All | Since 2020 | |
Citation | 172 | 110 |
h-index | 7 | 5 |
i10-index | 1 | 0 |
WJERT Citation 
Login
News & Updation
Abstract
ADAPTIVE FEDERATED LEARNING FOR SECURE AND EFFICIENT EDGE INTELLIGENCE IN MOBILE COMPUTING
Gift Aruchi Nwatuzie*
ABSTRACT
Federated Learning (FL) enables collaborative model training across distributed edge devices while preserving user data privacy. However, practical deployment faces challenges including heterogeneous device reliability, communication con- straints, and adversarial threats. This paper presents Adaptive Federated Model Optimization (AFMO), a novel framework that enhances the robustness, efficiency, and privacyof FL in mobile edge environments. AFMO introduces three core components: a Trust-Weighted Participation Index (TWPI) to dynamically prioritize reliable devices, a Homomorphic-Obfuscation Transfor- mation (HOT) that adaptively secures model updates with trust- adjusted noise and lightweight encryption, and Communication- Sparse Gradient Regulation (CSGR) to minimize communication overhead by selecting critical updates. Extensive experiments on edge-oriented datasets—CIFAR-Mobile, EdgeSpeech, and MHealth—demonstrate that AFMO improves model accuracy by up to 7%, reduces communication costs by 60%, and shows superior resilience to adversarial attacks compared to existing FL baselines.
[Full Text Article] [Download Certificate]