A federated learning method and a federated learning system

Abstract

The embodiments of this specification provide a federated learning method and a federated learning system. The method includes: before the target model used for training by the federated learning determines the feature configuration, the member objects of the federated learning preprocess the local private data according to the data structured configuration of the data warehouse of the federated learning; and, based on the multi-party secure computing protocol, the member objects of the federated learning send the preprocessed private data to the centralized computing platform, so that the centralized computing platform uploads it to the data warehouse. After the target model determines the feature configuration, the centralized computing platform extracts features from the private data recorded in the data warehouse according to the feature configuration of the target model to obtain feature data suitable for training the target model; and, based on the extracted feature data, the centralized computing platform trains the target model and sends the training results of the target model to the member objects of the federated learning.

Images