Summary of Demystifying Reinforcement Learning in Production Scheduling Via Explainable Ai, by Daniel Fischer et al.
Demystifying Reinforcement Learning in Production Scheduling via Explainable AI
by Daniel Fischer, Hannah M. Hüsener, Felix Grumbach, Lukas Vollenkemper, Arthur Müller, Pascal Reusch
First submitted to arxiv on: 19 Aug 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The abstract discusses applying two explainable AI (xAI) frameworks, SHAP and Captum, to a deep reinforcement learning (DRL) agent in a flow production setting. The goal is to understand the reasoning behind the agent’s scheduling decisions. While xAI methods provide explanations, they lack consistency and often fail to consider domain knowledge or real-world scenarios. To address this, the authors propose a hypotheses-based workflow for verifying explanations against domain knowledge and reward hypotheses. This approach can be applied to various DRL-based scheduling use cases. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper explores how explainable AI (xAI) techniques can help us understand why a deep reinforcement learning (DRL) agent makes certain decisions in a flow production setting. The authors test two popular xAI methods, SHAP and Captum, on the agent’s scheduling choices. They find that these methods have limitations, such as not considering domain knowledge or real-world scenarios. To fix this, they suggest a new approach that involves verifying explanations against what we know about the situation and why the agent was rewarded. This could be useful for other situations where DRL is used to make decisions. |
Keywords
* Artificial intelligence * Reinforcement learning
