Method and device for multi-party joint training neural network for security defense

Abstract

The embodiment of this specification provides a multi-party joint training neural network method for security defense, where the multi-party includes multiple clients corresponding to multiple data parties and a neutral server. Multiple clients use the secure multi-party technology MPC to jointly calculate one or more hidden layers on the basis of protecting data privacy, and then place the calculated hidden layers on a neutral server to perform other complex neural network calculations. In order to obtain the prediction result, it is used to compare with the sample label to determine the prediction loss. In addition, attacker models that simulate attackers are deployed in multiple clients, and attacker losses are calculated separately. Furthermore, the server adjusts the parameters of the rest of the complex neural network deployed on it according to the training loss determined based on the prediction loss and the attacker's loss, and multiple clients adjust the corresponding part of the parameters of the common computing hidden layer maintained by each according to the training loss, and adjust the parameters of the attacker model according to the attacker loss.

Description

technical field [0001] One or more embodiments of this specification relate to the field of information security technology, and in particular to a method and device for multi-party joint training of neural networks for security defense. Background technique [0002] At present, the collaborative training of machine learning models by multiple data parties has sparked a research boom. The difficulty lies in how to ensure the security of the data of all parties during the training process. For example, the payment platform has some characteristics and labels of users, and the bank has other characteristics of users. At this time, the payment platform and the bank hope to use the data of both parties to jointly build a machine learning model. However, due to the existence of laws and regulations on data security, or the unwillingness of both parties to disclose the data to the other party, it is necessary to jointly model in a form that can guarantee their own data security. ...

AI Technical Summary

Problems solved by technology

However, due to the cumbersome data processing, the modeling efficiency is not good

Especially when the model uses a neural network, because the model parameters are often large, the model training takes up a lot of computing resources and takes a lot of time

Images