Two-stage federated learning method and system

Abstract

The invention provides a two-stage federated learning method and system, and the method comprises the steps: repeating the first-stage federated learning according to a local model and a new global model based on a feature fusion algorithm according to a preset first round, so as to update the local weight of the local model and the parameters of the new global model, enabling the updated new global model to serve as a two-stage global model, carrying out convergence test on the second-stage global model through preset test data to obtain accuracy, if the accuracy is converged, taking the second-stage global model as a second-stage local model of the client, and repeatedly carrying out second-stage federated learning based on a federated average algorithm to update the local weight of the second-stage local model; and if the number of repetitions reaches a preset second round number, completing two-stage federated learning, so that a loss function of original model training is changed, a local model is closer to a global model received from a server side during updating, excessive influence of local data of a user is reduced, the model has generalization ability more quickly, and the convergence speed is improved.

Description

technical field [0001] The present invention relates to the technical field of blockchain, and more specifically, to a two-stage federated learning method and system. Background technique [0002] Machine learning is a data analysis method that automatically analyzes and builds models. The machine learning model is to learn the characteristics of the data through the data, and predict the new data, and complete the task of classification or regression. [0003] However, the machine learning model is driven by data, and its accuracy is closely related to the quantity and quality of training data. With sufficient and high-quality data training, the machine learning model will perform better in practical applications. . Therefore, how to provide a large amount of high-quality data for machine learning models has become a key issue in improving the performance of machine learning models. [0004] But at present, in most industries, data exists in the form of isolated islands....

AI Technical Summary

Problems solved by technology

[0007] In view of the above problems, the purpose of the present invention is to provide a two-stage federated learning method to solve the problem that the traditional federated average algorithm has a high communication cost between the server and the client, and the performance of the data is not independent and identically distributed.

Images