基于混合隐私的区块链高效模型协同训练共享方案

针对海量数据下，基于区块链的联邦学习数据共享平台面临的效率低下和隐私泄露问题，该文提出基于混合隐私的区块链高效模型协同训练共享方案。在该方案中，首先根据欧氏距离设计了一种基于相似度的训练成员选择算法来选择训练成员，组成联邦社区，即通过选取少量的高匹配训练节点来提高训练的效率和效果。然后，结合阈值同态加密和差分隐私，设计一种基于混合隐私技术的模型协同训练共享方案来保证训练和共享过程中的隐私性。实验结果和系统实现表明，所提方案可以在保证训练结果准确率的情况下，实现高效训练和隐私保护下的数据共享。

Efficient Model Collaborative Training and Sharing Scheme of Blockchain Based on Hybrid Privacy

Considering the problems of inefficiency and privacy leakage faced by the blockchain-based federated learning data sharing platform under massive data, an efficient model collaborative training and sharing scheme of blockchain based on hybrid privacy is proposed. In this scheme, a similarity-based training member selection algorithm according to Euclidean distance is first designed to select training members, forming a federated community, that is, to improve the efficiency and effect of training by selecting a small number of high-matching training nodes. Then, combined with threshold homomorphic encryption and differential privacy, a model collaborative training and sharing scheme based on hybrid privacy technology is constructed to ensure the privacy in the training and sharing process. The experimental results and system implementation show that the proposed scheme can achieve efficient training and data sharing under privacy protection while ensuring the accuracy of the training results.