Unveiling the Trade-offs: A Parameter-Centric Comparison of Synchronous and Asynchronous Federated Learning

Ishaan Gakhar, Satvik Rajesh, Rohit Verma, Rajeev Shorey

Abstract: Federated Learning (FL) has emerged as a transformative paradigm enabling decentralized model training while preserving data privacy. Within FL, synchronous and asynchronous training modes are two widely adopted approaches. This paper compares synchronous and asynchronous federated learning by examining their impact across accuracy and convergence epochs. Through a series of experimental evaluations, we demonstrate the strengths and limitations of each approach in diverse settings, offering insights into the trade-offs and applicability of synchronous and asynchronous FL modes for various distributed learning scenarios. Our findings aim to guide researchers and practitioners in selecting optimal FL configurations based on their system requirements and performance priorities.