Summary of Provably Unlearnable Data Examples, by Derui Wang et al.
Provably Unlearnable Data Examples
by Derui Wang, Minhui Xue, Bo Li, Seyit Camtepe, Liming Zhu
First submitted to arxiv on: 6 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 The paper addresses concerns about data privacy and intellectual property breaches in the age of artificial intelligence. Existing methods render shared data unlearnable by applying perturbations to create Unlearnable Examples (UEs). However, these methods lack mechanisms to verify the robustness of UEs against uncertainty in unauthorized models. The authors propose a mechanism for certifying the (q, )-Learnability of an unlearnable dataset via parametric smoothing. This paper improves the tightness of certified (q, )-Learnability and designs Provably Unlearnable Examples (PUEs) with reduced (q, )-Learnability. The proposed mechanism mitigates issues with evaluation metrics, such as empirical test accuracy, which may not accurately represent the quality of UEs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Artificial intelligence is raising concerns about data privacy and intellectual property. People are sharing data without thinking about how others might use it. To solve this problem, researchers have come up with a way to make shared data unlearnable for unauthorized models. However, there’s no way to check if the data is really unlearnable or not. This paper proposes a new method to verify that shared data is indeed unlearnable and designs special examples called Provably Unlearnable Examples (PUEs) that are even harder to learn. |
