NIKT 2024 Publication: Understanding Federated Learning from IID to Non-IID

Published: November 24, 2024

Understanding Federated Learning from IID to Non-IID dataset: An Experimental Study

Authors: Jungwon Seo, Ferhat Ozgur Catak, Chunming Rong

Conference: NIKT 2024: Norwegian ICT Conference for Research and Education
Host: University of Bergen
Location: Bergen, Norway
Dates: November 25-27, 2024

• Publication Page

Brief Summary

This study examines why federated learning performance drops under non-IID client data. The paper highlights that inconsistent client loss landscapes are a key cause of degradation, and organizes existing solutions into two broad strategies: adjusting parameter update paths and modifying client loss landscapes.