FedSAUC: A Similarity-Aware Update Control for Communication-Efficient Federated Learning in Edge Computing

TL;DR

FedSAUC tackles the high communication and energy costs of federated learning on edge devices by using similarity-aware update control: devices with similar gradient updates are clustered and only representatives from each cluster participate in updates for a window of rounds, reducing transmissions while maintaining model diversity. The approach combines a model-similarity analysis stage (clustering gradient updates with methods like K-means, Agglomerative, and Spectral) and an update-control policy (randomly selecting active representatives with probability τ). Evaluation on a FedML-based edge testbed with Raspberry Pi 4 and Jetson Nano devices using MNIST shows notable communication savings and only modest short-term accuracy loss, with accuracy converging close to standard FL in the long run. These findings indicate FedSAUC can extend the lifetime of battery-powered edge devices and lower network load without sacrificing learning quality in practical edge scenarios.

Abstract

Federated learning is a distributed machine learning framework to collaboratively train a global model without uploading privacy-sensitive data onto a centralized server. Usually, this framework is applied to edge devices such as smartphones, wearable devices, and Internet of Things (IoT) devices which closely collect information from users. However, these devices are mostly battery-powered. The update procedure of federated learning will constantly consume the battery power and the transmission bandwidth. In this work, we propose an update control for federated learning, FedSAUC, by considering the similarity of users' behaviors (models). At the server side, we exploit clustering algorithms to group devices with similar models. Then we select some representatives for each cluster to update information to train the model. We also implemented a testbed prototyping on edge devices for validating the performance. The experimental results show that this update control will not affect the training accuracy in the long run.

Figures (7)

• Figure 1: The scenario of federated learning.
• Figure 2: Scenarios of (a) traditional federated learning and (b) our proposed FedSAUC.
• Figure 3: The system architecture of FedSAUC.
• Figure 4: Testbed prototyping: purple, green, and blue dots indicate Raspberry Pi 4, NVIDIA Jetson Nano (clients), and NVIDIA Jetson Nano (server), respectively.
• Figure 5: Testing accuracy of FedSAUC-trained models by (a) training set 1 and (b) training set 2.