Decision-Focused Continual Learning for Seaport Power-Logistics Scheduling: Generalization across Varying Tasks
Researchers have developed a new decision-focused continual learning framework to improve power-logistics scheduling in seaports. This approach adapts online to a stream of varying scheduling tasks, addressing the poor generalization of existing methods. By using Fisher-information-based regularization and a differentiable convex surrogate, the framework enhances cross-task generalization while maintaining sustainable computational and memory requirements. Experiments at Jurong Port demonstrated improved decision performance and generalization compared to current methods. AI
IMPACT Enhances decision-making in logistics through adaptive AI, potentially improving efficiency in port operations.