Scholarlink

<p>In the era of the Internet of Things, smart devices are widely used in industries such as healthcare, finance, and telecommunications, generating massive and diverse amounts of data. The aggregation and mining of these data can provide strong business support. Still, they also face challenges such as privacy breaches and data heterogeneity, which hinder the construction of high-quality personalized models. For this reason, this paper proposes a federated learning solution based on cloud edge architecture, combines the advantages of edge computing and cloud computing, and constructs a personalized model with a privacy protection mechanism through condition generation adversary network (cGAN) and parameter sparsity technology. Specifically, the model on the edge server is divided into two parts: using cGAN to simulate the output features of the shallow part of the model on the edge side, sparsifying the local generator and deep network, and then uploading the results to the central cloud server. By sharing edge-side generators, common knowledge is aggregated to improve local network performance while hiding shallow parts of the model. The experimental results show that this method can maintain high prediction accuracy while protecting privacy and verifying its feasibility and effectiveness in practical applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing personalized model construction and privacy protection in federated learning with generative adversarial networks and parameter sparsification

  • 期刊论文

  • 发表日期：

    2025年3月1日

查看全文
  • Zhongyuan Jing,
  • Ruyan Wang

摘要

In the era of the Internet of Things, smart devices are widely used in industries such as healthcare, finance, and telecommunications, generating massive and diverse amounts of data. The aggregation and mining of these data can provide strong business support. Still, they also face challenges such as privacy breaches and data heterogeneity, which hinder the construction of high-quality personalized models. For this reason, this paper proposes a federated learning solution based on cloud edge architecture, combines the advantages of edge computing and cloud computing, and constructs a personalized model with a privacy protection mechanism through condition generation adversary network (cGAN) and parameter sparsity technology. Specifically, the model on the edge server is divided into two parts: using cGAN to simulate the output features of the shallow part of the model on the edge side, sparsifying the local generator and deep network, and then uploading the results to the central cloud server. By sharing edge-side generators, common knowledge is aggregated to improve local network performance while hiding shallow parts of the model. The experimental results show that this method can maintain high prediction accuracy while protecting privacy and verifying its feasibility and effectiveness in practical applications.