Myeongseob Ko, a Ph.D. student in computer engineering, focuses on trustworthy machine learning, particularly in the areas of training data extraction, memorization, machine unlearning, privacy, and data attribution.  

 “As chat models like ChatGPT become part of people’s daily lives -- even for those unfamiliar with AI research -- preventing them from providing harmful, toxic, or privacy-related content remains a critical challenge. By addressing this challenge, I am trying to deepen my expertise in fundamental machine learning methods, broaden the positive impact of their practical applications, and make a meaningful contribution to society,” said Ko, who is advised by Ruoxi Jia. 

Jia’s research, closely aligned with Ko’s, drew him to the Sanghani Center. “In working with Professor Jia during the past four years, I have gained invaluable insights and grown as a researcher,” Ko said. “And as a Sanghani Center student, I especially value the opportunity to learn from outstanding researchers who continue to inspire and support my academic growth.”

One specific example of Ko’s work is in the area of machine unlearning.

“I study methods that enable large-scale foundation models to forget harmful or toxic information learned from vast amounts of publicly available data without negatively impacting their performance in benign or beneficial tasks, thereby improving the responsible behavior of these models when interacting with humans,” said Ko.

He presented "Boosting Alignment for Post-Unlearning Text-to-Image Generative Models" during a poster session at the 2024 Annual Conference on Neural Information Processing Systems (NeurIPS) last December.  

Projected to graduate in Spring 2026, Ko will seek a position where he can continue his research and contribute to advancing technologies that positively impact human lives and is open to opportunities in both academia and industry.