H-PCFed: Hierarchical Privacy-Enhanced and Communication-Efficient using Federated Learning with Lightweight Deep Learning Model for Industrial IoTs
A C Ashwanthram, G Dhivyadarshan, Gautam R S, Ashwin S, S Kirthiga · 2025
In IIoT ecosystems, privacy, scalability, real-time decision-making, and communication overhead are the major challenges. Traditional federated learning faces issues such as inefficient model aggregation and difficulty in handling non-IID data distributions. To address these, the proposed framework integrates federated learning with lightweight deep learning models within a fog-cloud hierarchical structure. This combination does improve the efficiency of the aggregation of models, reduces time taken for communication, and ensures privacy using differential privacy techniques. Lightweight deep learning is a critical role in these models, enhancing scalability with robustness to data variety and classification accuracy, fitting the approach particularly well for real-time IIoT applications. This framework highlights privacy preservation, efficiency, and improved model performance to show the potential of FL and lightweight DL models in creating privacy-preserving, real-time IIoT solutions by addressing communication overhead challenges.