Homomorphic encryption neural network framework of PS and PL collaborative architecture and reasoning method

Abstract

The invention discloses a homomorphic encryption neural network framework of a PS and PL collaborative architecture and a reasoning method. The framework comprises a PL side and a PS side, the PL side comprises a structure parameter analysis unit, a plaintext * ciphertext unit and a structure parameter scheduling unit; the structure parameter analysis unit is used for receiving and analyzing the DNN model structure parameters sent by the PS side; the data parameter scheduling unit is used for caching the received weight parameters of the PS side and the order of a polynomial in a ciphertext domain, splicing and outputting to the plaintext * ciphertext unit; and the plaintext * ciphertext unit is used for performing polynomial multiplication operation on the received data in the ciphertext domain and sending a result to the PS. The PS side comprises a convolution summation unit, a partial sum accumulation unit, a BN unit, a data updating unit, a global average pooling unit and a full connection unit. According to the method, the PS side and the PL side work cooperatively, the execution efficiency of the picture classification tasks is improved, and the reasoning time is shortened.

Description

technical field [0001] The invention belongs to the technical field of communication, and in particular relates to a homomorphic encryption neural network framework and reasoning method of a PS and PL collaborative architecture. Background technique [0002] In the past few decades, Deep Neural Networks (DNNs) have developed at an astonishing rate and are gradually integrated into people's lives (such as ubiquitous IoT devices), including image classification, speech recognition, and object recognition. However, most of the neural network training at this stage is based on large data sets, and the reasoning work also involves the user's image data. Since user-transmitted image data can be viewed by the cloud, this increases the risk of misuse of unencrypted data by third parties. Especially when it comes to commercial or medical privacy data, this kind of data abuse will cause great adverse effects. [0003] Homomorphic encryption (HE), as a data encryption scheme, has gre...

AI Technical Summary

Problems solved by technology

[0005] When processing image classification, there are problems such as low processing efficiency, limited application scenarios, and difficult development.

Images