Federal learning privacy protection system and method based on hierarchical aggregation and block chain

Abstract

The invention provides a federated learning privacy protection system and method based on hierarchical aggregation and a block chain. The system comprises a trusted mechanism, the block chain, a federated learning module and an IPFS distributed storage system. The block chain is used for recording various data generated by the federal learning module and model addresses corresponding to training models of various versions iterated all the time in the training process; the federated learning module is used for realizing federated learning training; the user terminal obtains a current global model as a local model, and trains the local model on a local data set to obtain a new round of local model; and the IPFS distributed storage system is used for storing local model parameters and global model parameters. Based on the system provided by the invention, the technical problems of low efficiency, reduced model accuracy and large storage overhead of a federal learning privacy protection mechanism in the prior art are solved, the federal learning efficiency can be improved, and the global model accuracy can be improved.

Description

technical field [0001] The invention relates to the technical field of information security, in particular to a federated learning privacy protection system and method based on hierarchical aggregation and blockchain. Background technique [0002] Advances in deep learning technology have benefited from increased computing power and the vast amount of data available to train neural networks. However, the data used to train neural networks, especially the data generated by personal devices, usually contains user privacy, such as user location, medical records, transaction records, etc. Traditional deep learning methods usually need to collect these sensitive data and store them in a centralized location to train neural network models. However, in some specific application scenarios, such as hospitals and banks, users are unwilling to provide their sensitive data to third parties. In order to be able to train neural network models without obtaining users' sensitive data and ...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the above technical problems, the present invention proposes a federated learning privacy protection system and method based on hierarchical aggregation and blockchain, which is used to solve the problems of low efficiency of federated learning privacy protection mechanism, lower model accuracy, and storage overhead in the prior art. big technical problem

Images