Enhancing the Antidote: Improved Pointwise Certifications against Poisoning Attacks

Poisoning attacks can disproportionately influence model behaviour by making small changes to the training corpus. While defences against specific poisoning do exist, they in general not provide any guarantees, leaving them potentially countered novel attacks. In contrast, examining worst-case behaviours Certified Defences make it possible guarantees of robustness a sample adversarial modifying finite number samples, known as pointwise certification. We achieve this exploiting both Differential Privacy and Sampled Gaussian Mechanism ensure invariance prediction for each testing instance numbers poisoned examples. doing so, our provides that are more than twice large those provided prior certifications.