Blockchain Hybrid Federated Learning for Intelligent Secure Cloud Data Management

Abstract

The need for secure intelligent management of cloud data has become increasingly urgent with the advent of the technology that has raised the stakes and consequentially, greater risks to privacy and data integrity. This paper presents a new framework to apply Federated Learning (FL) to develop decentralized secure intelligent and privacy-preserving cloud data management (CDM) systems using Forms of Blockchain Technologies. FL allows for model training on distributed data without needing to share direct data access between endpoints, thus lowering the risks associated with data sharing. FL hybridizes well with Blockchain Technologies such as Dispersed Ledger Technologies (DLT) as it can record transactions of the FL model and who accessed the cloud data, allowing whoever owns the cloud data to audit the activity of accessing their cloud data. Using Blockchain provides a trustworthy method for sharing and for satisfying the taught information auditability (particularly for building trust among stakeholders). The proposed system architecture employs a secure aggregation protocol and lightweight blockchain consensus mechanism for best scalability and computational cost of a deployed application in a heterogeneous environment. The functional evaluations of the proposed methods showed an average accuracy of 94.7% against the multiple datasets from real-world scenarios, an improvement of 5.6% against secure privacy protection score, and a decrease of latency of 12.3% over the standard cloud management models. Therefore, this shows the framework's potential to provide confidential hyper-intelligent effective cloud data management and as a landmark for future work with decentralized parts of the cloud.