Rethinking Model Ensemble in Transfer-based Adversarial Attacks
This work addresses the challenge of generating more effective adversarial attacks for security testing in deep learning systems, particularly against adversarially trained models, with incremental improvements over existing ensemble methods.
The paper tackled the problem of improving the transferability of adversarial examples in black-box attacks by rethinking model ensemble methods, proposing a Common Weakness Attack (CWA) that promotes flat loss landscapes and closeness to local optima, resulting in enhanced transferability validated on image classification, object detection, and a large vision-language model like Google's Bard.
It is widely recognized that deep learning models lack robustness to adversarial examples. An intriguing property of adversarial examples is that they can transfer across different models, which enables black-box attacks without any knowledge of the victim model. An effective strategy to improve the transferability is attacking an ensemble of models. However, previous works simply average the outputs of different models, lacking an in-depth analysis on how and why model ensemble methods can strongly improve the transferability. In this paper, we rethink the ensemble in adversarial attacks and define the common weakness of model ensemble with two properties: 1) the flatness of loss landscape; and 2) the closeness to the local optimum of each model. We empirically and theoretically show that both properties are strongly correlated with the transferability and propose a Common Weakness Attack (CWA) to generate more transferable adversarial examples by promoting these two properties. Experimental results on both image classification and object detection tasks validate the effectiveness of our approach to improving the adversarial transferability, especially when attacking adversarially trained models. We also successfully apply our method to attack a black-box large vision-language model -- Google's Bard, showing the practical effectiveness. Code is available at \url{https://github.com/huanranchen/AdversarialAttacks}.