Principles and Components of Federated Learning Architectures

TL;DR

The paper surveys Federated Learning principles, architectures, and security considerations, emphasizing data localization, privacy techniques, and communication efficiency. It analyzes cooperative training workflows, model aggregation methods such as FedAvg, and privacy-preserving strategies, while examining scalability through hierarchical and edge-enabled designs. The work reviews XAI and Zero Trust Architecture as mechanisms to enhance security, accountability, and continuous verification in FL systems, and evaluates algorithmic performance under i.i.d. and non-i.i.d. data. It culminates in architectural patterns and a forward-looking roadmap that guides practitioners and researchers toward private, robust, and scalable FL deployments.

Abstract

Federated Learning (FL) is a machine learning framework where multiple clients, from mobiles to enterprises, collaboratively construct a model under the orchestration of a central server but still retain the decentralized nature of the training data. This decentralized training of models offers numerous advantages, including cost savings, enhanced privacy, improved security, and compliance with legal requirements. However, for all its apparent advantages, FL is not immune to the limitations of conventional machine learning methodologies. This article provides an elaborate explanation of the inherent concepts and features found within federated learning architecture, addressing five key domains: system heterogeneity, data partitioning, machine learning models, communication protocols, and privacy techniques. This article also highlights the limitations in this domain and proposes avenues for future work. Besides, we provide a set of architectural patterns for federated learning systems, which are derived from the systematic survey of the literature. The main elements of FL, the fundamentals of Federated Learning, and a few architectural specifics will all be better understood with the aid of this research.

Figures (17)

• Figure 1: General Outlook of Federated Learning lo2022architectural
• Figure 2: The standard FL workflow involves a federation of training nodes that receive a global model, periodically send partially trained models to a central server for aggregation, and continue training on a consensus model provided by the server. We call this process the FL aggregation server (a). (b) FL Peer-to-Peer: An alternative FL formulation where each training node performs its own aggregation and shares its partially learned model with some or all of its peers. A basic non-FL training approach, known as "centralized training" (c), involves data collection sites providing data to a central data lake, from which they retrieve data for independent local training rieke2020future
• Figure 3: Overview of different FL theme options. FL topology: communication architecture of federation. (a) Centralized: models are collected, aggregated, and distributed among training nodes (hub and spokes) by an aggregation server that also manages the training iterations. (b) Distributed: aggregation happens simultaneously at each training node connected to one or more peers. (c) Hierarchical: Peer-to-peer federations and aggregation server federations can be combined to create various sub-federations forming a federated network (d). FL computation plan: Passing the model through multiple partners. Cycles of transfer learning and sequential training. (f) Peer-to-peer, (g) aggregation server rieke2020future
• Figure 4: An example of federated learning running on caches and edge computing, where the aggregator can be an edge computing platform on an edge network (such as a wireless base station or an unmanned aerial vehicle) and the local learner can be an edge user (an autonomous vehicle in an autonomous vehicle network, or an augmented/virtual platform for a user reality headset) niknam2020federated
• Figure 5: Universal architecture for federated learning aledhari2020federated.