时间： 2018年12月27日 14：00-15:00
Title: Privacy in the Modern Era: The Cases of Online Social Network and Machine Learning Model
The advancement of information and communication technologies has resulted in a deluge of data being available. While serving as the fuel for the next-generation industrial development, the large-scale data also raises severe privacy concerns. In this talk, I will cover our recent work on assessing privacy risks stemming from social network data and machine learning models. First, I will focus on the phenomenal location check-ins and hashtags shared in online social networks. We show that location check-ins can be used as an effective attack vector to infer users' social relations. Moreover, I'll demonstrate how hashtags can be exploited to effectively pinpoint a user's fine-grained locations. Second, I will discuss the risks of training data privacy in machine learning models. We relax various assumptions of the adversary model and show that membership inference attacks against machine learning classifiers can be performed in a much cheaper and effective way. Our results further shed light on the necessity of a general evaluation methodology for machine learning models in the future. The corresponding research papers of this talk are published at CCS 2017, WWW 2018, and NDSS 2019.
Yang Zhang (https://yangzhangalmo.github.io/) is currently a postdoc working in the group of professor Michael Backes at CISPA Helmholtz Center for Information Security, Saarbruecken, Germany. From January 2019, he will be an independent research group leader at CISPA. His research mainly concentrates on privacy in modern society. Topics include machine learning privacy, biomedical privacy, social network privacy, and location privacy. Besides, he also works on urban informatics/computing, social media analysis, and data mining. Yang obtained his Ph.D. degree from University of Luxembourg on 2016.11. Prior to that, he obtained his bachelor (2009) and master (2012) degrees from Shandong University, China.