Multi-level Explanation of Deep Reinforcement Learning-based Scheduling
This work addresses the trust and interpretability gap for system administrators using DRL-based schedulers in clusters, though it is incremental as it builds on existing DRL scheduling methods.
The paper tackles the problem of interpreting complex Deep Reinforcement Learning (DRL)-based job scheduling policies, which are hard for administrators to trust despite performance gains, by proposing a multi-level explanation framework that dissects decisions into job and task levels using interpretable models and rules aligned with operational practices, revealing robustness issues in the scheduler's behavior.
Dependency-aware job scheduling in the cluster is NP-hard. Recent work shows that Deep Reinforcement Learning (DRL) is capable of solving it. It is difficult for the administrator to understand the DRL-based policy even though it achieves remarkable performance gain. Therefore the complex model-based scheduler is not easy to gain trust in the system where simplicity is favored. In this paper, we give the multi-level explanation framework to interpret the policy of DRL-based scheduling. We dissect its decision-making process to job level and task level and approximate each level with interpretable models and rules, which align with operational practices. We show that the framework gives the system administrator insights into the state-of-the-art scheduler and reveals the robustness issue in regards to its behavior pattern.