Friday, February 23, 2024 03:00PM

Center for Signals and Information Processing (CSIP) Seminar

Date: Friday, February 23, 2024

Time: 3:00 p.m. - 4:00 p.m.

Location: Centergy Building 5126. The associated zoom link is:

Speaker: Eli Chien

Speakers' Title: Georgia Institute of Technology Postdoctoral Fellow

Seminar Title: Machine Unlearning: Current Challenges and Beyond

Abstract: With the adoption of recent laws ensuring the “right to be forgotten”, the problem of machine unlearning has become of significant importance. In this talk, I outline a recent line of my research aimed at tackling the current challenges in theoretical machine unlearning, including the generalization to graph unlearning and non-convex problems. Firstly, I address the challenges of graph unlearning, where removing an edge or node may potentially impact the label predictions of all other nodes. I propose certified graph unlearning, which leveraging the graph analysis to establish certified graph unlearning guarantees. Secondly, I discuss the challenges of extending unlearning methodologies to non-convex problems, where existing methods rely on the strong convexity of the underlying loss. I propose Langevin unlearning, which not only enables us to tackle non-convex problems but also offers a novel perspective of machine unlearning, unveiling its smooth connections to differential privacy.

Bio: Eli Chien is currently a Postdoctoral Fellow at the Georgia Institute of Technology, working with Professor Pan Li. He obtained his Ph.D. from the University of Illinois, Urbana-Champaign, advised by Professor Olgica Milenkovic. His current research focuses on privacy in machine learning, including machine unlearning and differential privacy, as well as their applications to graph machine learning. His previous research centered on designing better Graph Neural Networks with theoretical guarantees. His work has been primarily published in top-tier machine learning, data mining, and information theory venues, including ICML, NeurIPS, ICLR, Transactions on IT, KDD, TheWebConf, AISTATS, AAAI and more.