Large data sets consisting of potentially sensitive private information of individuals are common, and there to be analyzed. Yet, analysis of them is fraught with risk of leakage of private information, which is undesirable for ethical, legal or business reasons. I will start with some examples of privacy breaches in recent times, making a case against ad hoc approaches to privacy.
I will then describe Differential Privacy, a formal mathematical definition due to Dwork, McSherry, Nissim and Smith. It gives a strong guarantee of privacy and has several additional desirable properties. Finally, I will discuss how many data analyses tasks can be done accurately while guaranteeing differential privacy.
Kunal Talwar is a Senior Researcher at Microsoft Research in Mountain View, CA. He has contributed to several aspects of Theoretical Computer Science, especially Algorithms and Privacy and received the Privacy Enhancing Technologies Award in 2009. He received his Bachelor of Technology from the Indian Institute of Technology, Delhi, and finished his PhD in Computer Science from UC Berkeley.
This talk is part of our Ethics of Big Data series. Please join us immediately after the talk for a reception. Hors d'oeuvres will be served in the common area outside of room 527.