Christine Task, A Practical Beginners' Guide to Differential Privacy

Differential privacy is a very powerful approach to protecting individual privacy in data-mining; it's also an approach that hasn't seen much application outside academic circles. There's a reason for this: many people aren't quite certain how it works. Uncertainty poses a serious problem when considering the public release of sensitive data. Intuitively, differentially private data-mining applications protect individuals by injecting noise which "covers up" the impact any individual can have on the query results. In this talk, I will discuss the concrete details of how this is accomplished, exactly what it does and does not guarantee, common mistakes and misconceptions, and give a brief overview of useful differentially privatized data-mining techniques. This talk will be accessible to researchers from all domains; no previous background in statistics or probability theory is assumed. My goal in this presentation is to offer a short-cut to researchers who would like to apply differential privacy to their work and thus enable a broader adoption of this powerful tool. About the speaker: Christine Task is a PhD candidate in the Computer Science department of Purdue University, and a member of CERIAS. She has five years experience teaching discrete math and computability theory at the undergraduate level. Her research interests are in differential privacy and its application to social network analysis, and her research advisor is CERIAS fellow Chris Clifton.