Differential privacy-enabled federated learning for secure neural synchronization in protecting industrial data streams

Differential privacy-enabled federated learning for secure neural synchronization in protecting industrial data streams

연구 분야: Infrastructure

학회: Cluster Computing

This research introduces an innovative secure Federated Learning (FL) method for industrial data streams through differential privacy. The growing dependence of industrial systems on machine learning for processing sensitive data presents substantial challenges in preserving privacy and security while maintaining model performance. Current FL strategies must deal with data privacy breaches in addition to expensive communication requirements and susceptibility to adversarial threats. The paper introduces an adaptive differential privacy framework that dynamically modifies noise levels according to data sensitivity to achieve the best trade-off between privacy protection and model performance. The implementation of hierarchical neural synchronization in this study helps minimize communication overhead while improving scalability for Industrial Internet of Things (IIoT) networks. Adversarial-aware gradient clipping is integrated into the system to defend against threats, including gradient inversion attacks and model poisoning attacks. The analysis of industrial datasets BoT-IoT and UNSW-NB15 demonstrates that the method surpasses existing methods with substantial performance gains. This framework delivers a 27% decrease in privacy loss while simultaneously improving the accuracy of the model by 15% and reducing communication costs by 22%. The findings demonstrate how this method can support secure and efficient FL processes for industrial applications which require strict privacy protection. The research explores essential privacy and scalability concerns along with adversarial robustness to enhance FL systems’ effectiveness and operational efficiency. The framework offers an ideal solution for IIoT environments that require secure and efficient data processing. This solution delivers a FL system that scales effectively while preserving privacy and ensuring robustness across multiple industries, including manufacturing healthcare and automation.