Training method, device, equipment and storage medium of federated neural network model

Abstract

Embodiments of the present application provide a training method, device, device, and storage medium for a federated neural network model, which relate to the fields of artificial intelligence technology and cloud technology. The method includes: inputting the sample data into the federated neural network model to process the sample data through the first lower layer model to obtain the output value of the lower layer model; the output value of the lower layer model and the interactive layer model generated by the first participant Parameters, and the encrypted model parameters obtained after encrypting the interactive layer model parameters based on the RIAC encryption method are input to the interactive layer respectively to obtain the output vector of the interactive layer; the output vector is input to the upper model to obtain the federal neural network The output value of the model; input the output value into the preset loss function to obtain the loss result; perform backpropagation processing on the federated neural network model according to the loss result. Through the embodiments of the present application, it is possible to greatly reduce calculation complexity, reduce calculation amount, and reduce time consumption.

Description

technical field [0001] The embodiment of the present application relates to the field of Internet technology, and relates to but not limited to a training method, device, equipment and storage medium of a federated neural network model. Background technique [0002] Federated Learning (FL, Federated Learning) can use multi-party data sources to train deep learning models (that is, artificial neural network model ANN) and provide model prediction (model prediction) services without the need for data to go out of the domain. Under the premise of ensuring user privacy and data security, federated learning makes full use of multi-party data cooperation to improve the performance of deep learning models, for example, to improve the accuracy of recommendation models, while ensuring that multi-party data cooperation meets the requirements of data protection laws and regulations . In particular, vertical federated learning can expand data feature dimensions or obtain data label inf...

AI Technical Summary

Problems solved by technology

[0004] However, the encryption and decryption process of the homomorphic encryption scheme based on the Paillier algorithm requires a large number of modular exponentiation and modular inverse operations, and its computational complexity is very high, the amount of calculation is too large, and the time is too long. Vertical federated neural network model training and model inference services

Images