SIS Research Seminar by Jun ZHAO | Privacy and Security Issues in Data Analytics and the Internet of Things

Technologies for data analytics and the Internet of Things (IoT) pose privacy and security issues while providing ever-smarter services. For instance, data analytics with sensitive information may endanger the privacy of users. About data analytics, I will first present my current research on differentially private deep learning. Differentially private algorithms generate noisy answers to protect sensitive data. Under the iterative process of stochastic gradient descent in deep learning, my approach reuses a fraction of the noise added to the gradient in earlier iterations for later iterations. The result improves the tradeoff between privacy and learning accuracy in prior work. Afterwards, I will introduce my mechanisms for differential privacy under correlated data and use the mechanisms to enable adaptive data analysis with correlated samples. In addition, I will outline future directions in attack-resilient data analytics. In particular, one goal is to prevent maliciously crafted samples from triggering misbehavior of machine learning systems. About IoT security, I will review my extensive studies on security in sensor networks employing key predistribution schemes and my research on remote authentication of embedded devices. Moreover, I will discuss future work on end-to-end security in IoT as well as resilient design of cyber-physical systems.
 

About the Speaker