ECE Seminar Series – Feb 21 (Fri) @ 2:00pm: "Advancing Efficiency and Fairness in Machine Learning," Taesup Moon, Professor, ECE, Seoul National U.

Location: Engineering Science Building (ESB), Room 2001
Come at 1:30p for Cookies, Coffee and Conversation!
DISTINGUISHED LECTURE at the ECE SEMINAR SERIES

Abstract

In this talk, I will present our recent work on two aspects of machine learning: enhancing training efficiency and promoting fairness. First, I will introduce techniques for enhancing training efficiency in two widely used learning frameworks—preference-based reinforcement learning (PbRL) and vision-language pre-training (VLP). I will show that constructing ranked lists of trajectories instead of sampling independent preference pairs in PbRL enables more efficient and effective learning. Likewise, in VLP, grouping similar image-text samples within a mini-batch to leverage hard negatives significantly accelerates training while improving downstream accuracy by capturing finer-grained information.

Next, I will shift the focus to fairness and will introduce a unified approach to achieving group fairness in classification through the distributionally robust optimization (DRO) framework. Additionally, I will provide theoretical and empirical evidence demonstrating that counterfactual fairness does not necessarily imply group fairness in an image classifier, highlighting key gaps between these fairness notions.

If time permits, I will also share ongoing and future research directions at the M.IN.D Lab at Seoul National University. 

Bio

Taesup Moon earned his B.S. degree in electrical engineering from Seoul National University, Seoul, South Korea, in 2002, and his M.S. and Ph.D. degrees in electrical engineering from Stanford University, Stanford, CA, USA, in 2004 and 2008, respectively. He is currently a Professor in the Department of Electrical and Computer Engineering at Seoul National University (SNU). Before joining SNU in 2021, he held various positions in academia and industry—serving as Assistant/Associate Professor at DGIST/Sungkyunkwan University (2015-2021), Research Staff Member at Samsung Advanced Institute of Technology (2013-2015), Postdoctoral Researcher at UC Berkeley (2012-2013), and Scientist at Yahoo! Labs (2008-2012).

He currently leads the Machine Intelligence and Data Science (M.IN.D) Lab at Seoul National University, focusing on adaptive and trustworthy machine/deep learning, signal processing, information theory, and various data science applications.

Hosted by: Distinguished Lecture at the ECE Seminar Series

Submitted by: Haewon Jeong <haewon@ece.ucsb.edu>