题目: Security, privacy, and blockchain research in deep learning and the Internet of Things
时间：4月12日上午9点
地点：网络安全学院501
摘要：Jun Zhao will present his research on security, privacy, and blockchains with applications to deep learning and the Internet of Things. Specifically, he will talk about privacy-aware machine learning, adversarial deep learning, and blockchains for vehicle security.
1. Privacy-aware learning: Jun will talk about his work on local differential privacy accepted to IEEE 35th Annual International Conference on Data Engineering (ICDE). Local differential privacy (LDP) is a strong privacy standard for collecting and analyzing data, which has been used, e.g., in the Chrome browser, iOS and macOS. In LDP, each user perturbs her information locally, and only sends the randomized version to an aggregator who performs analyses, which protects both the users and the aggregator against private information leaks. Jun's paper proposes novel LDP mechanisms which outperform existing solutions regarding worst-case noise variance. The proposed solutions are further used to build an LDP-compliant stochastic gradient descent algorithm (SGD), which powers many machine learning tasks such as linear regression, logistic regression, and support vector machines (SVM) classification. Experiments on real datasets confirm the effectiveness of the proposed methods, and their advantages over existing solutions.
2. Adversarial deep learning: Recent studies have shown the vulnerability of many deep learning algorithms to adversarial examples, which an attacker obtains by adding subtle perturbation to benign inputs in order to cause misbehavior of deep learning. For instance, an attacker can add carefully selected noise to a panda image so that the resulting image is still a panda to a human being but is predicted as a gibbon by the deep learning algorithm. As a first step to propose effective defense mechanisms against such adversarial examples, Jun analyzes the robustness of deep learning against adversarial examples. Moreover, Jun will discuss the ongoing research of using differential privacy to improve the robustness of deep learning to adversarial examples.
3. Blockchains for vehicle security: To improve the security of blockchain-enabled Internet of Vehicles (IoV) for data sharing among vehicles, Jun's paper accepted to IEEE Transactions on Vehicular Technology presents a two-stage enhancement including miner selection via reputation-based voting and block verification via a contract-based incentive mechanism.
讲座人背景 Biography: Jun Zhao is currently an Assistant Professor in the School of Computer Science and Engineering at Nanyang Technological University (NTU) in Singapore. He received a PhD degree in Electrical and Computer Engineering from Carnegie Mellon University (CMU) in the USA (advisors: Virgil Gligor, Osman Yagan; collaborator: Adrian Perrig) and a bachelor's degree from Shanghai Jiao Tong University in China. Before joining NTU first as a postdoc with Xiaokui Xiao and then as a faculty member, he was a postdoc at Arizona State University as an Arizona Computing PostDoc Best Practices Fellow (advisors: Junshan Zhang, Vincent Poor). His research interests include blockchains, security, and privacy with applications to the Internet of Things and deep learning. In terms of publications, he has over a dozen journal articles published/accepted in IEEE/ACM Transactions as well as over twenty conference/workshop papers. One of his first-authored papers was shortlisted for the best student paper award in IEEE International Symposium on Information Theory (ISIT) 2014, a prestigious conference in information theory. Positions for PhD students, postdocs, and visiting students/researchers are available in Jun's research group. Please contact him at JunZhao@NTU.edu.sg or add his Wechat at http://t.cn/R1SOkcW
邀请人：陈晶教授