.png){kind=icon}
返回
FBL-ET: A federated broad learning framework based on event trigger
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2023.110366.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Federated learning (FL) is proposed to exchange models between server and clients rather than data, which protects privacy. But one important problem in FL is that the clients may be unreliable. The clients may suddenly hang up. If the hung-up clients are selected by the server, the time consumption can be significant. To solve this problem, in this paper, the clients are set to have the communication initiative, so they can decide whether or not to upload their local models. Another problem in FL is that the quality of the clients' models cannot be determined. If some models have bad quality, they will have a bad impact on the quality of the global model. To improve the quality of models upload by clients, we set an upload condition for each client, and the client will upload the model only when the condition is satisfied. For both purposes, we introduce an event-triggered method into federated broad learning, i.e. FBL-ET. We set up a test set on each client, and based on the prediction accuracy, a specified dynamic trigger condition is designed for each client. Therefore, FBL-ET reduces time consumption and improves the quality of uploaded models by introducing an event-triggered mechanism. In addition, in order to improve the activity of clients and attract more high-quality clients to participate in global aggregation, FBL-ET introduces an incentive mechanism. Finally, experimental results show that FBL-ET has higher accuracy than the original federated broad learning method, and can effectively reduce the waste of communication consumption.(c) 2023 Elsevier B.V. All rights reserved.
Keyword:
Broad learning
Federated learning
Event-trigger
Incentive mechanism
期刊
K
IF:
7.6
论文数:
1.3W
被引数:
4.5W
机构
引用论文
Budgeted Online Selection of Candidate IoT Clients to Participate in Federated Learning预算在线选择候选物联网客户参与联合学习
OAAI
OAAI分享收藏Gradient-based learning applied to document recognition基于梯度的学习在文档识别中的应用.png)
PROCEEDINGS OF THE IEEE
PROCEEDINGS OF THE IEEEIF25.9
PREAI
分享收藏SAFA: A Semi-Asynchronous Protocol for Fast Federated Learning With Low OverheadSAFA: 一种低开销快速联邦学习的半异步协议OAAI
OAAI分享收藏FedMCCS: Multicriteria Client Selection Model for Optimal IoT Federated LearningFedMCCS: 优化物联网联邦学习的多准则客户端选择模型
PREAI
分享收藏
查看更多内容