Synthesis of Constraint-Based Local Search Algorithms from High-Level Models

The gap in automation between MIP/SAT solvers and those for constraint programming and constraint-based local search hinders experimentation and adoption of these technologies and slows down scientific progress. This paper addresses this important issue: It shows how effective local search procedures can be automatically synthesized from models expressed in a rich constraint language. The synthesizer analyzes the model and derives the local search algorithm for a specific meta-heuristic by exploiting the structure of the model and the constraint semantics. Experimental results suggest that the synthesized procedures only induce a small loss in efficiency on a variety of realistic applications in sequencing, resource allocation, and facility location