Summary of Enhancing Targeted Transferability Via Feature Space Fine-tuning, by Hui Zeng et al.
Enhancing targeted transferability via feature space fine-tuning
by Hui Zeng, Biwei Chen, Anjie Peng
First submitted to arxiv on: 5 Jan 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 paper proposes a method to improve the transferability of adversarial examples (AEs) across unknown models. AEs are generated by iterative attacks, but these can overfit to specific models, making them less effective against new targets. The proposed approach, fine-tuning in the feature space, starts with an AE generated by a baseline attack and then adjusts the features conducive to the target class while discouraging those leading to the original class. This method achieves nontrivial and universal improvements in targeted transferability with only a few iterations. The results also show that simple iterative attacks can be comparable or even better than resource-intensive methods, which require training target-specific classifiers or generators. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps keep your personal information private by making sure fake pictures of you won’t work on other computers. When someone tries to trick the computer with a fake picture, it’s called an “adversarial example”. The problem is that these fake pictures often only work on one specific computer, so they’re not very useful for protecting privacy. To fix this, the researchers developed a new way to make these fake pictures work better across different computers. They call it “fine-tuning” and it helps the fake picture adapt to new computers by changing what features of the picture are important. The results show that this method can be really effective in making these fake pictures work on many different computers, which is important for keeping your personal information safe. |
Keywords
* Artificial intelligence * Fine tuning * Transferability
