Augment then Smooth: Reconciling Differential Privacy with Certified Robustness
This addresses the problem of securing models against both privacy attacks and adversarial examples for practitioners, offering a simpler and more effective solution than previous complex methods.
The paper tackles the challenge of combining differential privacy and certified robustness in machine learning models, showing that standard differentially private training is insufficient for strong robustness guarantees. The proposed DP-CERT method achieves both privacy and robustness simultaneously, increasing certified accuracy by up to 2.5% on CIFAR10 compared to prior work.
Machine learning models are susceptible to a variety of attacks that can erode trust, including attacks against the privacy of training data, and adversarial examples that jeopardize model accuracy. Differential privacy and certified robustness are effective frameworks for combating these two threats respectively, as they each provide future-proof guarantees. However, we show that standard differentially private model training is insufficient for providing strong certified robustness guarantees. Indeed, combining differential privacy and certified robustness in a single system is non-trivial, leading previous works to introduce complex training schemes that lack flexibility. In this work, we present DP-CERT, a simple and effective method that achieves both privacy and robustness guarantees simultaneously by integrating randomized smoothing into standard differentially private model training. Compared to the leading prior work, DP-CERT gives up to a 2.5% increase in certified accuracy for the same differential privacy guarantee on CIFAR10. Through in-depth per-sample metric analysis, we find that larger certifiable radii correlate with smaller local Lipschitz constants, and show that DP-CERT effectively reduces Lipschitz constants compared to other differentially private training methods. The code is available at github.com/layer6ai-labs/dp-cert.