Learning to Break Symmetries for Efficient Optimization in Answer Set Programming

The ability to efficiently solve hard combinatorial optimization problems is a key prerequisite various applications of declarative programming paradigms. Symmetries in solution candidates pose significant challenge modern algorithms since the enumeration such might substantially reduce their performance. This paper proposes novel approach using Inductive Logic Programming (ILP) lift symmetry-breaking constraints for modeled Answer Set (ASP). Given an ASP encoding with statements and set small representative instances, our method augments ground programs auxiliary normal rules enabling identification symmetries existing tools, like SBASS. Then, obtained are lifted first-order ILP. We prove correctness evaluate it on real-world from domain automated configuration. Our experiments show improvements performance due learned constraints.