Speaker: Lydia Zakynthinou (UC Berkeley)
Title: Algorithmic Stability for Trustworthy Machine Learning and Statistics
The seminar will be available for in-person and Zoom participation. To participate online, please email inquiry-cs@barnard.edu to receive the Zoom link.
Data-driven systems hold immense potential to positively impact society, but their reliability remains a challenge. Their outputs are often too brittle to changes in their training data, leaving them vulnerable to data poisoning attacks, prone to leaking sensitive information, or overfitting to training data. An understanding of fundamental principles for designing algorithms that are both ‘stable’ and accurate is crucial towards mitigating these risks.
In this talk, I will focus on statistical estimation under differential privacy—a rigorous framework that ensures the privacy of individuals in the input dataset. I will present algorithmic techniques that build on robustness against data poisoning attacks, that allow us to take advantage of beneficial structure in the data and achieve optimal error for several multivariate statistical tasks. Lastly, I will highlight the deeper connections between differential privacy and robustness that underpin these results.
Lydia Zakynthinou is a FODSI postdoctoral research fellow in the Simons Institute for the Theory of Computing at UC Berkeley, hosted by Michael I. Jordan. She earned her Ph.D. in Computer Science from Northeastern University under the supervision of Jonathan Ullman and Huy Nguyen. Her research lies in trustworthy machine learning and statistics, with a focus on data privacy and generalization, and has been recognized with a Meta PhD fellowship and a Khoury PhD Research Award. She holds a diploma in Electrical and Computer Engineering from NTUA and a MSc in Logic, Algorithms, and Theory of Computation from NKUA in Greece.