Date of Completion

12-15-2016

Embargo Period

12-15-2016

Keywords

Constraint Programming, Operations Research, Hybrid Automation

Major Advisor

Laurent Michel

Associate Advisor

Alexander Russell

Associate Advisor

Alexander Schwarzmann

Field of Study

Computer Science and Engineering

Degree

Doctor of Philosophy

Open Access

Open Access

Abstract

Many critical real world problems, including problems in areas such as logistics, routing and scheduling are very difficult to solve computationally (often NP-hard). Various programming and algorithmic paradigms have been developed to deal with these problems, including Constraint Programming (CP), Integer Programming (IP) and Local Search (LS). These technologies are largely declarative in nature and rely on vastly different underlying mathematical and algorithmic approaches. Hence, each paradigm has inherent strengths and weaknesses making it more or less suitable to a given problem. For particularly difficult problems, it can often be beneficial to leverage a sophisticated “hybrid solver” technique. Such techniques include, for example, combining CP and IP solvers in a cooperative fashion or iteratively solving problem relaxations such as in Lagrangian Relaxation, Column Generation or Logic-Based Benders Decomposition. The development of such hybrids, however, is often technically difficult and requires a great deal of trial and error. This thesis introduces a new high-level framework for automating the generation of several important classes of hybrid solvers as well as proposing a new set of theoretical abstractions allowing high-level model descriptions to be transformed and combined into hybrids while maintaining semantic soundness. Among the new theoretical abstractions is a proposal for `Generic Lagrangian Relaxation', allowing a well-known Integer Programming technique to be generalized and applied to other technologies such as CP. Experimental results demonstrate the practical benefits of this new framework.

Download
COinS