Machine learning, security, and privacy: challenges and opportunities


Machine learning provides more and more powerful tools for data analytics. On the other hand, security and privacy attacks increasingly involve data. Therefore, machine learning and security & privacy naturally intersect with each other as they both involve data, and there are many interesting questions at the intersections: i) How machine learning impacts security and privacy analytics design? In this talk, I will first talk about machine learning for security and privacy in social networks, particularly, graph-based collective classification to detect fake accounts in social networks. A long-standing challenge in collective classification is that existing methods cannot learn accurate edge weights, thus resulting in limited detection performance in practice.