Efficient Privacy-Preserving Federated Learning for IIoT Using Dual Proxy Re-Encryption

Abstract

The Industrial Internet of Things (IIoT) is revolutionizing industries such as smart grids, healthcare, and predictive maintenance by harnessing big data and deep learning technologies. However, limited datasets in IIoT devices often result in suboptimal model performance and overfitting. Federated deep learning can mitigate this issue by leveraging distributed datasets across devices, but data privacy concerns persist, especially in sensitive applications like smart healthcare and energy management. Although numerous privacy-preserving federated learning schemes have been proposed, their vulnerabilities hinder widespread adoption due to insufficient guarantees for participant data privacy and the security of global model parameters. To address these challenges, we propose a novel deep learning framework that leverages proxy re-encryption techniques to enhance data privacy. Our scheme employs a dual proxy re-encryption mechanism to enhance data security, enabling each participant to securely access global model parameters without relying on a proxy server during training rounds. This not only prevents unauthorized access by the parameter server, but also resists collusion attacks between the parameter server and participants. Furthermore, the confidentiality of the proxy server’s private key is maintained, even in cases of collusion involving the parameter server and participants. A comparative analysis with existing schemes highlights the advantages of our approach, including reduced communication overhead and computational complexity, as demonstrated by experimental results.