Federated learning method and device based on differential privacy and storage medium

Abstract

The invention discloses a federated learning method and device based on differential privacy and a storage medium. The method comprises the steps that S1, a preset first model parameter is issued to auser side, so that the user side updates a local deep learning model of the user side based on a differential privacy technology according to the first model parameter and returns a second model parameter; wherein the first model parameter comprises a privacy budget; S2, performing parameter equalization on the second model parameter to obtain a third model parameter, and issuing the third modelparameter to the user side; and S3, taking the total execution times of the steps S1 to S2 as a model training round, and repeatedly executing the steps S1 and S2 when the model training round does not reach a preset threshold, otherwise, ending the model training. According to the invention, the data privacy safety can be ensured, and the accuracy of the training model is improved.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a federated learning method, device and storage medium based on differential privacy. Background technique [0002] At present, most users, such as enterprises or institutions, want to jointly train AI models with other users' data. For data privacy protection and security considerations, each user cannot directly exchange data, and it is difficult to achieve cross-user collaborative training of AI models. and Applying the federated learning method proposed by Google can solve the above problems. [0003] Federated learning establishes a common model through parameter exchange and optimization under encryption mechanism or perturbation mechanism on the premise that the user's respective data does not go out of the local area. This requires the user to use their own data to train the user model, and upload the model parameters of the user model to the serv...

AI Technical Summary

Problems solved by technology

[0005] 1. The user directly uploads the model parameters to the server, exposing the model parameters to the data transmission channel and the server, which cannot fully guarantee data privacy and security;

[0006] 2. Users with large or small data volumes all use the same model building method and training rounds, which makes it difficult to guarantee the accuracy of the model;

[0007] 3. Differential privacy deep learning algorithms such as differential privacy stochastic gradient descent algorithm (DP-SGD) and adaptive Laplace mechanism (AdLM) add large and fixed noise to the model, and it is difficult to optimize and adjust the noise for different training processes. Guarantee the accuracy of the model

Images