Title: Toward Adaptive AI Models: Continual Learning and Machine Unlearning in Practice

Speaker: Dr. Sungmin Cha @ New York University

Time : 10:00 - 11:00, Oct 14th, 2025

Location: Online

https://hli.skku.edu/InvitedTalk251014 

Language: English speech & English slides

Abstract:

In this talk, I will present my research efforts toward developing adaptive AI models that can efficiently learn and forget in dynamic environments. My work focuses on continual learning and machine unlearning—two core challenges in building models that can evolve over time while staying robust and efficient. I will introduce a practical algorithmic approach for scalable continual learning under realistic constraints, as well as a new evaluation protocol that better reflects generalization in the wild. I will also discuss recent work in knowledge unlearning for LLMs, emphasizing a method that enables robust and parameter-efficient unlearning. Finally, I will discuss ongoing/future works on several relevant topics.

Bio:

Sungmin Cha is currently a Faculty Fellow at New York University, working with Prof. Kyunghyun Cho, and he will join Meta as a research scientist in November 2026. His research focuses on adaptive AI models, with a particular interest in continual learning, machine unlearning, and efficient post-training strategies for large models. He earned his Ph.D. in Electrical and Computer Engineering from Seoul National University (SNU), where he was advised by Prof. Taesup Moon. During his Ph.D., he was a visiting researcher at Harvard University and held research scientist internships at NAVER AI and LG AI Research. Sungmin has been recognized with several prestigious honors, including the Qualcomm Innovation Fellowship Korea (2021), the Yulchon AI START Fellowship (2022), the Distinguished Doctoral Dissertation Award from SNU (2023), and the Best Doctoral Dissertation Award from the Korean Academy of Science and Technology (2024).