The Power of Decaying Steps: Enhancing Attack Stability and Transferability for Sign-based Optimizers
This addresses reliability issues in adversarial attacks for security testing of machine learning models, but it is incremental as it builds on existing sign-based methods.
The paper tackled the problem of non-convergence and instability in sign-based adversarial attack optimizers like I-FGSM and MI-FGSM, which degrade attack success rates with more iterations, by proposing MDCS algorithms with monotonically decreasing step-sizes, resulting in significantly improved transferability and stability, achieving an optimal convergence rate of O(1/√T).
Crafting adversarial examples can be formulated as an optimization problem. While sign-based optimizers such as I-FGSM and MI-FGSM have become the de facto standard for the induced optimization problems, there still exist several unsolved problems in theoretical grounding and practical reliability especially in non-convergence and instability, which inevitably influences their transferability. Contrary to the expectation, we observe that the attack success rate may degrade sharply when more number of iterations are conducted. In this paper, we address these issues from an optimization perspective. By reformulating the sign-based optimizer as a specific coordinate-wise gradient descent, we argue that one cause for non-convergence and instability is their non-decaying step-size scheduling. Based upon this viewpoint, we propose a series of new attack algorithms that enforce Monotonically Decreasing Coordinate-wise Step-sizes (MDCS) within sign-based optimizers. Typically, we further provide theoretical guarantees proving that MDCS-MI attains an optimal convergence rate of $O(1/\sqrt{T})$, where $T$ is the number of iterations. Extensive experiments on image classification and cross-modal retrieval tasks demonstrate that our approach not only significantly improves transferability but also enhances attack stability compared to state-of-the-art sign-based methods.