Scholarlink

The research proposes an Efficient Federated Learning System with Integrated Knowledge Distillation that can be deployed to constrained Ambient IoT environments. Locally trained autonomous client models and the overall combined global model were determined by federated averaging on the Edge-IIoT dataset, which has been split into 10 clients. The test accuracy of 92.65% and the AUC of 0.8729 were obtained by the produced global model prior to the knowledge distillation phase. Afterwards, a condensed student model to facilitate the optimum deployment on the terminal devices were distilled by the produced global model, which produced the test accuracy of 85.87% and the AUC of 0.8727. The experimental result verifies that the proposed system has high predictive accuracy with high compression of the model that can be efficiently and scalably deployed on low-power IoT devices.

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 Lightweight Federated Learning Framework for Resource-Constrained Devices Using Knowledge Distillation in Ambient IoT Environments

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • R. Josephine,
  • Soumya,
  • Praveen Giridhar Pawaskar,
  • Robin Rohit Vincent,
  • S. P. Anandaraj

摘要

The research proposes an Efficient Federated Learning System with Integrated Knowledge Distillation that can be deployed to constrained Ambient IoT environments. Locally trained autonomous client models and the overall combined global model were determined by federated averaging on the Edge-IIoT dataset, which has been split into 10 clients. The test accuracy of 92.65% and the AUC of 0.8729 were obtained by the produced global model prior to the knowledge distillation phase. Afterwards, a condensed student model to facilitate the optimum deployment on the terminal devices were distilled by the produced global model, which produced the test accuracy of 85.87% and the AUC of 0.8727. The experimental result verifies that the proposed system has high predictive accuracy with high compression of the model that can be efficiently and scalably deployed on low-power IoT devices.