Wendy Wang, assistant professor in the Department of Computer Science, recently received a $499,941 National Science Foundation (NSF) grant for her project “SaTC: CORE: Small: Securing Network Embedding against Privacy Attacks” The project seeks to develop rigorous-yet-practical techniques to mitigate any vulnerabilities created by network embedding.
Performing data analysis on complex systems in the form of large networks, like social and information networks is challenging. To overcome this challenge, numerous network representation learning (NRL) approaches have been designed to learn low-dimensional vector representations (embedding) that preserve the network’s information, but because network embedding inherently captures the structure and properties of the original network, it raises a serious concern about whether the embedding also encodes sensitive information in the input network data.
Wang’s project plans to tackle this by researching three fundamental problems. First, she and her team will design three types of privacy attacks – namely membership inference attacks, attribute inference attacks, and property inference attacks – to infer the sensitive membership, attributes, and properties in the original network data from network embedding. Next, they will analyze which types of data characteristics and model properties impact the privacy vulnerabilities of network embedding against the three types of attacks, and finally design effective defense mechanisms to secure network embedding.
The research outcomes of Wang’s project will be disseminated broadly through developing new courses, involving students at various levels into cutting edge research in machine learning, and training of female and underrepresented students.
“As the research of network embedding has shaped entirely new research fields at the intersection of computer science and relevant science fields such as social science, physics, and biology, providing applied research and quantitative tools to assess and mitigate privacy risks in NRL models are very much needed in order to establish and reinforce trust in using these systems in high-stakes domains,” explained Wang. “By deepening the understanding of privacy risks in machine learning on graph-structured data, the proposed research will significantly push the boundary of research and practice in data science, machine learning, and data privacy.”
Learn more about computer science at Stevens: