.png){kind=icon}
Return
Dynamic weight federated client selection using adaptive genetic algorithm
.png){kind=icon}
PRE
AI
DOI：10.1007/s10586-025-05492-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In federated learning, most of the participating clients will have significant heterogeneity in data distribution, and this difference will lead to many challenges in the training process of the global model. By selecting high-quality clients to participate in training, the adaptability of the model to different data characteristics can be improved. We propose a dynamic weighted federal client selection method (GAFL) based on an adaptive genetic algorithm, where genes on chromosomes represent different weight combinations of all clients, and the optimal client combinations are found by dynamically adjusting the crossover probability through adaptive genetics on chromosome populations combined with a dynamic weight assignment strategy. This dual dynamic mechanism makes our method more flexible and greatly improves the performance of the model. Through experiments, the GAFL method outperforms the traditional algorithm on several medical datasets.
Keywords:
Federated learning
Dynamic weight
Adaption genetic algorithms
Client selection
Journal
C
IF:
0
Papers:
691
Citations:
1
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveA Survey on Federated Learning Systems: Vision, Hype and Reality for Data Privacy and Protection
OAAI
OAAIShareSaveShareSaveShareSave
View more