Cheriton School of Computer Science Professor Gautam Kamath has received a 2023 Faculty of Mathematics Golden Jubilee Research Excellence Award. The $2,500 prize, established in 2017 to mark the 50th anniversary of the Faculty of Mathematics, is conferred to faculty members who have made outstanding research contributions.
“Congratulations to Gautam on receiving this much-deserved research excellence award,” said Raouf Boutaba, Professor and Director of the Cheriton School of Computer Science. “Gautam’s research is grounded strongly in theoretical computer science, but he and his students ensure that the work they conduct also has critically important practical applications, most notably by drawing useful insights from data while preserving privacy of sensitive personal information.”
Professor Kamath joined the Cheriton School of Computer Science in 2019 as an assistant professor after an appointment as a Microsoft Research Fellow at the Simons Institute for the Theory of Computing at the University of California, Berkeley. He has a PhD and SM in Electrical Engineering and Computer Science from MIT.
He leads The Salon, a research group of postdoctoral fellows, graduate students and undergrads who study statistics, algorithms, machine learning, and optimization. Earlier this year, he was named a Canada CIFAR AI Chair and Vector Institute Faculty Member in recognition of his contributions to differential privacy, machine learning and statistics.
Highlights of Professor Kamath’s research contributions
Among Professor Kamath’s most significant contributions is a solution to what had been to that point an open problem in classical statistics — that is, providing a computationally efficient algorithm for robust mean estimation, even in settings with high-dimensional data. Computationally inefficient methods for this problem have existed for four decades, but Professor Kamath’s method sparked an explosive line of work on algorithmic robust statistics, which has in turn led to works by the community addressing a number of other major related problems. These include learning mixtures of Gaussians, achieving sub-Gaussian rates for heavy-tailed distributions, and estimation under the constraint of differential privacy.
Professor Kamath’s results were published in a 2016 Symposium on Foundations of Computer Science (FOCS 2016) paper titled “Robust Estimators in High-Dimensions Without the Computational Intractability.” This paper received a number of further invitations to prestigious venues, among them the SICOMP Special Issue for FOCS 2016, Highlights of Algorithms 2017, and Communications of the ACM Research Highlights. With more than 425 citations as of July 2023, it remains the most-cited work from FOCS 2016.
Much of Professor Kamath’s work has focused on differential privacy, a rigorous notion of data privacy that protects statistics against disclosure of sensitive information about individuals. Starting with his paper “Privately Learning High-Dimensional Distributions” presented at COLT 2019, the 32nd Conference on Learning Theory, he has had a number of works investigating statistical estimation under the constraint of differential privacy. Many other researchers have been inspired by his work, and addressed adjacent problems.
Some of his most recent works — “Efficient Mean Estimation with Pure Differential Privacy via a Sum-of-Squares Exponential Mechanism” presented at the 2022 Symposium on Theory of Computing and “Robustness Implies Privacy in Statistical Estimation” presented at the 2023 Symposium on Theory of Computing — link to his earlier work on robust estimation. These show exciting new technical connections between robustness and privacy that have inspired other researchers to pursue further.
Another important work is “The Discrete Gaussian for Differential Privacy,” presented at the 2020 Conference on Neural Information Processing Systems. This research introduced a new mechanism for releasing quantities subject to differential privacy, by careful analysis of a fundamental distribution, the discrete Gaussian. This research has had immense impact. It was deployed in the 2020 US Census as the key primitive underlying their Disclosure Avoidance System. Professor Kamath’s algorithm was used on the sensitive data of more than 300 million people in the most high-profile deployment of differential privacy to date. This work also serves as an exemplar of how Professor Kamath’s research begins with rigorous theoretical analysis but also has enormous practical impacts.
Also of note is his paper “Differentially Private Fine-tuning of Language Models,” presented at the 2022 International Conference on Learning Representations. This research shows that recent advances in machine learning can be highly effective even with the addition of privacy constraints. This has had a significant impact in both academia and industry. Within academia, it has spurred an active line of work on private machine learning with the assistance of public data. This project was a collaboration with Microsoft, and the methods have seen significant deployment in industry.
Professor Kamath’s research contributions are matched by the outstanding guidance and mentorship he provides to his students. His previous master’s student, Mahbod Majid, received the 2023 Faculty of Mathematics Graduate Research Excellence Award. In addition, four of his current master’s students — Alex Bie, Matthew Regehr, Sabrina Mokhtari, and Jimmy Di — have each been awarded prestigious Vector Scholarships in Artificial Intelligence, and his undergraduate student, Nicholas Vadivelu, won the 2021 Jessie W.H. Zou Memorial Award and was a runner up for a Computing Research Association’s 2022 Outstanding Undergraduate Researcher Award.
Faculty of Mathematics Golden Jubilee Research Excellence Award
Professor Kamath is the fourth faculty member at the Cheriton School of Computer Science to receive a Faculty of Mathematics Golden Jubilee Research Excellence Award.
Past recipients are Professors Florian Kerschbaum (2022 recipient), Jesse Hoey (2019 recipient), and Daniel Vogel (2018 recipient).
