Scaling Data Analysis Services in an Edge-based Federated Learning Environment

Abstract

Federated Learning represents among the most important techniques used in recent years. It enables the training of Machine Learning-related models without sharing sensitive data. Federated Learning mainly exploits the Edge Computing paradigm for training data acquired from the surrounding environment. The solution proposed in this paper seeks to optimize all the processes involved within a Federated Learning client through transparent scaling across different devices. The proposed architecture and implementation abstracts the Federated Learning client architecture to create a transparent cluster that can optimize the complicated computation and aggregate the data to solve the heterogeneous distribution issue of the data in Federated Learning applications.