Title: Enhancing data security in massive data sets using blockchain and federated learning: a loosely coupled approach

Authors: Haiyan Kang; Bing Wu

Addresses: Department of Information Security, Beijing Information Science and Technology University, Beijing, China ' Department of Information Security, Beijing Information Science and Technology University, Beijing, China

Abstract: The properties of the blockchain are not suitable for the storage of sensitive privacy data, and the excellent characteristics of the blockchain will be seriously affected by the existence of massive amounts of data on the chain. To address the above issues, propose a Loosely Coupled Local Differential Privacy Blockchain Federated Learning method (LL-BCFL). First of all, a client selection mechanism is put forward to ensure honesty and positivity in joining the training client and the correct effectiveness of the final global model aggregation. Secondly, use federated learning to alleviate the 'data silos' phenomenon and achieve joint training of big data stored in distributed multi-parties. Additionally, a differential privacy method is introduced to act on federated learning networks to avoid inference attacks. Finally, the MNIST dataset was used to confirm the availability of the LL-BCFL method on balanced and unbalanced datasets.

Keywords: federated learning; blockchain; differential privacy; massive data processing.

DOI: 10.1504/IJIPT.2024.143767

International Journal of Internet Protocol Technology, 2024 Vol.17 No.1, pp.31 - 41

Received: 31 Jul 2024
Accepted: 02 Oct 2024
Published online: 06 Jan 2025 *