Summary of Prompt-driven Contrastive Learning For Transferable Adversarial Attacks, by Hunmin Yang et al.
Prompt-Driven Contrastive Learning for Transferable Adversarial Attacks
by Hunmin Yang, Jongoh Jeong, Kuk-Jin Yoon
First submitted to arxiv on: 30 Jul 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
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 a novel attack method called PDCL-Attack that leverages the capabilities of vision-language foundation models like CLIP to generate transferable adversarial examples. The approach utilizes a generative model-based framework and enhances the transferability of perturbations by formulating an effective prompt-driven feature guidance using semantic representations from text, specifically class labels. This method outperforms state-of-the-art methods in various cross-domain and cross-model settings. The paper demonstrates the effectiveness of leveraging powerful models like CLIP to tackle challenging vision tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research proposes a new way to trick artificial intelligence (AI) models that are good at recognizing pictures. The method uses a special kind of training data called “class labels” to make the AI models more confused and less accurate when trying to recognize pictures from different sources or models. The approach is tested on various types of images and shows that it can outdo other methods in making the AI models more uncertain. This is important because it allows researchers to better understand how these powerful AI models work and how they can be used for different tasks. |
Keywords
» Artificial intelligence » Generative model » Prompt » Transferability
