John Mitchell: 2016 Security Workshop

 

Monday, April 11, 2016
Location: McCaw Hall, Arrillaga Alumni Center

"Would That It Were So Simple: Yet Another Privacy Theory"

9:30am


Abstract:

A number of theories of data privacy have been proposed over the past fifteen or more years, including k-anonymity, L-diversity, and t-closeness. The most robust, differential privacy, is applicable to aggregate queries such as average or standard deviation of some quantity represented in a dataset, but is inherently unsuited to queries about particular individuals in the dataset. We formulate new definitions of privacy and utility for accessing information about unknown individuals identified by some form of token such as a cookie which is chosen randomly but correlated with web interaction. These definitions, unlike most current privacy definitions, take probabilistic prior information into account and are intended to reflect the use of aggregated web information for targeted advertising. In one intended application, a publisher placing content on a user's browser can send a token to an advertiser who bids on ad placement according to partial information about that user's possible preferences. We show that some negative results, such as the well-known Netflix privacy attack, carry over into this setting. We also describe scenarios allowing both privacy and utility. For example, when user preferences can be characterized by coarse-grained user traits and aggregate data about preferred products that have these traits, privacy arises from the similarities between many users. We illustrate this by discussing a restaurant recommendation scenario.


Joint work with: Avradip Mandal, Hart Montgomery and Arnab Roy


Bio:

John Mitchell is Professor of Computer Science, Vice Provost for Teaching and Learning, and the Mary and Gordon Crary Family Professor in the School of Engineering. His organization on campus supports excellence and innovation in teaching and learning. Between summer 2012 and the end of 2014, the prior office for online learning completed approximately 450 projects, each developing technology to support a Stanford class, a public online class, or revise a previous project. These projects were completed in collaboration with over 200 Stanford instructors. As a professor of computer science, Mitchell's research interests include computer security, privacy, programming languages, mathematical logic, and web technology.