Machine Learning for Integer Programming

Mixed Integer Programs (MIP) are solved exactly by tree-based branch-and-bound search. However, various components of the algorithm involve making decisions that are currently addressed heuristically. Instead, I propose to use machine learning (ML) approaches such as supervised ranking and multi-armed bandits to make better-informed, input-specific decisions during MIP branch-and-bound. My thesis aims at improving the overall performance of MIP solvers. To illustrate the potential for ML in MIP, I have so far tackled branching variable selection, a crucial component of the search procedure, showing that ML approaches for variable selection can outperform traditional heuristics.

Authors

• Elias B. Khalil Department of Mechanical and Industrial Engineering, University of Toronto

Keywords

No keywords are indexed for this paper.