基于欧几里得距离的客户端选择联邦学习方案

Euclidean-Based Client Selection in Federated Learning

摘要: 联邦学习(Federated Learning)作为一种先进的分布式机器学习框架，它允许多个参与者在各自数据不出本地的前提下，协同开展模型训练工作，保护数据隐私。然而，传统联邦学习在实际应用中存在数据异质性影响联邦学习系统的效率和收敛性，限制其广泛应用。针对上述问题，本文提出了ECS-FL (Euclidean-Based Client Selection in Federated Learning)框架。该框架融入基于欧几里得距离的客户端选择机制，通过依据客户端模型更新与全局模型的相似度来筛选客户端，有效应对了客户端差异与数据异质性带来的挑战，显著降低了模型训练过程中的偏差，大幅提升了模型的鲁棒性。在独立同分布和非独立同分布数据分布环境下开展的大量实验结果表明，ECS-FL框架能够切实有效地改善模型的收敛性、鲁棒性与准确性。

Abstract: Federated Learning, as an advanced distributed machine learning framework, allows multiple participants to collaboratively carry out model training work while keeping their respective data local, thus protecting data privacy. However, traditional federated learning has the problem in practical applications that data heterogeneity affects the efficiency and convergence of the federated learning system, restricting its widespread application. In response to the above issues, this paper proposes the ECS-FL (Euclidean-Based Client Selection in Federated Learning) framework. This framework incorporates a client selection mechanism based on Euclidean distance. By screening clients according to the similarity between the client model updates and the global model, it effectively addresses the challenges posed by client differences and data heterogeneity, significantly reducing the bias during the model training process and greatly enhancing the robustness of the model. A large number of experimental results conducted in both independent and identically distributed and non-independent and identically distributed data distribution environments show that the ECS-FL framework can effectively improve the convergence, robustness, and accuracy of the model.