Diffusion Denoising as a Certified Defense Against Clean-Label Poisoning Attacks

TL;DR — We show that our certified defense against data poisoning that leverages the diffusion denoising approach renders existing clean-label poisoning ineffective while preserving a model utility.

Abstract

We present a certified defense to clean-label poisoning attacks. These attacks work by injecting poisoning samples that contain $p$-norm bounded adversarial perturbations into the training data to induce a targeted misclassification of a test-time input. Inspired by the adversarial robustness achieved by $denoised$ $smoothing$, we show how a pre-trained diffusion model can sanitize the training data before a model training. We extensively test our defense against seven clean-label poisoning attacks and reduce their attack success to 0-16\% with only a small drop in the test time accuracy. We compare our defense with existing countermeasures against clean-label poisoning, showing that the defense reduces the attack success the most and offers the best model utility. Our results highlight the need for future work on developing stronger clean-label attacks and using our certified yet practical defense as a strong baseline to evaluate these attacks.