Researchers continue to improve consumer privacy protection
UNIVERSITY PARK, Pa. – As a graduate student researching cryptography, Adam Smith wanted to study problems that required new theoretical insights, but whose solutions would actually affect people’s lives in a positive way. The issue of privacy and data collecting particularly piqued his interest.
“The phrase big data wasn’t around yet, but it was a thing and it was becoming clear that if technology didn’t think carefully about this, we would get to a stage where data would be shared widely without regards to privacy,” said Smith, a computer science and engineering professor in the School of EECS. “I wanted to help find a way to get the benefits of this data without the cost to privacy.”
In 2006, Smith co-wrote a paper that was a major breakthrough in privacy protection for companies like Google, Apple, and the U.S. Census Bureau. The paper introduced the world to differential privacy and provided a solid mathematical foundation for a vast body of subsequent work on private data analysis.
“The term refers to a class of techniques, many of which were developed at Penn State, that will help organizations aggregate information about users’ behavior while ensuring that the raw information is never collected or stored,” Smith said.
Smith’s work started a large and growing area of research on algorithms for analyzing a sensitive data set while preserving the privacy of the individuals whose data it contains.
“The work we are now doing at Penn State will help organizations aggregate information about consumer behavior while ensuring that the raw information is never collected or stored by the company,” Smith said.
In addition to Smith, other researchers at Penn State, including computer scientists Sofya Raskhodnikova and Dan Kifer in the School of EECS and statistics Professor Aleksandra Slavkovic in the Eberly College of Science, are doing research in this area.
In 2013 and 2015, Google awarded Penn State researchers two grants, one to Kifer and one to Smith and Vitaly Shmatikov at Cornell, to investigate what deep learning systems can leak about sensitive inputs, as well as to develop a system for privacy-preserving deep learning.
Smith said some of this is funding differential privacy research that investigates false discovery, addressing the problem where some scientific findings seem valid but cannot be reproduced. There’s an assumption that the more privacy is implemented into the data, the less valuable it is, but Smith said algorithms can be used to adjust to the noise distraction, which would allow privacy to make the data more useful.
Kifer is on sabbatical this semester helping the U.S. Census Bureau implement disclosure control. He is helping to write a program that respects the constraints the agency needs when looking at data on various populations, but also makes sure the information being released is differentially private.