log in  |  register  |  feedback?  |  help  |  web accessibility
MC2 Internal Tech Seminar #2
Dr. Jennifer Golbeck and Dr. Elaine (Runting) Shi
Monday, November 12, 2012, 12:00-1:30 pm Calendar
  • You are subscribed to this talk through .
  • You are watching this talk through .
  • You are subscribed to this talk. (unsubscribe, watch)
  • You are watching this talk. (unwatch, subscribe)
  • You are not subscribed to this talk. (watch, subscribe)
Registration requested: The organizer of this talk requests that you register if you are planning to attend. There are two ways to register: (1) You can create an account on this site (click the "register" link in the upper-right corner) and then register for this talk; or (2) You can enter your details below and click the "Register for talk" button. Either way, you can always cancel your registration later.



Talk 1: Dr. Jennifer Golbeck

Title: Trust, User Profiling, and Privacy in Social Media


Abstract: Users share vast amounts of information about themselves in social media. Some is shared directly and other data can be inferred by learning over their language, behavior, and interactions. In this talk, I will present an overview of my work on computing trust in social networks and on profiling information about users based on their online behavior. I will follow that with some of our current work that integrates this and other techniques into applications for security, intelligence, and military applications. 

For more information on Dr. Golbeck see link:  http://www.cs.umd.edu/~golbeck/


Talk 2: Dr. Elaine (Runting) Shi

Title:  Privacy-preserving Distributed Data Analysis

Abstract: We consider a privacy-preserving distributed data analysis scenario, where data is distributed among users (or organizations), and users may not wish to entrust their sensitive data to a centralized party such as a cloud service provider. Participating parties are mutually distrustful, and a subset of the parties may be compromised and colluding.

We prove new lower bounds on the utility of distributed data analysis satisfying information theoretic differential privacy. To circumvent such lower bounds in practice, we consider computational differential privacy. We demonstrate new constructions that satisfy computational differential privacy while achieving asymptotically smaller error. Our constructions also offer several properties that will be desirable in a practical deployment scenario, including support for periodic aggregation, no peer-to-peer communication, and fault tolerance.

For more information on Dr. Shi see link:  http://www.cs.umd.edu/~elaine/


This talk is organized by Carolyn Flowers