APR 16: Dr. Debarghya Mukherjee, Boston University

“Learning From Data: Two Aspects of Fairness”

Wednesday, April 16, 2025
Hybrid Event
In-person: 801 Massachusetts Avenue
2nd Floor, Room 2128
Boston, MA 02119
12:00-1:00pm Seminar
1:00-2:00pm Luncheon

Registration is now closed. View the recording from this event here.

Bio: Dr. Debarghya Mukherjee obtained his Ph.D. under the supervision of Prof. Moulinath Banerjee and Prof. Ya’acov Ritov from the Dept. of Statistics, University of Michigan. Prior to joining BU in Fall 2023, he spent a year as a post-doctoral fellow at Princeton University under the tutelage of Prof. Jianqing Fan. His research interests span a variety of fields: he is keenly interested in the analysis of complex statistical models (e.g. deep neural networks), especially in high dimensions, where the geometry of the problem becomes extremely relevant. Although his previous research has dealt with independent data settings, he is now starting to explore models that capture data dependency (e.g., spatial or temporal dependence under some mixing conditions). He also actively works on algorithmic fairness, domain adaptation, and domain generalization, all of which fall within the purview of constrained classification/regression. One of his key goals is to delve into the connections among these fields of modern machine learning and develop methodologies for the optimal transfer of knowledge from one domain to another while preserving fairness.

Abstract: As modern machine learning systems increasingly influence high-stakes decision-making in areas such as hiring, lending, and law enforcement, ensuring fairness in these automated processes is critical. This presentation explores two key aspects of fairness—group fairness and individual fairness—and discusses their theoretical foundations, challenges, and practical implementations. We highlight how biases emerge in learned systems due to historical data imbalances and cognitive biases and talk about some strategies to mitigate them at different stages of model development, from data collection to post-deployment auditing. Finally, we explore several promising directions for advancing fairness in AI systems, highlighting potential contributions that can help develop more equitable and accountable models in the future.