Randomizing SVM Against Adversarial Attacks Under Uncertainty

Yan Chen, Wei Wang, Xiangliang Zhang

Abstract: Robust machine learning algorithms have been widely studied in adversarial environments where the adversary maliciously manipulates data samples to evade security systems. In this paper, we propose randomized SVMs against generalized adversarial attacks under uncertainty, through learning a classifier distribution rather than a single classifier in traditional robust SVMs. The randomized SVMs have advantages on better resistance against attacks while preserving high accuracy of classification, especially for non-separable cases. The experimental results demonstrate the effectiveness of our proposed models on defending against various attacks, including aggressive attacks with uncertainty.