Regularized Training and Tight Certification for Randomized Smoothed Classifier with Provable Robustness
This work addresses the challenge of enhancing certified adversarial robustness for deep learning models, providing incremental improvements to existing smoothing-based defenses.
The paper tackles the problem of training base classifiers for Gaussian-smoothed neural networks to improve both accuracy and robustness, and introduces a regularized training method and a tighter certification algorithm that achieve state-of-the-art probabilistic robustness guarantees on CIFAR-10 and ImageNet datasets.
Recently smoothing deep neural network based classifiers via isotropic Gaussian perturbation is shown to be an effective and scalable way to provide state-of-the-art probabilistic robustness guarantee against $\ell_2$ norm bounded adversarial perturbations. However, how to train a good base classifier that is accurate and robust when smoothed has not been fully investigated. In this work, we derive a new regularized risk, in which the regularizer can adaptively encourage the accuracy and robustness of the smoothed counterpart when training the base classifier. It is computationally efficient and can be implemented in parallel with other empirical defense methods. We discuss how to implement it under both standard (non-adversarial) and adversarial training scheme. At the same time, we also design a new certification algorithm, which can leverage the regularization effect to provide tighter robustness lower bound that holds with high probability. Our extensive experimentation demonstrates the effectiveness of the proposed training and certification approaches on CIFAR-10 and ImageNet datasets.