Fog enabled distributed training architecture for federated learning

TL;DR

The paper tackles bandwidth, latency, and privacy challenges of cloud-based ML for IoT by proposing a fog-enabled distributed training architecture that combines online fog learning with privacy-preserving federated learning across edge, fog, and cloud. Local losses on devices are defined as $F_i(w) = \frac{1}{|D_i|} \sum_{(x_j,y_j)\in D_i} f(h(w,x_j),y_j)$ and the global objective is $F(w) = \frac{1}{|D|} \sum_{i=1}^k |D_i| F_i(w)$, enabling coordinated model updates without sharing raw data. The approach is validated via a Docker-based simulation using FMCW radar IIoT data for safe/human-position detection in a shared workspace, demonstrating online training on rapidly changing data with high final accuracy. This work provides a practical, low-latency, privacy-conscious pathway for distributed ML in delay-sensitive IoT deployments, with future avenues including energy-efficient communication and strengthened privacy protections for parameter exchanges.

Abstract

The amount of data being produced at every epoch of second is increasing every moment. Various sensors, cameras and smart gadgets produce continuous data throughout its installation. Processing and analyzing raw data at a cloud server faces several challenges such as bandwidth, congestion, latency, privacy and security. Fog computing brings computational resources closer to IoT that addresses some of these issues. These IoT devices have low computational capability, which is insufficient to train machine learning. Mining hidden patterns and inferential rules from continuously growing data is crucial for various applications. Due to growing privacy concerns, privacy preserving machine learning is another aspect that needs to be inculcated. In this paper, we have proposed a fog enabled distributed training architecture for machine learning tasks using resources constrained devices. The proposed architecture trains machine learning model on rapidly changing data using online learning. The network is inlined with privacy preserving federated learning training. Further, the learning capability of architecture is tested on a real world IIoT use case. We trained a neural network model for human position detection in IIoT setup on rapidly changing data.

Figures (4)

• Figure 1: Decentralized federated learning training paradigm
• Figure 2: Distributed learning architecture
• Figure 3: Global model performance on test data: Loss value (a) and Accuracy (b) at every training round
• Figure 4: Model performance on all devices