Low-Cost Labels, Reliable Choices: Rollout-Calibrated Hyper-Heuristics for Job Shop Scheduling
Researchers have developed a new method for job shop scheduling that uses learning-assisted hyper-heuristics to select among dispatching rules. This approach aims to reduce the computational cost of generating labels, which is typically the most expensive part of the process. The system incorporates a gate that only switches from a default rule when the predicted gain is credible, using regret-normalized rollout labels and uncertainty estimates. Experiments on synthetic instances showed this method achieved significantly lower mean RPD compared to other learned selectors and reduced the RPD of random hyper-heuristics by over an order of magnitude. AI
IMPACT Introduces a novel algorithmic approach for optimizing complex scheduling tasks, potentially improving efficiency in manufacturing and logistics.