SCALE: Self-regulated Clustered federAted LEarning in a Homogeneous Environment

TL;DR

The paper tackles high communication costs and reliance on centralized infrastructures in federated learning by introducing SCALE, a self-regulated clustered FL framework that eliminates edge servers and uses a server-assisted Proximity Evaluation for dynamic clustering. It combines a Hybrid Decentralized Aggregation Protocol with a dynamically elected driver, decentralized driver selection, check-pointing, and health-status verification to drastically reduce global communication while preserving learning performance. Experimental validation on a breast cancer dataset shows SCALE achieving nearly tenfold reductions in communication overhead, reduced training latency, and lower energy consumption, while maintaining competitive accuracy and robustness. This approach offers a scalable, privacy-preserving path for federated learning in fluid edge environments with reduced infrastructure costs.

Abstract

Federated Learning (FL) has emerged as a transformative approach for enabling distributed machine learning while preserving user privacy, yet it faces challenges like communication inefficiencies and reliance on centralized infrastructures, leading to increased latency and costs. This paper presents a novel FL methodology that overcomes these limitations by eliminating the dependency on edge servers, employing a server-assisted Proximity Evaluation for dynamic cluster formation based on data similarity, performance indices, and geographical proximity. Our integrated approach enhances operational efficiency and scalability through a Hybrid Decentralized Aggregation Protocol, which merges local model training with peer-to-peer weight exchange and a centralized final aggregation managed by a dynamically elected driver node, significantly curtailing global communication overhead. Additionally, the methodology includes Decentralized Driver Selection, Check-pointing to reduce network traffic, and a Health Status Verification Mechanism for system robustness. Validated using the breast cancer dataset, our architecture not only demonstrates a nearly tenfold reduction in communication overhead but also shows remarkable improvements in reducing training latency and energy consumption while maintaining high learning performance, offering a scalable, efficient, and privacy-preserving solution for the future of federated learning ecosystems.

Figures (2)

• Figure 1: The system architecture elucidates various components in the SCALE approach, with a key focus on driver nodes, client nodes, and communication with the global server. Additionally, it provides a concise overview of the distinctive features of the global server. The significance of local aggregations and checkpoints has been underscored for each driver node, highlighting their essential role in the primary functionality of clusters.
• Figure 2: The comparison metrics between different scales of traditional federated learning and the SCALE approach are under scrutiny. These statistical measures are derived at the conclusion of randomly selected epoch rounds. This analysis aims to provide a nuanced understanding of how the two approaches differ in performance across various evaluation criteria.