arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1903.02585cs.LGcs.CRcs.CVstat.ML
收藏

GanDef:一种基于GAN的神经网络分类器对抗训练防御

GanDef: A GAN based Adversarial Training Defense for Neural Network Classifier

New Jersey Institute of Technology(新泽西理工学院);Qatar Computing Research Institute(卡塔尔计算研究所)

Guanxiong Liu, Issa Khalil, Abdallah Khreishah

摘要:

机器学习模型,特别是神经网络(NN)分类器,被广泛用于包括自然语言处理、计算机视觉和网络安全的许多应用中。它们在无攻击场景的假设下提供高准确性。然而,这一假设已经被引入对抗性示例所挑战-精心扰动的输入样本通常被错误分类。许多研究人员试图开发一种防御对抗性例子的方法;然而,我们离实现这一目标还很远。在本文中,我们设计了一个基于生成对抗网络(GAN)的对抗训练防御,称为GanDef,它利用竞争游戏来调节训练过程中的特征选择。我们分析表明,GanDef可以训练分类器,使其能够抵御对抗性的例子。通过对不同白盒对抗示例的广泛评估,GanDef训练的分类器显示出与最先进的对抗训练防御训练的分类器相同的测试准确度。更重要的是,GanDef的变体GanDef-Comb可以利用GanDef来实现正确分类原始示例和对抗示例之间的动态权衡。因此,当对抗性示例的比例超过41.7%时,它实现了最高的整体测试准确率。

英文摘要:

Machine learning models, especially neural network (NN) classifiers, are widely used in many applications including natural language processing, computer vision and cybersecurity. They provide high accuracy under the assumption of attack-free scenarios. However, this assumption has been defied by the introduction of adversarial examples -- carefully perturbed samples of input that are usually misclassified. Many researchers have tried to develop a defense against adversarial examples; however, we are still far from achieving that goal. In this paper, we design a Generative Adversarial Net (GAN) based adversarial training defense, dubbed GanDef, which utilizes a competition game to regulate the feature selection during the training. We analytically show that GanDef can train a classifier so it can defend against adversarial examples. Through extensive evaluation on different white-box adversarial examples, the classifier trained by GanDef shows the same level of test accuracy as those trained by state-of-the-art adversarial training defenses. More importantly, GanDef-Comb, a variant of GanDef, could utilize the discriminator to achieve a dynamic trade-off between correctly classifying original and adversarial examples. As a result, it achieves the highest overall test accuracy when the ratio of adversarial examples exceeds 41.7%.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑