Friendly Noise against Adversarial Noise: A Powerful Defense against Data Poisoning Attacks

A powerful category of data poisoning attacks modify a subset of training examples by small adversarial perturbations to change the prediction of certain test-time data. Existing defense mechanisms are not desirable to deploy in practice, as they often drastically harm the generalization performance, or are attack-specific and prohibitively slow to apply. Here, we propose a simple but highly effective approach that unlike existing methods breaks various types of poisoning attacks with the slightest drop in the generalization performance. We make the key observation that attacks exploit sharp loss regions to craft adversarial perturbations which can substantially alter examples' gradient or representations under small perturbations. To break poisoning attacks, our approach comprises two components: an optimized friendly noise that is generated to maximally perturb examples without degrading the performance, and a random varying noise component. The first component takes examples farther away from the sharp loss regions, and the second component smooths out the loss landscape. The combination of both components builds a very light-weight but extremely effective defense against the most powerful triggerless targeted and hidden-trigger backdoor poisoning attacks, including Gradient Matching, Bulls-eye Polytope, and Sleeper Agent. We show that our friendly noise is transferable to other architectures, and adaptive attacks cannot break our defense due to its random noise component.