DiffDefense: Defending against Adversarial Attacks via Diffusion Models

1 code implementation • 7 Sep 2023 • Hondamunige Prasanna Silva, Lorenzo Seidenari, Alberto del Bimbo

This paper presents a novel reconstruction method that leverages Diffusion Models to protect machine learning classifiers against adversarial attacks, all without requiring any modifications to the classifiers themselves.

Adversarial Defense