基于生成对抗网络的高端装备研制数据脱敏方法

针对高端装备研制数据的保密性和重要性,提出一种基于生成对抗网络(generative adversarial networks, GAN)的高端装备研制数据脱敏方法。面对高端装备研制数据在数据挖掘和数据共享时可能面临泄露的风险,利用GAN进行数据脱敏。在随机生成的高斯数据集上进行实验,通过比较源数据和脱敏数据的统计特征,证明GAN的数据脱敏方法能够有效实现数据脱敏过程中所要求的数据安全性、数据有效性和成本可控性。最后,在Yeast数据集上进行验证, GAN输出的脱敏数据同样在现实世界数据集上表现出色,能够准确地预测Yeast的分类,为高端装备研制数据的管理和分析提供了一种新的思路。

High-end equipment development data hyposensitization method based on generative adversarial networks

Aiming for the confidentiality and importance of high-end equipment development data, a data hyposensitization method based on generative adversarial networks (GAN) is proposed. Considering the high-end equipment development data may face the risk of sensitive data disclosure during data mining and data sharing, the GAN for data hyposensitization is used. Experiments are carried out on randomly generated Gauss data sets. By comparing the statistical characteristics of the original data and hyposensitization data, it is proved that the data hyposensitization method based on GAN can effectively realize the data security, data utility and cost controllability. Finally, on the Yeast dataset, the desensitized data output by the GAN also performs well in real-world datasets, which can accurately predict the classification of Yeast, providing a new idea for the management and analysis of high-end equipment development data.