Scholarlink

The Internet of Things has integrated seamlessly woven itself into different aspects of our lives, playing a crucial role in sectors such as security, transportation, smart homes, and healthcare. Within these areas, Internet of Things utilizes various sensors to generate substantial amounts of data, which are traditionally stored in centralized systems like Cloud Computing. However, this method brings about challenges, including high latency, response times, and security issues during data processing and analysis. To address these challenges, federated learning a groundbreaking branch of artificial intelligence has emerged. This new paradigm enables machine learning techniques to leverage decentralized data and computing power, primarily found in edge devices, ensuring a personalized user experience without compromising privacy. In this article, we propose a Federated CNN Model Based on Knowledge Distillation (MD_CNN) that clients train local models on private datasets, then share model weights with a central server, which uses these models for labeling and training a global model. Using the UCI-HAR dataset, achieving an accuracy of 83.46%, a precision of 83.62%, a recall of 83.46%, an AUC of 93.46%, and a loss of 1.23.

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

Federated Learning Using Knowledge Distillation for CNN-Based eHealth Data Analysis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chaimae Zaoui,
  • Faouzia Benabbou,
  • Chaimaa Bouaine,
  • Yassir Matrane

摘要

The Internet of Things has integrated seamlessly woven itself into different aspects of our lives, playing a crucial role in sectors such as security, transportation, smart homes, and healthcare. Within these areas, Internet of Things utilizes various sensors to generate substantial amounts of data, which are traditionally stored in centralized systems like Cloud Computing. However, this method brings about challenges, including high latency, response times, and security issues during data processing and analysis. To address these challenges, federated learning a groundbreaking branch of artificial intelligence has emerged. This new paradigm enables machine learning techniques to leverage decentralized data and computing power, primarily found in edge devices, ensuring a personalized user experience without compromising privacy. In this article, we propose a Federated CNN Model Based on Knowledge Distillation (MD_CNN) that clients train local models on private datasets, then share model weights with a central server, which uses these models for labeling and training a global model. Using the UCI-HAR dataset, achieving an accuracy of 83.46%, a precision of 83.62%, a recall of 83.46%, an AUC of 93.46%, and a loss of 1.23.