[1]袁 媛,袁 松*.一种区块链支持的联邦学习认知模型[J].计算机技术与发展,2023,33(11):215-220.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 032]
　YUAN Yuan,YUAN Song *.Federal Learning of Cognitive Model Supported by Blockchain[J].,2023,33(11):215-220.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 032]

点击复制

一种区块链支持的联邦学习认知模型()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年11期

页码:

215-220

栏目:

新型计算应用系统

出版日期:

2023-11-10

文章信息/Info

Title:

Federal Learning of Cognitive Model Supported by Blockchain

文章编号:

1673-629X(2023)11-0215-06

作者:

袁 媛¹ ; 袁 松^2*

1. 武汉体育学院 体育工程与信息技术学院,湖北 武汉 430079;

2. 杭州电子科技大学 电子信息学院,浙江 杭州 310018

Author(s):

YUAN Yuan1 ; YUAN Song2 *

1. Sports Engineering and Information Technology Academy,Wuhan Sports University,Wuhan 430079,China;

2. Electronic Information Academy,Hangzhou Dianzi University,Hangzhou 310018,China

关键词:

认知计算; 联邦学习; 区块链; 马尔可夫决策; Q 强化学习

Keywords:

cognitive computing; federal learning; blockchain; Markov decision; Q reinforcement learning

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 11. 032

摘要:

物联网数据的有效认知是实现智能制造的重要载体,针对物联网数据驱动认知性能和抗攻击力问题,提出一种区块链支持的联邦学习认知模型。 首先,该模型将中心化的参数服务器构建为去中心化的参数聚合器,构建了一个区块链支持的联邦学习数据驱动认知计算框架;然后,该模型利用共识算法构建激励机制和交叉验证机制,生成区块链支持的联邦学习数据驱动认知,用以维护物联网数据认知性能;最后,该模型利用改进马尔可夫决策过程完成模型聚合实现全局更新的最优策略,并提出 Q 强化学习方法支持联邦学习的认知计算,用以增强模型抗攻击能力。 实验结果表明:所提的模型不仅可以优化全局模型的准确率和损失率,还可以使平均存储开销和计算时延收敛稳定。 同时,所提的模型与联邦平均聚合算法相比准确率平均高 10% ,与区块链方法和联邦学习方法相比具有更有效的抗攻击能力,为实现物联网数据驱动提供了支持。

Abstract:

Effective recognition of data in internet of things is an important carrier for intelligent manufacturing. focusing on the data-driven cognitive performance and anti-attack ability?

in internet of things,a federal learning of cognitive model supported by blockchain( FB_DC) is proposed. Firstly,a centralized parameter server is constructed as a decentralized parameter aggregator to build a blockchain-supported federated learning cognitive computing framework. Then,federal learning of data - drive cognitive is generated by building　incentive mechanism and cross verification mechanism through consensus algorithm of blockchain, maintaining the data cognitiveperformance of the internet of things. Lastly,

the model aggregation is completed by using improved Markov decision process to realizethe optimal strategy of global update,and the Q reinforcement learning method is proposed to support the federated learning that realizingcognitive computing. Experimental results show that the proposed model can not only improve the accuracy rate and?

loss rate,but alsomake the average storage cost and computing delay converge and stabilize. While the accuracy is higher than 10% on average comparingwith the federal average aggregation algorithm ( FA) ,and the anti-attack ability is more effective comparing with the blockchain methodand federal method,which provides support for the realization of the internet of things data-driven.

常用功能