Summary of Instant Adversarial Purification with Adversarial Consistency Distillation, by Chun Tong Lei et al.
Instant Adversarial Purification with Adversarial Consistency Distillation
by Chun Tong Lei, Hon Ming Yam, Zhongliang Guo, Chun Pong Lau
First submitted to arxiv on: 30 Aug 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); 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 This paper proposes two novel methods for defending neural networks against subtle adversarial noise: One Step Control Purification (OSCP) and Gaussian Adversarial Noise Distillation (GAND). OSCP is a diffusion-based purification model that can purify adversarial images in one Neural Function Evaluation (NFE) iteration, using Latent Consistency Model (LCM) and ControlNet. This method is computationally friendly and time-efficient, achieving a defense success rate of 74.19% on ImageNet with only 0.1s per purification. GAND is a consistency distillation framework that reconciles the latent space dynamics between natural and adversarial manifolds, without requiring Full Fine Tune (FFT). Instead, LoRA or PEFT can be used for sufficient results. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps protect neural networks from fake or misleading images by developing two new ways to remove noise. One method is called One Step Control Purification (OSCP) and it uses a special process to quickly clean up noisy images. The other method, Gaussian Adversarial Noise Distillation (GAND), helps computer vision models understand the difference between real and fake images. |
Keywords
» Artificial intelligence » Diffusion » Distillation » Latent space » Lora
