The paper presents a solution to challenges in Federated Learning (FL) – data heterogeneity and model security. The solution, called Estimated Mean Aggregation (EMA), acts as a reference point for advanced aggregation techniques in FL systems. EMA enhances model security by effectively handling malicious outliers and uncovers data heterogeneity to ensure trained models’ adaptability across various client datasets. Through several experiments, EMA demonstrates high accuracy and area under the curve (AUC) compared to alternative methods, establishing itself as a robust baseline for evaluating FL aggregation methods’ effectiveness and security.
Publication date: 22 Sep 2023
Project Page: https://github.com/AnonymousUser08/EMA.git
Paper: https://arxiv.org/pdf/2309.12267
