Meta-Optimization
Abstract
This grant sought to investigate the use of meta-optimization in tackling broader classes of optimization problems, and solving existing optimization problems more efficiently. The research teams ultimate goal is to build solvers that learn over time, that exploit the common features found in repeated optimization problems, and that take account of the rich structure found in real world optimization. They investigated a number of approaches to improving solving using meta-optimization by (a) making use of a subsolver for particular class of important constraints (difference logic), (b) adding sub-problems that reason about objective bounds more effectively, (c) automating a logic-based Benders approach of problem decomposition and (d) exploiting the online nature of many natural resource allocation problems. They also tackled a number of important scheduling problems, and redefined the state-of-the-art for each of these problems. Their results have been presented in multiple peer-reviewed archival journals throughout the project duration.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 14, 2019
- Accession Number
- AD1085914
Entities
People
- Toby Walsh