Blockchain federated learning with sparsity for IoMT devices.

The recent advancements in the Internet of Medical Things (IoMT) have significantly contributed to improving personalized medicine and patient diagnosis and monitoring. Nonetheless, the implementation of IoMT may encounter obstacles due to security and privacy concerns. Federated learning emerges as a promising solution, enabling multiple devices to collaborate on training rich, heterogeneous datasets while preserving privacy. Despite its potential, traditional federated learning methods exhibit vulnerabilities such as single points of attack or failure and performance degradation with heterogeneous data. To this end, this paper proposes a blockchain federated learning system to address these limitations. In the proposed blockchain, a Proof-of-Contribution-Earned (PoCE) consensus protocol is designed for block propagation and miners’ selection using an improved addition tic-tac-toe game. To overcome the challenge related to heterogeneous data, a reward system based on a cooperation strategy is proposed to ensure that high-quality data is shared among health institutions. We employ a Convolutional Neural Network (CNN) where we replace the fully connected layers with sparse ones to minimize the number of parameters using an exponential random graph while maintaining model accuracy. The experimental results on real-world heterogeneous data demonstrate that the proposed system outperforms existing state-of-the-art systems in terms of accuracy and convergence rate. Security analysis reveals that the proposed system is robust against existing security and privacy-related attacks. [ABSTRACT FROM AUTHOR]