Rollout-Calibrated Hyper-Heuristics for Job Shop Scheduling
A recent preprint on arXiv (2605.23957) presents a learning-assisted hyper-heuristic aimed at solving the Job Shop Scheduling Problem (JSSP), focusing on issues of label cost and reliability. This innovative selector employs regret-normalized rollout labels and a contextual KNN uncertainty estimation, utilizing a gate mechanism that only deviates from a default rule when the predicted improvement surpasses an uncertainty-adjusted threshold. The research explores variations in rollout depth and breadth to evaluate trade-offs between cost and quality. In tests with synthetic JSSP instances, the gated selector demonstrates the lowest mean Relative Percentage Deviation (RPD) among learned selectors, closely aligning with the top fixed rule. Notably, this study does not pertain to art.
Key facts
- arXiv:2605.23957 introduces a learning-assisted hyper-heuristic for JSSP.
- The method uses regret-normalized rollout labels and contextual KNN uncertainty.
- A gate switches from a default rule only when improvement exceeds an uncertainty-adjusted margin.
- Rollout depth and breadth are varied to measure cost-quality trade-off.
- On synthetic JSSP instances, the gated selector achieves the lowest mean RPD among learned selectors.
- The work is not related to art.
Entities
—