Fair Federated Learning under Group-Specific Distributed Concept Drift

Priberam Machine Learning Lunch Seminar

Abstract:

Machine learning models can become unfair when different groups experience changes in data over time, a phenomenon called group-specific concept drift. This challenge is amplified in Federated Learning, where clients may encounter different drifts while training a shared model. This talk introduces the problem and presents a distributed, group-aware approach that detects and adapts to such drifts, helping maintain fairness in dynamic environments.

Bio:

Teresa Salazar is an ML/NLP Researcher at Priberam, where she conducts applied research on Natural Language Processing in an industrial setting. She holds a B.Sc. in Informatics Engineering from the University of Coimbra (2018), an M.Sc. in Informatics from the University of Edinburgh (2019), and a PhD from the University of Coimbra (2025), where her research focused on fairness in machine learning. In parallel with her industry work, she serves as an Invited Professor at the University of Coimbra.

www.priberam.com