
Kedad-Sidhoum, Safia
Rachelson, Emmanuel
contents Mixed-Integer Linear Programming (MILP) is a powerful framework used to address a wide range of NP-hard combinatorial optimization problems, often solved by Branch and Bound (B&B). A key factor influencing the performance of B&B solvers is the variable selection heuristic governing branching decisions. Recent contributions have sought to adapt reinforcement learning (RL) algorithms to the B&B setting to learn optimal branching policies, through Markov Decision Processes (MDP) inspired formulations, and ad hoc convergence theorems and algorithms. In this work, we introduce BBMDP, a principled vanilla MDP formulation for variable selection in B&B, allowing to leverage a broad range of RL algorithms for the purpose of learning optimal B\&B heuristics. Computational experiments validate our model empirically, as our branching agent outperforms prior state-of-the-art RL agents on four standard MILP benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2510_19348
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Markov Decision Process for Variable Selection in Branch & Bound
Strang, Paul
Alès, Zacharie
Bissuel, Côme
Juan, Olivier
Kedad-Sidhoum, Safia
Rachelson, Emmanuel
Machine Learning
Mixed-Integer Linear Programming (MILP) is a powerful framework used to address a wide range of NP-hard combinatorial optimization problems, often solved by Branch and Bound (B&B). A key factor influencing the performance of B&B solvers is the variable selection heuristic governing branching decisions. Recent contributions have sought to adapt reinforcement learning (RL) algorithms to the B&B setting to learn optimal branching policies, through Markov Decision Processes (MDP) inspired formulations, and ad hoc convergence theorems and algorithms. In this work, we introduce BBMDP, a principled vanilla MDP formulation for variable selection in B&B, allowing to leverage a broad range of RL algorithms for the purpose of learning optimal B\&B heuristics. Computational experiments validate our model empirically, as our branching agent outperforms prior state-of-the-art RL agents on four standard MILP benchmarks.
title A Markov Decision Process for Variable Selection in Branch & Bound
topic Machine Learning
url https://arxiv.org/abs/2510.19348