Scholarlink

<p>Synthetic data generation is essential for enhancing machine learning models while safeguarding data privacy. Wearable devices, which capture time series health data, greatly benefit from synthetic data due to the sensitivity and personal nature of this information. Generative Adversarial Networks (GANs) enable the creation of synthetic data but require significant computational resources, making edge computing – a distributed computing infrastructure – an attractive solution for privacy-sensitive tasks. Edge-based GANs demonstrate potential in mobile and wearable devices, improving both activity recognition and data privacy. This study proposes a novel workflow for edge-based synthetic data generation, showcasing its adaptability to various GAN architectures. We conducted a comparative analysis of GAN training performance in edge computing and cloud-based environments and evaluated the feasibility of model reuse across different users and activities within the same user context. The findings highlight the importance of integrating cloud computing with privacy-oriented strategies to achieve effective synthetic data generation using GANs in wearable technology.</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

A Workflow for Synthetic Biometric in-Edge Data Generation using GAN based models

  • 期刊论文

  • 发表日期：

    2025年6月17日

查看全文
  • L. Gustavo C. do Rêgo,
  • Ticiana L. Coelho da Silva,
  • Atslands R. Rocha,
  • Matheus R. Monteiro,
  • Lívia A. Cruz,
  • Karine Zeitouni,
  • Jorge B. Soares,
  • José Antônio F. de Macêdo

摘要

Synthetic data generation is essential for enhancing machine learning models while safeguarding data privacy. Wearable devices, which capture time series health data, greatly benefit from synthetic data due to the sensitivity and personal nature of this information. Generative Adversarial Networks (GANs) enable the creation of synthetic data but require significant computational resources, making edge computing – a distributed computing infrastructure – an attractive solution for privacy-sensitive tasks. Edge-based GANs demonstrate potential in mobile and wearable devices, improving both activity recognition and data privacy. This study proposes a novel workflow for edge-based synthetic data generation, showcasing its adaptability to various GAN architectures. We conducted a comparative analysis of GAN training performance in edge computing and cloud-based environments and evaluated the feasibility of model reuse across different users and activities within the same user context. The findings highlight the importance of integrating cloud computing with privacy-oriented strategies to achieve effective synthetic data generation using GANs in wearable technology.