Scholarlink

In recent years, institutions have explored synthetic data generation to retain statistical properties and enhance privacy. Mobile network data (MND), containing sensitive call records and SMS activity, exemplifies this need. Generative Adversarial Networks (GANs) are crucial for synthesizing tabular data like MND, with Copula Conditional Tabular GANs (Copula GANs) standing out for preserving spatial distribution and privacy. However, replicating the temporal structure of call couplings may require advanced models like Time GANs or Graph GANs, employing Recurrent Neural Networks (RNNs) or Transformers. These models enhance the replication of bivariate joint distributions by handling temporal dependencies. The study emphasizes Copula GANs’ promise for synthetic data generation and suggests exploring advanced GAN topologies for improved temporal structure replication.

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

Generation of Synthetic Data from Mobile Network Operators (MNO) Data Through Generative Adversarial Networks (GANs)

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Francesco Pugliese,
  • Angela Pappagallo,
  • Massimo De Cubellis

摘要

In recent years, institutions have explored synthetic data generation to retain statistical properties and enhance privacy. Mobile network data (MND), containing sensitive call records and SMS activity, exemplifies this need. Generative Adversarial Networks (GANs) are crucial for synthesizing tabular data like MND, with Copula Conditional Tabular GANs (Copula GANs) standing out for preserving spatial distribution and privacy. However, replicating the temporal structure of call couplings may require advanced models like Time GANs or Graph GANs, employing Recurrent Neural Networks (RNNs) or Transformers. These models enhance the replication of bivariate joint distributions by handling temporal dependencies. The study emphasizes Copula GANs’ promise for synthetic data generation and suggests exploring advanced GAN topologies for improved temporal structure replication.