Towards Convergence of Blockchain and Self-sovereign Identity for Privacy-Preserving Secure Federated Learning

Abstract

More and more researchers are eager to train their Machine Learning (ML) model by the distributed dataset in a federated manner. However, there are numerous privacy and security concerns in federated learning, i.e., adversarial attacks, authentication, and model inversion attacks. In this paper, we integrate self-sovereign identity for privacy-preserving secure federated learning. The proposed framework represents identity management and authentication scheme for edge devices, which are participated in FL. It guarantees participants’ privacy, where users’ private data are not shared with the centralized training module. The trust triangle and Blockchain ensure the authenticity of the distributed federated learning, where data privacy is handled with Differential Privacy. Result analysis is done based on measured score and time complexity. In terms of accuracy, BCWD, HDD, and DD achieve $94.69\%$, $81.05\%$ and $78.00\%$, respectively on standard ML, whereas $94.45\%$, $80.59\%$ and $76.43\%$ on $\epsilon$ DP-based FL with privacy budget, $\epsilon$ = 3. For the time complexity measurement, the total number of transactions employed in the system is 1000. Transactions time and their verification took 3323.15 s. Moreover, the proposed system shows robust performance in both cases.

This research work is supported by the Key-Area Research and Development Program of Guangdong Province under Grant No. 2019B010137002, and National Key Research and Development Program under Grant NO. 2020YFA0909100.