Regularized Training and Tight Certification for Randomized Smoothed Classifier with Provable Robustness

  • Huijie Feng Cornell University
  • Chunpeng Wu Alibaba Group US Inc
  • Guoyang Chen Alibaba Group US Inc
  • Weifeng Zhang Alibaba Group US Inc
  • Yang Ning Cornell University

Abstract

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 ℓ2 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.

Published
2020-04-03
Section
AAAI Technical Track: Machine Learning