School of CSE Seminar Series: Eli Chien

Speaker: Eli Chien, postdoctoral fellow at Georgia Tech
Date and Time: March 28, 2:00-3:00 p.m.
Location: Coda, Room 230
Host: Anqi Wu

Title: Machine Unlearning: The General Theory and LLM Practice of Privacy

Abstract: “The right to be forgotten” is the concept from GDPR that data holder (server) should erase the data and the corresponding derivatives whenever the original data providers (users) request for it. It is the common practice that LLM are trained on extensive and diverse dataset, which are usually generated from users. While retraining from scratch without those data is the gold standard, it is prohibitively costly. The goal of machine unlearning is to develop efficient approaches to approximate such gold standard, thus obeying the privacy regulation in laws like GDPR. In this talk, I will first introduce our empirical studies of unlearning for LLM, which highlight pitfalls of current empirical unlearning evaluation and weaknesses of heuristic methods. Then we will dive into the generic machine unlearning problem and my recent progress in unlearning theory.

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.