Litcius/Paper detail

FEDge-HAR: An Optimized Private Mobile Edge-Enabled IoT Paradigm for Privacy of Human Activity Recognition

Ateeq Ur Rehman, Mahnoor Farooq, Fazlullah Khan, Gautam Srivastava, Rakan Aldmour, Ryan Alturki, Bandar Alshawi

2024IEEE Internet of Things Journal12 citationsDOI

Abstract

Federated learning (FL) has emerged as a pivotal technology for the Internet of Things (IoT) that models distributed client data without compromising privacy. The IoT-based wearable generates data and FL running on a private edge performing human activity recognition (HAR). In this article, we proposed a novel technique to protect sensitive data during the training process and ensure the confidentiality of model updates before transmission to the edge server. The proposed technique integrates the El-Gamal encryption technique for data protection, and the FL process is rigorously optimized using pruning, quantization, and network slicing. Pruning removes redundant connections, which reduces model complexity and communication delays. On the other hand, quantization decreases the bit precision of model parameters, and network slicing strategically allocates resources solely for FL resulting in low latency and optimal bandwidth utilization. The results are evaluated in terms of accuracy and communication overhead, which is highly required in real-world applications. Furthermore, the HAR system within PEC shows better results by achieving an accuracy of 99% at 300 epochs that outperformed existing machine learning (ML) algorithms.

Topics & Concepts

Computer scienceInternet of ThingsEnhanced Data Rates for GSM EvolutionEdge computingComputer securityInformation privacyInternet privacyActivity recognitionMobile telephonyComputer networkHuman–computer interactionTelecommunicationsArtificial intelligenceMobile radioContext-Aware Activity Recognition SystemsPrivacy-Preserving Technologies in DataUser Authentication and Security Systems
FEDge-HAR: An Optimized Private Mobile Edge-Enabled IoT Paradigm for Privacy of Human Activity Recognition | Litcius