Abstract: The ability to selectively remove undesirable learned information (such as private data, copyrighted content, or harmful knowledge that could facilitate the misuse of generative models) is increasingly recognized as a critical capability for trustworthy AI. This process, known as machine unlearning (MU), has become essential as generative models are deployed in sensitive domains including healthcare, defense, personalized education, and autonomous systems. In this talk, I will present a systematic, rigorous, and safety-centered exploration of machine unlearning in modern generative AI systems, with a primary focus on large language models (LLMs). Rather than treating unlearning as an isolated task, we position it as a multidisciplinary frontier shaped by the co-design of optimization, data, and model principles.

Register Now

Sijia Liu is an Associate Professor in the Department of Computer Science and Engineering at Michigan State University, and an Affiliated Professor at IBM Research. His research focuses on scalable and trustworthy AI, including machine unlearning for vision and language models, scalable optimization for deep models, adversarial robustness, and data–model efficiency. Dr. Liu has received multiple honors, such as the NSF CAREER Award (2024), the INNS Aharon Katzir Young Investigator Award (2024), the MSU Withrow Rising Scholar Award (2025), the Best Paper Runner-Up Award at UAI (2022), and the Best Student Paper Award at ICASSP (2017). He is the co-founder of the New Frontiers in Adversarial Machine Learning Workshop series (ICML/NeurIPS 2021–2024) and has delivered numerous tutorials on trustworthy and scalable ML at major conferences such as NeurIPS, CVPR, AAAI, and ICASSP.