Graduate School awards Ph.D. student Liwei Song for work on machine learning privacy and security

Written by
Scott Lyon
April 23, 2021

The Graduate School has awarded Liwei Song a Wallace Memorial Fellowship in Engineering, funding his Ph.D. work for the 2021-2022 academic year.

The Graduate School selected 27 total honorific fellows, with six coming from the School of Engineering and Applied Science.

Honorific fellowships of this kind recognize advanced Ph.D. students whose research shows exceptional promise. Fellows do not teach classes during their award year, and the funding comes with no stipulated outcomes, giving Song an opportunity to pursue his research without the need to meet external benchmarks.

"Liwei has made exciting contributions at the intersection of machine learning, privacy and security," said Prateek Mittal, an associate professor of electrical and computer engineering and Song's adviser. "He has also demonstrated the ability to quickly master new areas and make deep contributions."

Song specializes in understanding how hackers work backwards through machine learning models to infer the sensitive data originally used to train those same models. He's published four recent conference papers, with two more under review, untangling what's called a membership inference attack.

In such attacks, adversaries infer the inputs (personal information about real people) by sophisticated analysis of patterns in the model's outputs. These attacks are particularly dangerous in privacy-critical applications such as the healthcare industry, according to Song. If a hospital wants to use a machine learning model to aid in diagnoses, engineers would first need to train that model using a set of actual patients' medical records. Song and his colleagues developed a way to block the attacks that would grant adversaries capability  to infer those records' data.

His privacy analysis techniques have been incorporated in Google’s TensorFlow Privacy system, and he was recently selected as a Rising Star in Data Science by the Center for Data and Computing at the University of Chicago.

Currently in his fifth year of the Ph.D. program, Song recently completed a research internship at Facebook, expanding on his research into the privacy risks of machine learning models. Last year he won the Best Paper Award at the 10th International Conference on the Internet of Things. Song is originally from Henan, China.