Scholarlink

<p>Integrating digital twins (DT) into 6G networks presents substantial prospects for improving federated learning (FL). Within this framework, we provide Digital Twin-Assisted Semi-Supervised Autoencoder-based Robust Network (DASSARN), a new method specifically developed to tackle the difficulties associated with FL in 6G systems. DASSARN utilizes data transfer-enabled FL to address challenges such as update noise and client selection, ultimately optimizing performance. We evaluate the performance of DASSARN by comparing it to well-known FL benchmarks, such as FL with FedAvg, FL with FedSGD, and several forms of self-supervised FL (SSFL). The evaluation is conducted on the MNIST and CIFAR-10 datasets. Our assessment, which includes independent and non-independent data scenarios, shows that DASSARN competes effectively with the top supervised FL methods and outperforms the most advanced semi-supervised FL techniques. This is especially important considering the difficulties in obtaining extensive labeled datasets in real-world applications. DASSARN accomplishes these outcomes by utilizing data augmentation and pseudo-labeling techniques to improve the diversity of client data. The results emphasize the potential of DASSARN to enhance FL in the developing 6G environment, showcasing its capacity to achieve strong performance with limited dependence on labeled data.</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

6G digital twin and CPS system promote the development of rural architectural planning

  • 期刊论文

  • 发表日期：

    2025年4月16日

查看全文
  • Zhai Binqing,
  • Yao Yicong,
  • Mohammad Khishe

摘要

Integrating digital twins (DT) into 6G networks presents substantial prospects for improving federated learning (FL). Within this framework, we provide Digital Twin-Assisted Semi-Supervised Autoencoder-based Robust Network (DASSARN), a new method specifically developed to tackle the difficulties associated with FL in 6G systems. DASSARN utilizes data transfer-enabled FL to address challenges such as update noise and client selection, ultimately optimizing performance. We evaluate the performance of DASSARN by comparing it to well-known FL benchmarks, such as FL with FedAvg, FL with FedSGD, and several forms of self-supervised FL (SSFL). The evaluation is conducted on the MNIST and CIFAR-10 datasets. Our assessment, which includes independent and non-independent data scenarios, shows that DASSARN competes effectively with the top supervised FL methods and outperforms the most advanced semi-supervised FL techniques. This is especially important considering the difficulties in obtaining extensive labeled datasets in real-world applications. DASSARN accomplishes these outcomes by utilizing data augmentation and pseudo-labeling techniques to improve the diversity of client data. The results emphasize the potential of DASSARN to enhance FL in the developing 6G environment, showcasing its capacity to achieve strong performance with limited dependence on labeled data.