Scholarlink

<p>Federated learning techniques have been successfully applied to solve multi-sensor data modeling problems. However, existing federated learning methods ignore the impact of local model’s performance on the global model. In this paper, we try to figure out the impact and propose a new federated learning algorithm to solve time series prediction problems. Firstly, we designed a new aggregation formula comprising importance levels of all local models. The importance level is used to represent individual information from the perspective of model performances. Secondly, we used this new aggregation formula to realize the update of local models and we named this method FedBiased. Thirdly, we designed a comprehensive framework comprising three scenarios of disturbances to verify and compared the strong robustness of FedBiased and other federated learning methods. Finally, a real case, air quality index prediction problem was adopted for experimental study. The experiment results show our algorithm’s effectiveness on both clean time series and noisy time series.</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 robust federated biased learning algorithm for time series forecasting

  • 期刊论文

  • 发表日期：

    2025年2月25日

查看全文
  • Mingli Song,
  • Xinyu Zhao,
  • Witold Pedrycz

摘要

Federated learning techniques have been successfully applied to solve multi-sensor data modeling problems. However, existing federated learning methods ignore the impact of local model’s performance on the global model. In this paper, we try to figure out the impact and propose a new federated learning algorithm to solve time series prediction problems. Firstly, we designed a new aggregation formula comprising importance levels of all local models. The importance level is used to represent individual information from the perspective of model performances. Secondly, we used this new aggregation formula to realize the update of local models and we named this method FedBiased. Thirdly, we designed a comprehensive framework comprising three scenarios of disturbances to verify and compared the strong robustness of FedBiased and other federated learning methods. Finally, a real case, air quality index prediction problem was adopted for experimental study. The experiment results show our algorithm’s effectiveness on both clean time series and noisy time series.