Summary of Data-dependent Stability Analysis Of Adversarial Training, by Yihan Wang and Shuang Liu and Xiao-shan Gao
Data-Dependent Stability Analysis of Adversarial Training
by Yihan Wang, Shuang Liu, Xiao-Shan Gao
First submitted to arxiv on: 6 Jan 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 proposes novel generalization bounds for adversarial training, a widely used defense against adversarial example attacks in deep learning. The authors fill a gap by incorporating data distribution information into the bounds, which is essential for understanding how changes in data distribution and adversarial budget affect robust generalization gaps. Specifically, they utilize on-average stability and high-order approximate Lipschitz conditions to examine the impact of data poisoning attacks on robust generalization. The derived bounds are at least as good as uniform stability-based counterparts that do not consider data distribution. This work has important implications for developing robust machine learning models in real-world scenarios. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how deep learning models can be fooled by fake examples, and how to make them more resistant to these attacks. The authors find new ways to calculate how well a model will perform on unseen data, taking into account the kind of mistakes it might make during training. They show that their approach is just as good as existing methods, but also accounts for important details about the data that was used to train the model. This research helps us understand how to build more reliable and secure AI systems. |
Keywords
* Artificial intelligence * Deep learning * Generalization * Machine learning
