【Author】
Qi, Minfeng; Wang, Ziyuan; Wu, Fan; Hanson, Rob; Chen, Shiping; Xiang, Yang; Zhu, Liming
【Source】INFORMATION SECURITY AND PRIVACY, ACISP 2021
【Abstract】Information Silo is a common problem in most industries, while Federated Learning (FL) as an emerging privacy-preservation technique aims to facilitate data sharing to solve the problem. It avoids data leakage by sharing the model gradient instead of the raw data. However, there are some challenges of FL, such as Single Point of Failure (SPoF), gradient privacy, and trust issues. This paper proposes a Homomorphic-integrated and blockchain-based FL model to address the above issues. It provides gradient privacy protection by employing Homomorphic, and uses a smart contract-based reputation scheme and an on/off-chain storage strategy to respectively solve FL trust and blockchain storage issues. In the end, it evaluates the proposed model by providing a qualitative privacy analysis and conducting preliminary experiments on model performance.
【Keywords】Blockchain; Smart contract; Homomorphic encryption; Federated learning; Privacy protection; Data sharing
【标题】用于保护隐私的区块链联邦学习模型：系统设计
【摘要】信息孤岛是大多数行业的常见问题，而联邦学习（FL）作为一种新兴的隐私保护技术旨在促进数据共享以解决该问题。它通过共享模型梯度而不是原始数据来避免数据泄漏。但是，FL 也存在一些挑战，例如单点故障 (SPoF)、梯度隐私和信任问题。本文提出了一种同态集成和基于区块链的 FL 模型来解决上述问题。它通过使用同态提供梯度隐私保护，并使用基于智能合约的信誉方案和链上/链下存储策略分别解决FL信任和区块链存储问题。最后，它通过提供定性隐私分析和对模型性能进行初步实验来评估所提出的模型。
【关键词】区块链；智能合约；同态加密；联邦学习；隐私保护;数据共享
评论