Year: 2025
Type: conference-paper
Abstract: Federated learning research primarily focuses on algorithm design, privacy protection, and communication efficiency optimization. Although it enables parameter exchange and update without disclosing data, there are still pressing issues to be addressed. Firstly, in practical applications, there are some resource-constrained devices that are unable to complete the large scale and complex federated learning training process, which restricts the development of federated learning in certain fields. Secondly, the problem of data heterogeneity seriously affects the performance of federated learning, leading to slow convergence and decreased performance of the global model. To tackle these problems, this paper proposes a model customized personalized federated learning method based on information quantity (FedMC). This method utilizes Fisher information to evaluate the information content of the model, thereby selecting important model parameters. Only the key information is transmitted between the clients and the server, significantly reducing the communication volume while ensuring model updates. In addition, this method customizes the model for each client based on the important information to better handle the data heterogeneity problem. Through experiments, we have verified the effectiveness and practicality of this research.
Institution: Henan University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4413602149