The Current State and Challenges of Fairness in Federated Learning
Sean Vucinich, Qiang Zhu
Abstract
The proliferation of artificial intelligence systems and their reliance on massive datasets have led to a renewed demand on privacy of data. Both the large data processing need and its associated data privacy demand have led to the development of techniques such as Federated Learning, a distributed machine learning technique with privacy preservation built-in. Within Federated Learning, as with other machine learning based techniques, the concern and challenges of ensuring that the decisions being made are fair and equitable to all users is paramount. This paper presents an up-to-date review of the motivations, concepts, characteristics, challenges, and techniques/methods related to fairness in Federated Learning from the literature. It also highlights open challenges and future research directions in evaluating and enforcing fairness in Federated Learning systems.