A Classifier to Decide on the Linearization of Mixed-Integer Quadratic Problems in CPLEX

• 成果类型: Article
• 署名作者: Bonami, Pierre; Lodi, Andrea; Zarpellon, Giulia
• 署名单位: Universite de Montreal; Polytechnique Montreal; Vector Institute for Artificial Intelligence
• 刊物名称: OPERATIONS RESEARCH
• ISSN/ISSBN: 0030-364X
• DOI: 10.1287/opre.2022.2267
• 发表日期: 2022
• 页码: 3303-3320
• 关键词: library

摘要

With the aim of fully embedding learned predictions in the algorithmic design of a mixed-integer quadratic programming (MIQP) solver, we translate the algorithmic question of whether to linearize convex MIQPs into a classification task and use machine learning (ML) techniques to tackle it. We represent MIQPs and the linearization decision by careful target and feature engineering. Computational experiments and evaluation metrics are designed to further incorporate the optimization knowledge in the learning pipeline. As a practical result, a classifier deciding on MIQP linearization is successfully deployed in CPLEX 12.10.0: to the best of our knowledge, we establish the first example of an end-to-end integration ofML into a commercial optimization solver and ultimately contribute a general-purpose methodology for combining ML-based decisions and mixedinteger programming technology.