Abstract:
A high-level abstract-datatype-based constraint modelling language opens the door to an automatable empirical determination -- by a synthesiser -- of how to 'best' represent the decision variables of a combinatorial optimisation problem, based on (real-life) training instances of the problem. the extreme case where no such training instances are provided, such a synthesiser would simply be non-deterministic. A first-order relational calculus is a good candidate for such a language, as it gives rise to very natural and easy-to-maintain models of combinatorial optimisation problems.