Modular Federated Learning: A Meta-Framework Perspective

TL;DR

This paper reframes Federated Learning as a modular meta-framework composed of interoperable building blocks across Infrastructure, Data, Threat Protection, Distributed Optimisation, and Model Design, with Aggregation and a novel Alignment operator guiding knowledge integration. It connects FL with historical distributed optimisation, introduces a structured eight-module architecture, and surveys practical Python frameworks, datasets, and deployment considerations. The work highlights risk areas such as privacy, security, convergence under non-IID data, and environmental impact, and proposes future directions including standardisation, resource-conscious learning, and trustworthy FL with robust uncertainty estimation. By formalising FL as a composable system, the authors provide a blueprint for designing adaptable, scalable, and privacy-preserving distributed learning solutions with broad applicability across healthcare, industry, space, and beyond.

Abstract

Federated Learning (FL) enables distributed machine learning training while preserving privacy, representing a paradigm shift for data-sensitive and decentralized environments. Despite its rapid advancements, FL remains a complex and multifaceted field, requiring a structured understanding of its methodologies, challenges, and applications. In this survey, we introduce a meta-framework perspective, conceptualising FL as a composition of modular components that systematically address core aspects such as communication, optimisation, security, and privacy. We provide a historical contextualisation of FL, tracing its evolution from distributed optimisation to modern distributed learning paradigms. Additionally, we propose a novel taxonomy distinguishing Aggregation from Alignment, introducing the concept of alignment as a fundamental operator alongside aggregation. To bridge theory with practice, we explore available FL frameworks in Python, facilitating real-world implementation. Finally, we systematise key challenges across FL sub-fields, providing insights into open research questions throughout the meta-framework modules. By structuring FL within a meta-framework of modular components and emphasising the dual role of Aggregation and Alignment, this survey provides a holistic and adaptable foundation for understanding and advancing FL research and deployment.

Figures (22)

• Figure 1: Standard centralised Federated Learning Setting. Each client $c$ learns a local model and shares it with a central server, aggregating the individual model parameters and then sending this new global model back to the clients. This process is repeated for several rounds until a certain criterion is met.
• Figure 2: Contrasting different infrastructure meta-types of Federated Learning architectures. Cross-Device FL utilises low-resource devices with uncontrollable training availability (e.g. smartphones, microcontrollers). Cross-Silo is based on a reliable infrastructure such as a collection of servers. Hierarchical FL mixes Cross-Device and Cross-Silo FL infrastructure.
• Figure 3: Abstract overview of the main differences between Horizontal and Vertical Federated Learning. Horizontal Federated Learning (left) expects the same features across clients, while Vertical Federated Learning (right) involves clients with different features.
• Figure 4: Meta-Framework overview of the Federated Learning paradigm as a fusion of key modules categorised into Infrastructure, Data, Threat Protection, Distributed Optimisation, and Model Design.
• Figure 6: Split-NN architecture overview. Distributed Learning network where each client trains an intermediate representation and shares it with a server, which then fuses them, applies backpropagation, and defuses the update with the clients.

Theorems & Definitions (1)

• Definition 1: Differential Privacy