Method and device for jointly training model based on multi-party private data

Abstract

The embodiment of the invention provides a method and device for jointly training model based on multi-party private data, and the method comprises the steps that any first party in a plurality of first parties utilizes the first privacy data and first class labels of all objects in a local first object set and the first privacy data and first class labels of other first parties, based on a transverse federated learning mode, a first parameter set shared by a plurality of first parties corresponding to the first sub-model is acquired; a common object in which a local first object set and a second object set of a second party are overlapped with each other is determined; a first parameter set corresponding to a local first sub-model is updated based on a longitudinal federated learning modeby utilizing first privacy data corresponding to a local common object, and second privacy data and a second category label of a second party; and a second party obtains a second parameter set corresponding to a second sub-model of the second party. The prediction accuracy of the model obtained after training can be improved.

Description

Technical field[0001]One or more embodiments of the present specification relate to the computer field, and in particular to a method and device for joint training models based on private data of multiple parties.Background technique[0002]With the general advancement of artificial intelligence and machine learning, it is possible to jointly train models based on private data from multiple parties, and use the trained models to perform corresponding prediction tasks. Data occupies an extremely important position in modeling, and more dimensional and richer data are conducive to the establishment of more accurate and effective models. Data is distributed among multiple parties, and there are certain privacy and security issues, and multiple parties cannot obtain other parties' private data. Due to issues such as privacy protection and data barriers, it is difficult to safely and comprehensively use multi-party private data to jointly train the model on the premise of protecting the pr...

AI Technical Summary

Problems solved by technology

The data is distributed in multiple parties, and there are certain privacy and security issues, and the private data of other parties cannot be obtained between multiple parties

Due to issues such as privacy protection and data barriers, it is difficult to safely and comprehensively use the private data of multiple parties to jointly train the model under the premise of protecting the private data of all parties, so the prediction accuracy of the model obtained after training is not good.

Images