Securing pervasive systems against adversarial machine learning
Brent Lagesse, Cody Burkard, Julio Perez · 2016
Applications and middleware in pervasive systems frequently rely on machine learning to provide adaptivity and customization that results in a seamless user experience despite operating in a dynamic environment. Machine learning algorithms have been shown to be vulnerable to covert, strategic attacks through the manipulation of training data. Machine learning algorithms in pervasive systems frequently train on data that could be manipulated by a malicious 3rd party. In this paper, we present our ongoing work to develop a security mechanism that is designed to work in the dynamic environments of pervasive computing as opposed to traditional security mechanisms that are designed for static environments. Furthermore, we present our modular testing framework that will be used to rapidly compare our work with other security mechanisms, applications and adversarial models.