Learning 2-Opt Heuristics for Routing Problems via Deep Reinforcement Learning

Abstract Recent works using deep learning to solve routing problems such as the traveling salesman problem (TSP) have focused on construction heuristics. Such approaches find good quality solutions but require additional procedures beam search and sampling improve achieve state-of-the-art performance. However, few studies improvement heuristics, where a given solution is improved until reaching near-optimal one. In this work, we propose learn local heuristic based 2-opt operators via reinforcement learning. We policy gradient algorithm stochastic that selects operations current solution. Moreover, introduce neural network leverages pointing attention mechanism, which can be easily extended more general k -opt moves. Our results show learned policies even over random initial approach faster than previous methods for TSP. also adapt proposed method two extensions of TSP: multiple TSP Vehicle Routing Problem, achieving par with classical heuristics methods.