DiffDefense: Defending Against Adversarial Attacks via Diffusion Models

This paper presents a novel reconstruction method that leverages Diffusion Models to protect machine learning classifiers against adversarial attacks, all without requiring any modifications the themselves. The susceptibility of models minor input perturbations renders them vulnerable attacks. While diffusion-based methods are typically disregarded for defense due their slow reverse process, this demonstrates our proposed offers robustness threats while preserving clean accuracy, speed, and plug-and-play compatibility. Code at: https://github.com/HondamunigePrasannaSilva/DiffDefence .