Regression model training method and system without trusted third party in longitudinal federated learning

Abstract

The embodiment of the invention provides a regression model training method and system without a trusted third party in longitudinal federated learning. A data provider sends a first inner product to a model initiator; the model initiator calculates an inner product sum of the first inner product and the second inner product, calculates a predicted value by adopting a preset function, calculates a residual error between the predicted value and a real label value, performs semi-homomorphic encryption on the residual error, and sends the semi-homomorphic encryption residual error and a semi-homomorphic encryption public key to the data provider; the data provider calculates a first gradient for the first characteristic parameter and adds a random mask under semi-homomorphic encryption; the model initiator decrypts the encrypted mask gradient according to the semi-homomorphic encrypted private key, and sends the mask gradient to the data provider; and the data provider removes the random mask in the mask gradient to obtain a first gradient, and updates the first feature parameter. The training of the regression model in the longitudinal federated learning is completed on the premise of not depending on a third-party mechanism and protecting the privacy data of the two parties.

Description

technical field [0001] This application relates to the technical field of machine learning, in particular to a regression model training method and system without a trusted third party in vertical federated learning. Background technique [0002] Today, big data-driven artificial intelligence technology has been widely used in finance, retail, medical and other fields. In order to get a better model, it often requires the support of a large amount of data, but in reality, the data is often distributed in different institutions. However, most artificial intelligence algorithms do not consider the protection of personal privacy information at the beginning of design. How to break the "data islands" and share and use these data while satisfying laws and treaties has become an urgent problem to be solved. [0003] Federated learning is a distributed learning paradigm proposed by Google. According to data distribution, federated learning can be divided into three scenarios: ho...

AI Technical Summary

Problems solved by technology

But in real scenarios, it is difficult to find a trusted third party recognized by both parties

Images