TEE-based federal recommendation method, apparatus and device, and medium

Abstract

The embodiment of the invention discloses a TEE-based federal recommendation method, apparatus and device, and a medium. The method comprises the steps of executing a DH key exchange protocol on any two edge devices based on a cloud edge frame, and storing a generated symmetric encryption key in a memory of a TEE; receiving a push request sent by the edge device; decrypting the local model stored in the local hard disk of each edge device in the TEE; and inputting local data of each edge device into the corresponding decrypted local model in the TEE, encrypting a generated intermediate result by using the corresponding symmetric encryption key, and sending the encrypted intermediate result to the edge device corresponding to the push request for decryption and reasoning calculation so as to perform push service. The invention further relates to a block chain technology, and the push request is stored in a block chain. Based on the digital signature technology, the DH protocol is executed in the TEE and encryption and processing of the data in the cloud edge end frame are performed, so that the possibility that the data is tampered in the federated reasoning process is avoided.

Description

technical field [0001] The present invention relates to the technical field of federated reasoning, in particular to a TEE-based federated recommendation method, device, equipment and medium. Background technique [0002] With the development of edge computing, cloud-edge-device collaborative integration has become a new trend in the field of artificial intelligence. Intelligent edge devices such as intelligent gateways and small edge cloud servers have been widely used. Edge devices run a series of artificial intelligence-related applications to provide data processing, reasoning and other services used in daily life, but edge devices are often in open network environments and insecure physical environments, and it is very easy to install sniffers on network devices To intercept data packets, such as intercepting the deep learning model pushed by the cloud to the edge device, or directly attacking the data in the memory through memory bus snooping, memory tampering, etc., n...

AI Technical Summary

Problems solved by technology

Edge devices run a series of artificial intelligence-related applications to provide data processing, reasoning and other services used in daily life, but edge devices are often in open network environments and insecure physical environments, and it is very easy to install sniffers on network devices To intercept data packets, such as intercepting the deep learning model pushed by the cloud to the edge device, or directly attacking the data in the memory through memory bus snooping, memory tampering, etc., not only leads to the leakage of user privacy, but also cannot guarantee that in the process of federation reasoning Data is not tampered with

Images