Safe Learning for Near-Optimal Scheduling

In this paper, we investigate the combination of synthesis, model-based learning, and online sampling techniques to obtain safe near-optimal schedulers for a preemptible task scheduling problem. Our algorithms can handle Markov decision processes (MDPs) that have $$10^{20}$$ states beyond which cannot be handled with state-of-the art probabilistic model-checkers. We provide probably approximately correct (PAC) guarantees learning model. Additionally, extend Monte-Carlo tree search advice, computed using safety games or obtained earliest-deadline-first scheduler, safely explore learned model online. Finally, implemented compared our empirically against shielded deep Q-learning on large systems.