Enhancing Adversarial Attacks: The Similar Target Method
This work addresses security concerns in deep neural networks by improving adversarial attack transferability, representing an incremental advancement over prior ensemble methods.
The paper tackles the problem of enhancing transferability of adversarial attacks by proposing the Similar Target method, which regularizes optimization direction to simultaneously attack multiple surrogate models, and it outperforms state-of-the-art attackers on 18 classifiers and adversarially trained models on ImageNet.
Deep neural networks are vulnerable to adversarial examples, posing a threat to the models' applications and raising security concerns. An intriguing property of adversarial examples is their strong transferability. Several methods have been proposed to enhance transferability, including ensemble attacks which have demonstrated their efficacy. However, prior approaches simply average logits, probabilities, or losses for model ensembling, lacking a comprehensive analysis of how and why model ensembling significantly improves transferability. In this paper, we propose a similar targeted attack method named Similar Target~(ST). By promoting cosine similarity between the gradients of each model, our method regularizes the optimization direction to simultaneously attack all surrogate models. This strategy has been proven to enhance generalization ability. Experimental results on ImageNet validate the effectiveness of our approach in improving adversarial transferability. Our method outperforms state-of-the-art attackers on 18 discriminative classifiers and adversarially trained models.