Robust Deep Learning Models Against Semantic-Preserving Adversarial Attack
This addresses the problem of improving model robustness against combined perturbations for AI security applications, representing an incremental advance over prior work on individual perturbations.
The paper tackles the challenge of making deep learning models robust against joint adversarial and natural perturbations by proposing a Semantic-Preserving Adversarial (SPA) attack and using it for adversarial training. Empirical results on four benchmarks show that the SPA attack causes a larger performance decline under constraints and that SPA-enhanced training outperforms existing defenses against such joint perturbations.
Deep learning models can be fooled by small $l_p$-norm adversarial perturbations and natural perturbations in terms of attributes. Although the robustness against each perturbation has been explored, it remains a challenge to address the robustness against joint perturbations effectively. In this paper, we study the robustness of deep learning models against joint perturbations by proposing a novel attack mechanism named Semantic-Preserving Adversarial (SPA) attack, which can then be used to enhance adversarial training. Specifically, we introduce an attribute manipulator to generate natural and human-comprehensible perturbations and a noise generator to generate diverse adversarial noises. Based on such combined noises, we optimize both the attribute value and the diversity variable to generate jointly-perturbed samples. For robust training, we adversarially train the deep learning model against the generated joint perturbations. Empirical results on four benchmarks show that the SPA attack causes a larger performance decline with small $l_{\infty}$ norm-ball constraints compared to existing approaches. Furthermore, our SPA-enhanced training outperforms existing defense methods against such joint perturbations.