机器学习系统毒化攻击综述

Overview on Poisoning Attacks against Machine Learning System

自机器学习被应用到许多关键性领域以来,机器学习系统的脆弱性也引起了人们的高度重视。其中,针对机器学习系统的毒化攻击得到了研究者的广泛关注,呈现了一些研究成果。因此,将系统地介绍当前机器学习系统毒化攻击的研究进展,对机器学习系统毒化攻击算法进行分类和总结,包括针对机器学习中的线性分类器、支持向量机、贝叶斯分类器和深度神经网络等几类常见模型的毒化攻击等攻击算法,目标是使现有的关于机器学习系统毒化攻击的研究成果更加清晰,为相关研究者的研究工作提供启发。

Since machine learning has been applied to many key areas, the vulnerability of machine learning systems has also attracted much attention. Among them, the poisoning attacks against machine learning systems have received great attention from researchers, and some research results have also been presented. Therefore, this paper systematically introduces the current research progress of poisoning attacks on machine learning systems, classifies and summarizes the algorithms of poisoning attacks on machine learning systems, including attack algorithms such as poisoning attacks on several common models such as linear classifiers, support vector machines, Bayesian classifiers, deep neural networks in machine learning. The goal is to make the existing research results on poisoning attacks against machine learning systems clearer, and to provide inspiration for related researchers’ research work.