Evolving Variable-Ordering Heuristics for Constrained Optimisation

In this paper we present and evaluate an evolutionary approach for learning new constraint satisfaction algorithms, specifically for MAX-SAT optimisation problems. Our approach offers two significant advantages over existing methods: it allows the evolution of more complex combinations of heuristics, and; it can identify fruitful synergies among heuristics. Using four different classes of MAX-SAT problems, we experimentally demonstrate that algorithms evolved with this method exhibit superior performance in comparison to general purpose methods.