Reproduce “Adversarial Patch Attacks and Defences in Vision-Based Tasks: A Survey” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Abhijith Sharma, Yijun Bian, Phil Munz, Apurva Narayan

This survey provides a comprehensive overview of adversarial patch attacks and corresponding defense mechanisms in vision-based deep learning models. It categorizes attack types and defense strategies, highlighting their evolution, real-world concerns, and current challenges faced by researchers.

Key insights

  • Adversarial patches pose a unique and practical threat to vision-based AI systems due to their localized, visible, and physically realizable nature, affecting tasks from classification to object detection.
  • The field is characterized by an ongoing 'arms race' between new attack methodologies and evolving defense mechanisms.
  • Current defense strategies, while varied, often struggle with scalability, computational efficiency, and adaptability to new attack variants, pointing to areas for future research.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.