Improvement of Federal Average Algorithm Based on CRITIC Weighting Method
Qiantong Yu, Nianqing Wei, Yiyang Mi · 2023
Federated learning is a machine learning technology that can protect data privacy. Its federated average algorithm updates the global model through weights. In view of the influence of subjective factors in the calculation of data weights by analytic hierarchy process (AHP), a federated average algorithm based on CRITIC weighting method is proposed to process multi-source data from the perspective of data diversity and volatility. In the experiment, the CRITIC weighting method is used to calculate the importance of each attribute in the data, and it is used as the model aggregation parameter to calculate the weight of each client data. By applying the federated average algorithm in horizontal federated learning to the data model training fusion, the accuracy of data model establishment is improved. In terms of both theoretical basis and practical research, experiments and analysis on the collected data sets show that the proposed method is executable. The experimental results show that the proposed method improves the accuracy by about 4 % compared with the federal average algorithm, reduces the communication overhead of the federal learning and training process, and improves the overall efficiency.