CS COLLOQUIUM: BO WAGGONER (University of Pennsylvania)
Title: Buying and Learning from User Data, Privately
Abstract: In this talk, I'll propose systems that purchase data from people for use in machine learning. The learning and budget objectives must be balanced with the goal of empowering users to be fairly compensated and have the privacy of their data formally protected. It will include a brief primer on *differential privacy* and overview of how we design differentially private "prediction markets" as building-blocks. At some point there will likely be a short rant on current company-user imbalances in privacy and value of data on the web.
Bio: Bo Waggoner is a postdoctoral fellow at the University of Pennsylvania's Warren Center for Network and Data Sciences. His work focuses on systems for gathering and predicting from data when people involved have strategic, privacy, or fairness consideration. His thesis, from Harvard in 2016 with advisor Yiling Chen, is titled "Acquiring and Aggregating Information from Strategic Sources".
Wednesday, December 13 at 11:00am to 12:00pm
STM 326 3700 Reservoir Road NW