Abstract: Modern communication networks such as online social networks give us a unique opportunity of observing, analyzing and better understanding large scale network structures. Networks data from these measurements often capture potentially private or confidential data. These privacy concerns are especially significant when researchers try to share these real network traces for research. Researchers need realistic graphs to perform realistic benchmark analysis of novel protocols and applications. Measurement-calibrated graph models is an attractive alternative to sharing real datasets. We investigated several graph models to generate synthetic graphs, and leveraged online social networks traces to validate whether the resulting synthetic graphs are statistically similar to real graphs. However, it is unclear whether extremely realistic synthetic data can leak information from the original dataset. In addition, the privacy of synthetic graphs is difficult to quantify, because we lack a practical and accurate metric able to quantify the resilience of a graph against privacy attacks. We propose a novel graph generation process, Pygmalion, which is able to bridge the natural tension between accuracy and privacy of synthetic graphs. Pygmalion integrates a statistical framework, called $\epsilon$-Differential Privacy, into a graph generative model, and provides a cohesive solution to share accurate synthetic graphs with strong and provable privacy guarantees. Short Bio: Alessandra Sala is a research associate in the Department of Computer Science at University of California Santa Barbara. She has received funding from the Cisco University Research program for her project on “Understanding and Modeling the Structure of Online Social Networks.” In her prior appointment, she worked for two years as post-doctoral fellow with the CurrentLab research group led by Prof. Ben Y. Zhao, and still collaborates with them closely. She received her Ph.D. in Computer Science from Università degli Studi di Salerno, Italy in May 2008.
This event is organised by the Seminar - Privacy.