Constructing Semantics-Aware Adversarial Examples with a Probabilistic Perspective

Andi Zhang, Mingtian Zhang, Damon Wischik

Keywords: Adversarial Examples, Probabilistic Generative Models, Diffusion Models, Energy-based Models

Abstract: We propose a probabilistic perspective on adversarial examples, allowing us to embed subjective understanding of semantics as a distribution into the process of generating adversarial examples, in a principled manner. Despite significant pixel-level modifications compared to traditional adversarial attacks, our method preserves the overall semantics of the image, making the changes difficult for humans to detect. This extensive pixel-level modification enhances our method's ability to deceive classifiers designed to defend against adversarial attacks. Our empirical findings indicate that the proposed methods achieve higher success rates in circumventing adversarial defense mechanisms, while remaining difficult for human observers to detect.

Primary Area: Safety in machine learning

Submission Number: 16554