会议专题

Security Analysis of Kolmogorov Complexity based Online Spam Filtering under Adversarial Impact

As a variety of machine learning algorithms have been applied to the field of network security, they become a very attractive target for attackers who seek to evade and disrupt them. This situation happens especially in spam filtering. Since spam evolves continuously in adversarial environment, it calls for fast, incremental, robust and secure learning algorithms. Currently security analysts for machine learning based system under adversarial impact becomes a very hot and important topic in researchers from the fields of information security and machine learning. More considering of the security of learning when people start to design machine learning based system in adversarial environment, more robustness and security will be achieved. Kolmogorov complexity is relatively new effective method proposed to measure individual randomness. It is used in spam filtering recently and gets very good performance. In this paper, we make a primarily security analysis of a classic model of kolmogorov complexity based spam filtering. We quantitatively analyze its vulnerability against malicious attacks by adversaries. Also the implications for security in complexity based filtering systems are discussed.

Spam filtering Adversarial learning Security learning Kolmogorov complexity

Wei Deng Zhiguang Qin

School of Computer Science and Engineering,University of Electronic Science and Technology of China,Chengdu, Sichuan 611731, China

国际会议

2010 International Conference on Information Security and Artificial Intelligence(2010年信息安全与人工智能国际会议 ISAI 2010)

成都

英文

311-314

2010-12-17(万方平台首次上网日期,不代表论文的发表时间)