Summary of Universal Black-box Reward Poisoning Attack Against Offline Reinforcement Learning, by Yinglun Xu et al.
Universal Black-Box Reward Poisoning Attack against Offline Reinforcement Learning
by Yinglun Xu, Rohan Gumaste, Gagandeep Singh
First submitted to arxiv on: 15 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
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 This research paper investigates a type of attack, known as reward poisoning, which can compromise the integrity of reinforcement learning models. The goal is to develop an attack strategy that can be applied universally across various machine learning algorithms and datasets. The proposed “policy contrast attack” manipulates the rewards associated with certain actions or policies, making them appear more or less effective than they truly are. This can lead to suboptimal or even catastrophic performance in the learning process. The study demonstrates the effectiveness of this approach against current state-of-the-art offline reinforcement learning algorithms on various benchmark datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research explores a way to trick artificial intelligence systems that learn from past experiences, called reinforcement learning. The attackers try to make the AI think some actions are better or worse than they really are. This can cause the AI to make poor decisions. The researchers propose a new method, “policy contrast attack,” which is designed to work with many different AI algorithms and datasets. They show that this attack works well against current best practices in offline reinforcement learning. |
Keywords
* Artificial intelligence * Machine learning * Reinforcement learning
