End-cloud cooperative training system for protecting end-side privacy

Abstract

The invention provides an end-cloud cooperative training system for protecting end-side privacy. An aggregation model storage part stores at least a pre-trained cloud side aggregation model, an end-side communication part sends an end-side aggregation model to cloud-side equipment, a cloud-side communication part receives the end-side aggregation model, a cloud-side self-encoding part processes acloud-side image to obtain a pseudo image, an aggregation model processing part is used for respectively inputting the pseudo image into the end-side aggregation model and the cloud-side aggregation model for processing to obtain an output end-side aggregation model and an output cloud-side aggregation model, and according to a plurality of loss iteration parts processed by the loss processing part, model parameters are updated through back propagation and repeated iteration to obtain a cloud side training pseudo image generator and an end cloud aggregation model. Therefore, the system can still efficiently and stably aggregate the model under the condition of protecting the privacy of the user, has the advantages of protecting the privacy of the user, being good in aggregation effect, good in robustness, good in generalization ability and the like, and is suitable for practical application such as model aggregation between user equipment and end-cloud collaborative training.

Description

technical field [0001] The invention relates to a terminal-cloud collaborative training system for protecting terminal-side privacy, and belongs to the technical field of machine learning. Background technique [0002] Federated learning is an emerging basic technology of artificial intelligence. Its design goal is to ensure information security, protect terminal data and personal data privacy, and ensure legal compliance between multiple participants or multiple computing nodes during big data exchange. Efficient machine learning has gradually led to the distinction between device-side and cloud-side, and then the idea of ​​device-cloud collaborative training exists. [0003] Due to the large difference between user domain data and development data and the development and training data storage department cannot cover all user scenarios, many unseen scenarios, category recognition errors or unsupported. In addition, due to the large differences in the data distribution of d...

AI Technical Summary

Problems solved by technology

The existing device-cloud model averaging method and encryption methods that directly average model weights are fundamentally unable to accurately protect user data and effectively protect device-side privacy.

[0006] Although the cloud model averaging method [3] that directly averages the model weight has a good effect on protecting user privacy, the violent direct averaging algorithm is not a good strategy for the model weight

Although the encryption method [4-5] can improve the security of end-side data to a certain extent, the end-side information still does not fundamentally guarantee user privacy because it leaves the device. In addition, the encryption operation also increases the computational complexity.

Images