Federated learning method based on trusted execution environment

Abstract

The invention belongs to the field of data security, and discloses a federated learning method based on a trusted execution environment, which comprises the following steps: generating a security areabased on the trusted execution environment; downloading the initialized model parameters from the cloud by a local user; loading the training algorithm, the training data set, the training data example number and the initialized model parameters returned by the cloud into a security area; obtaining a trained model parameter gradient and generating a digital signature, performing local user identity authentication through a group signature algorithm, and uploading the trained model parameter gradient, a model integration algorithm and the local user identity authentication to a cloud; and enabling the cloud to verify the local user identity authentication, obtain the uploaded model parameter gradient and model integration algorithm after successful verification, placing the model parametergradient and model integration algorithm in a cloud security area, integrating the model, and updating the model parameter gradient. According to the invention, the trusted execution environment is used to generate the security area, the user cannot bypass the training process to directly give the training result, and the training integrity and the user privacy protection are realized.

Description

technical field [0001] The invention belongs to the field of data security, in particular to a federated learning method based on a trusted execution environment. Background technique [0002] Machine learning based on private data has achieved good results in practical applications. Many companies such as Google, Facebook, Apple, etc. have collected massive training data from users and applied powerful GPU computing power to deploy deep learning algorithms. In order to obtain a deeper model, many companies are willing to collect complementary data and collaborate in training. However, this method of training machine learning models by directly gathering raw data sets from multiple users hides huge challenges: issues such as privacy data security, data poisoning, and false reporting of data volume. [0003] The invention patent with the publication number CN108717514A published on October 30, 2018 discloses a data privacy protection method and system in machine learning, wh...

AI Technical Summary

Problems solved by technology

The Trusted Execution Environment executes the machine learning code using at least one data oblivious program, wherein the data oblivious program is any of memory access patterns, disk access patterns, and network access The pattern makes it impossible to predict the process of classified data from the pattern

Similarly, its system protects private data very well, but it also does not solve the problem of the authenticity of training results and the amount of false data

Images