A joint learning method and system based on a public block chain and an encrypted neural network

Abstract

The invention discloses a joint learning method and system based on a public block chain and an encrypted neural network. The method comprises the following steps of: enabling the terminal to downloada universal neural network model pre-constructed on the server; performing Training using local data, obtaining gradient updating data and encrypting the gradient updating data; acquiring encrypted data and a first identification code, sending the encrypted data to a server; calculating a hash value of the first identification code and uploading the hash value to the block chain; enabling a server to decrypte encrypted data, obtaining the sum of the second identification code and the gradient updating data, calculating the hash value of the second identification code, then verifying whether the hash value of the first identification code is consistent with the hash value of the second identification code or not, if yes, updating the universal neural network model by using the sum of the gradient updating data, otherwise, not updating, and repeating the above steps until the model meets a preset convergence condition. In the process of completing deep network training, user informationcan be prevented from being leaked, and it is guaranteed that transmission data are not leaked or tampered.

Description

technical field [0001] The invention relates to the technical field of block chains, in particular to a joint learning method and system based on public block chains and encrypted neural networks. Background technique [0002] Deep learning is a common learning method in the field of machine learning, which is characterized by the use of deep neural networks and large amounts of data to train the model. Indispensable among them is training data. At present, many sources of training data are generated by users in the process of using the product, which involves some users' usage habits, such as frequency of use, time of use, personal preferences, and so on. Traditional deep learning projects collect user data on the server side and then conduct large-scale training, which makes users worry about the privacy of personal data. [0003] Federated learning (federated learning) is a kind of predictive model proposed by American company Google that enables multiple client computin...

AI Technical Summary

Problems solved by technology

[0007] In order to solve the problems of the prior art, the embodiment of the present invention provides a joint learning method and system based on the public block chain and the encrypted neural network, so as to overcome the problem of how to complete the deep network training in the prior art without leaking User information, and issues such as ensuring that the transmitted data is not leaked or tampered with

Images