SDM: A Powerful Tool for Evaluating Model Robustness
The paper introduces Sequential Difference Maximization (SDM), a new gradient-based attack method for evaluating model robustness. It addresses the limitations of previous methods by reconstructing the objective for adversarial example generation. Experimental results show that SDM outperforms state-of-the-art techniques in both attack performance and cost-effectiveness.
- ▪SDM is designed to maximize the difference between non-ground-truth label probability upper bound and ground-truth label probability.
- ▪The method employs a three-layer optimization framework consisting of cycle, stage, and step.
- ▪SDM utilizes a negative probability loss function and a Directional Probability Difference Ratio loss function during its optimization stages.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.20308 (cs) [Submitted on 19 May 2026] Title:SDM: A Powerful Tool for Evaluating Model Robustness Authors:Xinlei Liu, Tao Hu, Jichao Xie, Peng Yi, Hailong Ma, Baolin Li View a PDF of the paper titled SDM: A Powerful Tool for Evaluating Model Robustness, by Xinlei Liu and 5 other authors View PDF HTML (experimental) Abstract:Gradient-based attacks are important methods for evaluating model robustness. However, since the proposal of APGD, it has been difficult for such methods to achieve significant breakthroughs.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
