Edge-Based Communication Optimization for Distributed Federated Learning

Federated learning can achieve distributed machine learning without sharing privacy and sensitive data of end devices. However, high concurrent access to cloud servers increases the transmission delay of model updates. Some local models may be unnecessary with an opposite gradient from the global model, thus incurring many additional communication costs. Existing work mainly focuses on reducing communication rounds or cleaning local defect data, and neither takes into account latency associated with high server concurrency. To this end, we study an edge-based communication optimization framework to reduce the number of end devices directly connected to the parameter server while avoiding uploading unnecessary local updates. Specifically, we cluster devices in the same network location and deploy mobile edge nodes in different network locations to serve as hubs for cloud and end devices communications, thereby avoiding the latency associated with high server concurrency. Meanwhile, we propose a method based on cosine similarity to filter out unnecessary models, thus avoiding unnecessary communication. Experimental results show that compared with traditional federated learning, the proposed scheme reduces the number of local updates by 60%, and the convergence speed of the evaluated model increases by 10.3%.

I. Introduction

Machine learning (ML) [1] has been successfully applied in a wealth of practical artificial intelligence (AI) applications in the field of computer vision, natural language processing, healthcare, finance, robotics and many others. These applications enhance the operational efficiency of the entire manufacturing process and generate huge amounts of data daily [2], [3]. As a result of industry competition and data privacy, data is not shared in most industries [4], [5]. Even in the same company, data integration between different departments is faced with massive resistance [6], not to mention integrating data from various agencies, which is almost impossible in reality. Besides, with the further development of big data, the emphasis on data privacy and security has become a worldwide trend. As the essential technology of AI, federated learning (FL) [7], [8] is a promising approach to resolve this challenge. End devices participating in federated learning only need to train the model locally and send the trained model to the cloud server for aggregation. In a word, FL can achieve machine learning under the condition of protecting data privacy.