Abstract
Combining blockchain and federated learning has emerged as a promising solution for secure, privacy-preserving data sharing and collaborative training of machine learning models in decentralised settings. However, their methods suffer from scalability issues, computational overhead, and challenges in preserving privacy in dynamic environments such as IoT, healthcare, and smart cities. Although blockchain-based solutions ensure data confidence and provenance, they come at the cost of the high computational overhead involved in transaction validation and consensus. In a similar spirit to federated learning, where local models are trained on sensitive data, and only aggregation is performed without centralising the data, traditional approaches fail to provide efficient aggregation and do not allow for secure transmission. This paper proposes BLEND, a Blockchain- and federated-learning-enabled network for Data sharing, as a new framework to tackle this issue. It will integrate a new consensus protocol, adaptive encryption schemes, and smart contract-based aggregation to deliver outstanding security, scalability, and operational efficiency. The proposed framework automatically adjusts its encryption strength in real time based on threat intelligence, enabling secure data while enhancing model performance. We show through experiments that BLEND outperforms existing blockchain-based federated learning methods in terms of latency, computation cost, and model accuracy by several orders of magnitude. Exploiting this commonality, the frame can reduce the existing framework's computational overhead by as much as 20%, while maintaining around 90-95% of the original model's accuracy and achieving lower latency than traditional methods. The framework enables privacy-friendly practical scenarios that enable large-scale, distributed data sharing and model training. BLEND offers industries such as healthcare and IoT a performant, robust, and scalable instruction-level collaborative solution that does not compromise performance or privacy.