首页|基于联邦学习的移动边缘节点计算的数据智能分类问题研究

基于联邦学习的移动边缘节点计算的数据智能分类问题研究

扫码查看
目前利用监督学习来解决网络流量数据分类的方法过度依靠高品质标记的大规模数据采样,以此导致难以直接应用于移动边缘节点计算等问题.因此,研究结合联邦学习与生成对抗网络(Generate Adversarial Networks,GAN)提出数据分类方法,并引入自监督学习提出了改进模型,同时对其性能进行了验证.实验结果表明,模型对比中改进模型三个指标的数值分别为0.90、0.63和0.59,均显著高于对比模型,其与集中式的无监督分类模型相比,其实际的分类准确度的提升超过了20%.不同参数分析中,参数a、b、c、d下的指标N均在较低的训练轮次时开始收敛整体上维持在0.52~0.62之间.综合来看,研究提出的改进模型在数据智能分类上具备较高的性能,在实际移动边缘节点计算数据分类中具备有效性.
Research on Intelligent Data Classification for Mobile Edge Node Computing Based on Federated Learning
The current methods of using supervised learning to solve network traffic data classification overly rely on high-quality labeled large-scale data sampling,which makes it difficult to directly apply to mobile edge node computing and other issues.There-fore,this study proposes a data classification method combining federated learning with Generate Adversarial Networks(GAN),and introduces self supervised learning to propose an improved model,while verifying its performance.The experimental results show that the values of the three indicators of the improved model in model comparison are 0.90,0.63,and 0.59,which are significantly high-er than the comparison model.Compared with the centralized unsupervised classification model,its actual classification accuracy has improved by more than 20%.In the analysis of different parameters,the index N under parameters a,b,c,and d all began to con-verge at lower training rounds and remained between 0.52 and 0.62 overall.Overall,the improved model proposed in the study has high performance in intelligent data classification and is effective in practical mobile edge node computing data classification.

federated learningmoving edge nodesgenerate adversarial networksclassification accuracyself supervised learning

杨文彬

展开 >

广州松田职业学院,广州 511370

联邦学习 移动边缘节点 生成对抗网络 分类准确度 自监督学习

省级基金项目

GDJG2021314

2024

自动化与仪器仪表
重庆工业自动化仪表研究所,重庆市自动化与仪器仪表学会

自动化与仪器仪表

CSTPCD
影响因子:0.327
ISSN:1001-9227
年,卷(期):2024.(6)
  • 13