Learning to Control Local Search for Combinatorial Optimization

Combinatorial optimization problems are encountered in many practical contexts such as logistics and production, but exact solutions particularly difficult to find usually NP-hard for considerable problem sizes. To compute approximate solutions, a zoo of generic well problem-specific variants local search is commonly used. However, which variant apply particular decide even experts. In this paper we identify three independent algorithmic aspects algorithms formalize their sequential selection over an process Markov Decision Process (MDP). We design deep graph neural network policy model MDP, yielding learned controller called NeuroLS. Ample experimental evidence shows that NeuroLS able outperform both, well-known general purpose controllers from Operations Research latest machine learning-based approaches.