1: \begin{abstract}

2: Adversarial perturbations have drawn great attentions in various machine learning models. In this paper, we investigate the sample adversarial perturbations for nonlinear support vector machines (SVMs). Due to the implicit form of the nonlinear functions mapping data to the feature space, it is difficult to obtain the explicit form of the adversarial perturbations. By exploring the special property of nonlinear SVMs, we transform the optimization  problem of attacking nonlinear SVMs into a nonlinear KKT system. Such a system can be solved by various numerical methods.  Numerical results show that our method is efficient in computing adversarial perturbations.

3: %Adversarial perturbations have drawn great attentions in various machine learning models. Most of them are calculated by iteration, which takes a long time and can not be well explained. In this paper, we investigate the adversarial perturbations for support vector machines(SVMs). Through the Random Fourier Features method, we transform the optimization  problem of attacking nonlinear SVM into the optimization problem of attacking linear SVM in feature space, so as to obtain the exact directions of perturbations in feature space.

4: %Our method avoids a large number of iterations and possible non convergence in the iterative process. Numerical results show that our method is fast and effective in computing adversarial perturbations.

5: \end{abstract}

6: