A smart hybrid memory scheduling approach using neural models

Conclusion SmartS is a novel solution for hybrid memory scheduling using neural models. It proposes a novel collective-page prediction approach, effectively reducing training and inference costs. It also proposes a clustering-based approach to address the class explosion problem. Experiments show that SmartS improves hybrid memory effectiveness significantly. It also reduces the cost of neural models to allow their practical deployment in real-world hybrid, representing a substantial step towards practical neural-model-based scheduling.