Summary of Adversarial Guided Diffusion Models For Adversarial Purification, by Guang Lin et al.
Adversarial Guided Diffusion Models for Adversarial Purification
by Guang Lin, Zerui Tao, Jianhai Zhang, Toshihisa Tanaka, Qibin Zhao
First submitted to arxiv on: 24 Mar 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 A novel diffusion model (DM) based adversarial purification (AP) technique is proposed, which combines the power of guided diffusion and adversarial training to simultaneously maintain semantic information and remove adversarial perturbations. This approach leverages a pre-trained DM as the core component, while introducing an auxiliary neural network that provides adversarial guidance in the latent representation space. The proposed method, dubbed Adversarial Guided Diffusion Model (AGDM), is evaluated on three benchmark datasets: CIFAR-10, CIFAR-100, and ImageNet. Results show significant enhancements in robust accuracy, with a maximum improvement of 7.30% on CIFAR-10 compared to existing DM-based AP methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to make computers better at recognizing pictures has been developed. This method uses something called “diffusion models” to remove bad things that people put into pictures to try and trick the computer. The goal is to keep the important parts of the picture, like what it shows, while getting rid of the bad parts. The new approach combines two ideas: one that helps the computer understand the picture better and another that makes sure the picture isn’t changed too much. This method was tested on many pictures and worked well, improving how well computers recognize what’s in a picture. |
Keywords
* Artificial intelligence * Diffusion * Diffusion model * Neural network
