Multi-party collaborative business prediction model updating method for realizing privacy protection

Abstract

The embodiment of the invention provides a multi-party collaborative business prediction model updating method for realizing privacy protection. The multiple parties comprise a server and a pluralityof data parties; the server deploys a service prediction model for a service object. The method comprises the steps that any round of iterative updating is conducted on a service prediction model, andspecifically comprises the steps that a first data party participating in the round of iterative updating obtains a first current parameter part from a server, and the first current parameter part isobtained by conducting random selection based on current model parameters of the service prediction model; the first data party updates the first current parameter part by using local training data,and obtains a first updated parameter part based on at least one part of the updated parameter; the first data party partially uploads the first updating parameter to a server; and the server updatesthe service prediction model according to the first updating parameter part and other updating parameter parts received from other data parties participating in the current round of iterative updating.

Description

technical field [0001] One or more embodiments of this specification relate to the field of information security technology, and in particular, to a method and device for multi-party cooperative update of a service prediction model that realizes privacy protection. Background technique [0002] At present, the collaborative training of machine learning models by multiple data parties has sparked a research boom. For example, a large number of Internet of Things (IOT) devices hope to break through data silos, and by linking each IOT device together, they can learn and obtain corresponding machine learning models. The difficulty lies in how to ensure that the data privacy of each data party is not prying while collaborative modeling is used to maximize the value of data connection. [0003] However, the existing collaborative modeling methods are relatively single, and it is difficult to meet the higher privacy protection requirements in practical applications. Therefore, th...

AI Technical Summary

Problems solved by technology

[0003] However, the existing collaborative modeling methods are relatively single, and it is difficult to meet the higher privacy protection requirements in practical applications.

Images