Toward Improving the Robustness of Deep Learning Models via Model Transformation

Deep learning (DL) techniques have attracted much attention in recent years, and have been applied to many application scenarios, including those that are safety-critical. Improving the universal robustness of DL models is vital and many approaches have been proposed in the last decades aiming at such a purpose. Among existing approaches, adversarial training is the most representative. It advocates a post model tuning process via incorporating adversarial samples. Although successful, they still suffer from the challenge of generalizability issues in the face of various attacks with unsatisfactory effectiveness. Targeting this problem, in this paper we propose a novel model training framework, which aims at improving the universal robustness of DL models via model transformation incorporated with a data augmentation strategy in a delta debugging fashion. We have implemented our approach in a tool, called Dare, and conducted an extensive evaluation on 9 DL models. The results show that our approach significantly outperforms existing adversarial training techniques. Specifically, Dare has achieved the highest Empirical Robustness in 29 of 45 testing scenarios under various attacks, while the number drops to 5 of 45 for the best baseline approach.