Colloquium
Mila Hardt
Twitter
Monday, March 25, 2013
2250 WEB
Refreshments 3:20 p.m.
Lecture 3:40 p.m.
Title: Privacy-Preserving Personalization on Smartphones
Mila Hardt
Monday, March 25, 2013
2250 WEB
Refreshments 3:20 p.m.
Lecture 3:40 p.m.
Title: Privacy-Preserving Personalization on Smartphones
Abstract
Abstract: Mobile devices today are equipped with a range of sensors that can be effectively used to infer a user's context including her location, transportation mode and social state. This has empowered personalized applications such as the targeting of advertisements and the monitoring of physical and mental health. However, these context-aware mobile applications raise serious privacy concerns. People already believe that risks of sharing location information outweigh benefits in many location-based services.
In this talk, I will present scalable techniques to find more attractive tradeoffs of privacy and utility from personalization.
I will focus on targeted advertising as an application. Targeted advertising requires statistics over sensitive context and click data. To collect these large amounts of statistics, I will present an efficient distributed algorithm that preserves privacy. I will discuss two scalability challenges: (1) The number of users is large and the set of users is changing constantly. This makes it difficult to execute a distributed protocol. (2) The domain over which we collect click statistics can be huge. Thus, our protocol must scale sub-linearly in the domain size. These statistics are then used by our ad serving platform to efficiently deliver targeted advertisements to users based on the amount of personal information they were willing to share.
BIO
Mila Hardt works on search quality at Twitter. She received her PhD at Cornell University advised by Johannes Gehrke in 2012. Her research interests lie in the area of data management with focus on privacy. She is the recipient of a Microsoft Research Women Scholarship and a Google Engineering Intern Scholarship.
Abstract: Mobile devices today are equipped with a range of sensors that can be effectively used to infer a user's context including her location, transportation mode and social state. This has empowered personalized applications such as the targeting of advertisements and the monitoring of physical and mental health. However, these context-aware mobile applications raise serious privacy concerns. People already believe that risks of sharing location information outweigh benefits in many location-based services.
In this talk, I will present scalable techniques to find more attractive tradeoffs of privacy and utility from personalization.
I will focus on targeted advertising as an application. Targeted advertising requires statistics over sensitive context and click data. To collect these large amounts of statistics, I will present an efficient distributed algorithm that preserves privacy. I will discuss two scalability challenges: (1) The number of users is large and the set of users is changing constantly. This makes it difficult to execute a distributed protocol. (2) The domain over which we collect click statistics can be huge. Thus, our protocol must scale sub-linearly in the domain size. These statistics are then used by our ad serving platform to efficiently deliver targeted advertisements to users based on the amount of personal information they were willing to share.
BIO
Mila Hardt works on search quality at Twitter. She received her PhD at Cornell University advised by Johannes Gehrke in 2012. Her research interests lie in the area of data management with focus on privacy. She is the recipient of a Microsoft Research Women Scholarship and a Google Engineering Intern Scholarship.