基于联邦学习算法的复杂网络大数据隐私保护  

Privacy Protection of Big Data in Complex Network Based on Federated Learning Algorithm

作　　者：符太东 李育强 FU Tai-dong;LI Yu-qiang(Management Center of Big Data and Network,Jilin University,Changchun Jilin 130000,China;Information Center,University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China)

机构地区：[1]吉林大学大数据和网络管理中心,吉林长春130000 [2]电子科技大学信息中心,四川成都611731

出　　处：《计算机仿真》2024年第6期498-502,共5页Computer Simulation

摘　　要：网络数据中含有大量的数据信息,为避免大数据泄露用户的隐私信息,有效保证复杂网络大数据的安全性,提出一种基于联邦学习算法的复杂网络大数据隐私保护方法。将联邦学习算法和VAE技术结合,构建变分自编码器,训练本地数据集。通过训练好的数据建立参数聚合链,将训练过程中产生的中间参数设定为证据,利用激烈节点验证模型参数,删除虚假以及低质量参与节点。建立联合模型,通过模型参数的交互代替数据的直接交换,完成复杂网络大数据隐私保护。实验结果表明,所提方法可以有效保护大数据隐私,同时可以提升整体运行效率和抵抗病毒攻击能力,降低平均存储损耗。Network data contains a lot of data information.In order to avoid leaking users'privacy information and effectively ensure the security of complex network big data,this paper presented a privacy protection method for com⁃plex network big data based on federated learning algorithm.Firstly,federated learning algorithm was combined with VAE technology to build a variational autoencoder for training local data sets.Then,parameter aggregation chain was constructed through the trained data.Meanwhile,the intermediate parameters generated in the process of training were set as evidence.Moreover,model parameters were verified by using violent nodes.After that,false and low-quality participating nodes were deleted.Furthermore,a joint model was built.Then,the direct exchange of data was replaced with the interaction of model parameters,thus completing the privacy protection of complex network big data.The ex⁃perimental results show that the proposed method can effectively protect big data privacy while improving the operation efficiency and resistance to virus attacks,thus reducing the average storage loss.

关 键 词：联邦学习算法 复杂网络 大数据 隐私保护 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]