Summary of Verification-guided Shielding For Deep Reinforcement Learning, by Davide Corsi et al.
Verification-Guided Shielding for Deep Reinforcement Learning
by Davide Corsi, Guy Amir, Andoni Rodriguez, Cesar Sanchez, Guy Katz, Roy Fox
First submitted to arxiv on: 10 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 paper presents a novel approach to address the reliability issue in Deep Reinforcement Learning (DRL) policies. DRL has shown promise in solving real-world tasks, but current methods lack formal safety guarantees, hindering their deployment in critical domains. Two main approaches are shielding and verification. Shielding uses an external online component to override potentially dangerous actions, while verification is an offline process that identifies unsafe policies prior to deployment. This paper integrates both methods, using formal and probabilistic verification tools to partition the input domain into safe and unsafe regions. The approach employs clustering and symbolic representation procedures to compress unsafe regions, reducing runtime overhead while preserving safety guarantees. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper solves a problem in artificial intelligence called Deep Reinforcement Learning (DRL). DRL is good at solving real-world tasks, but it’s not always safe. Two ways to make DRL safer are shielding and verification. Shielding uses an extra step to stop the policy from doing something bad. Verification checks the policy before it’s used to see if it will do anything bad. This paper combines these two methods to make DRL safer while still being fast. |
Keywords
* Artificial intelligence * Clustering * Reinforcement learning
