Title: Fair Federated Learning: Are We Asking the Right Questions?

Abstract: Fairness is a key pillar of responsible AI, but defining and achieving it remains challenging—especially in decentralized settings like federated learning (FL). Unlike traditional machine learning, FL introduces new sources of bias and harm that require rethinking fairness beyond standard definitions.
In this talk, I will explore how fairness in FL is shaped not just by algorithms, but by the choices developers make throughout the FL lifecycle. I will examine existing fairness definitions in FL,highlighting which types of harms they address—and which they overlook.
I will also discuss how fairness interacts with other responsible AI concerns, such as privacy and robustness, revealing critical gaps in current research.
One key issue I will unpack is how fairness definitions in FL often prioritize system performance over protecting the most vulnerable. Through this discussion, I aim to challenge current approaches and offer insights into how we can design FL systems that are not just fair on paper, but in practice.