Neural network architecture search method, device, equipment and storage medium

Abstract

The invention relates to the technical field of neural networks, and discloses a neural network architecture search method, device, equipment and storage medium. Including: defining the neural network architecture search space; generating candidate model architectures based on the search network; sharing the data sets and candidate model architectures secretly to the computing server, so that the computing server can perform privacy-preserving neural network training and feedback training accuracy; update predictions based on training accuracy Network and search network, continue to search for candidate model architectures according to the updated search network until the model error of the prediction network reaches the threshold; perform accuracy prediction on the final candidate model architecture based on the updated prediction network, and perform candidate architecture evaluation on the final candidate model architecture . The data owner initiates the model architecture search task, uses the secret sharing method to send the data set, and outsources computing tasks such as model training to the computing server, which ensures the privacy of the model and solves the problem of model structure leakage caused by the architecture search.

Description

technical field [0001] The present invention relates to the technical field of neural networks, in particular to a neural network architecture search method, device, equipment and storage medium. Background technique [0002] In the existing technology, in order to realize adaptive adaptation of dense-state models, each candidate model is usually trained, and the candidate model score is determined according to its performance. However, the training of dense-state data models is very time-consuming, making model adaptive adaptation difficult for practical application. . And in the actual process, the model structure and parameters are the digital assets of the data owner. Direct model architecture search will use the plaintext structure of the model, resulting in the leakage of the model structure and affecting the reasonable use of dense model adaptation. [0003] The above content is only used to assist in understanding the technical solution of the present invention, and...

AI Technical Summary

Problems solved by technology

[0004] The main purpose of the present invention is to provide a neural network architecture search method, device, equipment and storage medium, aiming to solve the technical problem of model structure leakage caused by architecture search

Images