Multi-party joint training method, device, system and equipment for business model

Abstract

The invention provides a multi-party joint training method, device, system and equipment for a business model. The method comprises the following steps: in a multi-party joint training service model process, data holders cut off the gradient of the iteration and then send the cut-off gradient to the collaboration party; and the collaboration party calculates a first target gradient based on the truncation gradient sent by each data holder, and adds noise in the first target gradient to obtain a second target gradient, so that each data holder determines a model parameter of the current iteration based on the second target gradient and the model parameter of the last iteration. According to the method, the calculated gradient is cut off through the data holder, and noise is added to the first target gradient summarized by the collaboration party, so that personal data leakage is prevented in a differential privacy mode in the model training process.

Description

technical field [0001] This specification relates to the field of computer communication, and in particular to a multi-party joint training method, device, system and equipment for a business model. Background technique [0002] Model training is an important part of artificial intelligence technology. Model training usually relies on samples from users. In general, the more user samples and the richer the feature dimension, the higher the accuracy of the trained model. [0003] However, the samples held by a single data holder are limited, thus hindering model training. In order to improve the accuracy of model training, the industry has proposed federated learning. Federated learning proposes that multiple parties use their own samples to jointly train business models. Federated learning includes horizontal federated learning and vertical federated learning. [0004] The horizontal federated learning system includes multiple data holders and collaborators. Gradient in...

AI Technical Summary

Problems solved by technology

However, since the gradient is calculated based on the user's private data, it is easy to obtain the user's private data by inverting the gradient, which will lead to the leakage of the user's private data

Images