基于模型相似度的联邦学习本地模型可信计算研究OA

Model training based on blockchain and federated learning is conducive to monitoring the quality of the client's local mod-el while protecting the client's data privacy.However,the risk of malicious client training behavior in the federated learning model-ing process will lead to the degradation of the quality of the local model,which in turn affects the performance of the global model.In response to the above problems,an improved federated learning algorithm based on model similarity was proposed,which im-proved the performance of the federated learning global model through trusted computing of the client's local model.First,the client participated in federated learning for local model training.Then verify the client obtained the local model through the blockchain and model similarity was calculated.Finally,according to the verification results,qualification to participate in the federated learning ag-gregation process is given to the client,so that the client can perform trusted computation of the local model and improve the global model performance of federated learning.Taking the credit evaluation model as an example,the improved federated learning algo-rithm has an accuracy rate of 6.2%higher than the traditional global model,and the convergence is 8 rounds earlier,which effective-ly improves the global model performance of federated learning.