Practical Federated Learning without a Server

2025-03-07

AI Summary

To address the deployment limitations of federated learning (FL) imposed by reliance on a centralized parameter server, this paper introduces Plexus—the first serverless, large-scale decentralized FL system. Plexus eliminates the need for any central coordinator through four key innovations: (1) a distributed-consensus-based client sampling mechanism; (2) a local model aggregation strategy; (3) latency-aware dynamic topology construction; and (4) a lightweight communication protocol. Extensive experiments at scale—up to 1,000 participating nodes—demonstrate that Plexus achieves 1.4–1.6× training speedup over prior decentralized FL approaches, reduces communication overhead by 15.8–292×, and cuts computational resource consumption by 30.5–77.9×. These gains significantly enhance system scalability while strengthening privacy preservation by removing centralized trust assumptions and minimizing cross-node data exposure.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Multiagent Systems: Multiagent Learning
• Data Mining & Knowledge Management: Scalability, Parallel & Distributed Systems

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Systems and Infrastructure for Web, Mobile and WoT: Decentralized Web and Fediverse systems
• Security and Privacy: Large-scale security measurements

Abstract

Federated Learning (FL) enables end-user devices to collaboratively train ML models without sharing raw data, thereby preserving data privacy. In FL, a central parameter server coordinates the learning process by iteratively aggregating the trained models received from clients. Yet, deploying a central server is not always feasible due to hardware unavailability, infrastructure constraints, or operational costs. We present Plexus, a fully decentralized FL system for large networks that operates without the drawbacks originating from having a central server. Plexus distributes the responsibilities of model aggregation and sampling among participating nodes while avoiding network-wide coordination. We evaluate Plexus using realistic traces for compute speed, pairwise latency and network capacity. Our experiments on three common learning tasks and with up to 1000 nodes empirically show that Plexus reduces time-to-accuracy by 1.4-1.6x, communication volume by 15.8-292x and training resources needed for convergence by 30.5-77.9x compared to conventional decentralized learning algorithms.

Problem

• Decentralized Federated Learning without a central server
• Reduces communication volume and training resources
• Improves time-to-accuracy in large network environments

Innovation

• Decentralized Federated Learning without central server
• Distributed model aggregation and sampling among nodes
• Reduced time-to-accuracy and communication volume significantly