Learning to Perform Local Rewriting for Combinatorial Optimization
This addresses the time-consuming process of heuristic tuning for combinatorial optimization problems, offering a generalizable approach with incremental improvements over existing methods.
The paper tackled the problem of tuning heuristics for combinatorial optimization by proposing NeuRewriter, a method that learns a policy to pick heuristics and rewrite local components iteratively, showing strong performance in tasks like expression simplification, online job scheduling, and vehicle routing problems, with specific improvements such as outperforming Z3, DeepRM, Google OR-tools, and neural baselines.
Search-based methods for hard combinatorial optimization are often guided by heuristics. Tuning heuristics in various conditions and situations is often time-consuming. In this paper, we propose NeuRewriter that learns a policy to pick heuristics and rewrite the local components of the current solution to iteratively improve it until convergence. The policy factorizes into a region-picking and a rule-picking component, each parameterized by a neural network trained with actor-critic methods in reinforcement learning. NeuRewriter captures the general structure of combinatorial problems and shows strong performance in three versatile tasks: expression simplification, online job scheduling and vehicle routing problems. NeuRewriter outperforms the expression simplification component in Z3; outperforms DeepRM and Google OR-tools in online job scheduling; and outperforms recent neural baselines and Google OR-tools in vehicle routing problems.