Authors: Amira Guesmi, Ioan Marius Bilasco, Muhammad Shafique, Ihsen Alouani

Published on: March 03, 2023

Impact Score: 8.12

Arxiv code: Arxiv:2303.01734

Summary

• What is new: A novel approach to generate adversarial patches with semantic meaning using similarity loss, offering greater flexibility than GAN-based techniques.
• Why this is important: Current physical adversarial attacks using GANs are noticeable by humans and have a limited latent space, compromising naturalness or attack efficiency.
• What the research proposes: Introducing an additional loss term with semantic constraints to the optimization problem for generating naturalistic and inconspicuous adversarial patches.
• Results: Achieved a success rate of up to 91.19% in digital environments and 72% in smart cameras, outperforming GAN-based techniques.

Technical Details

Technological frameworks used: Generative Adversarial Networks (GANs) with an additional semantic constraint for optimization.

Models used: nan

Data used: nan

Potential Impact

Security and surveillance industry, autonomous vehicle manufacturers, AI-based authentication systems.

Want to implement this idea in a business?

We have generated a startup concept here: SeamlessGuard.