Seminar: Toward Responsible AI: Privacy, Fairness, and Explainability in Federated Learning

_Abstract

Federated learning (FL) promises a simple bargain: share models, not data. That promise has made FL a natural choice to privacy-sensitive domains such as healthcare, finance, and defense. Yet privacy is only one dimension of responsible AI, and in FL in often comes at a hidden cost. The noise that protects individuals can disproportionately harm under-represented groups; data partitioned across organizations conceals the very attributes needed to audit bias; and a model assembled from many clients is difficult for anyone to explain. This talk asks what it takes to build FL systems that are private, fair, and explainable at once, and argues that uncertainty awareness is key to navigating the tensions among these goals.

I begin with a sytematic empirical study showing that privacy, fairness, and accuracy cannot be tuned independently in FL. I then present a progression of uncertainty-aware methods: fair federated learning for healthcare, fairness under local differential privacy, and a general framework that jointly balances privacy, fairness, and utility. Next, I shift from groups to individuals, asking how vertical FL can treat each person consistently when no single party ever sees a complete record. Finally, I turn to explainability, showing how global and local explanations can be aligned so that a model no one holds in one place can still be understood and trusted. I close with open challenges and lessons for making responsible AI a design principle, not an afterthought, in decentralized learning.

_Biography

 Jin-Hee Cho is an Associate Professor of Computer Science at Virginia Tech and an Affiliate Professor at the Virginia Tech National Security Institute, where she directs the Trustworthy Cyperspace Lab (tClab). Before joining Virginia Tech in 2018, she spent nearly a decade as a computer scientist at the U.S. Army Research Laboratory. Her research spans trustworthy and responsible AI, cybersecurity, human-centered AI and LLM safety, and decision-making under uncertainty. She has published over 250 peer-reviewed papers, including recent work at NeurIPS, ICML, ICLR, FAccT, AIES, and CHI. She received the Presidential Early Career Award for Scientists and Engineers (PECASE) in 2013, the IEEE ComSoc William R. Bennet Prize in 2015 and several best paper awards. She serves as an Associate Editor for IEEE Transactions on Network and Service Management, IEEE Transactions on Services Computing, and the Journal of Computational Social Science. She received her Ph.D. in Computer Science from Virginia Tech.