arXiv cs.AI· Rahul Kumar Yadav, Tunir Ghosh, Sudhendu Ahir, Parthe Pandit, Arjun Bhagoji·· 12 小时前AI 评分31
核方法模型的对抗鲁棒性全面分析
Adversarial Robustness of Kernel Machines: A Comprehensive Analysis
AI 导读
研究通过白盒与黑盒逃逸攻击，对二分类和多分类任务中的核方法模型、随机特征分类器及递归特征机（Recursive Feature Machines）进行了全面的对抗鲁棒性实证评估，并提出一种核方法近似对抗训练算法。在图像和表格数据集上的实验表明，核方法模型原生鲁棒性强于神经网络，面对对抗扰动时性能下降更平缓。
正文 · 原文
Abstract:Kernel machines are an important class of models in machine learning. However, the question of how robust modern kernel machines are to adversarial examples has not been studied comprehensively. In this paper, we provide a thorough empirical investigation of the adversarial robustness of kernel-based models for both binary and multiclass classification tasks, via white-box and black-box evasion attacks. We evaluate the native robustness of kernel machines, random feature classifiers, and the relatively newer Recursive Feature Machines. In addition, we provide an algorithm for approximate adversarial training of kernel machines. Our experiments on classification tasks on image and tabular datasets strongly indicate that kernel machines are natively more robust than neural networks, i.e., their performance degrades more gracefully against adversarial perturbations.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.07474 [cs.LG] |
| (or arXiv:2611.07474v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.07474 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Rahul Yadav [view email]
[v1]
Thu, 5 Nov 2026 05:33:09 UTC (6,959 KB)
来源：arXiv cs.AI · arxiv.org