Researchers at MIT and ETH Zurich have developed a machine learning-based technique that speeds up the optimization process used by companies like FedEx to deliver packages. This approach simplifies key steps in mixed integer linear programming (MILP) solvers and uses company-specific data to tune the process, resulting in 30-70% speedups without sacrificing accuracy. This has potential applications in a variety of industries facing complex resource allocation problems.
The research conducted by Massachusetts Institute of Technology and ETH Zurich aims to address complex logistics challenges, including delivering packages, distributing vaccines, and managing power grids. The traditional software used by companies like FedEx to find optimal delivery solutions is called a Mixed Integer Linear Programming (MILP) solver, but it can be time-consuming and may not always produce ideal solutions.
The newly developed technique employs machine learning to identify important intermediate steps in the MILP solver, resulting in a significant reduction of time required to unravel potential solutions. By using company-specific data, this approach allows for custom tailoring of the MILP solver. This new technique results in speeding up the MILP solver by 30-70% without sacrificing accuracy.
Lead author Kathy Wu, along with co-lead authors Sirui Li, Wenbin Ouyang, and Max Paulus, highlights the potential of combining machine learning and classical methods to address optimization problems. The research will be presented at the Neural Information Processing Systems Conference. The team hopes to further apply this approach to solve complex MILP problems and interpret the effectiveness of different separation algorithms.
Source: scitechdaily.com
