This prestigious award supports early-career faculty in integrating education and research to establish a strong foundation for career development within their field.
Dr. Tianxi Ji, an assistant professor in the Edward E. Whitacre Jr. College of Engineering’s Department of Computer Science, has received a $585,000 National Science Foundation CAREER Award that will advance his research in data privacy protection.
“The importance of protecting private data is of critical concern for every organization that has responsibility for this kind of information,” Dr. Ji explains. “But the process to evaluate and interpret privacy is difficult.”
Ji’s NSF Career project “Differential Privacy Meets Random Geometry: Interpretation, Applications, and Trade-offs” aims to simplify our ability to visualize and interpret the strength of privacy protection as well as measure the tradeoff between privacy and usefulness in order to identify the protection techniques that maximize both.
To achieve privacy protection for sensitive data, a certain form of randomization has to be introduced so that the data to be studied cannot be traced back to the person. This randomness can be inserted through a variety of techniques that range from a simple “coin-flip” or adding a random number to the data record to much more complex methods.
However, when privacy is presented as a mathematical formula, it is hard to interpret the quality of the privacy protection. Ji’s project proposes to map the mathematical equations established for privacy to random triangles that can be uniquely mapped as point clouds on a sphere.
In his model, when the point clouds centralize along the equator of the sphere, the equation has a strong privacy guarantee. If the point cloud shifts away from the equator, the further away it moves the weaker the privacy guarantee.
Why is this visualization tool important? Privacy protection is not the only factor in designing data management systems: the usefulness, or utility, of the data set is extremely important. When randomness is applied to a data set to protect privacy it also limits usefulness.
The stronger the privacy protection, the more randomness that is introduced and the less useful the data becomes. Ji’s goal is to find the sweet spot—that space that maximizes the tradeoff between privacy and utility.
Ji’s project will study the geometric properties of the points on the sphere for each privacy technique. Then he will identify the latitude that strikes the optimal trade-off between privacy and utility and develop novel techniques that calibrate randomness when protecting sensitive data to meet the desired latitude on the sphere.
The proposed research will also evaluate convergence speed and learnability factors for training AI models.
The NSF grant supports Dr. Ji’s Statistically Enhanced Privacy lab, including five doctoral students in these research efforts. The research will then provide the basis for developing data security and privacy curriculum.
“One of the most exciting facets of this project is that we will use this research to develop privacy curriculum that gives students an end-to-end hands-on experience that starts with understanding privacy data and its importance and moves to developing privacy tools and training models to launching third-party usage,” says Ji.
In addition, area high school students in the TexPREP-Lubbock program hosted by TTU’s Mathematics Department and STEM Core, who are already learning Probability, Trigonometry, and Geometry, will be introduced to its application to privacy protection.
Dr. Ji’s work is timely. The US and Texas have each enacted data privacy protection legislation within the last two years and continue to invest in developing tools that better protect citizens’ privacy.