Federated learning private data processing method, device and system and storage medium

Abstract

The invention discloses a federated learning private data processing method, device and system, and a storage medium. The federated learning private data processing method comprises the steps: receiving the updating of a model parameter with a mask of a participating device, adding the mask to the model parameter updating obtained through training through the participating device based on a firstmask, and obtaining the updating of the model parameter with the mask; generating a second mask the same as the first mask in the TEE module, and updating the model parameter with the mask based on the second mask to remove the mask to obtain a model parameter update; adding a mask to the global model parameter update by adopting the generated third mask in the TEE module to obtain a global modelparameter update with the mask; and sending the global model parameter update with the mask to the participating equipment, and removing the mask from the global model parameter update with the mask by the participating equipment based on a fourth mask which is the same as the third mask to obtain a global model parameter update. According to the federated learning private data processing method,a security mechanism is realized, so that the information of the participating equipment is not leaked to the coordination equipment, and the requirement of remarkably increasing the communication bandwidth is not caused.

Description

technical field [0001] The present invention relates to the technical field of data processing, in particular to a federated learning privacy data processing method, device, system and storage medium. Background technique [0002] With the development of artificial intelligence, in order to solve the problem of data islands, people put forward the concept of "federated learning", so that both sides of the federation can also conduct model training to obtain model parameters without giving their own data, and can avoid data The issue of privacy breaches. [0003] In the actual application of horizontal federated learning scenarios, the local model parameter updates (for example, neural network model weights, or gradient information) sent by participants to the coordinating device will be obtained by the coordinator. In scenarios where the reliability of the coordinator cannot be guaranteed, Participants' privacy, data information, and trained machine learning models may be l...

AI Technical Summary

Problems solved by technology

[0004] However, using encryption technology will significantly increase the length of the information that needs to be transmitted. For example, using homomorphic encryption technology, using the most commonly used Paillier algorithm, the length of the obtained ciphertext (measured by the number of bits) is at least twice the length of the plaintext, That is, encryption at least doubles the communication bandwidth requirements compared to non-encryption

In some practical applications, such as IoT, mobile Internet, remote sensing, and commercial satellite communication links, the communication bandwidth is severely limited, and the additional communication bandwidth requirements brought by participants’ encryption operations are likely to be unable to be met, or at least significantly increase the latency of communication

Images