Privacy in clustering: applications and algorithms

Clustering is a fundamental unsupervised machine learning problem that lies at the core of several real-world applications. While traditional clustering algorithms have not considered the privacy of the users providing the data, recently private clustering has received significant attention. In this talk will cover recent research in clustering with differential privacy, a strong notion of privacy guarantee promising plausible deniability for user data. Will mostly cover work on clustering graph data. For graph clustering, will focus on our recent work (ICML 2023) where we show edge-differentially private hierarchical clustering algorithms with provable approximation guarantees.

Presenter: Alessandro Epasto

Biography

Alessandro Epasto is a staff research scientist at Google, New York working in the Graph Mining team part of the Algorithms and Optimization team lead by Vahab Mirrokni. Alessandro Epasto received a Ph.D. in computer science from Sapienza University of Rome, advised by Professor Alessandro Panconesi. Before joining Google, Alessandro Epasto was a postdoc at Brown University advised by Professor Eli Upfal. His research interests include problems in the areas of privacy, clustering, and large-scale data analysis.

Event organised by SmartData@PoliTO center.