Standard-Deviation-Inspired Regularization for Improving Adversarial Robustness
This work addresses adversarial vulnerability in deep neural networks, which is a critical security concern for AI systems, but it represents an incremental improvement over existing adversarial training variants.
The authors tackled the problem of improving adversarial robustness in deep neural networks by proposing a standard-deviation-inspired regularization term that enhances existing adversarial training methods, resulting in improved robustness against stronger attacks like CW and Auto-attack and better generalization.
Adversarial Training (AT) has been demonstrated to improve the robustness of deep neural networks (DNNs) against adversarial attacks. AT is a min-max optimization procedure where in adversarial examples are generated to train a more robust DNN. The inner maximization step of AT increases the losses of inputs with respect to their actual classes. The outer minimization involves minimizing the losses on the adversarial examples obtained from the inner maximization. This work proposes a standard-deviation-inspired (SDI) regularization term to improve adversarial robustness and generalization. We argue that the inner maximization in AT is similar to minimizing a modified standard deviation of the model's output probabilities. Moreover, we suggest that maximizing this modified standard deviation can complement the outer minimization of the AT framework. To support our argument, we experimentally show that the SDI measure can be used to craft adversarial examples. Additionally, we demonstrate that combining the SDI regularization term with existing AT variants enhances the robustness of DNNs against stronger attacks, such as CW and Auto-attack, and improves generalization.