Research on data fusion method based on Federated learning

Inhalt:
With the development of large data technology, the large variety and quantity of data put forward higher requirements on how to better use the data. The distribution of large data is characterized by "data island", "wide area dispersion", data dispersion, etc. Data types, data relationships and data quality are increasingly different, and contain a large number of unlabeled data, data sparse areas and domain knowledge. In order to solve the problem of multi-source heterogeneous data fusion, this paper innovatively introduces the Federal Learning algorithm, builds a Federal Learning Classifier, aggregates multiple fusion models of the same kind of data, and achieves the iterative optimization of data fusion without uploading data. The simulation results show that this algorithm model has certain advantages in data fusion stability and model iteration convergence speed.