Learning within Optimization
Abstract
Combinatorial optimization problems are ubiquitous and important. Such problems arise anywhere a set of interlinking decisions need to be made. For example, in a transportation problem, we might want to determine which trucks will deliver which goods in which order to minimize costs. Learning is an important but neglected aspect of combinatorial optimization. We want our solver to learn how to solve such optimization problems. In addition, as the world is dynamic and uncertain, we want our solver to learn how the constraints of our problem change. The goal then of this project was to develop a theoretical and practical understanding of how we can combine learning within optimization. In particular, we developed a powerful learning based search method called lazy-clause generation. This has now been shown to advance the state of the art in several domains, especially with the optimization subfield of scheduling.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 18, 2013
- Accession Number
- ADA575367
Entities
People
- Peter Stuckey
- Toby Walsh