Deep Learning for Smartphone Security and Privacy

Due to the growing popularity of smartphones with more than 3.5 billion users all around the world, their built-in sensors data has been excessively exploited in various fields of research including security and privacy. User identification and continuous user authentication are two representatives of challenging mobile sensing problems in security and privacy research areas. Based on the user's distinguishable behavioral patterns inferred from sensors data, continuous user authentication aims to verify whether the current user is the actual legitimate user or not, whereas in the user identification problem the goal is to identify the current user of the mobile device correctly among all users. Since deep learning algorithms have great potential for discriminative learning tasks, we utilize them to develop deep models advancing the current smartphone user authentication and user identification systems. First, we present DeepAuth as a generic framework for re-authenticating users in a mobile app that leverages sensors data and deep learning techniques to enable a frictionless and secure user experience in the application. Existing machine learning techniques for user identification are classification-oriented and thus are not amenable easily to large-scale, real-world deployment. In the second task, we present DeepFP that exploits metric learning techniques to address the challenges of current user identification techniques. In the last work, we investigate the possibility of transferring knowledge from a pre-trained user identification model on a source domain to a target domain.