基于端边云协同体系的联邦学习模型训练与优化OA

In response to the problem that federated learning training models are easily affected by data attributes,a federated learning model training and optimization method based on end-edge-cloud collaborative system is proposed.This method introduces credibility and dynamic learning rate to achieve self-learning and self-optimization of global model parameters.Experiments have shown that compared with other algorithms,the proposed algorithm fully considers the credibility of the edge,which can prevent the rapid decrease in accuracy caused by rapid changes in global model parameters due to data distribution or quality issues.In addition,due to the introduction of dynamic learning rate,the global model can adaptively adjust the learning rate based on the error of the local model during aggregation,which to a certain extent balances the global parameter update speed and algorithm stability.