Defacement Detection with Passive Adversaries

摘要: A novel approach to defacement detection is proposed in this paper, addressing explicitly the possible presence of a passive adversary. Defacement an important security measure for Web Sites and Applications, aimed at avoiding unwanted modifications that would result significant reputational damage. As many other anomaly contexts, algorithm used identify defacements obtained via Adversarial Machine Learning process. We consider exploratory setting, where adversary can observe detector’s alarm-generating behaviour, with purpose devising injecting will pass undetected. It then necessary make learning process unpredictable, so be unable replicate it predict classifier’s behaviour. achieve goal by introducing secret key—a key our does not know. The influence number different ways, are precisely defined paper. This includes subset examples features actually used, time testing, as well algorithm’s hyper-parameters. methodology successfully applied context, using system both real artificially modified sites. year-long experimentation also described, referred monitoring new Site major manufacturing company.