Many online services, such as recommender systems, email, and social networks collect user data, which is then used for both personalization and monetization. Although the latter enables services to be free, users are realizing that these services come at a hidden cost of potentially exposing their private data. In this talk I will show that even the common 5-star item-rating recommender system leaks private demographic information. Then, I will discuss methods for helping users preserve their privacy while getting accurate recommendations. Finally, a building block of many recommender systems, and an important machine-learning algorithm on its own, is linear regression. I will present a system that learns a linear model without learning anything about the private input data other than the output model.
Udi Weinsberg (PhD 2011) is a researcher and associate fellow at Technicolor Research in Palo Alto, CA. He studies privacy and security, focusing on enabling practical privacy-preserving machine learning algorithms in online services.