Federated Learning Method Based on Generative Adversarial Networks and Contrastive Learning
As a distributed learning mechanism,federated learning is mainly used to protect user data privacy and solve the problem of data islands.Existing research has problems such as the small scale of edge node data and the low accuracy of the model caused by the non-independent and identical distribution of data.Using the method of comparative learning can achieve the perfor-mance closest to the global model when the local model updates the model gradient parameters during the parameter aggregation up-date process of federated learning,which solves the problem of low accuracy of the edge model caused by the non-independent and identical distribution of data.The performance of marginal models is improved.Simulation results show that when the weight of the contrastive loss is 1,the accuracy of the model is increased by 1.05%on average,and when the weight of the contrastive loss is 10,the accuracy of the model is increased by 2.14%on average.Using the method of generative confrontation network to expand the data volume of edge nodes solves the problem of low model accuracy caused by the small amount of data of edge nodes.The simulation ex-periment results show that the accuracy of the model is increased by 0.76%and 1.12%on average after data expansion using the method based on the generative confrontation network.
国家科技期刊平台
NSTL
万方数据
