Secure aggregation for heterogeneous enterprise data based on federated meta-learning

This paper proposes a secure and efficient data aggregation approach for heterogeneous enterprise data, leveraging federated meta-learning (FML) and data consolidation. The proposed approach addresses critical challenges in enterprise data, including data privacy, heterogeneous data distributions, and communication constraints. Specifically, the proposed approach enables decentralized devices to collaboratively train local models, which are aggregated by the server into a global model using meta-learning. Then, a consortium blockchain is used to ensure secure, immutable storage of aggregated data through a dual-chain structure: lightweight fluffy chains for temporary storage and heavyweight bulky chains for permanent records. In further, the proposed approach incorporates robust security mechanisms, such as XF authentication, timestamp-based counters to thwart replay attacks, asymmetric encryption for secure key exchange, and a Hampel filter to detect and mitigate model poisoning. Simulations results under Secure Water Treatment (SWaT) dataset are finally provided to demonstrate the superiority of the proposed approach. Specifically, the proposed approach achieves a lower validation delay scaling efficiently with training history size over the competing ones. Additionally, the proposed approach can enhance the system security by increasing the attack failure probability up to 15% over the competing ones.