Federated learning privacy protection method and system based on homomorphic pseudo-random numbers

Abstract

The invention discloses a federated learning privacy protection method and system based on homomorphic pseudo-random numbers. The method comprises the steps that n clients generate a secret key s through employing a verifiable secret sharing VSS, the secret key s is divided into n parts, and each client obtains a secret share si; at least t clients participate in recovering the key s and send thekey s to a server, wherein n and t are both positive integers, and si represents the secret share of the ith client; each client performs federated learning, and each client locally uses respective data to perform machine learning model training to generate an updated gradient value; each client generates a random number F(si, x) by taking the secret share si as a seed and using a key homomorphicpseudorandom function; the updated gradient value is encrypted by using the random number F(si, x) to obtain an updated gradient value ciphertext, and then the updated gradient value ciphertext is sent to the server; and the clients receive the updated model fed back by the server.

Description

technical field [0001] The present application relates to the technical field of privacy protection, in particular to a method and system for federated learning privacy protection based on homomorphic pseudo-random numbers. Background technique [0002] The statements in this section merely mention the background art related to this application, and do not necessarily constitute the prior art. [0003] The emergence of new technologies such as big data, cloud computing, and deep learning has promoted the vigorous development of artificial intelligence and machine learning, but data security and privacy issues have seriously restricted the practical application of artificial intelligence and machine learning. At present, due to concerns about the risk of data leakage, the sharing and utilization of data by governments, companies, and individuals is extremely limited, and a large amount of data cannot be effectively utilized. Due to factors such as the approval process, busin...

AI Technical Summary

Problems solved by technology

[0005] Although federated learning does not need to transmit the original data, it only needs to transmit the update gradient value of the model, but these update gradient values ​​​​come from the original data, so these model update gradient values ​​may still leak data information

In order to solve this problem, Google proposed to use pseudo-random numbers to protect the model to update the gradient value, but the complexity of this method is too high to be suitable for large-scale federated learning scenarios

Images